W O R L D B A N K R E G I O N A L A N D S E C T O R A L S T U D I E S Economic Growth, Poverty, and Household Welfare in Vietnam EDITED BY PAUL GLEWWE NISHA AGRAWAL DAVID DOLLAR Economic Growth, Poverty, and Household Welfare in Vietnam WORLD BANK REGIONAL AND SECTORAL STUDIES Economic Growth, Poverty, and Household Welfare in Vietnam Edited by Paul Glewwe Nisha Agrawal David Dollar THE WORLD BANK Washington, D.C. © 2004 The International Bank for Reconstruction and Development/THE WORLD BANK 1818 H Street, NW Washington, DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org E-mail: feedback@worldbank.org All rights reserved. 1 2 3 4 07 06 05 04 The World Bank Regional and Sectoral Studies series provides an outlet for work that is relatively focused in its subject matter or geographic coverage and that con- tributes to the intellectual foundations of development operations and policy for- mulation. Some sources cited in this publication may be informal documents that are not readily available. The findings, interpretations, and conclusions expressed herein are those of the author(s) and do not necessarily reflect the views of the Board of Executive Directors of the World Bank or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of the World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this work is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of applicable law. The World Bank encourages dissemination of its work and will normally grant permission promptly. For permission to photocopy or reprint any part of this work, please send a request with complete information to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA, telephone 978-750-8400, fax 978-750- 4470, www.copyright.com. All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, World Bank, 1818 H Street, NW, Washington, DC 20433, USA, fax 202-522-2422, e-mail pubrights@worldbank.org. ISBN 0-8213-5543-0 Cover credits: World Bank Photo Library Library of Congress Cataloging-in-Publication Data Economic growth, poverty, and household welfare in Vietnam / edited by Paul Glewwe, Nisha Agrawal, David Dollar. p. cm. ­ (World Bank regional and sectoral studies) Includes bibliographical references and index. ISBN 0-8213-5543-0 1. Vietnam--Economic policy. 2. Income distribution--Vietnam. 3. Poverty-- Vietnam. 4. Households--Economic aspects--Vietnam. 5. Vietnam--Economic conditions--1975- I. Glewwe, Paul, 1958- II. Agrawal, Nisha. III. Dollar, David. IV. Series. HC444.E263 2004 330.9597--dc22 2003056887 Contents Foreword vii Acknowledgments ix Contributors xi Abbreviations and Acronyms xiii Map of Vietnam xvi 1. An Overview of Economic Growth and Household Welfare in Vietnam in the 1990s 1 Paul Glewwe Part I. Vietnam's Economic Performance in the 1990s 27 2. Reform, Growth, and Poverty 29 David Dollar 3. The Wage Labor Market and Inequality in Vietnam 53 John Luke Gallup 4. Household Enterprises in Vietnam: Survival, Growth, and Living Standards 95 Wim P. M. Vijverberg and Jonathan Haughton 5. Agriculture and Income Distribution in Rural Vietnam under Economic Reforms: A Tale of Two Regions 133 Dwayne Benjamin and Loren Brandt v vi Contents Part II. Poverty Reduction in Vietnam in the 1990s 187 6. The Static and Dynamic Incidence of Vietnam's Public Safety Net 189 Dominique van de Walle 7. The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting 229 Nicholas Minot and Bob Baulch 8. Ethnic Minority Development in Vietnam: A Socioeconomic Perspective 273 Bob Baulch, Truong Thi Kim Chuyen, Dominique Haughton, and Jonathan Haughton Part III. Progress in Health and Education in Vietnam in the 1990s 311 9. Poverty and Survival Prospects of Vietnamese Children under Doi Moi 313 Adam Wagstaff and Nga Nguyet Nguyen 10. Child Nutrition, Economic Growth, and the Provision of Health Care Services in Vietnam 351 Paul Glewwe, Stefanie Koch, and Bui Linh Nguyen 11. Patterns of Health Care Use in Vietnam: Analysis of 1998 Vietnam Living Standards Survey Data 391 Pravin K. Trivedi 12. Trends in the Education Sector 425 Nga Nguyet Nguyen 13. An Investigation of the Determinants of School Progress and Academic Achievement in Vietnam 467 Paul Glewwe Part IV. Other Topics 503 14. Child Labor in Transition in Vietnam 505 Eric Edmonds and Carrie Turk 15. Economic Mobility in Vietnam 551 Paul Glewwe and Phong Nguyen 16. Private Interhousehold Transfers in Vietnam 567 Donald Cox List of Figures, Maps, and Tables 605 Index 615 Foreword Vietnam's economic and social achievements in the 1990s are nothing short of amazing, arguably placing it among the top two or three performers among all developing countries. This success demands serious study in order to draw lessons for other developing countries. Fortunately, there are high-quality data available to undertake such a study, and this book has made full use of those data, especially the 1992­93 and 1997­98 Vietnam Living Standards Surveys, to document and understand Vietnam's experi- ence and to provide policy recommendations for other low-income countries. This volume offers a very broad array of studies of Vietnam's economy and society in the 1990s. It begins with four chapters on Vietnam's eco- nomic performance, each focusing on a different topic: macroeconomic growth, wage labor markets, household enterprises, and agriculture. Of course, economic growth can take many forms, with widely differing con- sequences for poverty reduction. The next three chapters focus on poverty reduction in the 1990s, examining the impact (or lack thereof) of various poverty programs, the spatial distribution of poverty, and poverty among ethnic minorities. The next five chapters examine health and education out- comes. Three chapters on health consider child survival, child nutrition, and use of health care services, and two chapters on education cover basic trends in enrollment and financing and the factors that determine school progress and academic achievement. The last three chapters examine topics of particular interest in Vietnam: child labor, economic mobility, and inter- household transfers. As a whole, this book constitutes a comprehensive study of economic and social development in Vietnam in the 1990s. The research presented in this book involves the collaboration of numer- ous individuals and organizations. The two Vietnam Living Standards Surveys used in the book were implemented by Vietnam's General Statistical Office, with financing from the United Nations Development vii viii Foreword Programme and the Swedish International Development Agency and tech- nical support from the World Bank. Funding for the research was obtained from the World Bank's Research Committee. The results were first presented at a workshop in Hanoi in May 2001 that was attended by a wide range of government officials, international organizations, and individual researchers. The extensive use made of household survey data in this study raises the question of what data will be collected in the future in Vietnam. Fortunately, Vietnam's General Statistical Office has developed, with assistance from the United Nations Development Programme and the World Bank, a plan for implementing similar household surveys every two years. The first survey, known as the Vietnam Household Living Standards Survey, was imple- mented in 2002 and preparations are now under way to implement another survey in 2004. This continued data collection will provide a sound foun- dation for study of Vietnam's social and economic progress in the first decade of the 21st century. François J. Bourguignon Chief Economist and Senior Vice President The World Bank Acknowledgments Many people in addition to the authors contributed to this book, and we appreciate their assistance. Funding for the two Vietnam Living Standards Surveys was provided by the United Nations Development Programme and the Swedish International Development Agency. The Social and Environmental Statistics Department of Vietnam's General Statistical Office implemented both surveys with a very high degree of enthusiasm and pro- fessionalism. Indeed, several of the authors of these chapters are from that department. Sarah Bales served as an outstanding consultant in the imple- mentation of the second survey. Financial support to undertake much of the research was obtained from the World Bank's Research Committee. The British Department for International Development funded a workshop to disseminate first drafts of the papers, which was held in Hanoi in May 2001, and also funded several of the papers. Very able editing and manuscript processing were provided by Alison Peña, Emily Khine, and Lucie Albert- Drucker. The World Bank's Office of the Publisher managed editorial and print production, including book design. Numerous other people con- tributed in many ways, but if we attempt to name them all, we are likely to omit several of them. Finally, we would like to thank all the households that participated in both surveys; by providing a large amount of information, they have helped us understand what has occurred in Vietnam in the 1990s. Hopefully, our research will lead to better policies that will improve their lives in the years to come. ix Contributors Bob Baulch Fellow, Institute of Development Studies, University of Sussex, Brighton, United Kingdom Dwayne Benjamin Department of Economics, University of Toronto, Ontario, Canada Loren Brandt Department of Economics, University of Toronto, Ontario, Canada Truong Thi Kim Chuyen Department of Geography, National University of Ho Chi Minh City, Vietnam Donald Cox Department of Economics, Boston College, Chestnut Hill, Mass. David Dollar Development Research Group, World Bank, Washington, D.C. Eric Edmonds Department of Economics, Dartmouth College, Hanover, N.H. John Luke Gallup Consultant to the World Bank Paul Glewwe Department of Applied Economics, University of Minnesota, St. Paul; and senior economist for the World Bank xi xii Contributors Dominique Haughton Department of Mathematical Sciences, Bentley College, Waltham, Mass. Jonathan Haughton Department of Economics, Suffolk University, Boston, Mass. Stefanie Koch Consultant to the World Bank Nicholas Minot Research Fellow, International Food Policy Research Institute, Washington, D.C. Bui Linh Nguyen General Statistical Office, Hanoi, Vietnam Nga Nguyet Nguyen Poverty Reduction and Economic Management, World Bank, Vietnam Country Office, Hanoi Phong Nguyen General Statistical Office, Hanoi, Vietnam Pravin K. Trivedi Department of Economics, Indiana University, Bloomington Carrie Turk Poverty Reduction and Economic Management, World Bank, Vietnam Country Office, Hanoi Dominique van de Walle Development Research Group, World Bank, Washington, D.C. Wim P. M. Vijverberg School of Social Sciences, University of Texas at Dallas Adam Wagstaff Health, Nutrition, and Population Team, World Bank, Washington, D.C. Abbreviations and Acronyms 2SLS Two-stage least squares 2SLSFE Two-stage least squares with fixed effects ANOVA Analysis of variance ASEAN Association of Southeast Asian Nations BMI Body mass index CEMMA Committee for Ethnic Minorities in Mountainous Areas CHC Commune health center CPI Consumer price index CPRGS ComprehensivePovertyReductionandGrowthStrategy CSI Comprehensive Student Insurance D Vietnamese dong (currency) DHS Demographic and Health Survey FDI Foreign direct investment FEs Fixed effects GDP Gross domestic product GER Gross enrollment rate GNI Gross national income GNP Gross national product GSO General Statistical Office HEPR Hunger Eradication and Poverty Reduction ICDS Inter-Censual Demographic Survey ICRG International Country Risk Guide ILO International Labour Organisation IMR Infant mortality rate IUD Intrauterine device IV Instrumental variable MARS Multiple adaptive regression spline MCI Multiple cropping index xiii xiv Abbreviations and Acronyms MDGs Millennium Development Goals MICS Multiple Indicator Cluster Survey MOLISA Ministry of Labor, Invalids and Social Affairs NERs Net enrollment rates NFHEs Nonfarm household enterprises NGO Nongovernmental organization NPK Nitrogen-potassium-phosphate compound fertilizer OECD Organisation for Economic Co-operation and Development OLS Ordinary least squares PCSI Propensity to consume out of social income PHF Private health facility plim Probability limit PROM Test of policy's Promotion of the poor PROT Test of policy's Protection of the poor PRSP Poverty Reduction Strategy Paper PTA Parent-teacher association ROC Receiver operating characteristics SCF U.K. Save the Children Fund, United Kingdom SIDA Swedish International Development Authority SMEs Small and medium enterprises SOEs State-owned enterprises TDY Thousands of dong per year U5MR Under-five mortality rate UNDP United Nations Development Programme UNESCO United Nations Educational, Scientific, and Cultural Organization VHI Vietnam Health Insurance VHSR Vietnam Health Sector Review VLSSs Vietnam Living Standards Surveys WHO World Health Organization WTO World Trade Organization Map of Vietnam 1 An Overview of Economic Growth and Household Welfare in Vietnam in the 1990s Paul Glewwe In the 1980s, Vietnam was one of the poorest countries in the world, and throughout most of that decade there was little indication that Vietnamese households had any hope of raising their level of welfare. Its gross domes- tic product (GDP) per capita in 1985 is estimated to have been US$130 per year, making it one of the world's five poorest countries. Although school enrollment rates were relatively high for such a poor country, they remained stagnant while school enrollment rates were increasing dramati- cally in nearby East Asian "miracle" countries. At the same time, while life expectancy was unusually high for such a poor country, exceptionally low incomes meant that the majority of Vietnamese children were malnour- ished. As a very poor country with scant prospects for a better future, Vietnam was in the same category as many of the poorest countries in Africa, Asia, and Latin America. Starting in the late 1980s and continuing through the 1990s, Vietnam transformed itself from an economic "basket case" into one of the most suc- cessful countries in the world in terms of economic growth, poverty reduc- tion, and increased household welfare. This transformation raises significant questions for anyone concerned with poverty in the poorest developing countries: What accounts for Vietnam's astonishing success? What can Vietnam do to ensure continued success? Finally, can other very poor coun- tries achieve this same success by following Vietnam's policies? This book seeks to answer these questions. It will do so by analyzing Vietnam's success in detail, using a variety of data from Vietnam and else- where. Vietnam is fortunate, not only because of its economic and social success but also because of the existence of an unusually large amount of high-quality data. The analyses in the chapters that follow make full use of 1 2 Economic Growth, Poverty, and Household Welfare in Vietnam these data and thus provide a wealth of information that can be used by researchers and policymakers in Vietnam and in other developing countries. This chapter sets the stage for the book. The first section describes the new economic policies that Vietnam has adopted since the late 1980s. The next section provides an overview of Vietnam's achievements in the 1990s. The following section summarizes the results from the various chapters, and a final section summarizes the conclusions and raises issues for future research. An appendix at the end of the chapter provides information on the 1992­93 and 1997­98 Vietnam Living Standards Surveys (VLSSs), the data that are used most frequently in the book.1 Doi Moi Policy Reforms In the 1980s, Vietnam was an extremely poor country, with a low rate of eco- nomic growth. Inflation rose dramatically as government deficits were fi- nanced by printing money; by 1986, the annual inflation rate had risen to 487 percent. Vietnam's response to this poor economic performance was the adoption of the Doi Moi ("renovation") policy reforms in the late 1980s.2 This process began with the Sixth Congress of the Communist Party, held in December 1986. At this meeting, the government explicitly adopted the goal of replacing central planning with a regulated market economy. A series of fundamental policy changes was quickly implemented in the following years, so that by 1989 most forms of private economic activity were legal and price controls had been removed for almost all goods and services. This section describes these policy changes, as well as several that were imple- mented in the 1990s, in more detail. The first important policy changes were implemented in the agricultural sector. In 1987 and 1988, price controls were gradually removed for agricul- tural goods, and farm households were allowed to sell any surplus products at whatever price the private market would bear. Another decisive change occurred when Decree Number 10 was issued in April 1988. That decree dis- mantled agricultural cooperatives and divided up almost all agricultural land among the rural households that had worked for those cooperatives. Those households were provided with leases that lasted for 15 or more years for the plots of land they received. Households were required to pay taxes for the right to use the land, but after taxes all output was the property of the households. These changes in agricultural policy, along with the lifting of many re- strictions on overseas exports in the late 1980s, helped Vietnam to become the world's third largest rice exporter by 1992, a dramatic change from its status as a rice importer in the mid-1980s. Yet land rights were still limited at the beginning of the 1990s; agricultural land could not be transferred to an- other household, nor could it be transferred in the form of an inheritance. In the 1990s, the government increased property rights for farming households and, more generally, reduced restrictions on agricultural markets. Decree Number 5 of 1993 (often referred to as the 1993 Land Law) granted more An Overview of Economic Growth and Household Welfare in Vietnam in the 1990s 3 land rights and security. Tenure lengths were extended to 20 years for an- nual cropland and 50 years for perennial cropland. Households were al- lowed to rent out and mortgage their land and to transfer land use rights, including transfer by inheritance. Another important policy change was Decree Number 140 of 1997, which relaxed restrictions on the internal trade of agricultural commodities. Most of the remaining export restrictions were removed in the 1990s. Sweeping policy changes were also made in other sectors of Vietnam's economy. To ensure macroeconomic stability, and in particular to reduce the rate of inflation, the central government reduced spending and modified the taxsystemtoraisemorerevenue.Thisreducedthecentralgovernmentbudget deficit from 8.4 percent of GDP in 1989 to 1.7 percent in 1992, one consequence of which was that the rate of inflation plummeted, as will be seen in the next section.Muchofthisspendingreductiontooktheformofclosingorsellingun- profitable state-owned enterprises (SOEs), and reducing the number of em- ployees at many of those that remained. Between 1989 and 1992, the number of SOEs was cut in half, from 12,000 to 6,000, and about 800,000 employees of SOEs (about one-third of the initial number) were laid off. The rapid growth of private sector employment opportunities, and the small share of SOE em- ployees in the total work force, helped Vietnam avoid a sizable increase in unemployment from this sharp reduction in public sector jobs. A third area of major policy changes was in foreign trade and invest- ment, although these changes were more gradual. One of the first steps took place in 1989, when the exchange rate was unified and then devalued. Bar- riers to exports and imports were gradually dismantled in the late 1980s and early 1990s, and the monopoly on foreign trade granted to a small number of state trading companies was ended.Alaw encouraging private investment was passed in 1987 and implemented in early 1988. It loosened regulations on joint ventures and allowed for 100 percent foreign-owned enterprises. Policy changes in the 1990s continued to remove trade and investment bar- riers. By 2003, import quotas existed for only two items, sugar and petro- leum products, and quantitative restrictions on exports applied to only a few items. Import tariffs gradually decreased, with the average tariff falling from 12.7 percent in 1996 to 9.3 percent in 2003. Social sector policies also experienced major changes under Doi Moi, es- pecially in the areas of health and education. A fundamental deregulation of the health care system was implemented in 1991. Doctors, nurses, and other health care personnel were allowed to establish private clinics, and private shops and individuals were permitted to sell a wide range of drugs. Both public and private health facilities were able to charge fees for medicines and health services. In 1994, the central government assumed the responsi- bility of paying employees in commune health centers, which previously had been the responsibility of the communes. In 1993, a health insurance program was started, which by 2001 covered 12 percent of the population. In education, changes were less radical, but they were still substantial. Private schools were legalized in 1989. Spending per pupil has increased 4 Economic Growth, Poverty, and Household Welfare in Vietnam dramatically in real terms, increasing from 1.8 percent of GDP in 1992 to 3.5 percent in 1998. At the same time, tuition fees were introduced at the sec- ondary and postsecondary levels. Government jobs were no longer guaran- teed for graduates of upper secondary and postsecondary schools. A more recent change is that entrance examinations are no longer used to limit stu- dent enrollment into lower secondary and upper secondary schools. Finally, several programs have recently been introduced to increase school enroll- ment among ethnic minorities. A final aspect of Vietnam's Doi Moi reforms has been integration into the international economy. In 1992, the country signed a preferential trade agreement with the European Economic Community. Diplomatic relations were reestablished with the United States in 1994, and in 2001, Vietnam and the United States signed a wide-ranging bilateral trade agreement. Vietnam joined the Association of Southeast Asian Nations (ASEAN) in 1995, which included membership in the ASEAN Free Trade Area. In 1995, it also sub- mitted an application to join the World Trade Organization, and negotia- tions started in earnest in 2002. Economic and Social Performance Vietnam's Doi Moi policy changes were followed by more than a decade of rapid economic growth. The average annual rate of real economic growth from 1988 to 2000 was 7.1 percent. In the early 1990s, Vietnam became the world's third largest exporter of rice, and in the late 1990s, it became the sec- ond largest exporter of coffee. This performance is all the more extraordinary given that Vietnam's main economic benefactor in the 1980s, the former Soviet Union, dissolved in 1991, ending a variety of subsidies that it had been providing to the Vietnamese economy. While the EastAsian financial crisis in 1997 and 1998 slowed economic growth somewhat, the slowdown was minor and short-lived. Although GDP growth dropped to 5.8 percent in 1998 and 4.8 percent in 1999, it increased to 6.8 percent in 2000 (World Bank, Asian Development Bank, and United Nations Development Programme 2000). Economic and Social Trends from 1984 to 2000 Table 1.1 provides economic and social data for Vietnam for four years: 1985, 1988, 1994, and 2000 (1985 is the earliest year for which comparable economic data are available). The period from 1985 to 1988 reflects conditions before the Doi Moi policies were in place; although some policies were adopted in 1987 and 1988, a year or two is usually needed before they have major ef- fects. During this period, Vietnam's real GDP grew at a respectable rate of 4.2 percent, but its population growth of 2.1 percent resulted in a per capita annual increase of only 2.0 percent. This rate of growth was far behind the growth rates of higher-performing East Asian economies such as China, Hong Kong, Malaysia, the Republic of Korea, Singapore, Taiwan (China), and Thailand. Disaggregation of total GDP into agriculture, industry, and (1993, 7.4 4.4 1.11 6.4 1.5 5.7 n.a. n.a. United 23.6 17.9 cent) 1994­2000 Bank (per rldo ollment: W enr rates trade: hool 6.9 3.9 7.5 8.5 1.8 5.0 n.a. n.a. sc 33.1 24.5 owth 1988­94 and 2b). gr (1999); 200 deficit Bank Annual 1996b, 4.2 1.8 8.1 3.6 2.1 2.0 n.a. n.a. budget 13.4 16.1 orld W 1990, 1985­88 (2002a); and (1987, Bank (1998) Bank 2000 273.58 63.35 96.92 13.311 77.64 2.8 3,524 ­1.6 14.45 14.07 0.38+ 37.4% (1998) 106 67 69 34% (1998) orld orld W W Litvack and and (2002) 1994 178.53 48.97 51.54 78.03 70.82 3.0 9.5 4.05 5.25 2,521 1.20- 58.1% expectancy: (1993) 311 41 68 51% (1993) Dollar GSO life rate: years); -- -- 1988 19.961 38.87 33.35 47.74 63.73 7.1 0.73 1.41 1,882 374.4 0.68- poverty 104 40 66 cent. population: per (various Performance or 2002); (2003); rate a 2000, Fund Social ganization 1985 106.18 36.83 26.40 42.95 59.87 1,774 -- 91.6 0.50 0.90 ­0.41 (1984) -- ~75.0% 103 43 65 eady Or and alr ([GSO] is fice Monetary Of Cultural oss) number and Economics prices) (gr dong, Statistical indicators International 1994 GDP) billion) original of cent) ratios Scientific, ietnam' social General V dong, (% e (thousand (per billion) billion) (US$ inflation: and (million) ollment rate applicable; GDP: 1.1. prices) deficit (US$ (US$ rate available. Educational, (trillion balance enr malnutrition Not ces: 2002a); expectancy Not ableT Agricultur Industry Services 1994 Economic GDP Population GDP/capita Budget Inflation Exports Imports radeT Primary Secondary (stunting) -- n.a. Sour Poverty School Life Child 1996a, Nations 5 6 Economic Growth, Poverty, and Household Welfare in Vietnam services reveals very slow growth in agriculture (1.8 percent, which is less than the population growth rate), modest growth in services (3.6 percent), and high growth in the industrial sector (8.1 percent). The high growth in Vietnam's industrial sector could give a false im- pression of success in this sector during the 1980s. Yet much of this "success" reflects large government subsidies to this sector, which had negative con- sequences for the economy as a whole. The 1980s were characterized by high and growing government budget deficits and consequent high and growing rates of inflation. By 1988, the budget deficit was 7.1 percent of GDP and the annual inflation rate was 374 percent. Another characteristic of Vietnam in the 1980s was low levels of exports and a large trade deficit; in 1988, the value of imports was almost double the value of exports (US$1,410 million and US$730 million, respectively). This mediocre economic performance was accompanied by high rates of poverty and little improvement in social indicators. A rough estimate is that 75 percent of Vietnamese were poor in 1984 (Dollar and Litvack 1998), in the sense that their consumption expenditures were insufficient to purchase a basket of food items that meet minimal caloric requirements (after allowing for purchase of essential nonfood items). The primary school enrollment rate was high, but the (gross) secondary school enrollment rate dropped slightly, from 43 percent in 1985 to 40 percent in 1988. Life expectancy was also high, but it was almost certainly accompanied by very high rates of child malnutrition. (Data for the 1980s are scarce, but the high rates in the early 1990s suggest even higher rates in the previous decade.) Vietnam's prospects dramatically improved in the late 1980s. The GDP growth rate increased rapidly, from 4.2 percent in the mid-1980s to 6.9 per- cent from 1988 to 1994 and 7.4 percent from 1994 to 2000. It is particularly remarkable that the East Asian financial crisis in the late 1990s had almost no effect on Vietnam's economic growth. Increases in per capita GDP growth are even more dramatic as a result of declining population growth; that growth rate nearly tripled, from 2.0 percent in the mid-1980s to 5.0 percent from 1988 to 1994 and 5.7 percent from 1994 to 2000. Dividing overall GDP into agriculture, industry, and services, the agricultural growth rate more than doubled, from 1.8 percent in the mid-1980s to 3.9 percent from 1988 to 1994 and 4.4 percent from 1994 to 2000. In contrast, industrial growth dropped slightly, from 8.1 percent in the mid-1980s to 7.5 percent from 1988 to 1994, but then it jumped to 11.1 percent from 1994 to 2000. Finally, changes in the growth in services are similar to those in agriculture; they more than doubled, from 3.6 percent in the mid-1980s to 8.5 percent from 1988 to 1994, after which growth was somewhat lower at 6.4 percent from 1994 to 2000. Thus, in the initial years after the Doi Moi policies were introduced, most of the increase in economic growth occurred in agriculture and services, and only in the latter half of the 1990s did economic growth in industry surpass the levels seen in the mid-1980s. While industrial growth in Vietnam after the Doi Moi policies were adopted may seem less impressive given the high growth rates in this sector An Overview of Economic Growth and Household Welfare in Vietnam in the 1990s 7 in the mid-1980s, it is important to recall that nearly 1 million employees of SOEs lost their jobs in the late 1980s and early 1990s, and more generally that the high growth in the late 1980s and early 1990s was maintained while government budget deficits (and thus government subsidies to industry) were shrinking. As seen in table 1.1, Vietnam's government budget deficits shrank from 7.1 percent of GDP in 1988 to about 3 percent in the 1990s, and inflation virtually disappeared (9.5 percent in 1994 and ­1.6 percent in 2000). Another sign of economic health is exports, which include both agricultural and industrial products. Vietnam's exports grew dramatically, at an annual rate of 33 percent from 1988 to 1994 and a slightly lower annual rate of 24 percent from 1994 to 2000. By 2000, Vietnam was running a trade surplus, with a 20-fold increase in exports over a 12-year period. This rapid economic growth was accompanied by a sharp decrease in poverty and dramatic improvements in social indicators. Specifically, the poverty rate declined from about 75 percent of the population in 1984 to 58 percent in 1993 and 37 percent in 1998. The sharp drop in only five years from 1993 to 1998 is an achievement that is rarely seen in any developing country, and the economic growth since 1998 suggests that the poverty rate has continued to decline into the 21st century. Turning to social indicators, the (gross) primary school enrollment rate increased somewhat from its al- ready high rates in the 1980s, and (gross) secondary school enrollment rates rose from 40 percent in the mid-1980s to 67 percent by 2000. The incidence of child malnutrition, as measured by stunting (low height for age) among children younger than five years of age, also declined dramatically, from 50 percent in 1993 to 35 percent in 1998. Finally, life expectancy continued its steady rise to rates usually seen only in high-income countries. In summary, Vietnam's economic and social performance in the 1990s was arguably better than that of any other developing country during the same period, with the possible exception of China. Yet despite these impres- sive gains, Vietnam remains a very poor country, and future success is far from assured. One issue that commands particular attention is trends in inequality. This is discussed further in the next subsection. Economic Growth and Inequality Ever since Simon Kuznets (1955) examined the relationship between eco- nomic growth and inequality in developed countries, many economists and other social scientists have investigated whether economic growth in- evitably leads to increased inequality. For Vietnam, the question is whether the rapid economic expansion that followed the adoption of the Doi Moi policies led to an increase in inequality and, if so, whether future economic growth will be accompanied by even greater inequality. Vietnamese policy- makers are genuinely concerned about inequality, because reductions in poverty brought about by economic growth are diminished by increases in inequality. There are no reliable data on inequality in Vietnam in the 1980s, but the 1993 and 1998 VLSSs show what happened in the 1990s. Table 1.2 8 Economic Growth, Poverty, and Household Welfare in Vietnam Table 1.2. Changes in Inequality in the 1990s Inequality index 1993 1998 Gini coefficient 0.329 0.352 Theil index 0.197 0.230 Decomposition of Theil index A. Urban/Rural Within 0.155 0.158 Between 0.042 0.072 B. Regions Within 0.171 0.180 Between 0.026 0.050 C. Education of head of household Within 0.181 0.197 Between 0.016 0.033 D. Ethnic group Within 0.177 0.206 Between 0.020 0.024 Source: Glewwe, Gragnolati, and Zaman (2000). presents information from those two surveys on inequality of consumption expenditures, and how that inequality changed, in the 1990s.3 The Gini coefficient is the most commonly used index of inequality. Table 1.2 shows that it increased by 7 percent from 1993 to 1998, from 0.329 to 0.352. Another commonly used measure of inequality is Theil's entropy measure; that index increased at a faster rate, rising by 17 percent (from 0.197 to 0.230). The Theil measure has the useful feature that when the total population is divided into a few groups, overall inequality is the sum of the (weighted) average inequality within each group plus inequality in the mean incomes (in this case, mean consumption expenditures per capita) across the different groups. To see the usefulness of this property of the Theil measure, consider the simplest decomposition: division of Vietnam's total population into the 20 percent that lives in urban areas and the 80 percent that lives in rural areas. The weighted average of inequality within urban areas and within rural areas increased only slightly over the five years, an increase of 2 percent from 0.155 to 0.158. In contrast, the inequality due to the difference between mean (average) urban expenditures and mean rural expenditures increased by 71 percent. Thus, almost all of the increase in in- equality as measured by the Theil index was due to the increased gap be- tween mean per capita expenditures in urban areas and mean per capita expenditures in rural areas; very little comes from increased inequality within urban and rural areas. Three other decompositions of the Theil inequality index are shown in table 1.2. When Vietnam is divided into its seven economic regions (North- ern Uplands, Red River Delta, North Central Coast, Central Coast, Central An Overview of Economic Growth and Household Welfare in Vietnam in the 1990s 9 Highlands, Southeast, and Mekong Delta), average inequality within each of these regions increased by only 5 percent, while average inequality be- tween regions almost doubled. Thus, about three-fourths of the increase in the Theil measure (0.024 out of 0.033) is accounted for by greater inequality in the mean expenditure levels across regions, as opposed to increased in- equality within regions. As noted later in this chapter, there is evidence that the returns to education have increased in Vietnam in the 1990s. One way of assessing the contribution of this change to the overall increase in inequality is to divide the population into groups according to the education level of the head of household (none, primary, lower secondary, upper secondary, and so forth) and apply the Theil decomposition property. This decomposi- tion shows that about half of the total increase in inequality (0.016 out of 0.033) is due to inequality within these groups and the other half is due to in- creases in equality in the mean expenditure levels of these groups. A final decomposition divides the population into different ethnic groups. In this case, increased gaps in mean expenditure levels across ethnic groups play very little role in explaining increased inequality, accounting for only about 12 percent (0.004 out of 0.033) of the increase in the Theil index. Income and expenditure levels are only one dimension of the quality of life; thus, inequality in other dimensions also merits attention. There are worrisome trends in health and education, but there are also some changes in a more egalitarian direction, as discussed further below. One worry in health is that declines in infant mortality rates appear to be concentrated among middle-income and better-off households, with little reduction in in- fant mortality among low-income households. A more positive result is that malnutrition, as measured by child stunting, has dropped for all income groups. In education, one bright spot is that primary school enrollment rates have increased fastest among poorer households, especially ethnic minority households, but at the secondary and postsecondary levels, large differences in enrollment rates have persisted and may even have increased. Overall, increased economic growth in Vietnam has been accompanied by a modest increase in inequality of consumption expenditures, as well as increased inequality in some, but not all, other dimensions. The decomposi- tions in table 1.2 provide some information on the nature of the increase in expenditure inequality, and they can provide some guidance for Vietnamese policymakers on how to prevent, or at least minimize, future increases in inequality. The next section reviews the chapters in this book, after which the policy conclusions are summarized (including a discussion of policies to reduce inequality) and suggestions are provided for future research. Summary of the Volume Vietnam's success in maintaining high economic growth and reducing poverty has already been described in other publications, such as a recent report by the World Bank and other donor agencies (World Bank 1999). The distinguishing characteristic of the chapters in this book is that they attempt 10 Economic Growth, Poverty, and Household Welfare in Vietnam to explain the reasons for this success and draw lessons for the future. They do so by going beyond simple descriptive exercises and presenting rigorous analyses of a wide variety of topics. Economic Performance Economic growth is important not only because it raises incomes but also because it leads to a wide array of better socioeconomic outcomes, such as increased school enrollment and better health. Chapters 2 to 5 in this book examine the impact of Vietnam's policies on economic growth and the na- ture of that growth in different sectors of the economy. Chapter 2, by David Dollar, looks at the economy as a whole; chapter 3, by John Luke Gallup, fo- cuses on the incomes of wage earners; chapter 4, by Wim P. M. Vijverberg and Jonathan Haughton, considers nonagricultural businesses; and chap- ter 5, by Dwayne Benjamin and Loren Brandt, looks at farming households. David Dollar begins chapter 2 by posing a question: How can Vietnam's impressive record of economic growth be reconciled with the fact that its eco- nomic policies are not necessarily better than those of many other developing countries? For example, in one index of economic freedom (O'Driscoll, Holmes, and Kirkpatrick 2000), Vietnam was ranked 144th out of 155 coun- tries. Dollar argues that Vietnam's policies in the 1980s were even worse than they were in the 1990s, and this improvement in policies explains its enviable economic performance in the last decade. Yet Dollar goes on to argue that the boost from this modest improvement in policies is likely to be temporary, and sustained economic growth (and the poverty reduction that comes with it) cannot continue unless additional pro-growth policies are adopted. Indeed, he argues that there is much room for improvement in Vietnam's policies. In the general area of property rights and governance, Vietnam's rank with respect to other countries is better than average for political stability, but it is worse than average in terms of property rights, government effectiveness, regulatory burden, and corrup- tion. In terms of market development, Vietnam's financial system and labor market are both rated as very weak by the international business commu- nity. Finally, although much progress has been made, there are still signifi- cant barriers to trade and foreign investment. To support his interpretation of Vietnam's current situation and its fu- ture prospects, Dollar uses cross-country growth regressions. This allows him to provide much more specific advice than Dollar and Litvack (1998) provided several years earlier. Yet such regression results must be treated with care, and the recommendations provided in this chapter are likely to be controversial. Even so, the debate that this chapter will provoke should prove fruitful to policymakers and researchers alike. The most important economic asset of the vast majority of Vietnamese householdsistheirlabor.Inchapter3,JohnLukeGallupprovidesanoverview oflabormarketsinVietnamandhowtheyhavechangedduringthe1990s,giv- ing particular attention to the inequality of labor income. Most Vietnamese An Overview of Economic Growth and Household Welfare in Vietnam in the 1990s 11 workers are farmers, but the proportion of workers who were farmers-- compared with workers who work for wages or who work in nonagricultural self-employment activities--slowly declined in the 1990s. Gallup focuses on wage earners, and chapters 4 and 5 examine self-employed workers. Gallup shows that real wage rates increased dramatically in the 1990s, at an average rate of 10.5 percent per year. This is particularly true of skilled nonagricultural workers, whose real wages increased by nearly 13 percent per year from 1993 to 1998. The number of hours worked also increased substantially. Gallup also shows that the returns to education are low in Vietnam, although they did increase during the 1990s. The increases in wages have a distinct regional dimension. Wages were highest, and increased the fastest, in the two largest cities in Vietnam, Hanoi and Ho Chi Minh City. This disparity is not explained by the fact that work- ers in these two cities tend to have more skills and education; it is still large even when comparing workers with the same skills and educational back- grounds, and regression analysis confirms that the gap is large and is not de- creasing. This likely reflects legal barriers that discourage migration from other areas into these two cities, which is consistent with recent qualitative research on migration in Vietnam. The rapid increase in wage income, the wage gap between Hanoi and Ho Chi Minh City and the rest of Vietnam, and the increase in the return to ed- ucation all raise the question of whether overall income inequality has in- creased in Vietnam, and what will happen in the future. Gallup shows that inequality of wage income declined modestly in the 1990s for Vietnam as a whole, although it did increase in Hanoi and Ho Chi Minh City. Despite that overall decline, the fact that agricultural income is much more equally dis- tributed than wage income, combined with the steady decline over time of agricultural income as a share of total income, led to a slight increase in in- equality in the 1990s. Using simple simulations based on data from the two household surveys, Gallup demonstrates that inequality is likely to increase in the future as wage work becomes a larger share of total employment in Vietnam. Although Gallup's focus on wage earners provides many insights, the lessons learned have limited implications because only about 20 percent of the working population works for wages. Chapters 4 and 5 examine income from self-employment. Wim Vijverberg and Jonathan Haughton begin chapter 4 by investigating the nature of nonagricultural self-employment in Vietnam. Small household enterprises are potentially very important for private sector economic growth in Vietnam because they are, traditionally, the first step in the development of a vibrant small and medium enterprise sector. Vijverberg and Haughton argue that the role of household enterprises in future economic growth may be much more modest than some have hoped. Instead of providing the basis for the expansion of the private sector, self- employed workers in household enterprises decreased from about 26 per- cent to about 24 percent of the labor force between 1993 and 1998. On a more positive note, household enterprises do appear to be a stepping-stone to 12 Economic Growth, Poverty, and Household Welfare in Vietnam more lucrative wage employment for Vietnamese workers. Although the re- cent policy changes have simplified the procedure for registering new en- terprises, the authors nevertheless conclude that household enterprises will play only a modest role in Vietnam's economic transformation. This assess- ment is somewhat more pessimistic than that given by Vijverberg (1998). About 60 percent of Vietnamese workers are self-employed farmers. Farming households are poorer than both wage earners' households and households operating nonagricultural enterprises, so the fate of farmers is closely tied with the prospects for future poverty reduction in Vietnam. In chapter 5, Dwayne Benjamin and Loren Brandt use the extensive data on agricultural activities in the 1993 and 1998 VLSSs to provide a detailed analysis of agriculture and income distribution in rural Vietnam. They focus on two important events that affected Vietnamese farmers in the 1990s--the large increase in the relative price of rice and the sharp drop in the price of fertilizer--both of which occurred as a result of the lifting of marketing and trade restrictions. Benjamin and Brandt show that the increase in rice prices and the de- crease in fertilizer prices raised the incomes of most rural households in Vietnam, which helps explain why poverty decreased in rural areas in the 1990s. In addition, poverty was reduced by the shifting patterns of agricul- tural production, with increasingly intensive cultivation of rice in southern areas and shifts into nonrice crops in northern areas. The authors also find that income inequality did not increase in rural areas of Vietnam; even though the price of rice increased, many rural households (including many low-income households) were net sellers of rice, so their incomes increased. They conclude that liberalized agricultural policies did not have any ad- verse effects on inequality or poverty in rural areas. Poverty Reduction Chapters 6 through 8 focus directly on poverty in Vietnam and on programs designed to reduce poverty. Chapter 6, by Dominique van de Walle, exam- ines the extent to which government safety net programs actually benefit the poor. Chapter 7, by Nicholas Minot and Bob Baulch, focuses on the spatial distribution of the poor and investigates whether there is potential to target assistance to the poor based solely on their area of residence. Chapter 8, by Bob Baulch, Truong Thi Kim Chuyen, Dominique Haughton, and Jonathan Haughton, investigates poverty among minority groups, who are much more likely to be poor than are ethnic Vietnamese (the Kinh). The Vietnamese government has in place many different programs that transfer resources to households and communities. Some are explicitly intended to reduce poverty, and others provide income support for groups that may or may not be poor, such as retired government workers and dis- abled military veterans. Funding for these programs more than doubled in the 1990s. The effectiveness of these programs at reducing poverty often depends on how they are implemented at the local level--indeed, some An Overview of Economic Growth and Household Welfare in Vietnam in the 1990s 13 programs must be financed at the local level. Chapter 6 examines a wide va- riety of programs, including school fee exemptions, pension and disability funds, assistance from nongovernmental organizations, and government transfer payments to disadvantaged households. In chapter 6, van de Walle's findings are both disturbing and highly in- formative. Most of these programs provide low benefits to a wide range of households instead of focusing most of their benefits on poor households. Indeed, the typical nonpoor household often receives more benefits than the typical poor household. An example of this is Vietnam's social insurance payments, which accrue mainly to urban households and better-off rural households. There is little coordination across programs, and many poor households in poor communities receive low benefits in part because the funding for many programs comes primarily from within those very same communities. Based on the panel data, the findings of chapter 6 suggest that Vietnam's social assistance programs played no role in the poverty reductions that oc- curred during the 1990s, and they typically failed to prevent households from falling into poverty. The chapter argues that Vietnam's Hunger Eradi- cation and Poverty Reduction program has produced "little discernable progress." Overall, van de Walle finds much room for improvement, and the chapter concludes with several recommendations to enhance the design of Vietnam's poverty reduction programs. In particular, it recommends less re- liance on local resources to finance local programs, more monitoring of the allocation of central government funds within provinces and districts, and restrictions on local discretion in implementing centrally mandated poverty reduction programs. In chapter 7, Nicholas Minot and Bob Baulch pose a fundamental ques- tion concerning poverty policies in Vietnam: To what extent is poverty con- centrated in certain geographic areas? This question is important because it is very difficult, in practice, to identify poor households. Yet if poor house- holds are concentrated spatially, then it may be more efficient to assist all households in poor communities, regardless of their income levels, instead of spending a large amount of time--and money--to distinguish poor from nonpoor households within each geographic area. Minot and Baulch point out that many of Vietnam's poverty reduction programs use some kind of geographic targeting, but they do so rather inef- fectively. This is consistent with van de Walle's findings in chapter 6. Chap- ter 7 combines the VLSS data with the data from Vietnam's 1999 census to estimate the incidence of poverty in each of Vietnam's 61 provinces. The authors find that poverty is concentrated in six provinces bordering China and the Lao People's Democratic Republic and in eight other provinces in the Northern Uplands, the North Central Coast, and the Central Highlands. They also find that the Vietnamese government's official list of poor com- munes is not very accurate. Chapter 7 concludes by recommending a method that combines census and VLSS data to estimate the incidence of poverty at the district level. 14 Economic Growth, Poverty, and Household Welfare in Vietnam About 15 percent of the Vietnamese population consists of ethnic mi- norities. With the sole exception of the Chinese, who are relatively well off, ethnic minorities in Vietnam are considerably poorer than the Kinh. They also have lower school enrollment, higher fertility, and less access to basic health services, although they do not appear to have higher rates of malnu- trition. In chapter 8, Bob Baulch, Truong Thi Kim Chuyen, Dominique Haughton, and Jonathan Haughton provide an in-depth analysis of the so- cioeconomic conditions of ethnic minorities in Vietnam, using both the VLSS data and data from the 1999 census. Particular care is taken to avoid aggregating minorities into a single group; instead, the authors distinguish among minority groups in different parts of Vietnam. Baulch and his coauthors find that most ethnic minorities have shared in many of the gains of the 1990s. Yet this is not the case for ethnic minorities in the Central Highlands, whose per capita expenditures did not increase in the 1990s. Another interesting (and potentially controversial) finding is that ethnic minorities that are more open to assimilation with the Kinh, both economically and culturally, have done relatively well. However, some eth- nic minorities, such as the Khmer and the Thai, appear to have assimilated economically while retaining their distinct ethnic and cultural identity. Chapter 8 uses regression analysis to investigate why ethnic minorities are poorer than the Kinh majority. The results suggest that observable dif- ferences between ethnic minority and Kinh households explain, at most, only one-third of the gap between these households. This implies that the "returns" to ethnic minority assets appear to be lower than the returns to the assets of Kinh households. Unfortunately, the underlying reasons for this cannot be examined until household survey data are collected that have much larger samples of ethnic minority households. Social Sectors Chapters 9 through 13 examine progress in health and education in Vietnam in the 1990s. Dramatic progress occurred in both sectors in that decade, but there is room for further progress, and there are some signs that inequality in health and education outcomes is increasing. Adam Wagstaff and Nga Nguyet Nguyen examine infant and child mor- tality in Vietnam in chapter 9. Compared with other low-income countries, Vietnam's performance in this area is unusually good, in that it has much lower infant and child mortality rates than other countries at its level of in- come (which is less than US$1 a day, or about US$300 per capita annually). In addition, its infant and child mortality rates dropped steadily in the 1990s. Despite this progress, Wagstaff and Nguyen point out a worrisome finding: Most of the reductions in these mortality rates occurred among nonpoor households, so that inequality in mortality rates across different income groups increased significantly in the 1990s. Indeed, it appears that there has been no reduction in infant mortality among the poorest 25 percent of the population. An Overview of Economic Growth and Household Welfare in Vietnam in the 1990s 15 The authors carefully examine the determinants of infant and child mor- tality in Vietnam. They find that many factors played a role, including increased incomes. These estimates provide suggestions for policies to re- duce infant mortality in the future. Specifically, increasing years of schooling of young women, improved sanitation, and increasing the proportion of births that occur in medical facilities, or at least are attended by trained med- ical personnel, should lead to substantial decreases in the infant mortality rate among the poor. Information campaigns may also be helpful, but the evidence in favor of this is indirect. Finally, chapter 9 uses the estimates to predict future infant and child mortality rates through to 2015. The authors find that it may be possible for Vietnam to reach its goal of reducing child mortality by two-thirds from 1990 to 2015, but this will depend on whether reductions in child mortality can be accelerated among poor households. Infant and child mortality is only one indicator of children's health sta- tus. Another important indicator is the health and nutritional status of the vast majority of children who survive. In chapter 10, Paul Glewwe, Stefanie Koch, and Bui Linh Nguyen examine progress in child nutrition in Vietnam in the 1990s. Vietnam was again fortunate in that children's nutritional sta- tus, as measured by height for age, was much higher in 1998 than in 1993. This chapter investigates the extent to which increased household incomes explain improvements in children's nutritional status and then investigates what role other factors may have played to bring about those improvements. Glewwe, Koch, and Nguyen find that growth in household incomes did not play a decisive role in reducing the incidence of stunting in Vietnam. Using a variety of different estimation methods, they find that increases in households' per capita expenditures always explain much less than one-half of the total reduction in stunting. This confirms the prediction of Ponce, Gertler, and Glewwe (1998), and it implies that something else, most likely changes in health services and health care policies, is primarily responsible for the improved nutritional status of young Vietnamese children. Chapter 10 employs regression analysis, using detailed data from com- mune health centers, to understand what aspects of health services and health care policies may be responsible for the improvement in children's health status. The results suggest that reducing the distance to private phar- macies could lead to better child nutrition outcomes, although the size of this effect is rather small. They also suggest that providing commune health centers with sanitary toilets and ample supplies of oral rehydration salts could have substantial positive impacts on child health in Vietnam. Unfor- tunately, these findings are tentative at best because of a variety of difficult estimation problems. Chapter 11 presents the third and final study of health in Vietnam--Pravin Trivedi's study of health care use. This study examines health care use by all household members, not just by children. It finds that poorer households, almost all of whom are found in rural areas, rely on commune health centers for medical services, and better-off households are much more likely to use 16 Economic Growth, Poverty, and Household Welfare in Vietnam public hospitals, which are almost always found in urban areas. Both rich and poor households spend most of their health care money on purchases of medicines from private providers, as opposed to spending it on medical consultations. This is consistent with the earlier findings of Gertler and Litvack (1998), but Trivedi goes beyond descriptive analysis by estimating the determinants of health care choices, giving particular attention to the role played by health insurance in the use of different medical services. He finds that households with health insurance are more likely to use public health facilities, espe- cially hospitals, and less likely to use private providers or to purchase med- icines without consulting medical personnel. He also estimates the income elasticity of health care expenditure and finds it to be quite high; this implies that as households' incomes increase, they spend a larger percentage of their income on health care. Trivedi also finds clear evidence that households are dissatisfied with commune health centers. Trivedi draws several policy conclusions from his analysis. First, he sug- gests three routes to reducing the high reliance of Vietnamese households on nonprescription purchases of antibiotics: increasing household incomes, in- creasing education levels, and expanding health insurance coverage. Sec- ond, he finds clear evidence of dissatisfaction with commune health centers, which are the first line of defense in Vietnam's health care system; the more other options are available, the less households use these facilities. Finally, he argues for a larger role for private health care providers. The next two chapters examine education in Vietnam. In chapter 12, Nga Nguyet Nguyen provides a broad overview of education in the country, with particular focus on school finance issues. School financing is a topic of particular interest in Vietnam. On the one hand, school fees have been elim- inated for primary schools. On the other hand, other costs associated with education are still a significant burden on households, and recent moves to decentralize education finance may prove to be particularly burdensome for poor communities. Nguyen shows that the Vietnamese government greatly increased spending on education in the 1990s, tripling the amount spent in real terms. Spending increases for primary and lower secondary education in Vietnam were even higher and were accompanied by particularly high increases in primary school enrollment among the poorest 20 percent of the population. Increases in school enrollment were widespread, affecting both boys and girls and all regions, ethnic groups, and income levels. This is in marked contrast with the situation in the late 1980s and the early 1990s, when school enrollment rates were declining (Glewwe and Jacoby 1998). However, significant problems remain. Enrollment gaps between rich and poor continue, and may even have increased, at the secondary and post- secondary levels. The quality of education may also vary dramatically, be- cause only better-off households can afford to pay for the extra classes and private tutors that compensate for the unusually short school day, which is typically only three or four hours long. The chapter ends with an analysis of An Overview of Economic Growth and Household Welfare in Vietnam in the 1990s 17 the rate of return to education, based on earnings regressions. Nguyen finds that rates of return to education increased in the 1990s, particularly at the upper secondary and postsecondary levels. In contrast, vocational educa- tion appears to have no impact on workers' wages. In chapter 13, Paul Glewwe investigates the determinants of school progress and academic achievement, as measured by test scores, for stu- dents in primary and secondary schools. As mentioned above, enrollment rates increased during the 1990s, especially at the secondary level. One of the most interesting and encouraging findings is that enrollment rates in- creased much more rapidly for ethnic minority groups than they did for the Kinh, closing much of the gap between the Kinh and the ethnic minorities. Indeed, regression analysis demonstrates that much--and perhaps most--of the remaining gap between the Kinh and ethnic minorities is due to differ- ences in the communities in which they live, as opposed to being due to any inherent traits. Two other findings of interest from the regression analysis have direct policy implications. First, there is no evidence that teaching ethnic minori- ties in their own languages has any positive effect on their school progress or academic achievement; indeed, negative impacts were found in some es- timates. Second, there is some evidence that shorter children, who presum- ably were malnourished in early childhood, do worse in school, but the effects are not strong and often lose statistical significance. On a more disappointing note, the regression analysis in chapter 13 did not produce clear findings regarding what schools can do to improve schooling outcomes. This result is due to serious estimation problems that are difficult to overcome. Much more work and more intensive data collec- tion are needed before useful advice can be provided to Vietnamese policy- makers on ways to improve primary and secondary schools. Other Topics The last three chapters of this book treat a diverse set of subjects that attract high interest in Vietnam and elsewhere. In chapter 14, Eric Edmonds and Carrie Turk examine an issue that has received increased attention in the last decade: child labor. In many developing countries, children in rural areas help their parents with farm work, and in some countries, children are found working in factories or other institutions in which the working condi- tions may fall far below the minimal standards set in developed countries. There is also concern that child labor reduces school enrollment, although the direction of causality is not always clear. Edmonds and Turk find that child labor in Vietnam declined in the 1990s, and they argue that this came about in large part as a result of in- creased economic growth. Yet they also find that child labor is still common in Vietnam, and the hours worked per week by some children are quite high. They also point out that girls are more likely to be child laborers than boys, and ethnic minority children are more likely to work than Kinh children. 18 Economic Growth, Poverty, and Household Welfare in Vietnam Finally, they find evidence that current regulations limiting child labor are not enforced. Chapter 14 concludes with some specific recommendations for govern- ment policy that, if implemented, would further reduce child labor. First, any policy changes should be made with the participation of the families who will be affected; outright banning of child labor could worsen the situation by making poor households even poorer. Second, to reduce child labor among girls, reductions in tuition--or even direct subsidies--could be offered for girls to encourage parents to send them to school. Third, special attention needs to be given to children of households that migrate to urban areas; poli- cies to stem such migration, such as excluding children of unregistered mi- grants from public schools, may increase child labor among migrant children. Fourth, proposals to increase the length of the school day should be consid- ered cautiously, because many children who work also go to school, and in- creasing the length of the school day may cause some of them to leave school. In chapter 15, Paul Glewwe and Phong Nguyen examine economic mo- bility in Vietnam. More specifically, they examine the extent to which house- holds change their position in the distribution of income over time. This issue is an important one because transitory poverty is probably less worri- some than chronic poverty. Similarly, for a given level of inequality at a point in time, increased economic mobility reduces long-run inequality. At first glance, Glewwe and Nguyen appear to have found a large amount of economic mobility in Vietnam. When the population is divided into quintiles (five equal groups, with the poorest 20 percent in the first group, the second poorest 20 percent in the second group, and so on), it ap- pears that many households were in a different quintile in 1998 than they were in 1993. Only 41 percent of the population remained in the same quin- tile in both years. About 40 percent moved up or down by one quintile, and 19 percent moved up or down by two or more quintiles. Yet these figures, which are based on expenditure data, ignore the fact that household consumption expenditures are almost certainly measured with error. This will lead to overestimation of the extent of economic mobil- ity. Glewwe and Nguyen use several methods to correct for measurement error and find that about one-half of the estimated mobility is simply mea- surement error, which implies that actual mobility is much lower than it initially appears to be. This implies that poverty is more of a permanent condition than a casual look at the data indicates. Finally, in chapter 16, Donald Cox examines transfers of income from one household to another. He finds that such transfers were quite common throughout the 1990s, as found in earlier work by Cox, Fetzer, and Jimenez (1998).Unlikealmostallotherdevelopingcountries,suchtransfersinVietnam primarily flow from the young to the old. These transfers can have signifi- cant effects on poverty and other socioeconomic outcomes, so it is important to understand how they work. Cox finds that interhousehold transfers flow from wealthier households to poorer ones and thus equalize income distribution. They also serve as an An Overview of Economic Growth and Household Welfare in Vietnam in the 1990s 19 important source of income for retired households, an essential function in Vietnam, where pensions are rare. Analysis of the panel data shows that many households changed roles in the 1990s--many recipients in the 1993 survey were donors in the 1998 survey and vice versa. Such changes in transfer flows appear to respond to changes in household circumstances and thus serve as a kind of informal (though hardly comprehensive) insur- ance mechanism. A final point is that about half of all households do not participate in transfers either as senders or as recipients. It is unclear whether these households can depend on transfers should they become poor or need money for some other reason. Conclusions and Suggestions for Future Research It is clear that Vietnam was one of the most successful countries in increasing economic growth and reducing poverty in the 1990s. This success raises the three questions posed at the beginning of this chapter: What accounts for Vietnam's astonishing success? What can Vietnam do to ensure continued success? And can other very poor countries achieve this same success by fol- lowing Vietnam's policies? Although it is almost impossible to provide com- plete answers to these questions, the research in this book provides some answers. The following paragraphs summarize those answers, after which issues for future research are raised. What accounts for Vietnam's astonishing success? Most of the credit for the dramatic decrease in poverty, as measured by per capita expenditures, must go to Vietnam's broad-based economic growth. In contrast, chapters 6 and 7 show that social assistance and antipoverty programs have had very little effect on poverty; indeed, the official list of poor communes is not par- ticularly accurate and excludes a large proportion of the poor. As explained in chapter 5, the key policies for raising incomes--and thus reducing poverty--in rural areas were the lifting of marketing and trade restrictions on rice and fertilizer. These policies increased the price of rice and reduced the price of fertilizer, which raised the incomes of most rural households, including many poor rural households. More generally, chapter 2 cites Vietnam's reduction of inflation and the government budget deficit, removal of barriers to trade and foreign investment, and financial liberaliza- tion as the main reasons for its increased economic growth. Yet economic growth may not explain all improvements in household welfare, nor does it guarantee improvements in every dimension of a na- tion's standard of living. Chapter 9 shows that increases in the expenditure levels of the poorest households had little effect on child mortality rates, and chapter 10 finds that growth in household expenditure levels had very little direct effect on child nutrition, although an indirect effect may have oper- ated through increased government expenditures on health services. Finally, the relationship between economic growth and school attainment is rather complex; in the late 1980s and early 1990s, secondary school enrollment rates dropped even though economic growth was quite high, and only in the 20 Economic Growth, Poverty, and Household Welfare in Vietnam mid- to late 1990s was continued economic growth accompanied by large increases in school enrollment rates. What can Vietnam do to ensure continued success? The chapters in this book provide extensive advice. Chapter 2 gives general advice on economic growth, warning the government and donor agencies that Vietnam's eco- nomic growth will decline if no further policy reforms are adopted. It argues that additional pro-growth policies are needed to strengthen property rights and government effectiveness, reduce regulatory burden and corruption, develop an efficient and modern financial system, improve the operation of labor markets, and reduce barriers to trade and foreign investment. For the agriculture sector, where most poor households are found, chapter 5 cau- tions that, unlike the past, future increases in rice prices may not have the same beneficial effect because many rural households are moving out of rice production and into other crops. On a more pessimistic note, chapter 4 argues that it is unlikely that non- farm household enterprises will play a decisive role in raising economic growth in the future, because their share of the total labor force slowly declined in the 1990s. Indeed, it appears that their main role is to serve as a temporary source of employment for workers who will eventually find wage work. Thus, policies to encourage these firms may be justified, but they are unlikely to have dramatic effects on Vietnam's overall economic performance. Economic growth will continue to play a critical role in reducing poverty, and there is also a role for government programs to play, a role for which there is much room for improvement. Chapters 6, 7, and 8 give specific sug- gestions on how to do so. Vietnam's social assistance programs should reduce their reliance on local resources to finance local programs, because poor communes have few resources to contribute to their own poverty reduction programs. Better monitoring is needed of the allocation within provinces and districts of central government funds, and local jurisdictions should have less discretion on the implementation of centrally mandated poverty reduction programs. Combining household survey data with recent census data can provide a much more accurate picture of which of Vietnam's 522 districts, and perhaps even its more than 10,000 communes, are the poor- est in the country. Finally, while many ethnic minority households, particu- larly the Khmer and those in the Northern Uplands, have shared in the benefitsofeconomicgrowth,otherethnicminorities,suchasthoseinthe Cen- tral Highlands, are being left behind. Special programs appear to be neces- sary for these groups, but more needs to be learned about the barriers that prevent them from participating in overall economic growth. Much more can also be done in the areas of health and education. The analysis of infant mortality in chapter 9 suggests that increasing the years of schooling of young women, improving sanitation, and increasing the proportion of births that occur in medical facilities (or at least are attended by trained medical personnel) can significantly reduce the infant mortality rate among the poor. To reduce child malnutrition, regression analysis in An Overview of Economic Growth and Household Welfare in Vietnam in the 1990s 21 chapter 10 suggests that providing commune health centers with sanitary toilets and ample supplies of oral rehydration salts could have substantial positive impacts on child health in Vietnam, but these results are tentative. Finally, chapter 11 concludes that, in the long run, increases in household incomes and education levels should reduce the worrisome reliance of Vietnamese households on nonprescription purchases of antibiotics. In the short to medium term, expansion of health insurance coverage should also reduce the high use of antibiotics. Useful policy advice was harder to come by in the area of education, as explained in chapters 12 and 13. Two current worries are that school days are very short (about three hours) and that many schools offer "extra classes" that are open only to parents who can afford to pay. Yet increasing the length of the school day may have the unintended consequence that some children who work while attending school may decide to drop out. Another issue is the value of vocational education, which does not appear to have any effect on workers' wages. Regression analysis also shows no discernible impact of providing lessons in ethnic languages on school enroll- ment or academic achievement, which raises doubts about the merits of such policies. Child labor is also a topic of great interest in Vietnam. Fortunately, it is decreasing--but it is still quite common. Chapter 14 argues that policies to reduce child labor must be chosen carefully. An outright ban could hurt poor households by depriving them of a source of income. One possibility to reduce child labor among girls is tuition reductions (or direct subsidies) to encourage parents to send their daughters to school. Finally, special atten- tion needs to be given to children of households that migrate to urban areas. Finally, consider the issue of inequality, which is a serious concern among Vietnam's policymakers. Chapter 3 argues that Vietnam is likely to experi- ence an increase in equality as the labor force continues the shift from self- employment in agriculture to nonagricultural wage work. One way to reduce inequality would be to relax restrictions on migration from rural to urban areas, because those restrictions almost certainly exacerbate inequality between the countryside and the cities. Chapter 15 points out that Vietnam's worries about increasing inequality should not be dismissed because there is a high degree of mobility--simple estimates of mobility greatly overestimate its true value. On a more positive note, chapter 16 points out that interhouse- hold transfers, which are very common in Vietnam, appear to equalize the distribution of income. The policies recommended to improve Vietnam's social assistance programs can reduce inequality at the lower end of the income distribution, especially the recommendation to reduce the reliance on local resources to finance local social assistance programs. A last recom- mendation regarding inequality in Vietnam is that policies are needed that raise the incomes of residents in rural areas; as seen earlier in this chapter, these areas had much lower growth than urban areas, and the increased gap between urban and rural areas accounts for almost all of the increase in inequality in Vietnam in the 1990s. 22 Economic Growth, Poverty, and Household Welfare in Vietnam Can other very poor countries achieve Vietnam's success by following its policies? This is the most difficult question to answer. David Dollar's analy- sis in chapter 2 is based on cross-country data, thus it should apply to other countries as much as it does to Vietnam. This suggests that Vietnam's poli- cies for stimulating macroeconomic growth should be successful elsewhere. Yet lessons from such analyses are, at best, averages across a large number of countries, which implies that Vietnam's policies may not be as effective in some countries. A more specific lesson for the few remaining countries that still pursue a planned economy, such as Cuba and the Democratic People's Republic of Korea, is that privatization should begin in agriculture. The two countries that followed this path, China and Vietnam, had a much more suc- cessful transition to a market economy than the countries of Eastern Europe and the former Soviet Union, most of which did not begin their privatization policies with the agricultural sector. A final potential lesson for other coun- tries is the role played by education. Relative to other poor countries, Vietnam has long had very high levels of education. It may be that its tran- sition to a market economy was smoothed, and rapid increases in inequality were avoided, by these high education levels. This suggests that other coun- tries should focus on raising education levels as early as possible. Although the research in this book answers many questions about Vietnam, many questions remain unanswered and thus more research is needed. Fortunately, the VLSS data used in almost all of the chapters of this book are available for use by researchers, and new household survey data are now being collected every two years, beginning in 2002, in the new Vietnam Household Living Standard Survey. Another development that augurs well for future research in Vietnam is that there are now many more well-trained Vietnamese researchers than there were in the 1990s. An indi- cator of this is the number of Vietnamese authors in this book. Specifically, in an earlier book based on the 1993 VLSS (Dollar, Glewwe, and Litvack 1998), none of the chapters had Vietnamese authors. In contrast, in this book, 5 of the 16 chapters have Vietnamese authors or coauthors. This demon- strates Vietnam's increased capacity to do research; this increased ability in Vietnam is not only an end in itself but will prove invaluable for designing policies and, ultimately, increasing the quality of life in Vietnam. What issues require immediate attention of Vietnamese and interna- tional researchers? In the area of economic growth, more detailed advice is needed on how to improve government effectiveness and reduce corrup- tion, as well as how to develop a well-functioning financial system. Another topic for further research is labor markets in Vietnam, such as the impact of migration barriers on inequality. The impact of government policies on non- farm household enterprises is also largely unknown; a better understanding of these firms would be useful, but such research is fraught with difficulties and thus will prove to be challenging. Research is also lacking on the pros- pects for income growth for farming households in Vietnam; per capita ex- penditures in the country have increased at a much higher rate in urban areas than in rural areas. An Overview of Economic Growth and Household Welfare in Vietnam in the 1990s 23 Future research on poverty reduction in Vietnam should investigate whether poorly targeted programs within provinces are due to weak imple- mentation capacity or to deliberate diversion of funds to less needy groups. More analysis is also needed on how best to monitor local implementation of centrally mandated programs. Additional study is needed to see whether combining household survey data and census data can be used to estimate accurately the incidence of poverty in Vietnam's more than 10,000 com- munes. Finally, special studies are needed to understand why ethnic mi- norities in the Central Highlands have not benefited from overall economic growth in Vietnam. Several issues in the health and education sectors require further research. First, the underlying causes for the apparent increase in the inequality in in- fant mortality rates across rich and poor households are still not well under- stood. Second, more analysis is needed on what factors other than higher household income led to better child nutrition outcomes; the recently com- pleted Vietnam Health Survey has the potential to shed much light on both of these questions. Third, a better understanding is needed of why households are dissatisfied with commune health centers, which are often bypassed by households that are able to visit public hospitals or private health care providers, and more research is needed on the role of private health care providers in expanding and increasing the quality of health care services. Fourth, the apparent inability of vocational education to raise workers' wages deserves further scrutiny. Fifth, much more research is needed on what de- termines learning in Vietnamese schools; regression analysis is hampered by many estimation problems, which suggests that randomized trials may be useful for assessing the impact of education policies on students' academic achievement. There are also several research priorities on other topics. The relationship between child labor and school attendance is complex and merits further study. Policy questions of particular interest are whether increasing the length of the school day will raise dropout rates and whether children of mi- grants face significant barriers to schooling that cause their parents to put them to work. More generally, much remains to be learned concerning the relatively high rates of child labor of girls and ethnic minority children. Eco- nomic mobility is not well understood, and its relationship with the long- run distribution of income is also unclear; panel data are needed that follow households over a long time to make further progress on this issue. Finally, much more remains to be learned about interhousehold transfers in Vietnam, particularly whether government programs to provide income support and pensions may crowd out informal systems that are already in place. Appendix 1A The Vietnam Living Standards Surveys Much of the analysis in this book is based on the 1993 and the 1998 Vietnam Living Standards Surveys (VLSSs). To avoid needless repetition in the dif- ferent chapters, this appendix briefly describes both surveys. 24 Economic Growth, Poverty, and Household Welfare in Vietnam The 1993 and 1998 VLSSs were conducted by Vietnam's General Statisti- cal Office (GSO), with financial assistance from the United Nations Devel- opment Programme and the Swedish International Development Agency and technical assistance from the World Bank. The 1993 VLSS covered 4,800 households, and the 1998 VLSS surveyed 6,000 households. Both surveys are nationally representative. About 4,300 households were covered in both surveys and thus constitute a large, nationally representative panel dataset. The 1993 VLSS used three questionnaires: one for households, one for rural communities, and one for collecting community price data. The house- hold questionnaire covered a wide variety of topics, including education, health, employment, migration, housing, fertility, agricultural activities, small household businesses, income and expenditures, and credit and sav- ings. The community questionnaires were completed only in rural areas (where about 80 percent of Vietnamese households live), and detailed price questionnaires were completed in both urban and rural areas. The 1998 VLSS used the same three questionnaires as in 1993 and two ad- ditional questionnaires. School questionnaires were completed for all pri- mary, lower secondary, and upper secondary schools in rural areas covered by the survey. The school questionnaire collected data on teachers, physical facilities, recent examination results, and finance. Commune health center questionnaires were also completed in rural areas. They collected data on clinical staff, medical equipment and services, supply of medicines, and fees charged. Many chapters in this book make use of a variable constructed from the VLSS data on total consumption expenditures. This variable was created by a team of GSO and World Bank staff. It includes explicit expenditures on food and nonfood items, the value of food produced and consumed at home, and the estimated annual rental value of durable goods and the household's dwelling. More detailed information on the survey and the constructed variables are given in World Bank (2000). This document and the questionnaires used in both surveys are found on the World Bank's Living Standards Measure- ment Study Web site: . That Web site also contains information on how to obtain access to the VLSS data. Future research will also be possible using data now being collected by the GSO. At the time of this writing (mid-2002), the GSO is conducting the Vietnam Household Living Standards Survey, which will have a much larger sample size but a smaller household questionnaire (and thus less information for each household). The GSO plans to implement this survey every 2 years for the next 10 years. Notes I would like to thank Bob Baulch, David Dollar, Jonathan Haughton, Dominique van de Walle, and two anonymous reviewers for comments on previous versions of this chapter. An Overview of Economic Growth and Household Welfare in Vietnam in the 1990s 25 1. The 1992­93 survey spanned a full year, starting in October 1992; the 1997­98 survey began in December 1997 and also lasted one year. For brevity's sake, refer- ence is made to the surveys as the 1993 VLSS and 1998 VLSS, respectively. 2. For more details on Vietnam's Doi Moi policies, see Litvack and Rondinelli (1999) and World Bank (1993, 2002a). 3. These data on inequality are from Glewwe, Gragnolati, and Zaman (2000), which provides a much more detailed analysis of inequality and poverty in Vietnam. Bibliography Cox, Donald, James Fetzer, and Emmanuel Jimenez. 1998. "Private Transfers in Vietnam." In D. Dollar, P. Glewwe, and J. Litvack, eds., Household Welfare and Vietnam's Transition. Washington, D.C.: World Bank. Dollar, David, and Jennie Litvack. 1998. "Macroeconomic Reform and Poverty Reduction in Vietnam." In D. Dollar, P. Glewwe, and J. Litvack, eds., Household Welfare and Vietnam's Transition. Washington, D.C.: World Bank. Dollar, David, Paul Glewwe, and Jennie Litvack, eds. 1998. Household Welfare and Vietnam's Transition. Washington, D.C.: World Bank. General Statistical Office (GSO). 2000. Statistical Yearbook 1999. Hanoi. _____. 2002. Statistical Yearbook 2001. Hanoi. Gertler, Paul, and Jennie Litvack. 1998. "Access to Health Care during Tran- sition: The Role of the Private Sector in Vietnam." In D. Dollar, P. Glewwe, and J. Litvack, eds., Household Welfare and Vietnam's Tran- sition. Washington, D.C.: World Bank. Glewwe, Paul, Michele Gragnolati, and Hassan Zaman. 2000. "Who Gained from Vietnam's Boom in the 1990s? An Analysis of Poverty and In- equality Trends." Policy Research Working Paper 2275. World Bank, Policy Research Department, Washington, D.C. Glewwe, Paul, and Hanan Jacoby. 1998. "School Enrollment and Comple- tion in Vietnam: An Investigation of Recent Trends." In D. Dollar, P. Glewwe, and J. Litvack, eds., Household Welfare and Vietnam's Tran- sition. Washington, D.C.: World Bank. International Monetary Fund. 2003. World Economic Outlook: Growth and Institutions. Washington, D.C. Kuznets, Simon. 1955. "Economic Growth and Income Inequality." American Economic Review 45(1): 1­28. Litvack, Jennie, and Dennis Rondinelli. 1999. Market Reform in Vietnam: Build- ing Institutions for Development. Westport, Conn.: Quorum Books. O'Driscoll, G. P., K. R. Holmes, and M. Kirkpatrick. 2000. Index of Economic Freedom. Washington, D.C.: The Heritage Foundation. 26 Economic Growth, Poverty, and Household Welfare in Vietnam Ponce, Ninez, Paul Gertler, and Paul Glewwe. 1998. "Will Vietnam Grow Out of Malnutrition?" In D. Dollar, P. Glewwe, and J. Litvack, eds., Household Welfare and Vietnam's Transition. Washington, D.C.: World Bank. United Nations Educational, Scientific, and Cultural Organization (UNESCO). Various years. Statistical Yearbook. Paris. Viet Nam Living Standards Survey (VNLSS). 1992­1993. Web site: www.worldbank.org/lsms/country/vn93/vn93bid.pdf. Vietnam Living Standards Survey (VLSS). 1997­98. Web site: www. worldbank.org/lsms/country/vn98/vn98bif.pdf. Vijverberg, Wim P. M. 1998. "Nonfarm Household Enterprises in Vietnam." In D. Dollar, P. Glewwe, and J. Litvack, eds., Household Welfare and Vietnam's Transition. Washington, D.C.: World Bank. World Bank, Asian Development Bank, and United Nations Development Programme. 2000. Vietnam Development Report 2001. Hanoi: World Bank. World Bank. 1987. World Development Report. Washington, D.C. _____. 1990. World Development Report: Poverty. Washington, D.C. _____. 1993. Viet Nam: Transition to the Market. Washington, D.C. _____. 1996a. Vietnam: Fiscal Decentralization and Delivery of Rural Services. Washington, D.C. _____. 1996b. World Development Report: From Plan to Market. Washington, D.C. _____. 1999. Vietnam: Attacking Poverty. Joint Report of the Government- Donor-NGO Working Group. Hanoi: World Bank. _____. 2000. "Vietnam Living Standards Survey (VLSS) 1997­98: Basic Infor- mation." World Bank, Development Research Group, Washington, D.C. _____. 2002a. Vietnam, Delivering on Its Promise: Development Report 2003. Hanoi. _____. 2002b. World Development Report: Building Institutions for Markets. Washington, D.C. Part I Vietnam's Economic Performance in the 1990s 2 Reform, Growth, and Poverty David Dollar Vietnam has been one of the fastest-growing economies in the world in the 1990s, and yet by many conventional measures it has economic policies that are mediocre at best. In the Heritage Foundation's Index of Economic Freedom 2000 (O'Driscoll, Holmes, and Kirkpatrick 2000), for example, Vietnam is ranked 144th out of 155 countries on a measure that seeks to capture the en- vironment for investment and growth.1 The objective of this chapter is to explain this apparent anomaly, and to do so in a way that provides useful guidance to policymakers in Vietnam on the institutional and policy reforms needed for sustained growth and poverty reduction. Between the 1980s and 1990s Vietnam carried out significant economic reforms, notably stabilization, introduction of positive real interest rates, trade liberalization, and initial property rights reform in agriculture. If these changes are related to the empirical growth literature, Vietnam's growth ac- celeration is seen to be about what would be predicted. Conditional conver- gence suggests that the country's high growth rate will decelerate unless further reforms are taken. The section titled "Determinants of Growth" provides a framework for the chapter by briefly reviewing modern growth theory and the empirical growth literature. The empirical results on important institutions and policies for growth, as well as the phenomenon of conditional convergence, are stressed. Policy improvements generally lead to growth accelerations, but without further reform the growth rate will tend to slow over time. "Experience with Reform" then examines a number of indicators of Vietnam's reform in the late 1980s and 1990s, notably the macroeconomic re- forms of stabilization, positive real interest rates, trade liberalization, and initial property rights reform in agriculture. These changes are related back to the empirical growth literature with an estimate of what growth effect 29 30 Economic Growth, Poverty, and Household Welfare in Vietnam Vietnam should have received from its reform compared with what actually transpired, to see if its performance really is an anomaly. The next section, "Vietnam Compared with Other Emerging Markets," looks at the level of institutional and policy development in Vietnam, com- paring it with other emerging market economies. While Vietnam's policies have improved, they have done so starting from a truly low base. Hence, it can be simultaneously true that Vietnam's policies have improved a great deal and yet are mediocre in comparative perspective. A comparison of gov- ernance indicators, financial sector issues, and the infrastructure of interna- tional integration reveals serious institutional weaknesses in Vietnam that need to be addressed if a high growth rate is to be sustained. A final section briefly sums up. Determinants of Growth There is a vast empirical literature that investigates the determinants of growth. Much of this was spurred by the endogenous growth theories of Romer (1986) and Lucas (1988). The new growth models emphasized the im- portance of creating a good environment for firms to innovate--either through research and development to generate truly new products or processes, or through transfer or imitation of advanced technologies from other countries, which is a type of innovation for the local economy. Researchers have looked at a wide range of different variables that may affect growth. This cross-country empirical literature needs to be approached with some caution. The number of countries in the world is not very large, so this work uses relatively small samples. Many of the variables that researchers have looked at are correlated among themselves; as a result it is difficult to precisely identify the effects of different policies. There are issues surrounding the way that the causality runs. Nevertheless, this empirical lit- erature is useful for summarizing important patterns in the growth data. Empirical studies of growth are based on the "standard" cross-country growth regression in equation 2.1: (2.1) yct = 0 + 1 · yc,t-k + 2 Xct + c + t + ct where yct is log-level of per capita gross domestic product (GDP) in country c at time t, yc,t-kis its lag k years ago, and Xct is a set of control variables measured as averages over the decade between t - k and t. Subtracting lagged income from both sides of the equation gives the more conventional formulation in which the dependent variable is growth, regressed on initial income and a set of control variables. The disturbance term in the regression consists of an unobserved country effect that is constant over time (c), an unobserved period effect that is common across countries (t), and a compo- nent that varies across both countries and years that is assumed to be uncorrelated over time (ct). Most of the early empirical studies considered growth over a very long period (k = 25 years or more), so that there is only one observation per country. As a result, all of the effects of interest are estimated using only the Reform, Growth, and Poverty 31 cross-country variation in the data. Some studies consider shorter periods, such as decades or quinquennia, and typically combine the cross-country and within-country variation in the data in an ad hoc manner. Caselli, Esquivel, and Lefort (1996) provide a useful critique of conventional panel growth econometrics and a proposed solution, which is to estimate equation 2.1 in differences, using appropriate lags of the right-hand-side variables as instruments. In particular, they advocate estimating the regres- sion as shown in equation 2.2: (2.2) yct - yc,t-k= 1 · (yc,t-k- yc,t-2k) + 2 (Xct - Xc,t ) + (t - t ) -k -k + (ct - c,t ). -k On the one hand, this is nothing more than a regression of growth on lagged growth and on changes in the set of explanatory variables. On the other hand, subtracting lagged growth from both sides of the equation reveals changes in growth from one decade to the next as a function of initial growth and changes in the explanatory variables. Much of the recent work on growth adopts this approach, which has the advantage of controlling for country-specific effects that do not change over time. Elaborations of these techniques involve jointly estimating a system of two equations, in levels (equation 2.1) and in differences (equation 2.2), and using lagged changes of endogenous variables as instruments for levels in the former (Arellano and Bover 1995). This approach can yield important efficiency gains (Blundell and Bond 1998). Five results are fairly robust in the empirical growth literature that draws on these techniques. First, Fischer (1993) finds that high inflation is bad for growth. This commonsense result is hard to dispute. To some ex- tent, inflation reflects shocks and other things beyond the government's control, but truly high inflation (above, say, 40 percent) obviously reflects monetary mismanagement. In addition, there is a clear negative relation- ship between government consumption and growth that was first noted by Easterly and Rebelo (1993). Some recurrent government expenditures are socially productive, but countries with very high government spending usually have inefficient bureaucracies and high levels of corruption. A number of studies--most recently Frankel and Romer (1999) and Dollar and Kraay (2001)--find that openness to trade and foreign direct invest- ment (FDI) accelerate growth. These latter findings are very much in the spirit of the new growth models, which emphasize the importance of the size of the market for creating a fine division of labor and stronger incen- tives to innovate. In addition to macroeconomic and trade policies, financial development is also a spur to growth (Levine, Loayza, and Beck 2000). After controlling for other variables, it has been found that countries that have more devel- oped stock markets or deeper banking systems, or both, tend to grow fast. Finally, measures of the strength of property rights, rule of law, or level of corruption are highly correlated with growth (Kaufmann, Kraay, and Zoido-Lobatón 1999; Knack and Keefer 1995). The existing measures of property rights or corruption come from surveys of private businesses and 32 Economic Growth, Poverty, and Household Welfare in Vietnam reflect the extent to which investors perceive there to be problems with ha- rassment, corruption, and inefficient regulation. There are many other variables that researchers have used in empirical studies of growth. There are some studies that link infrastructure deficien- cies in telecommunications or power to poor growth performance (for ex- ample, Easterly and Levine 1997). However, it can be inferred from reading the evidence that those variables are not that robust if there are also controls for all of the institutions and policies noted above. Infrastructure provision is the result of good public institutions and policies. While disputes remain about the importance of individual policies, in general there is broad agreement among economists that growth is promoted by the policy package of private property rights, sound rule of law, macroeconomic stability, government spending that is not exces- sive and is well focused on public goods, and openness to foreign trade and FDI. A final important result from the growth literature is "conditional con- vergence." Holding institutions and policies constant, there is a tendency for the growth rate to slow over time. This means that, cross-sectionally among a group of economies with similar institutions and policies, poorer economies will grow faster and hence "converge" on richer economies. This convergence can be seen, for example, among Organisation for Economic Co-operation and Development (OECD) countries. The ones that were rela- tively poor in 1960 (such as Japan and Italy) grew rapidly in the 1960s and 1970s, but then their growth rates slowed. The growth rate of the productiv- ity leader, the United States, has been stable at about 2 percent per capita an- nually. As a result of this convergence among OECD economies, the overall growth rate of the rich countries has slowed decade by decade and in the 1990s was at about 2 percent. A growing number of developing countries have acted on this growth- oriented agenda in the past 15 years, and in general they have seen good results from reform. For example, Dollar and Kraay (2001) identified a group of "globalizers" (the top one-third of developing countries in terms of increased participation in international trade over the past two decades). This group has had a particularly large increase in trade to GDP: 104 per- cent, compared with 71 percent for the rich countries. What is striking is that the remaining two-thirds of developing countries have actually had a decline in trade to GDP over this period. The globalizing group has also cut import tariffs significantly, 34 points on average, compared with 11 points for the nonglobalizers. The list of post-1980 globalizers includes some well- known reformers--Bangladesh, China, Hungary, India, Malaysia, Mexico, the Philippines, and Thailand. These countries have been moving on the broad policy agenda outlined above. These recent globalizers have experi- enced an acceleration of their growth rates, from 1.4 percent per year in the 1960s to 3.5 percent in the 1980s and 5 percent in the 1990s (figure 2.1), while the growth rates of rich countries slowed over this period. What about developing countries not in the "globalizing" group? They had a decline in Reform, Growth, and Poverty 33 Figure 2.1. Per Capita GDP Growth Rates, 1960s to 1990s Percent 6 5.0% 4 3.5% 2.9% 2 1.4% 0 1960s 1970s 1980s 1990s Source: World Bank 2000. the average growth rate from 3.3 percent per year in the 1970s to 0.8 percent in the 1980s and 1.4 percent in the 1990s. While there has been growing consensus over the past few years about what policies are good for growth, the old debate about growth and in- equality has simultaneously resurfaced. It is common to hear claims that this era of globalization is leading to mounting inequality within countries, so that growth is not benefiting the poor. Mazur (2000), for example, says, "Globalization has dramatically increased inequality between and within nations. . . ." Neither part of this claim is true. The globalizing developing countries grew at 5 percent in the 1990s and hence converged on the rich countries, which were growing at only 2.2 percent (figure 2.2). The develop- ing countries that are not embracing globalization are being left behind. What about the claim that globalization and growth in general lead to higher inequality within countries? To test whether growth is associated with higher inequality within countries (that is, whether growth is biased against the poor), Dollar and Kraay (2002) put together a large dataset on in- come inequality, compiled from a variety of existing sources (primarily the dataset constructed by Deininger and Squire [1996]) with several updates using more recently available data). The data, which cover 137 countries, consist of Gini coefficients for a large number of countries and years, and five points on the Lorenz curve for most of these country-year observations. As noted by these and other authors, there are substantial difficulties in comparing income distribution data across countries. Countries differ in the concept measured (income versus consumption), the measure of income 34 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 2.2. Convergence and Divergence in Per Capita GDP Growth Rates in the 1990s Percent 6 5.0% 4 2.2% 2 1.4% 0 Nonglobalizers Rich countries Post-1980 globalizers Source: World Bank 2001. (gross versus net), the unit of observation (individuals versus households), and the coverage of the survey (national versus subnational). Dollar and Kraay restrict attention to distribution data based on nationally representa- tive sources identified as high quality by Deininger and Squire (1996) and perform some simple adjustments to control for differences in the types of surveys. Dollar and Kraay use these data to try to understand what is happening to the income of the bottom 20 percent (that is, the lowest quintile) of the income distribution as globalization proceeds. There is a one-to-one rela- tionship between the growth rate of income of the poor and the growth rate of per capita income, but a great deal of variation around that average relationship (figure 2.3). In other words, percentage changes in incomes of the poor, on average, are equal to percentage changes in average incomes. A useful way of interpreting these results is to say that they are equivalent to the finding that changes in the distribution of income are not systematically associated with the growth rate. Can deviations around the one-to-one relationship, which reflect changes in inequality, be explained? The hypothesis that greater trade openness leads to growing household inequality is equivalent to the hypothesis that grow- ing openness leads to points "below the line" in figure 2.3: that is, that growth of the poor's income is less than proportionate to per capita GDP growth. Dollar and Kraay considered a variety of possible variables that might explain cross-country differences in the extent to which growth accrues to Reform, Growth, and Poverty 35 Figure 2.3. Economic Growth and Income of the Poor (percent) Growth in annual per capita income of the poor 20% 10% Growth in annual per 20% 10% 10% 20% capita income 10% y 1.17x 0.00 R2 0.52 20% Source: Dollar and Kraay 2002. those in the bottom quintile, but with little success. One of the variables considered was trade volumes, but Dollar and Kraay found no evidence whatsoever of a systematic relationship between changes in trade and changes in inequality. There is simply no association between changes in trade to GDP and changes in the Gini measure of inequality (figure 2.4). Cer- tainly trade and investment liberalization has distributional consequences-- that is, there are "winners" and "losers" in the short run. However, the losers do not come disproportionately from the poor. While it is heartening to note that this is true, nevertheless, it is a concern that some poor households are hurt in the short run by trade liberalization. It is thus important to com- plement open trade policies with effective social protection measures such as unemployment insurance and food-for-work schemes. (Closed economies obviously need safety nets also, because households are subject to shocks from business cycles, technological change, weather, and disease.) To the extent that trade openness raises national income, it strengthens the fiscal ability of a society to provide these safety nets. Increased trade generally goes hand in hand with more rapid growth and no systematic change in household income distribution; thus, increased trade generally goes hand in hand with improvements in the poor's well- being. Dollar and Kraay (2002) similarly examine whether or not other 36 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 2.4. Increased Trade and Changes in Inequality (percent) Change in Gini coefficient 15 10 5 Change in trade to GDP 0.4 0.2 0.2 0.4 5 10 Source: Dollar and Kraay 2002. institutions and policies that are good for growth tend to affect inequality. For trade openness, rule of law, and financial development, the distribution effects are all very small and not significantly different from zero. In the case of government consumption and inflation, there are more significant distributional effects. In general, high inflation and large consumption by the government are especially bad for the poor. Such policies create a poor environment for growth and tend to harm the poor disproportionately. Based on empirical evidence, the view that growth-enhancing policies, in- cluding integration with the global economy, do not work for the poor can be rejected. Experience with Reform Cross-country growth analysis is useful for summarizing what is true in general or on average. However, individual country experiences can vary a great deal from what would be expected or predicted from the cross- country regressions. This section looks at some indicators of Vietnam's reform between the late 1980s and the late 1990s, and asks whether the Reform, Growth, and Poverty 37 Figure 2.5. Indicators of Vietnam's Reforms: Mid-1980s to Late 1990s (percent) 1988 167% 1997 83% 1990s 1992 4.1 64% 1980 s 3.0 1999 27% 1997 1984 5% 8% Inflation ICRG rule Financial Trade to (annual rate) of law index development purchasing power parity GDP Source: World Bank 2003. country got the expected boost in growth or whether it is an anomaly in either direction (that is, whether there was higher or lower growth than would be predicted). There is a growing descriptive literature of Vietnam's economic reform (Dapice 1995; de Vylder and Fforde 1996; Dollar 1994; Le Dang Doanh 1995; Ljunggren 1993). It is widely recognized that stabilization from high inflation was one key aspect of the reform. The inflation rate declined from almost 170 percent a year in 1988 to 5 percent in 1997 (figure 2.5). This stabilization was the result of fiscal adjustment and monetary restraint in the early 1990s. Together with the stabilization there were some initial financial sector reforms, raising interest rates to positive real levels and introducing some element of competition into the system. A measure of financial development that has been used in the growth literature is commercial bank assets rela- tive to total bank assets. In Vietnam, this measure of financial development increased from 0.64 in 1992 to 0.83 in 1997 (unfortunately, the data before 1992 are not available). Liberalization of foreign trade and FDI has been an important part of Vietnam's reform. The trade system was highly restricted through the mid- 1980s. Reform has included dismantling of nontariff barriers and tariff reductions (Dollar and Ljunggren 1997). It is difficult to measure the extent of trade policy reform. One good indicator is the volume of trade in constant prices relative to purchasing power parity GDP. This ratio increased from 0.08 in 1989 to 0.27 in 1997. Today trade is closely related to FDI. Vietnam also liberalized its policies toward FDI. Flows of FDI averaged more than 5 per- cent of GDP in the second half of the 1990s, up from virtually zero in the 1980s. 38 Economic Growth, Poverty, and Household Welfare in Vietnam Some of the most important reforms in Vietnam involved strengthening of property rights. The initial land reform gave use rights over land to peasant families, and in practice there is quite an active market in land. Company law and FDI law improved the property rights over plant and equipment. It is difficult to measure this change in the extent of property rights. One indicator is the International Country Risk Guide (ICRG) rule of law index, with a scale from 1 (complete lack of property rights) to 6 (OECD standards). This measure shows Vietnam increasing from 3.0 to 4.1 between the 1980s and 1990s, one of the largest improvements recorded (though not quite as good as China's improvement from 3.0 to 4.5 over the same period). If the changes in figure 2.5 are taken as rough indicators of the major re- forms that have occurred in Vietnam, what would the effect of reform be in light of the empirical growth literature? In table 2.1, the parameter estimates are taken from a recent panel growth regression (Dollar and Kraay 2002) and applied to the changes in policies observed in Vietnam between the late 1980s and the late 1990s. The result is that trade liberalization is estimated to account for an increase in the growth rate of 0.1 percentage point; disinfla- tion, 1.9 percentage points; financial deepening, 1.0 percentage point; and Table 2.1. Estimated Growth Effect of Vietnam's Reforms Growth regression (2) Annual Change (1) Standard effect during reform (5) = Indicator Coefficient error (3) = (1)/7 (4) (3)×(4) Control variables Lagged income 0.67 0.17 Lagged inequality -0.09 0.06 Secondary education 0.10 0.06 Institutions/ policies Trade volume 0.05 0.07 0.007 0.19 0.001 Inflation -0.15 0.13 0.02 -0.93 0.019 Government consumption -0.97 0.42 0.14 0.009 0.000 Financial development 0.37 0.17 0.05 0.19 0.010 Rule of law 0.18 0.08 0.026 0.64 0.016 Note: The numbers in parenthesis are column numbers. The growth regression is taken from Dollar and Kraay (2002, table 6). The time periods used there are seven years, so the parameter estimates are divided by seven to give the annual effects. The changes in Vietnam are taken from figure 2.5. The change in inflation is the log difference. Reform, Growth, and Poverty 39 property rights reform, 1.6 percentage points.2 These are each independent effects: The total effect of these reforms would thus be an increase in the growth rate of nearly 5 percentage points. One should, of course, view these point estimates with a certain amount of skepticism. It is nevertheless interesting that the estimated effect of reforms from the cross-country growth literature--a large boost in growth--accords quite well with Vietnam's experience. For much of the 1980s, real per capita growth was zero; for the 1992­98 period, it was 5.4 percent. Thus, the actual increase in Vietnam's growth rate was quite similar to what would have been expected from its reforms based on the empirical growth literature. Vietnam has seen the expected results from economic reform in terms of growth. What about developments in inequality and poverty? Here, too, it turns out that Vietnam is typical. The last section noted that in general there is no strong relationship between the kinds of reforms that Vietnam has carried out and changes in household income inequality. The 1992­93 Vietnam Living Standards Survey (VLSS) found relatively low inequality in Vietnam (with a Gini coefficient of 0.33), and the follow-up VLSS in 1997­98 found only a slightly higher degree of inequality (Gini coefficient of 0.35).3 Thus, reforms in Vietnam have led to a dramatic increase in income of the poor. The poverty headcount rate (based on a 2,000-calorie poverty line) declined from 75 percent of the population in 1988 to 58 percent in 1993 and then to 37 percent in 1998 (figure 2.6). Poverty was halved in a decade. It is Figure 2.6. Effect of Economic Reform on Poverty, 1988­98 Poverty headcount (percent) 75% 75 58% 50 37% 25 0 1988 1993 1998 Source: World Bank 2000. 40 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 2.7. Poverty Reduction and Growth Rate in India, Vietnam, and China, 1992­98 Percent per year (1992­98) 10 9.2% 8.4% 8 7.5% 6.4% 6 5.4% 4.6% 4 2 0 India Vietnam China Growth rate Poverty reduction Source: World Bank 2001. interesting to put Vietnam's poverty reduction experience in perspective by comparing it with that of China and India. In both China and India, reform has coincided with mounting inequality, hampering the extent to which growth has translated into poverty reduction (figure 2.7). While Vietnam's growth rate has been distinctly lower than China's, its rate of poverty re- duction has been nearly the same.4 Thus, Vietnam's experience so far has been quite positive. There were large improvements in policies between the 1980s and the 1990s, an acceler- ation in growth, and a dramatic decline in poverty. The sectoral composition of this growth has been similar to that experienced in other rapidly growing East Asian economies. Because of land reform and price liberalization in agriculture, that sector grew well by historical standards between 1990 and 2000, at 4.3 percent per year. However, Vietnam is a densely populated country whose main comparative advantage lies in labor-intensive manu- factures. Over the same period, industry was the leading engine of growth, expanding at 10.9 percent per year. The industrial growth was spurred by the opening to the international market and led by products such as garments and footwear. The service sector also expanded rapidly from 1990 to 2000, at 7.5 percent per year. Although Vietnam has done well for a decade, conditional convergence is the fly in the ointment and cannot be disregarded. Vietnam started out in Reform, Growth, and Poverty 41 Figure 2.8. Conditional Convergence, 1995­2035 Growth rate (percent per year) 8 6 4 2 0 1995 2005 2015 2025 2035 Source: Dollar and Kraay 2001. the 1980s as an extremely poor country with extremely poor policies. In that context, its initial reforms were large improvements in policies, and, starting from an extremely low base, Vietnam was able to grow very rapidly. However, if policies remain the same, Vietnam's growth rate will slow as the country gets richer. Dollar and Kraay (2001) estimate a convergence coeffi- cient of 0.67; that is, a per capita growth rate of 6 percent in one decade will be 4 percent in the next, assuming policies remain the same (figure 2.8). The next section turns to the question of the level of Vietnam's policies compared with other emerging market economies. This comparison is a useful way of highlighting key areas in which Vietnam needs to move in its reform if it is to sustain a high growth rate. Vietnam Compared with Other Emerging Markets This section focuses on the level of policies in Vietnam in the late 1990s, comparing it with other emerging markets in the areas of property rights and governance, financial and labor markets, and openness to foreign trade and FDI. Property Rights and Governance Kaufman, Kraay, and Zoido-Lobatón (1999) combine dozens of indicators from 13 sources under five categories that measure broad dimensions of 42 Economic Growth, Poverty, and Household Welfare in Vietnam governance. The categories are: · Government effectiveness measures bureaucratic delays, the compe- tence of officials, the quality of public service delivery, and the inde- pendence of the civil service from political pressures. This grouping of indicators covers the elements needed for government to design and implement good policies. · Regulatory burden includes the number of regulations within a mar- ket, the number of markets that are regulated, competition policy measures, and price controls. This captures more of the outcomes of the policies and provides a sense of the extent to which the invest- ment climate is market friendly. · Rule of law captures the extent of crime, property rights, tax evasion, and the legal system's effectiveness. It indicates the enforceability of contracts and the predictability of rules. · Graft measures include the frequency and size of irregular payments. · Political instability and violence measures the incidence of coups d'état, assassinations, riots, and armed conflicts, and provides a measure of the likelihood of a violent overthrow of a governing party. The political instability measure is primarily concerned with the probability of violent shifts in political power rather than constitutionally sanctioned shifts in policy stance. The other four categories are highly corre- lated, with correlations of 0.69 to 0.93. These indicators capture different di- mensions of property rights: basic contract enforcement and the rule of law, political stability and freedom from violence, and freedom from corruption. At first glance, the link from the other two measures--government effec- tiveness and regulatory burden--to property rights may not be obvious. To- gether, though, they capture the extent to which there is an efficient division of labor between the public and private sectors, where an inefficient division leads to poor use of resources in both the public and private sectors. Figure 2.9 shows governance in Vietnam and four other Asian countries (China, India, Myanmar, and Thailand). For each indicator in the figure, the mean for all countries in the world is zero and the standard deviation is 1. Vietnam is thus perceived as having above-average political stability, but it is well below average on the other four dimensions of governance. In terms of basic property rights and rule of law, Myanmar is very poor, at about the level Vietnam was at 10 years ago. Vietnam is about one-half a standard deviation better than Myanmar, but significantly worse than China or India. Thailand, in turn, is significantly better than China or India. (Countries in the develop- ing world that score especially high on this measure are Chile and Singapore; in making comparisons, those countries are disregarded to concentrate on countries that are not too far from Vietnam's level of development.) In terms of regulatory framework for business and government effec- tiveness in providing services, Vietnam also lags significantly behind China and India. In terms of corruption, China, India, and Vietnam are all per- ceived to be about the same. Reform, Growth, and Poverty 43 Figure 2.9. Governance Pentagons: Vietnam, India, Thailand, Myanmar, and China Vietnam Rule of law 1 Regulatory 0 framework Corruption 1 Government Political effectiveness stability China India Rule of law Rule of law 1 1 Regulatory 0 Regulatory 0 framework Corruption framework Corruption 1 1 Government Political Government Political effectiveness stability effectiveness stability Myanmar Thailand Rule of law Rule of law 1 1 Regulatory 0 Regulatory 0 framework Corruption framework Corruption 1 1 Government Political Government Political effectiveness stability effectiveness stability Source: Kaufmann, Kraay, and Zoido-Lobatón 1999. Because these are indexes, the interpretation of these differences is com- plex. The measures of rule of law, regulatory burden, and government effectiveness are highly correlated. For all of them, Vietnam lags behind China by about one-half a standard deviation. In growth analysis, that large difference is "worth" about 1.5 percentage points in growth. Factor Markets Another useful source of comparative information is the Global Competitive- ness Report, which ranks a large number of countries in terms of different dimensions of competitiveness. These are subjective assessments of the international business community and should be treated with some caution. Still, it is interesting to see that Vietnam is ranked among the lowest of the countries rated (figure 2.10). Vietnam is seen to be far less competitive than China (which is in the middle of the rankings). Two areas that are singled 44 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 2.10. Global Competitiveness Report Rankings: Vietnam, China, and India (ranking out of 60 countries) Global Competitiveness Report rankings Overall Government Institutions Technology Labor Finance Management Infrastructure Openness 0 10 20 30 40 50 60 Best Poor Vietnam China India Source: World Economic Forum 2002. out as particular weaknesses in Vietnam relative to China are financial mar- kets and labor markets. Vietnam initiated reforms in its financial sector but then has done little in the past five years to follow up. The banking sector is still dominated by a few (and poorly managed) state banks. Much of the outstanding credit of these institutions is to state-owned enterprises (SOEs). Some of the SOEs, especially those working with foreign partners, are doing well, but many are in financial difficulty. Hence, the financial sector is saddled with a large number of nonperforming loans. This results in high spreads and poor ability to channel funds to the most productive investments. To the extent that all of this is explicitly or implicitly backed by the government, it also represents a contingent liability that weakens the government's fiscal position. Reform, Growth, and Poverty 45 A similar situation prevailed in Thailand before its 1997 financial crisis. The government's fiscal situation looked good on the surface; however, it was the guarantor of a risky and poorly managed financial system. A mod- est shock then precipitated a large financial and exchange rate crisis (Burnside, Eichenbaum, and Rebelo 2001). Vietnam has a relatively closed capital account, but it would be naive to think that this alone will insulate it from a potential financial or currency crisis. As Vietnam's economy develops and is more integrated with the global economy on the real side (trade and FDI), it becomes increasingly difficult to maintain capital controls. Thus, re- forming the financial system is a high priority, both to help create a good en- vironment for growth and to provide insurance against a financial or cur- rency crisis that would really set back the country's development. Openness Vietnam has been negotiating to join the World Trade Organization (WTO), and that process has revealed that the country still has significant barriers to trade. Following through on its plans to join the WTO will require address- ing these barriers. Participating more in the global economy has been an important part of Vietnam's success to date, but to deepen that participation will require not only trade reforms but also improvements in both the hard and soft infrastructure of trade: ports, customs administration, insurance, and finance. There are indications that Vietnam is lagging behind its com- petitors in these key areas. For example, there are good data available on shipping costs on all ship- ments into the United States. For shipments of garments in 1998, the ship- ping cost from Ho Chi Minh City to the west coast of the United States was nearly 10 percent of the value of the shipment, compared with 4 percent for similar shipments from Shanghai or Bangkok (figure 2.11). There are small variations in the distances from these different cities to the United States, but they cannot explain the variations in shipping costs (Ho Chi Minh City is actually closer than Bangkok to the United States). Rather, these differences result, to a large extent, from the level of port efficiency, which leads to de- lays and high costs in the case of Vietnam. In the case of garments, Vietnam has great potential as a producer and exporter, especially as it joins the WTO and the Multifibre Arrangement is phased out. The shipping cost difference between Ho Chi Minh City (nearly 10 percent) and Shanghai (4 percent) may not seem like much, but it means a great deal to a producer in a competitive industry in which the margins are very thin. If there are other bottlenecks in domestic transport (to get to the port) or in customs administration to get inputs into and outputs out of the country, then each of these will be an additional "tax," further eroding the potential profit. Thus, in comparing Vietnam with other emerging market economies, a number of deficiencies in the investment climate that will influence invest- ment and growth are revealed. The underlying weaknesses are institutional, 46 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 2.11. Maritime Transport to the United States (West Coast): Garments Costs as share of value of exports 0.10 0.08 0.06 0.04 0.02 0 Saigon Bombay Madras Shanghai Bangkok Kao Hsiung (6,996) (9,865) (9,001) (5,475) (7,470) (5,820) Source: Clark, Dollar, and Micco 2002. Figure 2.12. Foreign Direct Investment as a Share of GDP, 1998 Percent of GDP 10 8 6 4 2 0 Chile China India ThailandMalaysia Brazil Vietnam Mexico PhilippinesArgentina Indonesia Source: World Bank 2003. Reform, Growth, and Poverty 47 Figure 2.13. Foreign Direct Investment in Vietnam in the 1990s Percent of GDP 8 6 4 2 0 1990 1991 1992 1993 1994 1995 1996 1997 1998 Source: World Bank 2003. concerning the protection of property rights and efficient regulation of mar- kets, the supervision of the financial system, and the policy framework for in- frastructure development. Vietnam's reforms initially attracted a great deal of enthusiasm from foreign investors. In 1998, the level of FDI, relative to GDP, in Vietnam was comparable to other emerging markets, such as Brazil and China, though it was well behind Chile, Malaysia, or Thailand (figure 2.12). The time trend of FDI into Vietnam, however, raises some concerns. Vol- umes rose sharply from 1990 through 1995, but then leveled off and declined (figure 2.13). It might appear that the decline in 1998 was partly related to the overall East Asian financial crisis. However, FDI worldwide actually in- creased during the crisis years (while portfolio flows dropped sharply). Hence, in an overall improving global climate for FDI, Vietnam has been getting diminished attention from investors. Conclusions In the 1990s, Vietnam was one of the fastest-growing economies in the world, although the level of institutions and policies in Vietnam was mediocre com- pared with other emerging market economies. These two, apparently dis- cordant, facts can be reconciled within the framework of modern growth theory and evidence. The key to the apparent anomaly lies in Vietnam's ini- tial conditions in the mid-1980s. The country has a good location and good human resources. However, in the mid-1980s, it had very weak economic 48 Economic Growth, Poverty, and Household Welfare in Vietnam policies that translated into an extremely low per capita income. Starting from such a low base, a modest set of initial reforms will have a large impact and generate a very high growth rate (especially in a good location, such as Vietnam's). The initial reforms are quite feasible technically for a low-income country. They involve macroeconomic policy changes such as price liberalization, devaluation, trade liberalization, and interest rate increases. Vietnam also carried out land reforms in agriculture that were important but simple (such as returning land to peasant families). All of these reforms require political will, but they are not technically difficult. Hence, Vietnam received the kind of growth boost that is to be expected from an initial set of macroeconomic reforms in a low-income environment. But another well-established, empirical regularity of the growth literature is conditional convergence: Holding policies constant, there is a tendency for the initially high growth rate in a low-income country to slow as it develops. Thus, the key to sustained high growth over several decades is continual up- grading of economic institutions and policies: legal reform, regulatory im- provements, deepening of the financial system, and regulation that allows efficient infrastructure investment and thus reduces transactions costs. The highly successful economies--such as Taiwan, China--demonstrate this kind of continual improvement over several decades. The slowdown in FDI into Vietnam, and the fact that the country's growth rate remains well below that of neighboring China, suggest that Vietnam has not followed its initial reforms with sufficient institutional upgrading. These institutional dimensions are hard to measure, but there are various informative indicators. This chapter has presented a number of these indicators, painting a consistent picture of weaknesses in Vietnam's governance, financial sector, and transport infrastructure that identify im- portant areas in need of further reform. Without these reforms, Vietnam's growth rate will probably decelerate. A balanced assessment of Vietnam in the 1990s is that initial successful reforms produced good results and have led to an important decline in extreme poverty. However, there has been some tendency for the government to rest on its laurels in recent years, leading to a decline in FDI and growth. For Vietnam to keep up with other competitive economies in the region, the government will have to "get back on the reform horse." Notes This chapter was presented as a paper at the workshop on "Economic Growth and Household Welfare: Policy Lessons for Vietnam," Hanoi, May 16­18, 2001, and benefited from helpful comments from attendees and from Paul Glewwe and Nisha Agrawal. I am grateful to the Research Support Budget for financial assis- tance and to Ximena Clark and Pablo Zoido-Lobaton for excellent research assistance. 1. This index may not be an unbiased assessment of Vietnam's investment climate, but its low rating of Vietnam is mirrored in other indicators that are taken up Reform, Growth, and Poverty 49 in the section of this chapter titled "Vietnam Compared with Other Emerging Markets." 2. The growth regression used for this counterfactual estimate is in table 6 of Dollar and Kraay (2002). The rule of law index used in that regression averages the ICRG rule of law index with other, similar measures. By construction it has a stan- dard deviation of 1.0. Vietnam's improvement on the ICRG rule of law index be- tween the 1980s and the 1990s was 0.64 of a standard deviation. 3. The 1992­93 survey spanned a full year, starting in October 1992; the 1997­98 survey began in December 1997 and also lasted a year. For brevity's sake, reference is made to the surveys as the 1993 VLSS and 1998 VLSS, respectively. 4. This mapping from growth to poverty reduction depends not only on changes in the distribution of income, but also on the initial level of per capita income and the initial degree of inequality. Bibliography Arellano, M., and O. Bover. 1995. "Another Look at the Instrumental- Variable Estimation of Error-Components Models." Journal of Econometrics 68(1): 29­51. Blundell, Richard, and Stephen Bond. 1998. "Initial Conditions and Moment Restrictions in Dynamic Panel Data Models." Journal of Econometrics 87(1): 115­43. Burnside, Craig, Martin Eichenbaum, and Sergio Rebelo. 2001. "Prospective Deficits and the Asian Currency Crisis." Journal of Political Economy 109(6): 1155­97. Caselli, Francesco, Gerardo Esquivel, and Fernando Lefort. 1996. "Reopen- ing the Convergence Debate: A New Look at Cross-Country Growth Empirics." Journal of Economic Growth 1(1): 363­89. Clark, Ximena, David Dollar, and Alejandro Micco. 2002. "Maritime Trans- port Costs and Port Efficiency." Policy Research Working Paper 2781. World Bank, Development Research Group, Washington, D.C. Dapice, David, ed. 1995. In Search of the Dragon's Tail: Economic Reform in Vietnam. Cambridge, Mass.: Harvard Institute for International De- velopment/Harvard University Press. de Vylder, Stefan, and Adam Fforde. 1996. From Plan to Market: The Transition in Vietnam 1979­1994. Boulder, Colo.: Westview. Deininger, Klaus, and Lyn Squire. 1996. "A New Dataset Measuring Income Inequality." The World Bank Economic Review 10(3): 565­91. Dollar, David. 1994. "Macroeconomic Management and the Transition to the Market in Vietnam." Journal of Comparative Economics 18(3): 357­75. Dollar, David, and Paul Collier. 2001. Globalization, Growth, and Poverty. Washington, D.C.: World Bank and Oxford University Press. 50 Economic Growth, Poverty, and Household Welfare in Vietnam Dollar, David, and Aart Kraay. 2001. "Trade, Growth, and Poverty." World Bank Policy Research Working Paper 2615. World Bank, Policy Research Department, Washington, D.C. ------. 2002. "Growth Is Good for the Poor." Journal of Economic Growth 7(3): 195­225. Easterly, William, and Ross Levine. 1997. "Africa's Growth Tragedy: Policies and Ethnic Divisions." Quarterly Journal of Economics 112(4): 1203­50. Easterly, William, and Sergio T. Rebelo. 1993. "Fiscal Policy and Economic Growth: An Empirical Investigation." Journal of Monetary Economics 32(3): 417­58. Fischer, Stanley. 1993. "The Role of Macroeconomic Factors in Growth." Journal of Monetary Economics 32(3): 485­512. Frankel, Jeffrey A., and David Romer. 1999. "Does Trade Cause Growth?" The American Economic Review (June): 379­99. Kaufmann, Daniel, Aart Kraay, and Pablo Zoido-Lobatón. 1999. "Gover- nance Matters." Policy Research Working Paper 2196. World Bank, Policy Research Department, Washington, D.C. Knack, Stephen, and Phillip Keefer. 1995. "Institutions and Economic Per- formance: Cross-Country Tests Using Alternative Institutional Mea- sures." Economics and Politics 7(3): 207­27. Le Dang Doanh. 1995. "Doi Moi in Review: A Snapshot of 1994." Vietnam Economic Times (December/January): 18­19. Levine, Ross, Norman Loayza, and Thorsten Beck. 2000. "Financial Interme- diation and Growth: Causality and Causes." Journal of Monetary Eco- nomics 46(1): 31­77. Ljunggren, Börje, ed. 1993. The Challenge of Reform in Indochina. Cambridge, Mass.: Harvard Institute for International Development/Harvard University Press. Lucas, Robert. 1988. "On the Mechanics of Economic Development." Journal of Monetary Economics 22(1): 3­42. Mazur, Jay. 2000. "Labor's New Internationalism." Foreign Affairs (Janu- ary/February) 1: 79­94. O'Driscoll, G. P., K. R. Holmes, and M. Kirkpatrick. 2000. Index of Eco- nomic Freedom. Washington, D.C.: The Wall Street Journal/Heritage Foundation. Romer, Paul. 1986. "Increasing Returns and Long-Run Growth." Journal of Political Economy 94(5). Viet Nam Living Standards Survey (VNLSS). 1992­1993. Web site: www.worldbank.org/lsms/country/vn93/vn93bid.pdf. Reform, Growth, and Poverty 51 Vietnam Living Standards Survey (VLSS). 1997­98. Web site: www.world- bank.org/lsms/country/vn98/vn98bif.pdf. World Bank. 2000. "Vietnam: Development Report: Attacking Poverty-- Country Economic Memorandum." Economic Report, Poverty Reduction and Economic Management Sector Unit (EASPR), PREM Sector Department, East Asia, World Bank, Washington, D.C. ------. 2003. World Development Indicators. Washington, D.C.: World Bank. World Economic Forum. 2002. Global Competitiveness Report. New York: Oxford University Press. 3 The Wage Labor Market and Inequality in Vietnam John Luke Gallup Vietnam has had rapid economic growth since the implementation of the Doi Moi (renovation) policies in the late 1980s, despite a sometimes fitful market reform process. This fitfulness has been due to ideological doubts about moving toward a market-based economy, as well as concern about the social impact of a market-based transformation. The changes have indeed been dramatic. Output per person grew at an average rate of 5.5 percent per year from 1988 to 1998, raising the average level of output per person by 73 percent in a decade (see figure 3.1).1 This transformation has been of such magnitude that it has touched all facets of society and dramatically reduced poverty. Family farms and small household enterprises still dominate the income-generating activities of the population, and much of the initial change due to Doi Moi occurred in the small farm and small household en- terprise sectors. But wage employment is the future. Historically, the process of economic development has always caused a transition out of small farms and household enterprises into wage employment, as worker productivity increases and nonhousehold enterprises dominate the economy. Scrutiniz- ing the evolution of the labor market in the 1990s gives us clues about how economic development in Vietnam will continue to affect households and society in the coming decades. The economic transformation in Vietnam, despite its positive impact on poverty, could increase inequality. If that occurs, the labor market is likely to be the source of the disparity. The development of urban private enterprise could polarize workers between those in high-paying skilled jobs and others, often immigrants from the countryside, who are eligible only for low- skilled, low-paying jobs. If employment creation concentrates in the cities, it widens the gap between rural and urban dwellers. In a situation where there 53 54 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 3.1. Economic Growth in Vietnam GNP per capita (1995 US$) 350 300 250 200 150 100 50 0 1986 1988 1990 1992 1994 1996 1998 Source: World Bank 2000b. is a lack of opportunity for the poor and an increasing disparity of income, the economic transformation could contribute to many social problems. This chapter uses the two completed rounds of the Vietnam Living Stan- dards Survey (VLSS)2 to evaluate the contribution of wage employment to inequality and income growth over the period of rapid economic growth in the 1990s that followed market reforms. It will address the question: Has the expansion of wage employment in Vietnam exacerbated social inequalities, despite its contribution to income growth? If Vietnam is able to sustain its economic development, wage employment will become an increasingly im- portant source of household income as family farms and small household enterprises become less prevalent. Comparing the recent evolution of wage employment to farm and nonfarm self-employment provides clues as to how economic development will change Vietnamese society, in particular its impact on income inequality within and between communities. This chapter shows that standard methods for calculating income in- equality can be severely biased by measurement error when decomposing the contribution of different sectors, regions, or groups to overall inequality. An easily calculated, alternative, consistent method for decomposing income inequality is proposed. The first section reviews developments in the Vietnamese labor market in the 1990s, focusing on wage employment as opposed to self-employment. The following section explores the determinants of wages to find the indi- vidual and community characteristics that explain why different people are paid different wages. A presentation of standard inequality measures fol- lows. This section considers the important impact of measurement error on inequality statistics and proposes a new statistic that is not subject to the The Wage Labor Market and Inequality in Vietnam 55 biases of standard inequality measures. The section on Vietnamese wage inequality describes wage inequality in the country and how it has changed during the last decade. This is followed by a discussion of the contribution of wage employment to overall income inequality. Finally, the relationship between income sources and overall inequality is used to predict future changes in income inequality, followed by the conclusion. The Vietnamese Labor Market This section evaluates how the rapid economic changes in Vietnam have af- fected labor force participation, unemployment, sectoral shifts in employ- ment, and the growth and regional differences in wages. Labor Force Participation and Unemployment In Vietnam, a high percentage of the working-age population works. Even when housework is excluded (but work on the household farm or house- hold business is included), 81 percent of all Vietnamese women and 85 per- cent of men ages 16 to 60 were working in 1993 (see table 3.1).3 As income rose during the 1990s, participation rates rose by 2 percent for women and were unchanged for men. In rural villages, participation rates are even higher--83 percent for women and 87 percent for men in 1993. Rural participation rates jumped by 4 percent for women and stayed unchanged for men, so that both rural men and rural women had the same 87 percent participation rate in 1998. Participation rates were lower in urban areas than in rural areas in 1993, at 74 percent for urban women and 78 percent for urban men. Unlike rural Table 3.1. Labor Force Participation, Ages 16­60 (percent) Change, Population 1993 1998 1993­98 Female Rural 82.8 87.0 4.2 Urban 73.4 70.4 -3.4 Total 80.5 82.8 2.3 Male Rural 87.1 87.1 0.0 Urban 77.7 77.3 -0.4 Total 84.7 84.7 0.0 Note: Labor force participation is the percentage of the working age population (16­60) who were working in the previous seven days or who were looking for work. Source: Author's calculations from the 1993 and 1998 VLSSs. 56 Economic Growth, Poverty, and Household Welfare in Vietnam Table 3.2. Unemployment Rates (percent) Sector 1993 1998 Rural 0.5 0.4 Urban 1.6 1.5 Note: The unemployment rate is the percentage of people ages 16 to 60 in the labor force in the seven days before the survey who were not working and who were looking for work. Source: Author's calculations from the 1993 and 1998 VLSSs. women, urban women's participation fell substantially by 3 percent to 70 per- cent in 1998. Urban men's participation also fell slightly to 77 percent in 1998. Unemployment--people looking for work who do not have a job--is very low in Vietnam. In 1993, the unemployment rate was 0.5 percent in rural areas and 1.5 percent in urban areas (table 3.2). Unemployment fell by 0.1 of a percentage point from 1993 to 1998, an insignificant change. The virtual absence of unemployment is probably due to a combination of fac- tors, including the fact that Vietnam's income level is too low to permit people to be without work while job searching, and the wide availability of self-employment on a household farm or enterprise. The Composition of Employment Vietnam is still a highly agricultural country, with half of all workers work- ing on family farms, but the share of agricultural employment is shrinking. "Self-employment" is defined here in an unconventional way because of inconsistencies between the two rounds of the VLSS. The survey question asking whether the survey respondent was self-employed in his or her main job in the previous seven days was changed from the 1993 to the 1998 survey.4 Using this self-employment question results in dramatic, but spuri- ous, changes in type of self-employment between 1993 and 1998. To ensure comparability, tables 3.3, 3.4, and 3.5 use the questions about whether the re- spondent has worked in any job in the three sectors (wage employment, nonagricultural self-employment, and agricultural self-employment) dur- ing the previous week, so that the sector designations are not mutually exclusive. To make the sectors mutually exclusive, "wage employment" in- cludes anyone who has worked in wage employment, whether it was the person's main job or not. "Nonagricultural self-employment" excludes any- one who has participated in any wage employment, and "agricultural self- employment" excludes anyone with any other employment. This tends to exaggerate the number of wage employees and nonagricultural self- employed at the expense of the agricultural self-employed, but it should not bias the rate of change over time. The share of agricultural self-employment (family farms) fell from 52 per- cent in 1993 to 50 percent in 1998 (table 3.3). The 2 percentage point drop in The Wage Labor Market and Inequality in Vietnam 57 Table 3.3. Sectoral Composition of Employment (percent) Change, Employment sector 1993 1998 1993­98 Wage employment 23.9 24.9 1.0 Agricultural self-employment 51.6 49.7 -1.9 Nonagricultural self-employment 24.5 25.4 0.9 Note: Sectoral employment is the percentage of workers ages 16 to 60 employed in each of the three sectors in the previous seven days. See text for definition of sector. Source: Author's calculations from the 1993 and 1998 VLSSs. Table 3.4. Rural Composition of Employment, by Sex (percent) Change, Sector 1993 1998 1993­98 Men Wage employment 24.5 26.3 1.8 Agricultural self-employment 57.4 54.8 -2.6 Nonagricultural self-employment 18.0 18.9 0.9 Women Wage employment 13.0 12.7 -0.3 Agricultural self-employment 67.1 66.1 -1.0 Nonagricultural self-employment 20.0 21.2 1.2 Source: Author's calculations from the 1993 and 1998 VLSSs. Table 3.5. Urban Composition of Employment, by Sex (percent) Change, Sector 1993 1998 1993­98 Men Wage employment 50.9 53.0 2.1 Agricultural self-employment 12.9 8.2 -4.7 Nonagricultural self-employment 36.2 38.8 2.6 Women Wage employment 35.0 37.7 2.7 Agricultural self-employment 15.0 12.7 -2.3 Nonagricultural self-employment 50.0 49.6 -0.4 Source: Author's calculations from the 1993 and 1998 VLSSs. 58 Economic Growth, Poverty, and Household Welfare in Vietnam the employment share of agricultural self-employment was equally shared by increased wage employment and nonagricultural self-employment. Wage employment and nonagricultural self-employment each grew to 25 percent of total employment. Rural men's employment changed more decidedly than overall employ- ment toward the wage labor market. Of the 3 percentage point fall in rural men's agricultural self-employment from 1993 to 1998, 2 percentage points went into wage employment, but only 1 percentage point went into non- agricultural self-employment. Rural women's employment is the most highly agricultural, with two- thirds of employed women working on family farms. It also fell the least, by 1 percentage point. Rural women's wage employment also fell slightly, while nonagricultural self-employment expanded by 1 percentage point. The changes in the urban sectoral composition of employment were more dramatic (table 3.5). For urban men, agricultural self-employment still accounted for 13 percent of employment in 1993, but it had fallen 5 percent- age points by 1998. There was a roughly equal transfer of employment into nonagricultural self-employment and into wage employment, which made up 53 percent of all employment by 1998. Urban women saw the greatest increase in wage employment, almost 3 percentage points from 1993 to 1998. The expansion of the wage employ- ment came from a sharp reduction in agricultural self-employment and a small reduction in nonagricultural self-employment. Agricultural employment shrank for everyone between 1993 and 1998. It fell twice as much in urban areas as rural areas, and twice as much for men as for women. Where the agricultural workers went differed for men and women. Men's employment expanded roughly equally into wage employ- ment and nonagricultural self-employment in both urban and rural areas. Rural women's movement out of agricultural work was entirely absorbed by nonagricultural self-employment, while urban women's reduced participa- tion in agriculture went entirely into wage employment. This chapter will focus on wage employment of adults and its impact on inequality. Questions of agricultural and nonagricultural employment are left aside, but other chapters in this volume focus on agriculture (see chapter 5, by Dwayne Benjamin and Loren Brandt) and household enter- prise (see chapter 4, by Wim P. M. Vijverberg and Jonathan Haughton). Wage Growth As shown above, there was a steady expansion of wage employment in Vietnam in the 1990s for men and urban women. What happened to wages as this expansion occurred? Despite the increase in the supply of wage workers, wages in Vietnam grew extremely rapidly. Average hourly wages increased by 10.5 percent per year in real terms between the 1993 and the 1998 VLSSs (table3.6).Overthecourseofjustfiveyears,wagelevelsjumpedbytwo-thirds. Wage growth was considerably faster than the growth in household income The Wage Labor Market and Inequality in Vietnam 59 Table 3.6. Wage Levels and Growth, by Region Annual change, Percent of 1998 1993 1998 1993­98 Ho Chi Minh Region wage wage (percent) City wage Ho Chi Minh City 2.60 4.70 11.9 100.0 Hanoi 1.94 4.91 18.6 104.5 Medium urban 1.64 3.10 12.8 66.0 Small urban 1.81 2.91 9.5 61.9 Rural north 1.54 2.39 8.8 50.9 Rural central 1.50 2.41 9.5 51.2 Rural south 1.93 2.65 6.4 56.4 Total 1.85 3.13 10.5 66.6 Note: Wages are mean hourly compensation in thousand 1998 dong. Source: Author's calculations from the 1993 and 1998 VLSSs. per person in the same period, which grew at 8.7 percent per year, as well as outputperpersonintheeconomyasawhole,whichgrewat6percentperyear.5 Regional Wage Differences Wage growth and wage levels were distributed unevenly across different regions of the country, with a split between the two primary cities--Ho Chi Minh City and Hanoi--and the rest of the country (table 3.6).6 The two urban centers had higher wages than the rest of the country in 1993, and their wages grew much faster from 1993 to 1998. In 1998, average Ho Chi Minh City and Hanoi wages were more than 50 percent higher than in all the other regions. Ho Chi Minh City wages started out 37 percent higher than all other regions in 1993, and grew 1 per- centage point faster than overall wages from 1993 to 1998. Wages in Hanoi were not much higher than wage levels in other parts of the country in 1993, but grew by 19 percent per year to become the highest wages in the country. In 1998 wages in Hanoi were even 5 percent higher than they were in Ho Chi Minh City. Ho Chi Minh City and Hanoi together make up a significant part of Vietnam's wage labor market, accounting for 25 percent of all wage jobs in 1998, although they account for only 8 percent of the country's population (18 percent of all wage employment is in Ho Chi Minh City and 7 percent is in Hanoi).7 Outside Ho Chi Minh City and Hanoi, average wages were surprisingly similar across regions in 1993, and they had become even more similar by 1998. Regions with lower wages in 1993 are the ones that saw the largest wage increases in the 1993­98 period, and the regions with the highest wages initially tended to grow more slowly. Medium-size urban and rural central regions had the lowest average wages in 1993 and saw the largest 60 Economic Growth, Poverty, and Household Welfare in Vietnam Table 3.7. Skilled, Private, Nonagricultural Wages, by Region Annual change, Percent of 1998 1993 1998 1993­98 Ho Chi Minh Region wage wage (percent) City wage Ho Chi Minh City 2.34 5.57 17.3 100.0 Hanoi 1.50 5.75 26.8 103.1 Medium urban 1.53 3.16 14.5 56.7 Small urban 1.82 2.95 9.7 53.0 Rural north 1.90 2.58 6.2 46.4 Rural central 2.57 3.02 3.2 54.2 Rural south 2.23 3.62 9.7 65.1 Total 2.01 3.78 12.6 67.9 Note: Wages are mean hourly compensation in thousand 1998 dong. Source: Author's calculations from the 1993 and 1998 VLSSs. increase in wages by 1998. The rural south, which had the highest wages outside the two largest cities in 1993, saw the lowest increase in the follow- ing five years. With the exception of the two largest cities, labor markets seem to have been equilibrating over this period, with wages becoming more similar over the course of the decade. The divergence of the largest cities from the rest of the country could be due to the fact that the demographic characteristics of workers and the kinds of employment are different in different parts of the country. If the best- educated, most productive workers are drawn to the main cities, the wage disparity could be due to differences in worker characteristics rather than differences in wages for the same kinds of workers. Table 3.7 makes wage rates more comparable across regions by restricting wages of those workers with at least a lower secondary school education, working for a private nongovernmental or non-state-owned enterprise, and a nonagricultural business. Wages for skilled, private, nonagricultural work show even faster growth and a stronger convergence across regions over the 1993­98 period than overall wages, again with the exception of the two primary cities. Hanoi had the lowest level of skilled private wages of any region in 1993 but caught up so fast that it had the highest wages of all by 1998, with wage growth of 27 percent per year.8 When considering workers with high educa- tion and private, nonagricultural, employers, Hanoi and Ho Chi Minh City still have more than a 50 percent wage premium. Except for the rural south, the two largest cities have a 75 percent wage premium over the rest of the country for skilled private employment. Hours Worked At the same time that hourly wages grew extremely rapidly and employ- ment shifted toward the wage labor market, hours worked in wage labor also increased rapidly, all of which contributed to the large rise in wage income. The Wage Labor Market and Inequality in Vietnam 61 Table 3.8. Average Annual Hours Worked in Wage Employment Change, 1993­98 Region 1993 1998 (percent) Ho Chi Minh City 2,176 2,365 8.7 Hanoi 2,022 2,113 4.5 Medium urban 2,027 2,184 7.7 Small urban 1,816 2,169 19.4 Rural north 1,113 1,460 31.2 Rural central 1,321 1,576 19.3 Rural south 1,276 1,628 27.6 Total 1,572 1,862 18.4 Source: Author's calculations from the 1993 and 1998 VLSSs. On average, hours worked in wage employment increased by 18 percent from 1993 to 1998 (table 3.8). In 1993, there was a sharp difference between rural and urban wage employment in terms of total hours worked. Workers in medium-size and large cities worked, on average, about 50 percent more hours per year than rural wage laborers. Workers in Ho Chi Minh City worked the longest hours in 1993, averaging 2,176 hours per year. Assuming an eight-hour workday and a five-day workweek, this works out to 272 work days, or 54 workweeks.9 These are long hours by any standard. Despite this, by 1998, working hours in Ho Chi Minh City had grown by 9 percentage points. Ho Chi Minh City, Hanoi, and medium-size urban centers had the highest average working hours in 1993, but their increases in working hours were more modest than the increases in rural areas and small urban centers. Rural areas and small urban centers partially caught up with the working hours in the medium-size and large cities between 1993 and 1998, with very large increases in rural wage labor working hours. State-Owned Enterprise Employment The Vietnamese government has made plans for the reform of state-owned enterprises (SOEs) since the beginning of Doi Moi in the late 1980s. Despite plans for staff reductions, the political commitment to reform has not always been clear because SOE employees are an important political constituency for the government. All of Vietnam's 5,740 SOEs are scheduled to be priva- tized or restructured by 2005 (Belser and Rama 2001), so it is useful to review what was accomplished between 1993 and 1998. SOE personnel account for 5 percent of the labor force and more than 15 percent of wage employment. Surprisingly, SOE employment as a share of wage employment grew by almost 1 percent from 1993 to 1998 in the VLSS sample, from 15.9 to 16.7 percent (table 3.9). The estimates of SOE employ- ment in the VLSS should be reasonably accurate because there is a sample of 340 SOE employees in 1993 and 563 employees in 1998. 62 Economic Growth, Poverty, and Household Welfare in Vietnam Table 3.9. State-Owned Enterprise Employment SOE employment Change, 1993­98 measures 1993 1998 (percent) SOE share of wage employment (percent) 15.9 16.7 0.8 Average hourly wage in SOEs (thousand 1998 dong) 1.72 2.90 68.4 Share of SOE employees, female (percent) 7.6 7.4 -0.2 Share of all wage employees, female (percent) 43.0 39.8 -3.1 Source: Author's calculations from the 1993 and 1998 VLSSs. The increasing share of SOE employment in wage employment occurred while wage employment was growing as a fraction of total employment (table 3.3), while employment participation was growing (table 3.1), and while the total population was growing. If we apply the total population growth of 8.3 percent from 1993 to 1998 (World Bank 2002, p. 108) to the working population, that implies that SOE employment grew by 20.2 per- cent from 1993 to 1998, or 3.7 percent per year. Wage levels at SOEs mirrored the changes in other wage employment. SOE wages increased by 68 percent in real terms from 1993 to 1998, while all wages grew by 69 percent (table 3.9). SOE wages remained slightly lower than average wages in other jobs, at a steady 93 percent of average wages in both 1993 and 1998. Although SOE wages were typical of overall wages, the rapid increase in SOE wages in the 1990s is surprising during a period of planned retrenchments of SOEs. The VLSS data show that SOE employment is predominantly male. Only 7.6 percent of SOE workers were women in 1993, and this stayed essentially the same in 1998. The small participation of women in SOEs is untypical of overall wage employment, where 43 percent of workers were women in 1993, although this fell to 40 percent by 1998. Wage Employment Patterns In sum, wage employment shows remarkable growth in wage levels and hours worked, as well as a more modest increase in the fraction of workers employed outside the home. Hanoi and Ho Chi Minh City maintain large wage premia, probably because the residency permit restrictions in these cities exclude people who were not born in the city. SOE employment levels showed a surprising 20 percent increase between the surveys, and SOE wages kept up with the rapid increase in overall wages. The Wage Labor Market and Inequality in Vietnam 63 Determinants of Wages Which individual characteristics or employment characteristics determine how much an individual earns in the labor market? Previous work shows that wages are typically positively correlated with education levels and with work experience (at a decreasing rate).10 In other countries, wages are typi- cally negatively correlated with being female or being a member of an eth- nic minority.11 Wages also typically vary by regions within a country. In the case of Vietnam, where much of wage employment is still offered by the state and there is still a large amount of agricultural employment, indicators of private and nonagricultural employment also explain wage levels. Returns to Education and Experience The determinants of wages can be explored with a simple earnings equation: log(Wagei) = 0 + 1X1i + ··· + K XKi where Wagei is the wage of individual i, X1i, ... , XKi are the K correlates of wages (such as education, experience, and so on), and 0, 1, ... , K are the effects of the correlates on wages. Log(·) is the natural logarithm. With cer- tain assumptions, the coefficient on education (1) can be interpreted as the internal rate of return to an additional year of schooling (Berndt 1991, p. 162). The estimated effect of education and experience on wages is shown in table 3.10. The estimated rate of return to schooling in Vietnam in 1993 was quite low, just 2.9 percent. The rate almost doubled to 5 percent in 1998, but it is still very low compared with other developing countries. Psacharopoulos (1985, p. 588), for example, reports an average rate of return of 11 percent for Table 3.10. Wage Regressions: Estimated Effect of Education and Experience Difference, Independent variables 1993 1998 1998­93 Schooling (years) 0.029 0.050 0.021 (6.29)*** (14.61)*** (3.84)*** Experience (years) 0.033 0.025 -0.008 (5.42)*** (4.80)*** (0.93) Experience squared -0.001 -0.001 0.000 (5.37)*** (4.52)*** (0.66) Constant 7.269 7.757 0.488 (91.40)*** (128.23)*** (4.76)*** Observations 2,007 3,033 R2 0.04 0.08 ***Significant at 1 percent level. Note: Absolute value of t statistics in parentheses. Source: Author's calculations from the 1993 and 1998 VLSSs. 64 Economic Growth, Poverty, and Household Welfare in Vietnam Asia, 13 percent for Africa, and 14 percent for Latin America, averaging over many similar studies using this model of wage determinants. The increase in the rate of return to schooling from 1993 to 1998 of 2.1 percent is statisti- cally significant. The average rate of return to a year of education was quite low in the 1990s, but there was considerable variation in the return to different levels of schooling, at least for private employment, as shown in chapter 16 of this volume. Particularly for the small number of university graduates in the sample, the return to a year of university in private employment was negli- gible in 1993 but had become a very good investment in 1998. "Experience," which is actually years since completion of schooling, is strongly positively correlated with wages, but at a decreasing rate, as ex- pected, and its effect shows no sign of changing from 1993 to 1998. Table 3.11 shows a broader group of correlates of wages: being female, a member of a non-Chinese ethnic minority, having Chinese origins, working for a nonagricultural employer, working for a nongovernmental employer, and indicators of living in the two primary cities in Vietnam. The correlation of real wages with years of schooling is still strongly positive, though even lower with the inclusion of other correlates, and it still has a statistically sig- nificant increase from 1993 to 1998. Experience has a stable, positive correla- tion with wages. Women in Vietnam earn much less than men with the same observable characteristics, although the difference in earnings decreased between 1993 and 1998. In 1993, wages for women were 31 percent less than for their male counterparts, even after controlling for education and experience.12 The gap between men's and women's wages in Vietnam became smaller by 1998, when women's wages were 17 percent smaller. The wage gap between men and women halved between 1993 and 1998, a statistically significant change. Non-Chinese ethnic minorities do not show lower wages in 1993, but they do show a 10 percent lower wage in 1998. Ethnic Chinese in Vietnam had a 25 percent wage premium in 1993, but this all but disappeared by 1998. In both 1993 and 1998, nonagricultural employers paid higher wages, and in 1998, private employers paid a statistically significant higher wage than state employers. The regression in table 3.11 properly tests whether the two primary cities, Ho Chi Minh City and Hanoi, have significant wage premia, other things being equal. Residents of both cities earn much higher wages than residents of other regions with the same characteristics. Workers in Ho Chi Minh City earned a remarkable 80 percent higher wage than rural or small and medium-size urban area residents with the same education, experience, and so on in 1998. Hanoi residents earned 47 percent higher wages in 1998 than Vietnamese living outside the two largest cities. There is no sign that the wage premia of Ho Chi Minh City and Hanoi fell during the period of 1993 to 1998. The Wage Labor Market and Inequality in Vietnam 65 Table 3.11. Wage Regressions: Broader Group Correlates Difference, Independent variable 1993 1998 1998­93 Schooling (years) 0.019 0.035 0.016 (3.27)*** (8.01)*** (2.19)** Experience (years) 0.027 0.028 0.001 (4.72)*** (6.04)*** (0.15) Experience squared -0.001 -0.001 -0.000 (4.71)*** (5.55)*** (0.29) Female -0.370 -0.182 0.188 (10.69)*** (6.83)*** (4.27)*** Non-Chinese ethnic minority 0.030 -0.107 -0.137 (0.39) (2.05)** (1.58) Chinese origins 0.224 0.005 -0.220 (2.66)*** (0.07) (2.17)** Nonagricultural employment 0.126 0.289 0.163 (2.79)*** (6.42)*** (2.45)** Private (nongovernmental) employer 0.001 0.082 0.081 (0.01) (2.38)* (1.37) Ho Chi Minh City 0.609 0.589 -0.020 (11.19)*** (16.24)*** (0.32) Hanoi 0.296 0.384 0.088 (4.15)*** (6.58)*** (1.06) Constant 7.353 7.530 0.177 (71.91)*** (91.24)*** (1.24) Observations 2,007 3,033 R2 0.18 0.22 *Significant at 10 percent level. **Significant at 5 percent level. ***Significant at 1 percent level. Note: Absolute value of t statistics in parentheses. Source: Author's calculations from the 1993 and 1998 VLSSs. It should be noted that only 18 percent and 22 percent of the variation of wages in 1993 and 1998, respectively, were explained by the correlates in table 3.11 (as measured by the R2 statistics), so the largest part of wage vari- ation is due to other unidentified factors. Table 3.12 shows the effect of correlates on wage levels in 1998 separately by seven regions. The regional differences are strong. The rates of return to schooling are especially low in the rural central region, in small urban areas, and in the rural south, where the rate of return is not significantly different from zero. The rate of return to schooling in Hanoi and the rural north is more than double the level in the three lowest regions, approaching a respectable 8.5 percent in Hanoi and 6.9 percent in the surrounding rural north. The disadvantage of being female is similar around the country, except in the rural north and the rural central region, where women come closer to Rural south 0.015 (1.59) 0.035 (3.54)*** 0.001- (3.90)*** 0.249- (4.47)*** 0.202- (2.25)** 0.352 (5.12)*** 0.057 (0.78) 7.727 658 0.16 (51.21)*** Rural central 0.027 (2.54)** 0.020 (1.63) 0.000- (1.43) 0.066- (1.02) 0.067- (0.71) 0.300 (3.48)*** 0.406 (5.07)*** 7.384 433 0.15 (36.55)*** Rural north 0.069 (4.63)*** 0.026 (2.28)** 0.001- (2.44)** 0.134- (1.54) 0.138 (1.17) 0.896 (4.70)*** 0.260 (2.65)*** 6.312 270 0.23 (22.52)*** Small urban 0.032 (2.44)** 0.047 (3.15)*** 0.001- (2.72)*** 0.196- (2.84)*** 0.301- (1.16) 0.502 (3.02)*** 0.052- (0.46) 7.205 402 0.12 (25.19)*** 16 499 urban 0.045 Medium (3.86)*** 0.022 (1.79) 0.000- (1.06) 0.204- (3.19)*** 0.1 (0.51) 0.235 (1.74) 0.154- (1.98)** 7.774 0.13 (32.66)*** 1) Hanoi 0.085 (4.20)*** 0.029 (1.58) 0.001- (1.43) 0.284- (2.40)** 1.561 1.218 215 (15.18)*** (5.50)*** 0.014 (0.1 6.480 0.19 (12.30)*** 1998 Chi City VLSSs. 1)*** Ho 0.054 (6.1 0.019 (1.76) Minh 0.000- (1.25) 0.241- (3.99)*** 0.345- (3.36)*** 0.245 (0.87) 0.012 (0.15) 8.093 556 0.13 1998 (24.73)*** Region, and by 1993 the employer entheses. omfr level. level. par in Regressions, minority cent cent age ed employment per per 5 1 calculations W 's variables ethnic at at statisticst (years) (years) squar 3.12. (nongovernmental) Author Robust ce: bleaT **Significant ***Significant Note: Sour 2 Independent Schooling Experience Experience Female Non-Chinese Nonagricultural Private Constant Observations R 66 The Wage Labor Market and Inequality in Vietnam 67 obtaining their male counterparts' earnings. The disadvantage of being a woman is greatest in Hanoi, where average wages for women are 25 percent lower than wages for men with similar characteristics. The disadvantage of being a non-Chinese ethnic minority varies widely across regions.13 Only in Ho Chi Minh City and the rural south did ethnic mi- norities have statistically significantly lower wages in 1998--29 percent lower in Ho Chi Minh City and 18 percent lower in the rural south. In Hanoi, the wages of ethnic minorities were, on average, almost four times higher than the wages of Kinh and Chinese workers with the same observable characteristics. This may be due to a small sample effect, because this reflects just 16 persons out of the small Hanoi sample of 215 wage earners, but the coefficient remains significant and of similar size in a quantile regression (not shown). Nonagricultural employers paid higher wages in all the regions, but the effect was most pronounced in Hanoi, the rural north, and small urban areas. Private (nongovernmental) employers clearly paid a wage premium only in the rural north and the rural central region; this wage premium could be due to the traditionally strong communist roots in these two regions, which could motivate workers to take government jobs despite low government wages. The most notable results from exploring the determinants of wage levels are very high wage premia in Hanoi and Ho Chi Minh City, and low returns to schooling in Vietnam. With the same observable characteristics, workers in the two primary cities earned 50­80 percent higher wages than similar workers elsewhere in the country. The high wages in the primary cities create disparities with the countryside and drive up the cost of doing business in two crucial markets. The wage disparities are underpinned by the system of household registration, which prevents nonnatives from obtaining residency permission in the primary cities except for those obtaining gov- ernment jobs or (with difficulty) those sponsored by their employer. The distorting impact of this system is suggested by anecdotal evidence that res- idency papers for Hanoi and Ho Chi Minh City improve young people's marriageability and are even the basis for marriages of convenience. The wage premia in the primary cities reported here are for registered residents. The VLSS data are based on household registration records, so un- fortunately they leave out unregistered migrants to the primary cities. The unregistered migrants almost certainly find higher wages than they would have in the countryside, or they would not have moved to the city, but their wages are much lower than the wages of legal residents. The difficulty in ob- taining residency status in the primary cities has created a strong disparity between native residents and the class of illegal residents who are dispos- sessed by law. The unregistered migrants are not allowed to use government services, including education and health care, nor can they obtain jobs in most registered businesses. The labor market rate of return to schooling in Vietnam is quite low, though it seems to have improved during the 1990s. Figure 3.2 shows the strong correlation between estimated rates of return to schooling and average level of schooling across regions. It points to a vicious cycle in regions with 68 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 3.2. Levels of Schooling versus Labor Market Returns to Schooling by Region, 1998 Estimated rate of return to schooling (percent) 9 Hanoi 8 7 Rural north 6 Ho Chi Minh City 5 Medium urban 4 3 Small urban Rural central 2 Rural south 1 0 5 7 9 11 13 Average years of schooling Source: Author's calculations from 1998 VLSS data. low education, because the rate of return is lowest in those parts of the coun- try with the lowest education levels, and the rate of return to education is highest in the regions with highest education. This pattern may be explained by poor education quality in regions with low educational attainment. There is also a significant male-female wage gap, although it diminished during the 1990s. Methods for Measuring Inequality The previous sections explored the remarkable growth of Vietnamese wages in the 1990s. The rest of the chapter will examine the impact of wage em- ployment on inequality in Vietnam, starting with a consideration of a num- ber of different measures of inequality that have desirable properties. Inequality measures are chosen according to three criteria: (a) they satisfy the principle of transfers, (b) they are additively decomposable across sub- groups, and (c) they can handle negative income values.14 The principle of transfers is the intuitively appealing requirement that a transfer of income from a poorer to a richer person will increase the measure of inequality, as long as the transfer is not so large as to reverse the two persons' relative po- sitions. All the commonly used inequality measures satisfy the principle of transfers. In particular, the Gini coefficient, the generalized entropy measures, and the Atkinson inequality measures all adhere to the principle of transfers. The Wage Labor Market and Inequality in Vietnam 69 The decomposition of inequality across a set of groups is useful for assessing how much of total inequality is due to differences within the groups and how much is due to differences between the groups. These groups can be any mutually exclusive subgrouping of the population, such as region of residence or a household characteristic. Among inequality mea- sures with standard characteristics, only the generalized entropy measures of inequality are additively decomposable, where the inequality within sub- groups and the inequality between subgroups sum to total inequality (Shorrocks 1984). The decomposition of inequality across the source of income is a differ- ent problem because the sources are not mutually exclusive categories. Many households have income sources from more than one sector, for ex- ample, from farming and wage employment. The additive decomposition of inequality across sources or uses of income is possible for any inequality index, and Shorrocks (1982) shows that there is only one rule for decompos- ing the inequality that satisfies a small number of reasonable properties. Only two of the common inequality measures that satisfy the principle of transfers are well defined for negative income levels, such as occur when there are year-on-year losses to farm and enterprise self-employment: the Gini coefficient and one of the generalized entropy measures (I2, which is half the squared coefficient of variation). This study uses four measures of inequality: the Gini coefficient and three generalized entropy inequality measures (see appendix 3A for formal definitions of the inequality measures). The Gini coefficient is probably the most commonly used inequality measure, and it can be defined as a multi- ple of the covariance of individual income and the rank of individual in- come divided by average income. The Gini coefficient ranges between zero (perfect equality) and one (perfect inequality). The generalized entropy measures of inequality are designated I, where the more positive the parameter, the more sensitive the index is to differ- ences at the top of the income distribution rather than the bottom. This chap- ter uses I0 (also known as the mean logarithmic deviation), I1 (also known as the Theil index), and I2 (one-half the squared coefficient of variation). For mutually exclusive groups of people, a generalized entropy income inequality index for the whole population decomposes into a weighted sum of the inequality indexes of the groups that make up the whole. Generalized entropy indexes I can be written as the sum I = IW + IB of the total within-group inequality IW and between-group inequality IB.15 The for- mulas for IW and IB are in appendix 3A. The solution for decomposing the share of inequality from income sources is simpler and more elegant (see equation 3.1). For any J sources of income, overall income inequality can be decomposed into the inequality contributed by each source: J cov(yj, y) (3.1) I = I, j=1 var(y) 70 Economic Growth, Poverty, and Household Welfare in Vietnam where cov(·) is the sample covariance and var(·) is the sample variance. Be- cause the inequality index I appears on both sides of the equation, it implies that the shares of inequality sum to one, and that they are independent of any particular inequality index chosen. Our interest is the relative contribu- tion of each income source; thus, the actual inequality index can be dispensed with entirely. Note that if the covariance of the income from a par- ticular source is negatively correlated with total income, that income source makes a negative contribution to total inequality. In particular, income sources characteristic of poor households may contribute negatively to inequality--when these sources predominate in the household, total income is lower, resulting in a negative correlation. Measurement Bias Measurement error is a serious problem when studying inequality, even more serious than when studying other issues. Averages and growth rates of averages are typically unbiased in the presence of random measurement error, due to the law of large numbers. Inequality measures, in contrast, are typically biased and inconsistent in the presence of measurement error. Inequality is a measurement of variability, which is systematically increased by errors. Positive and negative errors balance out in a sample average, but both positive and negative errors add to the variance. More important for this study than biases in the estimated level of in- equality is that different income sources are likely to suffer from very differ- ent levels of measurement error. Wage incomes are usually well known by the survey respondents and others, making them easier to report accurately. Income from household farms and household enterprises is very difficult for the household and the researcher to calculate correctly. If household self- employment income has systematically large measurement error, household enterprises would appear (spuriously) to contribute a great deal to inequal- ity compared to wage employment, even when true income from both sources has the same inequality. Most inequality indexes depend on the variance of some transformation of income. The bias to inequality measures caused by measurement error is easily seen in the case of the I2 inequality measure, which is the sample vari- ance of income divided by its sample mean squared. If measured income yi is assumed to be equal to actual income yi plus a mean zero measurement error i, with variance , then: 2 yi = yi + i. Actual income yi has mean µ and variance 2 and is uncorrelated with the measurement error. Average measured income y¯ is unbiased, E(y¯) = E(y¯) = µ, but the sample variance of measured income is biased, var(y) = var(y) + var() = 2 + . This causes inequality measures such 2 The Wage Labor Market and Inequality in Vietnam 71 as I2 to be inconsistent when income is measured with error: I2 = var(y) 2y¯2 plim I2 = 2 + 2 2 2µ2 > 2µ2= plim I2. The larger the measurement error, the larger the bias in the inequality index.16 In a similar way, if the relative contribution to inequality is calculated from different income sources using Shorrocks' formula in equation 3.1, the calculated contribution to inequality of income from sources that are poorly measured would be greater than the calculated contribution of income sources measured more accurately, simply as a result of measurement error. This is shown formally in appendix 3B. Measurement Error in Income Measurement of household income is fraught with error, especially in low- income countries. Survey respondents may be reluctant to state their true in- comes; this is especially true where household farm and nonfarm enter- prises predominate, as in Vietnam, and households may not even know their precise incomes. Calculating net revenue for household enterprises re- quires aggregating large numbers of recurrent input and labor costs and product sales, as well as addressing intractable practical and conceptual problems. How does one account for home production, barter arrange- ments, and, especially, purchases of expensive capital equipment that will provide services over many years? The VLSS makes a valiant effort to mea- sure all of these items across hundreds of categories of inputs and outputs, but the overall aggregation of household net revenues nonetheless in- evitably contains substantial errors. There are quite extreme positive and negative outliers in farm and nonfarm household net revenues. There is no practical way to ensure that answers to all these questions about compo- nents of household net revenues add up to a consistent inventory of costs and revenues (although it would be interesting to confront the survey re- spondents with the calculated net result to see if they felt it corresponded to reality!). Researchers using the calculated household net income often resort to more or less arbitrary ways of trimming the outliers, but this does noth- ing to solve the problem of statistical bias caused by measurement error.17 Researchers usually work around this problem by ignoring the income data and using instead the household expenditure data that are more accu- rately measured. In fact, household expenditure--that is, consumption--is what should be measured, because it is a direct measure of the material well- being attained by the household. Household expenditure can be thought of as an estimate of "permanent income" (Friedman 1957). Household consumption decisions and well-being 72 Economic Growth, Poverty, and Household Welfare in Vietnam depend on the household's assessment of the smoothed expected income rather than on the fluctuating annual transitory income. Because we care about the inequality in household well-being rather than inequality of tran- sitory annual income, the measured annual income could be viewed as sub- ject to two kinds of measurement error. One is the mismeasurement of annual income due to imperfect collection of household information; the other is the mismeasurement of permanent income using accurately mea- sured annual income data. In a context of highly variable annual income from year to year, inequality in well-being using household income data will be overestimated when households are able to smooth consumption.18 The simple solution of using household expenditure data rather than income data for inequality calculations does not work when studying income sources. Expenditure does not tell us how wage employment, relative to other sources, contributes to household income. Using income data directly for in- vestigating the contribution of wages to inequality is especially problematic, because measurement errors are large for household self-employment earn- ings, but the errors in wage earnings data are probably much smaller. Survey respondents usually know precisely what they are paid, and they know that people around them already have a good idea of what they earn, so they have less reason to hide what they earn.19 This makes calculating the contribution to income inequality of wage employment versus household production mis- leading. Even with exactly the same distribution of income from wage em- ployment as from household self-employment, measurement errors in the self-employment data would spuriously show that self-employment con- tributesmuchmoretoincomeinequality,givingthefalseimpressionthatwage employment is an equalizing force. In addition, household farm and nonfarm production revenues are inherently variable because of natural weather and market fluctuations, whereas wage payments are relatively stable. The distrib- ution of well-being is what is important here; thus, estimating the inequality contribution of household production income versus wage employment income, even with perfectly measured annual income data, would also spuri- ously show that household production was disequalizing, even when perma- nent income actually had the same inequality across the income sources. Nevertheless, the contribution of wage employment versus household production to the distribution of household income can be consistently estimated by combining the data on source-specific income with total ex- penditure data. Because the measurement errors for income are generally uncorrelated with the measurement errors in expenditure, we can derive an estimator for which measurement errors cancel out as in averages and do not accumulate as in variances. The result is a consistent estimate of the con- tribution of income sources to inequality, as shown formally in appendix 3B. Vietnamese Wage Inequality The methods described in the previous section can be applied to look em- pirically at the relationship between wages and distribution in Vietnam using the two VLSSs, but first we look at average wages by quintile. The Wage Labor Market and Inequality in Vietnam 73 Table 3.13. Wages, by Quintile 1993 1998 Annual percent Quintiles wage wage change Poorest quintile 0.59 1.18 13.8 Middle three quintiles 1.60 2.58 9.6 Richest quintile 3.94 7.01 11.5 Total 1.87 3.13 10.3 Note: Wages are mean hourly wages in thousand 1998 dong. Source: Author's calculations from the 1993 and 1998 VLSSs. Table 3.14. Changes in Wages, by 1993 Quintile 1993 1998 Annual change Quintiles in 1993 wage wage (percent) Poorest quintile in 1993 0.61 2.07 24.4 Middle three quintiles in 1993 1.60 3.03 12.8 Richest quintile in 1993 3.82 4.80 4.6 Note: Wages are mean hourly wages in thousand 1998 dong. Source: Author's calculations from the 1993 and 1998 VLSSs. The distribution of wages equalized to some extent during the 1990s (see table 3.13). Wage earners can be divided into the "rich," the highest 20 percent of wages (the highest quintile), the "middle class" with the middle 60 percent of wages (the three middle quintiles), and the "poor" with the lowest 20 percent of wages (the lowest quintile). The wages of the poor grew at 14 percent per year, which was faster than the growth in the wages of the middle class (which was 10 percent per year) and the rich (12 percent per year). Note that households in the rich, middle, and poor categories in 1993 are not necessarily the same households in those categories in 1998. Since the VLSS reinterviewed the same households in 1998 that were in the original 1993 survey, evidence of what happened to those who were in the poorest or richest quintile in 1993 can be seen (table 3.14). There was a high degree of earnings mobility, both up and down.20 Of those who started out in the poorest quintile in 1993, only 34 percent were among the poorest in 1998. Because of this wage mobility, average wages of the poorest 20 per- cent in 1993 grew, on average, 24 percent per year. However, of those in the richest 20 percent in 1993, only 54 percent were still among the richest wage earners in 1998, and average wages of the richest in 1993 grew by only 5 per- cent per year. It is also possible to look back from the perspective of those who ended up in the richest or poorest quintile of earners in 1998 (see table 3.15). The picture is quite different from this angle. Those who ended up the poorest actually 74 Economic Growth, Poverty, and Household Welfare in Vietnam Table 3.15. Changes in Wages, by 1998 Quintile 1993 1998 Annual percent Quintiles in 1998 wage wage change Poorest quintile in 1998 1.25 1.20 -0.7 Middle three quintiles in 1998 1.78 2.66 8.1 Richest quintile in 1998 2.81 6.40 16.4 Note: Wages are mean hourly wages in thousand 1998 dong. Source: Author's calculations from the 1993 and 1998 VLSSs. saw their average wage decline by 1 percent in the previous six years, while those who were the richest saw their wages grow by 16 percent per year. How can the prospective view of the richest and poorest in 1993 look so different from the view in hindsight in 1998? The poor in 1993 were not the same group as the poor in 1998. Most of the poorest wage earners in 1993 did not stay poor, and most of the poor in 1998 did not start out poor in 1993. Only 34 percent of those who were the poorest in 1998 had started out poor in 1993. This great churning within the labor market showed that those who held a job through- out this period had many opportunities to succeed and to fail. There was a strong tendency of regression to the mean--receiving especially low or espe- cially high wages was usually a transitory phenomenon. The simple table of wages by lowest, middle, and highest quintiles in table 3.13 suggests that the distribution of wages became more equal over the 1990s. This is confirmed by the summary measures of wage inequality in tables 3.16 and 3.17. The statistics calculated are I0, I1, I2, and the Gini coeffi- cient, as defined in appendix 3A. Overall wage inequality in Vietnam indeed fell from 1993 to 1998 by all measures except I2, which most strongly weights high-wage earners. The overall pattern of decreasing wage inequality contrasts sharply with what happened in the medium-size and large cities. Hanoi, Ho Chi Minh City, and the second-tier medium-size urban centers had by far the lowest wage inequality in 1993. They are also the only regions of the country to see a rise in wage inequality over the 1990s. Wage inequality in Hanoi and Ho Chi Minh City rose quickly, while it stayed the same in medium-size urban areas and fell sharply in the rest of the country. It is likely that this unusual pattern in wage inequality, similar to the wage premia in the two largest cities, is influenced by the residency permit restrictions that are enforced there. In fact, wage inequality was probably worse in 1998 than these statis- tics show: The VLSS sample does not properly cover illegal migrants into the two largest cities, who lack residency permits, because the survey sam- ple was drawn from local residency records. The illegal migrants almost surely have among the lowest wages in the large cities. Another factor that could explain the low inequality in 1993 and the high inequality in 1998 for the medium-size and large cities is the prevalence of government employment there. In 1993, government employment was likely a large part of the wage labor market, which tends to equalize wages, The Wage Labor Market and Inequality in Vietnam 75 Table 3.16. Inequality Measures of 1993 Annual Wages, by Region Regions I0 I1 I2 Gini Ho Chi Minh City 0.204 0.178 0.221 0.312 Hanoi 0.188 0.182 0.210 0.332 Medium urban 0.239 0.236 0.318 0.365 Small urban 0.351 0.306 0.388 0.420 Rural north 0.310 0.261 0.301 0.393 Rural central 0.242 0.231 0.276 0.372 Rural south 0.322 0.263 0.303 0.394 Total within group 0.280 0.238 0.304 n.a. Between groups 0.036 0.039 0.044 n.a. Overall 0.316 0.277 0.348 0.403 n.a. Not applicable. Note: The Gini coefficient cannot be consistently decomposed into total within-group and between-group changes. Source: Author's calculations from the 1993 VLSS. Table 3.17. Inequality Measures of 1998 Annual Wages, by Region Regions I0 I1 I2 Gini Ho Chi Minh City 0.264 0.259 0.359 0.378 Hanoi 0.243 0.232 0.311 0.355 Medium urban 0.223 0.216 0.281 0.351 Small urban 0.241 0.243 0.356 0.362 Rural north 0.190 0.163 0.170 0.314 Rural central 0.158 0.144 0.162 0.290 Rural south 0.235 0.211 0.250 0.351 Total within group 0.222 0.218 0.329 n.a. Between group 0.039 0.042 0.046 n.a. Overall 0.261 0.260 0.375 0.377 n.a. Not applicable. Source: Author's calculations from the 1998 VLSS. because government wage levels tend to be similar across jobs. These cities are also the locations that saw the greatest structural transformation during the 1990s; thus, government employment played a much smaller role in 1998. The three generalized entropy inequality indexes--I0, I1, and I2--allow us to compare the within-region inequality versus the between-region inequality in wage rates. Despite the wage premia in the two largest cities, between-region inequality accounted for only between 12 percent and 15 percent of overall wage inequality in 1998, depending on the index (table 3.17). The balance of the inequality is due to variation within regions. Eliminating the cross-regional differences in wages would have some effect on wage inequality, but most of it would remain. 76 Economic Growth, Poverty, and Household Welfare in Vietnam Wages and Income Inequality If wage inequality fell between 1993 and 1998, what happened to the contri- bution of wages to overall income inequality? In this section, the impact of wages on income inequality is measured, using both the simple, but biased, method in equation 3.1 and the consistent method derived in appendix 3B. The inequality of household expenditure per person (as a measure of per- manent income) rose from 1993 to 1998 (table 3.18) at the same time that the inequality of wages was declining. Depending on the inequality measure, in- equality rose by 5 percent to 13 percent. Wages are the major source of income for only a minority of Vietnamese households (only 18 percent of households in 1998, accounting for 21 percent of household expenditure per person--see table 3.19). Fifty-four percent of households depend on farming for their main income source, and another 18 percent rely on income from a house- hold enterprise for their main income. The remaining 10 percent get most of their income from other sources in a given year, primarily overseas remit- tances and other gifts, with some interest and leasing income. Farming households have the lowest average per capita expenditure, only 60 percent of the income of predominantly wage employment house- holds. But the average expenditures of wage-earning households are not the Table 3.18. Inequality of Household Expenditure Per Capita Year I0 I1 I2 Gini 1993 0.188 0.211 0.308 0.339 1998 0.207 0.235 0.347 0.357 Source: Author's calculations from the 1998 VLSS. Table 3.19. Per Capita Household Expenditure, Broken Down by Main Income Source, 1998 Mean Ratio of income Share household source to overall of total expenditure average household household Income Population per capita expenditure expenditure source share (dong) per capita per capita Farming 0.541 2,157 0.722 0.391 Household enterprise 0.176 4,019 1.345 0.237 Wage employment 0.179 3,570 1.195 0.214 Other income sources 0.103 4,574 1.531 0.158 Source: Author's calculations from the 1998 VLSS. The Wage Labor Market and Inequality in Vietnam 77 Table 3.20. Decomposition of Household Income Inequality, by Source, 1993 No correction Consistent estimates Share of Share of inequality Share of inequality income Income source (percent) (percent) (percent) Farming 4.0 -5.2 36.3 Household enterprise 40.2 40.4 25.0 Wage employment 15.1 17.0 22.9 Other income 40.6 47.9 15.8 Source: Author's calculations from the 1993 VLSS. Table 3.21. Decomposition of Household Income Inequality, by Source, 1998 No correction Consistent estimates Share of Share of inequality Share of inequality income Income source (percent) (percent) (percent) Farming 15.6 -3.4 39.0 Household enterprise 39.3 38.9 24.1 Wage employment 17.0 29.3 19.3 Other income 28.1 35.2 17.6 Source: Author's calculations from the 1998 VLSS. highest. Households running their own businesses had an average per person expenditure 13 percent higher than wage-earning households, and households that receive their main incomes from other sources do even better, on average. As discussed above, household income data suffer from substantial mea- surement errors, especially for farms and household enterprises, because income must be netted out of the large number of costs and revenues. Mea- surement errors tend to be large, and less obvious to survey respondents, when income is the difference between much larger numbers, the costs and revenues. Tables 3.20 and 3.21 (in the columns labeled "No correction") present the biased calculation of the share of inequality attributed to each income source using the formula in equation 3.1. The "Consistent estimates" column in tables 3.20 and 3.21 presents the consistent calculation of inequality shares using the formula in appendix 3B. There are large differences between the uncorrected estimates and the consistent estimates of inequality shares. Farming's contribution to inequal- ity is overestimated by the biased calculation, and the contribution of wage 78 Economic Growth, Poverty, and Household Welfare in Vietnam employment and other income is underestimated. In both 1993 and 1998, the biased estimate shows that farming contributes to inequality, while the con- sistent estimate shows that farm income actually reduces income inequality. In 1998, the differences are especially dramatic. The biased calculation gives the impression that farming contributes about as much as wage employ- ment to total income inequality. In fact, the consistent estimates show that wage employment contributes 33 percentage points more than farming in- come of total inequality. Whereas wage employment accounts for 29 percent of total income inequality in 1998, farming income reduces inequality by 3 percent. The consistent estimates also show how important "other income"--largely overseas remittances--is to income inequality, contribut- ing almost one-half of all inequality in 1993, despite providing the smallest share of income, at 16 percent. The income share of other income rose to 18 percent in 1998, but its contribution to inequality fell to one-third of the total, presumably because remittances were being spread more equally among households in 1998. The share of inequality due to wage employment increased from 26 per- cent in 1993 to 33 percent in 1998, at the same time that the inequality of wages declined. This is possible because wage income was strongly nega- tively correlated with other income sources in 1993, but essentially uncorre- lated with other income sources in 1998.21 In other words, comparing 1998 with 1993, high-wage earners are more likely to be in households with high incomes from other sources, such as farming and household businesses. In fact, all income sources have become both less negatively correlated with other sources and less variable from 1993 to 1998, but the negative correla- tion of wage income fell more than for the other income sources. This pat- tern suggests a diversification of economic activity within the household that one would expect from the period of rapid development that Vietnam has experienced in the past decade. Another way of quantifying the contribution of wage employment to household inequality that proves useful for projecting future inequality is to divide households according to their primary income sources and decom- pose household expenditure inequality by type of household. Tables 3.22 and 3.23 show that households whose income source is predominantly farming have expenditure much more equally distributed than nonfarm households. Inequality of household expenditure among other kinds of nonfarm households is roughly similar, with the highest inequality among households whose income source is predominantly wage employment. Wage employment households had the highest inequality among household types in 1998, even though the previous analysis showed that wage em- ployment income is responsible for a smaller share of household inequality than enterprise income (table 3.21), because high-income wage employment households are more likely to have extra income from nonwage sources. Households that earn income predominantly from wage employment had higher inequality in 1998 than in 1993 as a result of earned income from other sources, not from high inequality of the wages themselves. The Wage Labor Market and Inequality in Vietnam 79 Table 3.22. Inequality Measures of Household Expenditure, Per Capita, by Main Income Source, 1993 Main income source I0 I1 I2 Gini Farming 0.115 0.124 0.161 0.263 Household enterprise 0.186 0.199 0.267 0.336 Wage employment 0.185 0.192 0.236 0.337 Other income sources 0.269 0.284 0.401 0.401 Within group 0.154 0.176 0.271 n.a. Between group 0.033 0.035 0.037 n.a. Total 0.188 0.211 0.308 0.339 n.a. Not applicable. Source: Author's calculations from the 1998 VLSS. Table 3.23. Inequality Measures of Household Expenditure, Per Capita, by Main Income Source, 1998 Main income source I0 I1 I2 Gini Farming 0.103 0.106 0.125 0.251 Household enterprise 0.198 0.213 0.286 0.349 Wage employment 0.250 0.267 0.375 0.390 Other income sources 0.220 0.234 0.308 0.367 Within group 0.158 0.186 0.297 n.a. Between group 0.048 0.048 0.049 n.a. Total 0.207 0.235 0.347 0.357 n.a. Not applicable. Source: Author's calculations from the 1998 VLSS. The consistent estimates of the contribution of each income source to total income inequality show that wage employment contributes almost one-third of income equality--about on par with household enterprise in- come and other income, even though it contributes only about 20 percent of total income. Farm income, instead of being a substantial contributor to in- equality as the uncorrected estimates make it appear, actually reduces household income inequality. The decomposition of household expenditure inequality by predominant income source also shows that predominantly wage employment households, household enterprise households, and other income households have similar levels of inequality, while farm households are substantially more equal. Projections of Future Inequality The previous section showed that predominantly wage income households are roughly similar to predominantly household enterprise and to "other income" households, both in terms of inequality and in terms of income 80 Economic Growth, Poverty, and Household Welfare in Vietnam levels. Farming households, however, are quite distinct. They have much less within-sector inequality, and they have a much lower average income level. One of the clearest historical patterns of economic development is the shrinking role of the agricultural sector as the economy grows, both as an em- ployer of labor and as a share of output. Farms now make up one-half of all households in Vietnam. Farm households are very different from other house- holds, they are a large part of the economy, and their relative number will diminish as the economy grows. This means that inequality can be expected to change in the near future. Inequality in Vietnam will rise over time as the proportion of equally distributed farming households falls in the economy. Empirical predictions of how fast inequality will change can be made as the Vietnamese economy develops by examining the relationship between the decline of agriculture and economic development in other countries around the world. The relationship between income growth and the share of labor in agriculture and the relationship between income growth and the ratio of agricultural income to total income are both well-established inter- national patterns. By estimating these relationships from historical data-- and assuming that inequality within the farming and nonfarming sectors remains unchanged--the future course of inequality can be predicted as a function of economic growth. The predictions of future changes in inequality in this section have a mechanical quality to them, but they result from a mechanism from which Vietnam will find it very difficult to escape. As long as agriculture grows more slowly than the rest of the economy, overall income inequality will worsen unless inequality in the nonagricultural part of the economy de- clines sharply over time. Over the 1990s, in fact, inequality within house- holds that earned income predominantly from household enterprise and wage employment increased (tables 3.22 and 3.23). There is no reason to be sanguine that inequality in nonagricultural households will decrease in the near future sufficiently to overcome the worsening inequality due to the gradual transition out of agriculture. The predictions are predicated on a continuation of Vietnam's recent economic growth. If growth occurs more slowly, the worsening inequality due to the transition out of relatively equal agriculture will occur more slowly than shown in these predictions. Statistics from the previous section confirm that farm households have more equally distributed income and lower income levels than nonfarm households, and that nonfarm households have similar income distribu- tions and levels across sectors. The generalized entropy indexes of inequal- ity in table 3.22 are one-half the level for farming than for other income sources. Predominantly farming households have 54 percent of the average income of nonfarming households (table 3.19). The inequality indexes for households whose main income sources are household enterprise, wage employment, or other income sources all have inequality indexes within 8 percent of each other (table 3.22) and average incomes within 30 percent of each other (table 3.19). The Wage Labor Market and Inequality in Vietnam 81 Figure 3.3. Agricultural Labor Force Compared with Income Level Share of labor force in agriculture (percent) 100 50 0 500 1,000 2,000 5,000 10,000 20,000 GDP per capita (1985 PPP$) Note: GDP = Gross domestic product. PPP = Purchasing power parity. Data points are country-specific least squares trend lines for 114 countries between 1960 and 1990. Source: World Bank 2000b. To estimate the relationship between income level (as measured by gross domestic product [GDP] per person) and share of labor force in agriculture, and between income level and agricultural output relative to total output, cross-country time-series data are used.22 Figure 3.3 shows the cross-country relationship between GDP per capita and the share of labor force in agricul- ture. The data come from 111 countries with an average of 30 observations per country. To emphasize the patterns over time, the data shown for each country in figure 3.3 are actually points on country-specific log linear trend lines to highlight the relationship in each country. Only 4 percent of the countries had positive income growth and an increasing share of the labor force in agriculture. Of the 15 percent of countries with an upward sloping trend to the share of labor in agriculture, 11 percent were countries with shrinking incomes, so the share of labor force in agriculture fell even though the economy was getting poorer. This suggests that the share of agriculture falls over time independently of income growth due to technical change. The estimated relationship between the share of labor force and GDP per capita using cross-country data shows a strong decline in the share of agri- cultural labor as the income level rises (table 3.24, second column). The regression includes country-specific constants and a time trend, which shows a clear but small decline in the share of agricultural labor over time of 0.5 percentage point per year. The simple regression is able to explain 73 per- cent of all the cross-country variation in the labor force share in agriculture. 82 Economic Growth, Poverty, and Household Welfare in Vietnam Table 3.24. Agriculture, in Relation to Income, across Countries Agricultural Agricultural output Independent variables labor sharea per capita ratiob GDP per capita (log of 1996 PPP, US$) -0.056 -0.200 (20.65)*** (26.33)*** Time (year 1960 = 1) -0.0045 0.0042 (57.59)*** (20.48)*** Constant 9.87 -6.18 (69.67)*** (16.41)*** Observations 3,399 2,926 Number of countries 114 106 R2 0.73 0.21 *** Significant at 1 percent level. Note: PPP = purchasing power parity. Absolute value of t statistics in parentheses. Both regressions include country-specific constants. a. Agricultural labor share is the fraction of agricultural workers in the total labor force. b. Agricultural output per capita ratio is the ratio of agricultural value added per worker to GDP per capita. Source: Author's calculations from data described in the text. The ratio of agricultural income per person to average income per person is also strongly correlated with the level of income per person and time (table 3.24, third column). Agricultural incomes are lower relative to non- agricultural incomes at higher GDP per capita levels, but agricultural in- comes become more similar to nonagricultural incomes over time, probably as a result of technical changes in agriculture. The net effect for a country growing at the rate of Vietnam in the 1990s is a decrease in agricultural in- comes relative to nonagricultural incomes. The agriculture income­share regression explains 21 percent of the cross-country variation. The inequality projection is based on the assumption that the Vietnamese economy continues to grow at the rate at which it has for the previous decade: GDP per capita grew 5.5 percent per year from 1988 to 1998. In other words, this is a projection of what would happen to income inequality if eco- nomic growth in Vietnam were to continue as it has in the recent past. The inequality projections are shown in table 3.25. The first column shows the actual figures for 1998, and the second and third columns show the projections for 2003 and 2008, respectively. The projection of the propor- tion of labor in agriculture depends on the first regression estimates from table 3.24, and it shows a decline of 3.8 percent every five years. The projec- tion of future agricultural GDP per worker depends on the second regres- sion estimates in table 3.24, and this shows that agricultural GDP per worker as a fraction of overall GDP per capita declines by 3.4 percent every five years. The within agriculture--and within nonagriculture--income inequali- ties are assumed to stay constant at the 1998 level and are measured by the I0 inequality index. The total within-inequality measure depends on the The Wage Labor Market and Inequality in Vietnam 83 Table 3.25. Inequality Projections, 2003 and 2008 1998 2003 2008 Indicator (actual) (projected) (projected) GDP per capitaa 325 428 563 GDP per capita growthb (percent) 5.5 5.5 5.5 Proportion of labor in agriculturec (percent) 54.1 50.3 46.5 Agricultural GDP per workerd 234 293 367 Nonagricultural GDP per workerd 432 564 734 Inequality of agricultural income (I0)e 0.103 0.103 0.103 Inequality of nonagricultural income (I0)e 0.223 0.223 0.223 Total within inequality (I0) 0.158 0.163 0.167 Inequality between agriculture and nonagriculture (I0) 0.047 0.052 0.058 Total income inequality (I0) 0.205 0.215 0.225 Change in inequality (since five years before) (percent) 9.6 4.9 4.4 a. For 1998, this is the purchasing power parity (PPP) gross national product (GNP) per capita estimate for Vietnam (World Bank 2000b). The 1998 level of GNP per capita is not used in the projections. b. The rate for Vietnam for 1988­98 is 5.5 percent real growth of GDP per capita (World Bank 2000b). c. The 1998 value was estimated by proportion of working-age individuals in predominantly agricultural households in the 1998 VLSS (table 3.18). d. The 1998 value was estimated from the ratio of average expenditures per person of predominantly agricultural and nonagricultural households (table 3.18). e. All values are set to the 1998 value estimated from the VLSS (table 3.22). Source: Author's calculations from data described in the text. proportion of households in agriculture (which is taken to be equal to the proportion of labor in agriculture). As the proportion of the labor force in agriculture declines going forth to 2003 and 2008, total within-inequality in- creases because agricultural households have lower inequality and their share of households is shrinking. Between-inequality also increases because average agricultural and nonagricultural incomes are diverging. Taking the sum of the within and be- tween measures of inequality, total inequality increases by 4.9 percent from 1998 to 2003, and 4.4 percent from 2003 to 2008. This compares with an actual increase to I0 of 9.6 percent from 1993 to 1998. This projection shows with numbers that the secular decline in agricul- ture as the economy grows will increase inequality because of a shift in the composition of households, because Vietnamese agricultural households 84 Economic Growth, Poverty, and Household Welfare in Vietnam have much less inequality than nonagricultural households. The second effect captured in the projections is that the relative incomes of the average agricultural household and the average nonagricultural household will con- tinue to diverge as the economy grows, causing the between-sector inequal- ity to grow. This second effect actually accounts for more than half of the change in inequality in the projections, as well as more than half of the actual change from 1993 to 1998. The predicted future inequality increases for 1998 to 2008 are only half of the actual increase for 1993 to 1998. This is primarily because agricultural in- comes have fallen behind nonagricultural incomes more rapidly from 1993 to 1998, compared with the change that had been predicted by the cross- country evidence. The regression estimates predict that the ratio of agricul- tural GDP per worker to overall GDP per capita will fall by 3.4 percent for the five-year periods 1998­2003 and 2003­08. The actual fall in expenditures per person of predominantly agricultural households relative to the average expenditures per person for all households fell by 5.3 percent for 1993­98. If the 1993­98 rate of decline in the ratio of agricultural incomes to total in- comes were to be extended to the future periods, projected total income inequality would increase by 8.9 percent from 1998 to 2003 and 8.4 percent from 2003 to 2008, which is quite similar to the 9.6 percent rise in inequality for 1993­98. The projected rise in inequality is not inevitable if nonagricultural income distribution equalizes in the future. During the 1993­98 period, however, the earnings of both predominantly wage-earning households and nonagricultural enterprise households became less equal, although the earnings of predominantly "other income" households became more equal (tables 3.22 and 3.23). A sharp improvement in the equality of nonagricul- tural earnings would be surprising. If the divergence of agricultural and nonagricultural incomes is slower than predicted in table 3.25, but income distribution within agriculture and nonagriculture remains unchanged, then income distribution will worsen more slowly than predicted in table 3.25. The rising inequality that looks likely for Vietnam is a consequence of the decliningroleofagricultureastheeconomydevelops,andtheunusuallyequal distribution of incomes among Vietnamese farm households. In low-income countries with unequally distributed agriculture, economic development im- proves overall income distribution as the share of agriculture declines. The increase in inequality due to the decline of agriculture can probably be delayed only by deliberately slowing economic growth and higher in- comes in Vietnam. A way out of rising inequality (other than a sharp im- provement in the equality of nonagricultural incomes) would be the rapid growth of incomes within agriculture while the equality of income distribu- tion among agricultural households is preserved. Sustained growth in agri- cultural incomes on par with income growth in the rest of the economy is a historical anomaly. The unfortunate consequence of Vietnam's escaping poverty will most likely be some increase in inequality, because, historically, The Wage Labor Market and Inequality in Vietnam 85 sustained development has almost always meant a shrinking role for agri- culture. The inequality in the nonagricultural sector in Vietnam is not par- ticularly high by international standards, however, and, if unchanged, it provides the upper limit for the effect of the declining share of agriculture on inequality. Conclusion The labor market in Vietnam saw very rapid change in the 1990s. Average real hourly wages grew by 10.5 percent per year between 1993 and 1998, faster than income per capita in the economy. At the same time, there was a substantial increase in hours worked, especially in rural areas, and a gradual increase in the share of the labor force in wage employment. SOE employment showed a marked rise of 20 percent from 1993 to 1998 at a time when the government was planning to restructure state firms, which are generally perceived to be overstaffed (Belser and Rama 2001). SOE em- ployees also fully shared in the rapid wage increases of the mid-1990s. The rate of return to schooling is very low in Vietnam, although it in- creased from 2 percent in 1993 to 4­5 percent in 1998. The lowest rates of re- turn to schooling are in the regions with the lowest education levels, and the highest rates of return are in the regions with the highest levels of schooling. Women face significant wage discrimination after controlling for their schooling and work experience, but the estimated wage gap fell by half over the period. After controlling for worker characteristics, workers in Hanoi and Ho Chi Minh City receive a very large wage premium over the rest of the coun- try, with average wages 50­80 percent higher than the other regions in the country. The primary-city wage premium suggests that the residency permit restrictions contribute to inequality between the two largest cities and the rest of the country. Qualitative evidence also suggests that residency permits contributed to a worsening of inequality within Ho Chi Minh City. Although households without residency permits are excluded from the VLSS sample design, participatory poverty assessment interviews indicate that poor, un- registered residents of Ho Chi Minh City perceived no improvement in their standard of living in the 1990s, unlike poor residents of other parts of the country (World Bank and Department for International Development 1999). This suggests that the residency permit requirement for formal sector jobs has kept all the benefits of rapidly rising wage levels from unregistered res- idents of Vietnam's largest city. In China, higher wages due to restrictions on urban residency have been a source of the sharp rise in income inequality between rural and urban areas (Yang 1999). Even though within-region in- come differences in China, as in Vietnam, are larger than differences across regions, Chinese rural-urban income disparities have been an important source of rural discontent. Inequality of wages fell modestly in the 1990s despite the rapid growth of wages. However, households with high wage income were more likely to 86 Economic Growth, Poverty, and Household Welfare in Vietnam have high incomes from other sources, so in this sense, wage employment has contributed to inequality. Agricultural households, for instance, are less likely to be engaged in wage employment. This effect will diminish as higher proportions of the population are engaged in wage employment. A new method for consistent decomposition of inequality by income source shows that, contrary to the results of uncorrected methods, wage em- ployment contributes a roughly similar amount to overall income inequality as other nonagricultural income sources (primarily household enterprise and remittances). Agricultural income actually reduces overall income in- equality because inequality between agricultural households is much lower than inequality between nonagricultural households, and agricultural in- come has a low correlation with other income sources. The much lower inequality and income level in agriculture allows us to predict future inequality change in Vietnam. A declining share of agriculture as the economy grows in Vietnam will raise income inequality unless within-sector inequality in the nonagricultural sectors falls substantially. From 1993 to 1998, the within-sector nonagricultural income inequality rose somewhat. If within-sector inequality does not change, the declining share of agriculture will increase inequality by 5­10 percent each five years for the next decade, after an increase of 9.4 percent from 1993 to 1998. This rising inequality due to the shrinking share of agriculture will be difficult to avoid without giving up economic growth and rapid poverty re- duction in Vietnam. Keeping a large proportion of the Vietnamese popula- tion on household farms would keep inequality from continuing to rise over time, but it would also keep the majority of Vietnamese at very low incomes because there are not good prospects for substantial rises in farm self- employment income without a major movement of labor out of farming. Im- provement of off-farm employment was a particular priority of the poor in a recent participatory poverty assessment in Vietnam (World Bank and Department for International Development 1999), indicating that the poor themselves would not be happy with efforts to keep them on the farm. The most notable results of this overview of the Vietnamese labor market in the 1990s are that Hanoi and Ho Chi Minh City have very large wage premia; SOE employment rose substantially; labor market rates of return to education are very low, especially in the regions with the lowest education levels; and inequality will continue to rise modestly as a result of the com- positional shift of the economy away from agriculture. Appendix 3A Inequality Measures The Gini coefficient can be defined as a multiple of the covariance of indi- vidual income and the rank of individual income divided by average income (Pyatt, Chen, and Fei 1980): G = 2 cov(y, r) Ny¯ The Wage Labor Market and Inequality in Vietnam 87 where y is a vector of individual incomes, yi, r is a vector of the ranks of individuals i when the population is ordered by increasing income, cov(·) is the sample covariance, and y¯ is average income. The generalized entropy measures of inequality have the form (Sen 1997, p. 140): N I = 1 1 yi (1 - ) N 1 - , = 0, = 1 i=1 y¯ with limit cases N I1 = 1 yi yi ln N i=1 y¯ y¯ and N I0 = 1 y¯ ln . N i=1 yi The I2 index simplifies to I2 = var(y) 2y¯2 which is one-half the squared coefficient of variation. For mutually exclusive groups, the overall generalized entropy indexes decompose into a weighted sum of the inequality indexes of the groups that make up the whole. Assume there are several groups k with population Nk, average group income y¯k, and within-group inequality index Ik. Then a generalized entropy index I can be written as the sum I = IW + IB of the total within-group inequality IW and between-group inequality IB (Shorrocks 1984). IW = Nk y¯k Ik N y¯ k is a weighted sum of the within-group indices. Between-group inequality IB has the form of I with y¯k substituted for yi. Appendix 3B Inconsistency of Inequality Shares in the Presence of Income Measurement Error For simplicity, take the case where income comes from only two possible sources: wage employment (w) and self-employment (s). Self-employment income is observed with measurement error, and wages are observed with- out error (relaxed below). Total income for person i is yi = ywi + ysi , where ywi is wage income and ysi is income from self-employment. Observed 88 Economic Growth, Poverty, and Household Welfare in Vietnam income for person i is yi = ywi + ysi , where ysi is measured income from self-employment. ysi = ysi + i where i is the measurement error. Actual in- comes have a constant mean and variance and may be correlated across source: E(ywi) = µw; E(ysi) = µs 2(ywi) = w;2(ysi) = s 2 2 (ywi ysi) = ws. Self-employment measurement error has a zero mean, constant variance, and is uncorrelated with incomes or across time: E(i) = 0; 2(i) = 2 (i ysi) = (i ywi) = 0 (ij) = 0i = j . The estimated contribution of wage employment income to inequality rela- tive to self-employment income from Shorrocks' equation 3.1 in the text is: cov(yw, y) cov(ys , y) cov(yw, y) var(y) var(y) = cov(ys , y) . does not provide a consistent estimate of the ratio of income source con- tributions to inequality, : cov(yw, y) 2 2 (3.B1) plim = plim w + s w w + s w cov(ys , y) = s + s + < 2 2 w s + s 2 w = plim cov(yw, y) = plim. cov(ys, y) Wage Income Also Measured with Error In real life, of course, wage income will also be measured with some error, though usually much less than the errors in imputed self-employment income. Wage income certainly has "errors" if the target is the inequality of permanent income, so that annual income would be viewed as an estimate of permanent income containing substantial error, even if annual income itself were measured perfectly accurately. This section shows that, as long as the variance of self-employment income is large enough relative to wage income, then the same result as in equation 3.B1 still holds. Let yi = ywi + ysi , where ywi is measured income from wages. ywi = ywi + i , where i is the measurement error. All other variables are the The Wage Labor Market and Inequality in Vietnam 89 same as above, and E(i) = 0; 2(i) = 2 (i ysi) = (i ywi) = 0 (ij) = 0 i = j (ii) = 0. Then the estimated contribution of wages relative to self-employment for income inequality is biased: plim = plim cov(yw, y) cov(ys , y) = w + s + 2 2 w w + s 2 w cov(yw, y) s + s + = = plim = plim. 2 2 w s + s 2 w cov(ys, y) plim < plim s + s 2 2 w > w + s 2 2 w so plim < plim if > and s w. That is, wages will appear to 2 2 2 2 make a greater contribution to income inequality than they really do when measurement error is worse for self-employment income than wage income, and actual self-employment income is more variable than wage income, both of which are likely to be true. Consistent Estimator If a second estimate of income is observed whose measurement error is un- correlated with the measurement error in yi , then the contribution of each in- come source to total inequality can be estimated consistently. That is, j = cov(yj , y), and hence also the ratio of the contribution of two sources of in- come, var(can be estimated consistently. Household consumption expenditure y) , per capita, ei, can be used as a second estimate of income. ei = yi + i, where E(i) = 0; 2(i) = 2 (ei yi) = 0 (ij) = 0 i = j (ii) = (ii) = 0. The estimator of the contribution of wage income to total inequality used here is ^w = cov(yw, e) cov(y, e) . cov(yw, e) = cov(yw + , y + ) = cov(yw, y) + cov(, y) + cov(yw, ) + cov(, ) 90 Economic Growth, Poverty, and Household Welfare in Vietnam cov(y, e) = cov(y + + , y + ) = var(y) + cov(, y) + cov(, y) + cov(y, ). Because , , and are uncorrelated with y and yw, and and are uncorre- lated with each other: plim[cov(, y)] = (, y) = 0 plim[cov(yw, )] = (yw, ) = 0 plim[cov(, )] = (, ) = 0 plim[cov(, y)] = (, y) = 0 plim[cov(y, )] = (y, ) = 0 so plim ^w = plim[cov(yw, y)] = w + s 2 w = plimw . plim[var(y)] w2 + s + 2s 2 w For ^ = ^w ^s , it follows that 2 plim ^ = w + s w s + s = plim. 2 w So combining poorly measured, source-specific income data with total household expenditure allows us to obtain consistent estimates of the rela- tive contribution of different income sources to total inequality. Notes 1. Output per person is measured by real gross national product (GNP) per capita in 1995 U.S. dollars (World Bank 2000b). 2. The 1992­93 survey spanned a full year, starting in October 1992; the 1997­98 survey began in December 1997 and also lasted one year. For brevity's sake, refer- ence is made to the surveys as the 1993 VLSS and 1998 VLSS, respectively. The sur- vey includes the information to make a detailed calculation of household income as well as a full household expenditure survey. The 1993 survey sampled households proportionally to the population in each region, but the 1998 survey oversampled certain areas, requiring the use of sampling weights to calculate representative sta- tistics. The VLSS is described in more detail in the appendix to chapter 1 of this vol- ume and in World Bank (1995) and (2000a). 3. Laborforceparticipationusedintable3.1referstothesevendaysbeforethesur- vey interview. If people are included who were not working during the past week, but who worked at some time during the year, total participation rises to 89 percent in 1998. 4. The 1993 VLSS asked a single question about whether the household mem- bers were self-employed for their main work in the past seven days. Whether they were self-employed on the family farm or in the family business could be only inaccurately inferred from other questions. The 1998 VLSS asked separate questions about whether household members were self-employed on the farm or self- employed in a family business. The Wage Labor Market and Inequality in Vietnam 91 5. Household income per capita growth is calculated from the 1993 VLSS and the 1998 VLSS data. Output per capita growth is measured by real GNP per capita growth in 1995 U.S. dollars (World Bank 2000b). 6. Wages here are hourly total compensation from the main job in the past seven days (or from the main job in the past 12 months if there was no main job in the past seven days). This includes the value of compensation in kind, as well as money wages. 7. The urban population of Ho Chi Minh City was 4.4 million (5.6 percent of the population) and of Hanoi, 1.6 million (2.1 percent of the population), out of a total population of Vietnam in 2000 of 77.9 million (World Bank 2000b). 8. Some of the apparent regional changes in table 3.7 may be due to small sam- ple sizes. In the worst case, there were only 20 skilled, private, nonagricultural wage observations for Hanoi in 1993. Averages in all other tables come from sample sizes of more than 200 observations, with the exception of Hanoi in 1993, for wage-related tables with a sample size of 136. 9. Fifty-four weeks is more than a year, so the workers must have been working more than five days a week or more than eight hours a day. At the time of the two surveys, the standard Vietnamese workweek was six days. 10. See Berndt (1991, chapter 5) for an accessible explanation of wage determi- nants and the regression specification used in this discussion. 11. About 86 percent of the population in Vietnam is ethnically Vietnamese (Kinh). Another 2 percent is Chinese, and the remaining 13 percent is spread across a wide variety of groups found mostly in remote rural areas. 12. The estimated effect of a dummy variable in the wage regression is eb - 1, where b is the coefficient estimate. 13. An indicator for Chinese ethnicity was not included in the regional wage re- gressions in table 3.12 because a number of regions had no Chinese in the survey sample. 14. Deaton (1997, pp. 134­40) provides a good, short explanation of inequality measures. 15. The author thanks Paul Glewwe for pointing out that the weights used to decompose the generalized entropy measure I2 do not sum to one (unlike the decomposition weights for I0 and I1), which makes the decomposition of I2 hard to interpret. 16. This has interesting implications for cross-country inequality comparisons. Countries with income data that are measured less accurately (typically, poorer countries) will have a spuriously larger calculated inequality index. 17. This is certainly not a criticism of the collection of detailed household pro- duction data in the VLSS and similar surveys. As noted, the mean income estimates are still unbiased. These data are very valuable for studying important questions of household production per se, and it is often possible to use the imprecise income data to calculate consistent estimates of income inequality, as has been done in this chapter. 18. Consumption smoothing is harder in Vietnam because financial institutions are poorly developed and, over the years, have undermined their credibility with ar- bitrary behavior. However, this does not prevent consumption smoothing through the saving of commodities and durable goods, informal credit and debt arrange- ments within the village, and cash savings, which are often in the form of gold buried under the house. 19. Household rice harvests and certain other staple crop yields are probably common knowledge in rural Vietnam, to other villagers as well as to the village tax 92 Economic Growth, Poverty, and Household Welfare in Vietnam authorities, but input costs are not, and much of the profit in household farming in Vietnam is in nonstaple agricultural production, which is difficult to observe. Non- farm household enterprises are also very difficult for outsiders to observe. 20. As shown in chapter 15 by Glewwe and Nguyen, measured mobility is highly sensitive to errors, so part of the apparent high wage mobility could be due to measurement errors. Glewwe and Nguyen's critique does not apply to the other in- equality measures used in this chapter. The general issue that income mismeasure- ment can bias inequality measures is addressed at length above. 21. The share of inequality of one income source compared with another de- pends on the ratio of the covariance of each income source with total income (see ap- pendix 3B). For the case of wage income, its share of inequality is greater when the covariance of wage income with total income is greater. Since cov(yw, y) = var(yw) + cov(yw, ys) = var(yw) + cov(yw, ys) where yw is wage income, y is total income, and ys is income from other sources, the share of wage income in inequality depends on both the inequality of wage income itself (var[yw]) and how correlated wage in- come is with other income sources. Because income is measured with error, the con- sistent estimates of inequality use total expenditure as an instrument for total income in the covariance calculations. 22. The cross-country data for share of labor force in agriculture and the ratio of agricultural value added to total output are from World Bank (2000b). The purchas- ing power parity GDP per capita data are from the Penn World Table 6.0 (2002). Bibliography Belser, Patrick, and Martín Rama. 2001. "State Ownership and Labor Redundancy: Estimates Based on Enterprise-Level Data from Viet- nam." World Bank Working Paper 2599. Washington, D.C. Berndt, Ernst R. 1991. The Practice of Econometrics, Classical and Contemporary. Reading, Mass.: Addison-Wesley. Deaton, Angus. 1997. The Analysis of Household Surveys: A Microeconomic Approach to Development Policy. Washington, D.C.: World Bank. Friedman, Milton. 1957. Theory of the Consumption Function. Princeton, N.J.: Princeton University Press. Penn World Table. 2002. "Data Appendix for a Space-Time System of National Accounts: Penn World Table 6.0 (PWT 6.0)." Accessed from http://webhost.bridgew.edu/baten, May 30, 2002. Psacharopoulos, George. 1985. "Returns to Education: A Further Interna- tional Update and Implications." Journal of Human Resources 20(4): 583­604. Pyatt, Graham, Chau-Nan Chen, and John Fei. 1980. "The Distribution of Income by Factor Components." Quarterly Journal of Economics 95(3): 451­73. Sen, Amartya Kumar. 1997. On Economic Inequality. Oxford: Clarendon Press. Shorrocks, A. F. 1982. "Inequality Decomposition by Factor Components." Econometrica 50(1): 193­212. The Wage Labor Market and Inequality in Vietnam 93 _____. 1984. "Inequality Decomposition by Population Subgroups." Econo- metrica 52(6): 1369­86. World Bank. 1995. "Vietnam Living Standards Survey: Basic Information Doc- ument." Development Research Group. World Bank, Washington, D.C. Web site: www.worldbank.org/lsms/country/vn93/vn93bid.pdf. _____. 2000a. "Vietnam Living Standards Survey (VLSS) 1997­98: Basic Information." Poverty and Human Resources Division, World Bank, Washington, D.C. Web site: www.worldbank.org/lsms/country/ vn98/vn98bif.pdf. _____. 2000b. World Development Indicators 2000 CD-ROM. Washington, D.C.: International Bank for Reconstruction and Development. _____. 2002. Vietnam Development Report 2002. Washington, D.C.: World Bank. World Bank and Department for International Development. 1999. Voices of the Poor: Synthesis of Participatory Poverty Assessments. Hanoi: World Bank. Yang, Dennis Tao. 1999. "Urban-Biased Policies and Rising Income Inequal- ity in China." American Economic Review 89(2): 306­10. 4 Household Enterprises in Vietnam: Survival, Growth, and Living Standards Wim P. M. Vijverberg and Jonathan Haughton Vietnam aims to double its gross domestic product (GDP) over the coming decade, an objective that the World Bank has called "ambitious but attain- able" (World Bank 2000, p. 18). To achieve this end, the private nonagricul- tural sector will need to grow even more rapidly than in the recent past. For instance, industrial GDP will need to rise by 10 percent annually, and the output of manufacturing small and medium enterprises (SMEs) may have to rise by as much as 18­25 percent every year. This may need "a vibrant pri- vate sector" (World Bank 2003, p. 133). There are many nonfarm household enterprises (NFHEs) in Vietnam. In the Vietnam Living Standards Survey (VLSS) of 1997­98, every 100 house- holds operate about 57 enterprises that together provide employment to 107 individuals (adults as well as children), indicating that this sector has the potential to contribute significantly to the achievement of the economic ob- jectives of Vietnam. Almost one-quarter of adults worked in NFHEs. Based on household panel data from the VLSSs of 1992­93 and 1997­98, there is some evidence that operating an enterprise leads to greater affluence.1 The enterprises are found across all sectors of the economy: To list some of the more important activities, 26 percent of the enterprises manufacture various items, especially food-related products, textiles, garments, wood, and furniture; 36 percent are in retail trade; 6 percent are in the transport business; 9 percent provide various services; 4 percent are in the hotel and restaurant sector; 2 percent are found in the construction industry; and some 11 percent are active in agriculture-related activities (fishing, logging, and so forth). NFHEs are most likely to be operated by urban households, by those with moderately good education, and by the children of proprietors. Based on a panel of NFHEs constructed over the two survey years, it was found that 39 percent of enterprises operating in 1993 were still in business in 1998. 95 96 Economic Growth, Poverty, and Household Welfare in Vietnam Those NFHEs in the (more affluent) south of the country were less likely to survive, as were smaller and younger businesses. A pattern emerges. In poor areas, the deficiencies of education, credit, and effective demand limit the development of NFHEs. In rich areas, the at- traction of wage labor competes with NFHE development. NFHEs are thus most important in the period of transition, when agriculture is declining in importance but before the formal sector dominates. We expect that NFHEs will continue to play only a modest supporting role in fostering economic growth in Vietnam. NFHEs are embryonic SMEs, and the success of Vietnam's growth plans will depend in large part on the vigor of these small firms (NFHEs and SMEs). On the one hand, some authors are skeptical about these firms' ca- pacities. In a comparison with China, Perkins (1994) wonders where the pri- vate enterprises in Vietnam are, or from whence they will emerge. On the other hand, the environment in which small firms operate has become friendlier. In 2000, partly as a result of easier procedures (Nguyen 2000; Phan 2000a), the number of new firm registrations almost doubled to 14,400 (Asia Pulse 2001), and this pace continued into 2001, when a further 21,000 were registered (Economist Intelligence Unit 2002). Based on a survey in mid-2001, the Vietnam Chamber of Commerce and Industry estimates that about 70 percent of newly registered firms are "truly new," and the rest were preexisting enterprises (McKinley 2001). The broad issue addressed by this chapter is whether NFHEs are up to the task of spawning enough promising firms and creating jobs in their own right. This analysis is largely based on the information collected by the VLSSs of 1993 and 1998. The next section provides the basic background information on the pattern of ownership of NFHEs and argues that participating in an NFHE, on balance, improves household expenditure levels. Attention then turns to the question of who operates an NFHE, where the logistic regression results indicate that a household is more likely to operate an NFHE if it is located in an urban area, if local wage rates are high, and if the household has a history of operating an enterprise. The most innovative part of the chapter consists of the construction of a panel of NFHEs, observed in both 1993 and 1998. Having constructed the panel and determined that attrition bias is not serious, it is possible to ask why some NFHEs survive and others fail. This is followed by a discussion of why enterprises are born in the first place. The penultimate section looks at the determinants of NFHE performance--as measured by profits--over time and is followed by a summary of the main conclusions. Household Enterprises and Living Standards A concern about the sources of economic growth is not the only reason for looking more closely at NFHEs. They may also influence the distribution and level of income--between poor and rich households, urban and rural areas, ethnic Vietnamese (Kinh) and other ethnic groups, north and south. This study, therefore, begins with an analysis of these distributional effects before turning to the determinants of firm survival and formation. Household Enterprises in Vietnam: Survival, Growth, and Living Standards 97 Just over one-quarter of all adults worked in NFHEs in 1993, as table 4.1 shows;2 this was true for both men and women. Over the subsequent five- year interval, GDP rose by 8.9 percent a year (Haughton 2000), and the struc- ture of employment also changed, with a sharp decline in the number of adults involved in agriculture--from 67.1 percent in 1993 to 60.7 percent in 1998--and almost all of the decline concentrated in households in the top two quintiles of the expenditure distribution. Perhaps surprisingly, the proportion of adults working in NFHEs also fell, from 25.7 percent to 24.2 percent, although the proportion relying on this as their sole source of earnings actually rose (9.5 percent to 10.2 percent). In very poor and very rich societies, NFHEs are rare. Between these two ex- tremes, NFHEs first gain in importance and then get pushed aside as better economic opportunities arise. Employment in NFHEs is perhaps best thought of as playing a bridging role, providing an attractive alternative to farming, but one that is less appealing than most wage-paying jobs. The unexpected finding for Vietnam is that the importance of NFHEs appears to have peaked already, although they remain a very important source of Table 4.1. Labor Market Participation, by Residence and Gender Based on 1993 VLSS Labor market activity Total Urban Rural Male Female Wage employment 25.7 34.1 23.3 33.8 18.6 Farming 67.1 20.1 80.6 68.0 66.3 Nonfarm self-employment 25.7 36.6 22.6 25.1 26.3 Only activity 9.5 27.1 4.4 8.4 10.5 With farming only 12.3 5.4 14.3 11.5 12.9 With wage employment only 1.3 2.9 0.9 1.6 1.1 With farming and wage employment 2.7 1.2 3.1 3.7 1.7 Not employed 13.5 24.7 10.2 11.2 15.4 Number of observations 14,297 3,205 11,092 6,643 7,654 Based on 1998 VLSS Labor market activity Total Urban Rural Male Female Wage employment 25.7 32.6 23.3 33.9 18.4 Farming 61.7 14.8 77.5 61.7 61.7 Nonfarm self-employment 24.2 34.1 20.8 23.7 24.6 Only activity 10.2 27.6 4.3 9.4 10.9 With farming only 11.3 3.8 13.8 10.7 11.8 With wage employment only 1.2 2.4 0.8 1.6 0.9 With farming and wage employment 1.4 0.3 1.8 1.9 1.0 Not employed 16.9 29.0 12.9 14.7 18.9 Number of observations 18,698 5,673 13,019 8,808 9,890 Source: Authors' calculations using the 1993 and 1998 VLSSs. 98 Economic Growth, Poverty, and Household Welfare in Vietnam employment and income. With rapid growth in the formal sector (that is, wage employment and large-scale private enterprises), we speculate that employment in NFHEs will continue to lose ground over the coming decade. Table 4.1 also shows that adults were much more likely to be employed in an NFHE in an urban area (36.6 percent in 1993, 34.1 percent in 1998) than a rural area (22.6 percent in 1993, 20.8 percent in 1998). Rural households are far more likely than urban ones to combine NFHE employment with other activities, particularly farming, and fewer than 5 percent of rural adults re- lied on an NFHE as their sole source of support. Women find employment in NFHEs as often as men do. Particularly low participation rates in NFHEs are found in the Central Highlands and Northern Uplands, as well as among ethnic minority (non-Kinh) households (see table 4.2),3 which tend to be found in the more inaccessible parts of the country (see chapter 7 of this text). Participation in an NFHE is associated with a higher standard of living, as the numbers in table 4.2 make clear. In the poorest quintile (as measured by expenditure per capita), just 15 percent of adults worked in an NFHE, compared with 32 percent in the richest quintile. This raises the possibility that participation in an NFHE is associ- ated with greater economic mobility. Table 4.3 is designed to explore this Table 4.2. Labor Market Participation by Quintile, Region, and Ethnicity Wage Number of NFHE employment Farming observations Indicator 1993 1998 1993 1998 1993 1998 1993 1998 Expenditure per capita (quintile) Poor 17.8 14.9 24.6 27.3 81.9 80.3 2,396 2,844 Poor-middle 21.9 19.4 23.8 26.6 79.6 75.9 2,608 3,114 Middle 24.1 23.1 25.0 24.8 75.5 72.9 2,817 3,580 Middle-upper 27.7 27.9 26.1 22.8 67.8 60.4 3,114 4,171 Upper 34.0 32.1 28.2 27.3 39.0 28.6 3,362 4,983 Regions Northern Uplands 20.5 19.1 16.8 15.2 80.2 77.1 2,139 2,564 Red River Delta 28.4 28.3 24.4 23.5 71.2 66.8 3,203 3,268 North Central Coast 24.3 27.1 18.9 23.3 84.1 75.8 1,776 2,037 Central Coast 25.6 21.8 23.5 27.6 58.1 54.8 1,715 2,471 Central Highlands 9.9 10.8 24.5 22.8 85.7 86.0 384 1,143 Southeast 28.4 27.1 32.0 36.2 33.9 25.4 1,918 3,495 Mekong Delta 27.7 23.7 34.4 29.9 67.2 60.0 3,162 3,714 Ethnic group Kinh 27.4 26.0 26.5 26.2 66.2 59.6 12,186 15,962 Hoa (Chinese) 37.2 31.9 30.9 31.6 9.7 12.1 392 518 Other ethnic minorities 11.1 10.5 18.5 21.2 86.2 84.5 1,719 2,218 Source: Authors' calculations using the 1993 and 1998 VLSSs. Household Enterprises in Vietnam: Survival, Growth, and Living Standards 99 Table 4.3. Percentage of Households with a Nonfarm Household Enterprise, 1993 and 1998 Expenditure per capita quintile in 1998 Expenditure per capita quintile Middle- in 1993 Poorest Low-middle Middle upper Upper Total Percentage of households with an NFHE in 1993 Poorest 30.6 30.8 39.5 37.7 25.0 778 Low-middle 34.6 38.1 38.9 34.4 52.6 851 Middle 41.8 37.4 41.6 44.4 47.7 848 Middle-upper 35.7 35.4 49.5 50.8 62.4 899 Upper 52.9 47.4 49.5 57.0 61.7 928 Total 730 828 908 947 891 4,304 Percentage of households with an NFHE in 1998 Poorest 26.4 35.1 40.3 28.3 62.5 778 Low-middle 31.4 38.1 42.0 45.0 50.0 851 Middle 39.8 39.0 42.8 45.3 52.3 848 Middle-upper 45.2 42.5 41.0 53.0 57.6 899 Upper 47.1 26.3 41.0 53.3 55.6 928 Total 730 828 908 947 891 4,304 Source: Authors' calculations using the 1993 and 1998 VLSSs. possibility. It considers only the 4,304 households that were surveyed in both 1993 and 1998 and creates a matrix with expenditure per capita quintile in 1993 on one axis and the quintile in 1998 on the other. Each cell shows the percentage of households with an NFHE in 1993 or 1998 (table 4.3). The first point that stands out in this table is that poor households are less likely than rich ones to participate in an NFHE in either year. There is an- other way to make this point more forcefully. Define a household as chroni- cally poor if it fell into one of the bottom three quintiles in 1993 and one of the bottom two quintiles in 1998,4 and define a household as affluent if it was in one of the top two quintiles in both years. Then, affluent households are seen as being far more likely to participate in NFHEs than chronically poor house- holds; about one-third of chronically poor households had an NFHE, compared with more than half of affluent households. Specifically, the per- centages of households with an NFHE in 1993 and 1998 are: Household 1993 1998 Chronically poor 35.6 percent 35.0 percent Affluent 58.0 percent 54.9 percent 100 Economic Growth, Poverty, and Household Welfare in Vietnam Put another way, the persistently affluent are more likely to operate an NFHE. What is not clear is whether this result is because NFHEs make households better off or whether better-off households are more likely to start NFHEs (for instance, because they have better access to credit). A review of table 4.3 helps one to understand the direction of causality: Households that moved up the income distribution were more likely to get involved in an NFHE. This, too, can be dramatized: Define households that rise at least two quintiles between 1993 and 1998 as "shooting stars" and those that fall at least two quintiles as "sinking stones" (this is the terminol- ogy used by D. Haughton and others [2001]). Table 4.3 demonstrates that sinking stones (who were more affluent to begin with) have reduced their involvement in NFHEs, while shooting stars (who were poorer at the start) have increased their participation. Specifically, the percentages of house- holds with an NFHE in 1993 and 1998 are: Household 1993 1998 Sinking stones 43.8 percent 40.3 percent Shooting stars 39.5 percent 46.4 percent This suggests that participating in an NFHE does, on balance, improve household expenditure levels. It then becomes important to explore why some households operate NFHEs and others do not, because it helps clarify the roots of both income distribution and income mobility in Vietnam. This issue is examined in detail in the next section. The Dynamics of Nonfarm Household Enterprises In seeking to understand the dynamics of NFHE creation and survival, it is natural to start by asking who operated households at the beginning of the period (that is, 1993); this is the question posed in box 1 in figure 4.1, and it is answered in the next section. Some of the households surveyed in 1993 dropped out of the sample by 1998. This raises the possibility of attrition bias, an issue that must be tack- led before moving on to two key questions. First, why did some of the en- terprises that operated in 1993 survive to 1998, while others did not? Second, what factors led households to start an NFHE between 1993 and 1998? To answer these two questions, a panel of enterprises must be con- structed; this is possible because of the unique way in which the VLSSs are designed. The questions themselves are then addressed by estimating a se- ries of logistic models. Who Operates Nonfarm Household Enterprises? What determines why some households operate NFHEs and others do not? Some basic numbers are set out in table 4.4. They show that adults are more likely to participate in NFHEs if they are moderately well educated (6­12 years of school) or at prime age (26­55). Employment in NFHEs appears to Household Enterprises in Vietnam: Survival, Growth, and Living Standards 101 Figure 4.1. Household Choices in 1993 and 1998 1. Operate an enterprise in 1993? Yes No 2a. 2b. Respond to 1998 survey? Respond to 1998 survey? Yes No No Yes 3a(j). 3b. 3c. (j 1, 2, 3) Continue Start a new enterprise Start a new enterprise enterprise (j) until 1998? between 1993 and 1998? between 1993 and 1998? Yes No Yes No Yes No Table 4.4. Labor Market Participation, by Age and Schooling Level Wage Number of NFHE employment Farming observations Indicator 1993 1998 1993 1998 1993 1998 1993 1998 Age 16­25 23.9 18.6 28.4 29.4 69.5 58.7 4,409 5,424 26­35 32.2 31.4 34.3 34.8 73.1 68.7 3,560 3,835 36­45 33.1 34.2 31.7 32.4 71.8 70.6 2,339 3,705 46­55 27.4 27.5 20.1 22.6 71.8 66.2 1,448 2,153 56­65 17.0 17.0 9.8 8.9 61.1 62.3 1,356 1,747 >65 6.9 8.1 2.9 2.4 31.9 32.5 1,185 1,834 Years of schooling 0 12.6 7.5 14.2 11.9 56.5 42.7 1,888 3,222 1­5 24.2 23.3 22.1 21.8 71.7 68.7 4,667 6,078 6­9 29.5 29.4 27.0 31.9 71.8 67.2 2,474 2,715 10­12 30.8 31.5 28.4 31.8 69.3 65.5 4,479 6,101 >12 26.1 22.5 55.3 65.7 39.3 21.5 789 493 Note: NFHE = nonfarm household enterprise. Source: Authors' calculations using the 1993 and 1998 VLSSs. be less attractive to those with some university education, probably because this group is able to find wage employment more easily. Between 1993 and 1998 there was a sharp drop in self-employment among two groups. Those with no schooling were working fewer jobs or 102 Economic Growth, Poverty, and Household Welfare in Vietnam stopping work, and were probably older workers, and those with more than 12 years of schooling were more likely to be working for a wage (and work- ing at just one job). There was also a noticeable drop in self-employment, and jobs overall, among young workers (ages 15­25), primarily because more of them are staying in school longer. Although tabulations of data, such as the one in table 4.4, are useful, they suffer from a limitation: It is possible to see the effects of only one variable at a time. Amore rigorous answer to the question, which would allow measure- ment of the effect of a variable while holding all other influences constant, calls for the estimation of a logistic model. Here the dependent variable is set equal to one if a household operated an NFHE in 1993 and to zero otherwise. The estimation results are set out in table 4.5; a similar model is found in Vijverberg (1998b, p. 149). Several of the variables that are used in this model to capture the effects of the rural environment are innovative, and they are de- fined more fully in the appendix. In addition, the variable "local producer price of rice" in table 4.5 was constructed by Benjamin and Brandt (2004) and captures both the attractiveness of farming as a source of income and the level of income in the rural community that drives the demand for nonfarm com- modities; these forces work in opposite directions. The first two groups of variables in table 4.5--"regional variables" and "in rural areas"--work in tandem. There are seven regions in Vietnam, but there are no urban locales in the Central Highlands. Thus, the six urban dummy variables (such as urban Southeast) serve to compare each urban region against a baseline rural area, and because the urban dummy variables exhaust the regional alternatives, the south dummy offers a comparison between southern rural households and northern rural households. It is pertinent that the rural communities are differentiated by the variables in the second group according to their features, such as accessibility, electrification, and presence of market institutions. These data come from the community questionnaire and are available only for rural areas. Thus, the baseline rural area is one with zero values for all of these variables.5 To aid in the interpretation of the results, the final column in table 4.5 shows the probability that a household will operate an NFHE, assuming that the baseline probability is 0.45 and that the independent variable in question has increased by one unit. A number of themes emerge. Perhaps the most important is that geog- raphy matters. Households in urban areas are more likely to engage in self-employment. Within rural areas, NFHEs are less common where agricul- tural extension programs are more active, perhaps a proxy for the greater profitability of farming in these areas. The presence and quality of local roads has an unexpected negative sign, although this variable is somewhat prob- lematic: The 1993 VLSS questionnaire did not specify clearly what constitutes a viable road, and the model does not control for waterway access, which in some areas in Vietnam is important. The presence and frequent operation of a local market has a positive effect--if there is such a market, the probability that a household would operate a business increases from an (assumed) Household Enterprises in Vietnam: Survival, Growth, and Living Standards 103 Table 4.5. Logistic Model of Operation of an Enterprise, 1993 New probability Variable Coefficient t Statistic (base = 0.45) Dependent variable: "Household operated an enterprise in 1992­93" Intercept -0.371 0.99 Regional variables: South -0.128 1.39 Urban Northern Uplands 0.629 1.56 Urban Red River Delta 0.552 1.41 Urban North Central Coast -0.377 0.87 Urban Central Coast 0.630 1.61 Urban Southeast -0.010 0.03 Urban Mekong Delta 0.429 1.13 In rural areas Availability of lower and upper secondary schools 0.042 0.23 Agricultural extension index -0.442 6.25*** 0.345 Presence and quality of roads -0.578 2.71*** 0.315 Availability of public transportation 0.000 0.09 Use of electricity and piped water 0.201 1.16 Presence and frequency of local market 0.491 2.80*** 0.572 Presence of market in nearby community 0.194 0.96 Local wage index 0.063 4.79*** 0.466 Dummy, = 1 if local wage index unknown 2.003 5.48*** 0.858 Local producer price of rice 0.050 1.11 Dummy, = 1 if local price of rice unknown -0.085 0.23 Household characteristics Number of women ages 16 years and older 0.107 1.71* Persons ages 16­25 years -0.143 2.17** 0.415 Persons ages 26­35 years 0.035 0.48 Persons ages 36­45 years 0.029 0.37 Persons ages 46­55 years -0.039 0.44 Persons ages 56­65 years -0.217 2.65*** 0.397 Persons ages >65 years -0.399 5.02*** 0.354 Persons with 1­3 years of schooling 0.215 3.37*** 0.504 Persons with 4­5 years of schooling 0.282 4.55*** 0.520 Persons with 6­9 years of schooling 0.334 5.90*** 0.533 Persons with 10­12 years of schooling 0.369 5.47*** 0.542 Persons with postsecondary schooling -0.245 3.65*** 0.390 Persons with technical training 0.047 0.50 Persons with completed apprenticeships 0.275 4.39*** 0.519 (table continues on following page) 104 Economic Growth, Poverty, and Household Welfare in Vietnam Table 4.5. (continued) New probability Variable Coefficient t Statistic (base = 0.45) Characteristics of parents of head of household Average years of schooling 0.023 1.99** Dummy, = 1 if years of schooling unknown -0.021 0.20 Major occupation: farmer -0.792 6.37*** 0.270 Major occupation: manager 0.558 0.73 Major occupation: proprietor 1.165 3.19*** 0.724 Major occupation: supervisor -0.397 0.33 Dummy, = 1 if major occupation unknown -0.148 0.14 Number of observations 4,800 Proportion affirmative 0.451 Average log-likelihood value -0.6134 Likelihood ratio test of slopes 717.79 *Significant at 10 percent level. **Significant at 5 percent level. ***Significant at 1 percent level. Note: Final column shows probability of household operating an enterprise, given a baseline value of 0.45 and then assuming that the independent variable changes by one unit. These figures are shown only for variables with statistically significant coefficients. In this and other tables in this chapter, the omitted categories against which comparisons are made are a baseline rural area (see text), persons with zero years of schooling, and parents of the head of household who were laborers. Source: Authors' calculations using the 1993 and 1998 VLSSs. baseline of 45 percent to 57 percent, a large 12 percentage point jump. The real price of rice is unrelated to the probability that a household operates an enterprise. The second key theme is that the local wage rate is important and raises the likelihood of self-employment. A negative sign might have been expected here, on the grounds that when wage labor pays better, self- employment is relatively less attractive. A higher wage may well reflect a more dynamic nonagricultural sector, however, inviting more households to participate in it, or raising living standards with an attendant higher demand for items such as restaurants and retail services. The third important theme is that family history is important. The chil- dren of proprietors are much more likely to be proprietors themselves. As expected, households are more likely to operate an NFHE if their members are better educated or of prime age. Household Enterprises in Vietnam: Survival, Growth, and Living Standards 105 Constructing a Panel of Enterprises It is well known that NFHEs frequently do not survive for long. More than half of the enterprises reported by the 1998 VLSS had been founded during the previous five years, yet the number of enterprises per household was no higher in 1998 than in 1993. This essentially means that for every NFHE that was started during the five-year period, another one failed. Why do enterprises succeed or fail? An answer to this question might facilitate the design of policies that would help enterprises stay in business. The VLSS data are unusual in that they allow the construction of a panel of enterprises, with information for each of these enterprises for 1993 and 1998.6 This then allows a rigorous exploration of the determinants of success (or at least survival). The construction of the panel proved to be more complex than expected. In both the 1993 and 1998 VLSSs, the interviewers collected information from the "most knowledgeable" household member on the age of each household enterprise and its area of activity. The interviewer also had a household roster for each year. In principle, this allows matching of specific enterprises across survey years. In reality, the situation was more ambiguous. The 1998 round used a different set of industrial codes. The respondents were decidedly imprecise about the enterprise's age. There were changes in the identity of the "most knowledgeable" household member. It was also not uncommon for one household member to be the respondent for several household enterprises. Last but not least, a household could list up to three enterprises in 1993 and up to four in 1998. For this study, the match of enterprises across survey years was made on the basis of the three most obvious pieces of information: enterprise age, in- dustry code, and identity of the entrepreneur. Table 4.6 summarizes the out- come of the matching process. The 1993 round yielded 2,795 enterprises, of which 311 occurred in households that disappeared in the next round and 765 were located in households that did not report any enterprises in the next round. This left 1,719 enterprises in households that also reported NFHE activities in 1998. The 1998 round had a sample of 3,439 enterprises, of which 1,042 were operated by households that were not part of the earlier round and of which 697 occurred in households that did not have an enter- prise in 1993. This left 1,700 enterprises that could possibly be matched with one in 1993 ("enterprises potentially in panel"). A problem arises: If the industry code must be identical, the identity of the entrepreneur must be the same, and the enterprise age must match within a margin of two years, then only 174 enterprises are matched from the 1993 VLSS to the 1998 VLSS. The criteria were relaxed, therefore, by re- quiring only that the entrepreneur and the industry code be the same; this change yielded 514 automatic matches. Cases where there was no match on any dimension were then eliminated, and the remaining cases were in- spected manually. This turned up 455 cases where there was a reasonable 106 Economic Growth, Poverty, and Household Welfare in Vietnam Table 4.6. Accounting for the Panel Enterprises Type of Indicator 1993 1998 enterprise Total enterprises surveyed 2,795 3,439 Household was not included in 1998 sample 47 Household was not included in 1993 sample 1,042 Household dropped out of sample in 1998 (attrition) 264 Attrited Enterprises potentially matchable 2,484 2,397 Household had no enterprise in 1998 765 Terminated Household had no enterprise in 1993 697 Start up Enterprises potentially in panel 1,719 1,700 Household had another enterprise in 1993 but not in 1998 83 Terminated Household had another enterprise in 1998 but not in 1993 96 Startup No match at all on industry code, entrepreneur, or age among 1998 enterprise 322 Terminated No match at all on industry code, entrepreneur, or age among 1993 enterprises 309 Startup Manual inspection found no possible match among 1998 enterprises 345 Terminated Manual inspection found no possible match among 1993 enterprises 326 Startup Matched 969 969 Panel Of which Automatic match between 1993 and 1998 enterprise 514 514 Manual match between 1993 and 1998 enterprise 455 455 Source: Authors' calculations using the 1993 and 1998 VLSSs. match between an enterprise in 1993 and another enterprise in 1998: Per- haps the entrepreneur was the same, but the industry code was slightly dif- ferent, or the age and industry code were consistent but the entrepreneur was possibly different. The net result was a panel of 969 enterprises. This implies a survival rate of 39 percent (969/2,484). To measure the survival rate satisfactorily, it is necessary to have panel data, obtained by observing the enterprise once and then a second time after a few years. Most of the available evidence on survival rates comes from industrial country studies. For example, Storey and Wynarczyk (1996) examine a sample of microenterprises from 1985 to 1994 in the United Kingdom, 60 percent of which had fewer than five employees. These enter- prises were drawn from all sectors of the economy and from all age groups Household Enterprises in Vietnam: Survival, Growth, and Living Standards 107 (rather than startups only). Of these, 70 percent survived until 1988 and 41 per- cent until 1994. Often, evidence on enterprise survival refers to newly established, larger firms (with at least 10 or even 20 employees) in the man- ufacturing sector in developed economies, and so it is not directly compara- ble to the Vietnamese numbers. For example, Audretsch (1995) reports a 35.4 percent 10-year survival rate among U.S. manufacturing firms during 1976­86. Baldwin and Gorecki (1991) report an annual 6.5 percent exit rate, suggesting a 71 percent five-year survival rate, in the Canadian manufactur- ing sector in the 1970s. Among manufacturing enterprises in the Netherlands in the 1980s, the five-year survival rate was approximately 64 percent (Audretsch, Houweling, and Thurik 2000). Littunen (2000) cites evidence that 45 percent of European firms close within the first five years of business and reports on Finnish data that show a survival rate of at least 55 percent after six years. Panel data on enterprises in developing countries are rare; therefore, lit- tle comparable evidence on survival rates exists. There are indications that Vietnam's survival rate of 39 percent is typical for developing countries. First, indirect evidence comes from the age distribution of NFHEs in the VLSSs, which is very similar to those found, based on the living standard measurement surveys, for Peru in 1985, Côte d'Ivoire in 1985­86, and Ghana in 1987­89 (Vijverberg 1998b). This suggests, but does not prove, that enter- prise survival rates in Vietnam are in line with those found elsewhere. Sec- ond, consider that a five-year survival rate of 39 percent implies an annual exit rate of about 17 percent. In the Dominican Republic, the exit rate was 29 percent in 1992 and 22 percent in 1992­93 (Cabal 1995; Mead and Liedholm 1998). In Zimbabwe, the exit rate was 11.5 percent in the early 1990s, but this was likely an underestimate because the whereabouts of almost half of the enterprises could not be verified in the second round of the survey (Daniels 1995; Mead and Liedholm 1998). Third, in a study of four countries in southern Africa, McPherson (1995) reported estimates that would imply a five-year survival rate of 81 percent, but this is based on cross-sectional data that most likely undersampled enterprises that had been closed. Fourth, Mead and Liedholm (1998) computed an average firm closure rate of 12.9 percent in five African countries in the early 1990s on the basis of retrospective questions in a cross-sectional survey. Overall, therefore, the survival rate of NFHEs as recorded by the VLSSs is fairly comparable to that found in other studies of developing countries and below the rates recorded in industrial countries. However, it should be pointed out that the lack of comparability makes it difficult to conclude that the enterprise survival rate is particularly low. Vietnam's survival rate esti- mate may be too low if some enterprises were misclassified in the 1998 round as startups rather than as enterprises continuing in a different line of business. If there was, indeed, more enterprise turnover in Vietnam between 1993 and 1998 than in other developing countries, it would be consistent with Goreski's (1995) finding that in a turbulent economic environment, 108 Economic Growth, Poverty, and Household Welfare in Vietnam there are high rates of both firm entry and firm exit. Rapid growth yields many opportunities for new firms while making existing firms obsolete more quickly. The characteristics of the panel of enterprises in 1993 and 1998 are sum- marized in table 4.7, where they are also compared with attrited enterprises (that is, those that had dropped out of the sample), terminated enterprises, and start-up businesses. When compared with the other enterprises that operated in 1993, the panel enterprises are older and better established. They were more likely to be open for business at the time of the interview, for more months a year and more days per month, and to operate from a fixed location. Panel B shows that enterprises in retail sales and in the hotel and restaurant business appear to survive longer; those in textiles, other manufacturing, services, and the "other" category are more likely to be ter- minated. Panel C of the table reveals small residence and regional differ- ences. Panel D examines enterprise performance; by all definitions, panel enterprises are larger and more profitable. None of these findings are sur- prising, but they do attest to the reasonableness of the panel matching procedure. In comparing panel enterprises between 1993 and 1998, three features are worth comment. Real household expenditures, or performance mea- sures such as real sales revenue or enterprise income, rose less quickly than did expenditure in Vietnam as a whole--where real GDP grew 53 percent between 1993 and 1998 and per capita GDP increased by 40 percent.7 The relatively slow growth of NFHE-related income is unexpected; one might have anticipated that dynamic NFHEs would lift their owners at least as quickly as the overall economic tide. It is also surprising that the reported age of panel enterprises rose by just 3.8 years, on average, even though the two surveys were 5 years apart. This age variable is notoriously unreliable, particularly when the "most knowl- edgeable" household respondent changes between the two surveys. The most curious figure relates to gender: In 1993, 81 percent of the panel enterprises were operated by women, but the 1998 survey indicated that only 57 percent of these same enterprises were operated by women. Note that the identity of the entrepreneur within the household is indicated by the response to the question, "Who among the household members is most knowledgeable about the activities of the enterprise?" Table 4.1 showed that there are a roughly equal number of men and women engaged in nonfarm self-employment. The increase in the number of male entrepreneurs shown in table 4.7 may reflect any of a number of phenomena: (a) the high number of women entrepreneurs in 1993 may be largely an artifact of the survey pro- cedures used in 1993; (b) men "take over" successful household enterprises; or (c) over time, men have taken on a more prominent role in NFHEs. Of these, answer (a) is not entirely likely; Vijverberg (1998b) showed that women contributed many more hours of nonfarm self-employment than did men and thus may indeed be "more knowledgeable" about enterprise operations. (A similar comparison of hours of work in 1998 is difficult page) 1,428) 5.6 3.3 7.4 49.6 83.3 8.5 22.9 61.5 2,936 2,381 10.4 7.6 7.2 3.7 3.0 3.5 29.2 2.6 5.7 = Start-up enterprises (N following on 1998 969) 9.0 7.0 8.7 5.8 6.4 2.5 0.8 3.4 4.6 3.8 Panel = 1.71 58.1 89.1 10.1 24.9 72.7 3470 2776 47.4 continues Households enterprises (N (table Attrited 969) 7.9 4.4 7.1 in Panel = 81.2 86.8 9.2 24.7 67.7 9.5 7.2 3.7 3.9 0.9 2.2 7.8 3.3 2,604 2,090 43.2 enterprises (N Enterprises 1,515) 6.7 3.5 7.3 7.4 9.4 9.1 3.2 7.7 1.1 2.2 4.4 4.0 1993 = 67.3 69.8 21.7 53.8 2,403 1,919 24.0 and erminatedT enterprises (N 264) Enterprises, attrited = 7.6 4.0 7.5 71.0 78.0 8.7 24.7 62.9 5.7 5.3 4.6 4.6 0.0 2.3 2,962 2,246 39.4 6.4 2.3 Enterprises in households (N Nonpanel cent cent cent cent cent cent cent cent cent cent cent cent Mean Median Mean Per Per Mean Mean Per Mean Median Per Per Per Per Per Per Per Per Per Enterprises, capita Panel per of eneur es epr oundsr ocessing location transport pr characteristics entr two operation operation in in fixed eneur a expenditur food/beverage textiles wood other pipeline Comparison epr year between omfr Enterprise schooling, month stauranter oad, 4.7. enterprise entr per Industry A: of of per household B: uction sales and railr ableT Indicator Panel Age earsY Female Operating Months Days Operating Real Panel Manufacturing: Manufacturing: Manufacturing: Manufacturing: Constr Wholesale Retail Hotel Road, 109 1,428) 1.71 7.5 8.1 23.2 16.7 23.0 17.5 10.2 1.1 12.3 19.2 466 962 539 619 = 3,517 4,605 1,176 4,129 1,647 Start-up enterprises (N 1998 969) 4.5 7.3 4.8 9.8 1.6 Panel = 33.6 25.3 14.3 13.0 15.4 20.5 728 882 5,853 1,363 7,283 2,438 6,735 1,974 1,245 enterprises (N 969) 6.4 0.0 Panel = 1.81 33.6 9.8 25.3 14.3 13.0 1.6 15.4 20.5 555 352 4,169 1,138 6,388 1,776 4,520 1,412 2,053 enterprises (N 1,515) 0.0 0.9 243 898 692 907 433 103 1993 = 10.4 24.6 27.7 16.3 23.9 12.0 10.9 12.4 23.6 2,420 3,710 2,526 erminatedT enterprises (N 264) attrited = 12.1 0.0 17.4 43.6 8.0 22.0 7.2 13.6 3.0 20.1 26.1 718 441 578 3,010 4,662 1,537 3,586 1,174 1,371 Enterprises in households (N cent cent cent cent cent cent cent cent cent cent cent Per Per Per Per Per Per Per Per Per Per Per Mean Median Mean Median Mean Median Mean Median Mean a,c a,b year) ent) year) utilities ent) curr, whole, ) curr, whole, mining, performance (monthly) (monthly (monthly e, es continued( Coast (monthly (monthly e Uplands Delta income income Residence 4.7. agricultur Coast Highlands Delta Enterprise C: Central D: River expenditur evenuer evenuer ableT Indicator Services Aquacultur Other: Panel Urban Northern Red North Central Central Southeast Mekong Panel otalT Sales Sales Enterprise Enterprise 110 and home 334 666 347 586 313 213 183 1.32 0.24 1.77 419 6,367 of egionsr oss value acr the ons and 438 935 509 891 461 271 243 1.46 0.26 1.85 487 indk 10,899 variati in price interviews. for . payments second 317 792 385 714 349 280 243 1.57 0.31 1.98 300 8,287 deflatede and survey plus ar first the e paid, e the values befor wer year 255 537 263 465 222 220 183 1.44 0.19 1.71 160 3,800 between the expenses Monetary period over after values. 1998 venueer over left 276 736 392 671 332 282 213 1.51 0.28 1.84 220 two-week 8,594 with the costs. monthly having " during eportr comparability operating typical" Median Mean Median Mean Median Mean Median Mean Mean Mean Mean Median evenuer VLSSs. eneurs for minus epr eportedr 1998 1.5087 eportedr entr on and by evenuer on that unweighted. 1993 c,d e based ent) ar sales is the b,d year) curr inflatede based as is amount ent) ar using Statistics venue)er the as curr, whole, 1993 (monthly) (value, defined venue)er (or workers omfr is (or defined labor workers stock months. is calculations' income income (monthly (monthly family paid workers values income family year of of of capital Dong sampling ent evenuer Authors of evenuer evenuer of ce: Enterprise Curr Whole Net Net Net Hours Number Number Number alueV Note: a. b. c. d. Sour between consumption. 111 112 Economic Growth, Poverty, and Household Welfare in Vietnam because of the structure of the new questionnaire.) Answer (c) is plausible in the light of the similar percentages in the columns for 1998 panel and start- up enterprises. Explaining Attrition of Households with Nonfarm Household Enterprises Ten percent of the households that operated enterprises in 1993 had dropped out of the sample by 1998. This attrition raises the possibility that the panel of enterprises may be biased and the households (and their enter- prises) that dropped out of the sample were atypical. Table 4.7 allows a comparison of the characteristics of the attrited enter- prises with those that either went out of business or were part of the panel. The enterprises that dropped out of the sample were more likely to be in urban areas, in southern Vietnam, and operated by better-off households. The performance measures of attrited firms do not stand out from those of other businesses, however. The determinants of attrition have also been captured in a logistic model, where the dependent variable is one if the household also responds in 1998 and zero otherwise. The results of estimating this model, which is condi- tional on the presence of an enterprise, are shown in the middle columns of table 4.8. A similar approach can also be used to model attrition among households that did not operate a business in 1993 (that is, answered "no" to question 2B in figure 4.1); these results are shown in the last two columns of table 4.8. The estimates show that, overall, urban households were less likely to re- main in the sample, and households with older members were more coop- erative. Other determinants are more sporadic. By and large, households in the north were less likely than households in the south to drop out of the sample between 1993 and 1998. Human capital variables matter little. There is a suggestion that better-off households are more cooperative, all other things being equal, and that those with higher-earning enterprises are less responsive, but the effect of the financial variables, which are in logarithmic form to reduce the impact of outliers,8 is not statistically significant. For all practical purposes, attrition is sufficiently small, and its correlation with enterprise performance so minimal, that attrition bias is unlikely to be a serious concern. Thus, the observed sample of enterprises in panel house- holds may be viewed as representative of the population of panel enterprises. Which Enterprises Survived? It is now possible to address the first of the two key questions mentioned in "The Dynamics of NFHEs": Why did some of the enterprises that operated in 1993 survive to 1998, while others did not? Note that the unit of observation is the enterprise, not the household. Some households operate more than one business, and one might surmise that the survival of one household enterprise might depend on the exis- tence and performance of the other enterprises within that household. Table 4.8. Determinants of the Attrition Process: A Logistic Model Households with Households without enterprise in 1993 enterprise in 1993 Variable Coefficient t statistic Coefficient t statistic Dependent variable: "Household responds to 1998 survey" Intercept -0.601 0.43 -0.774 0.59 Regional variables: Urban residence -0.708 3.93*** -1.235 6.03*** Northern Uplands 1.767 3.21*** 0.196 0.41 Red River Delta 1.156 2.32** 0.600 1.25 North Central Coast 1.439 2.66*** 1.654 2.91*** Central Coast 0.897 1.74* 0.868 1.68* Southeast 0.708 1.40 -0.140 0.28 Mekong Delta 0.806 1.65* -0.553 1.21 Household characteristics Number of women ages 16 years and older 0.019 0.13 -0.205 1.19 Persons ages 16­25 years -0.043 0.25 0.500 2.67*** Persons ages 26­35 years 0.161 0.86 0.526 2.63*** Persons ages 36­45 years 0.268 1.35 0.721 3.32*** Persons ages 46­55 years 0.395 1.75* 0.551 2.29** Persons ages 56­65 years 0.388 1.87* 0.914 3.88*** Persons ages >65 years 0.229 1.11 0.321 1.61 Persons with 1­3 years of schooling 0.158 0.89 -0.065 0.38 Persons with 4­5 years of schooling 0.119 0.72 0.083 0.46 Persons with 6­9 years of schooling 0.148 0.99 -0.051 0.33 Persons with 10­12 years of schooling 0.073 0.45 -0.369 2.02** Persons with postsecondary schooling -0.146 1.00 0.107 0.65 Persons with technical training -0.236 1.45 -0.253 1.14 Persons with completed apprenticeships -0.055 0.49 -0.158 0.95 Financial performance Log (real household expenditures) 0.210 1.22 0.279 1.76* Log (total enterprise income) -0.097 1.37 Number of observations 2,128 2,576 Proportion affirmative 0.905 0.924 Average log-likelihood value -0.2980 -0.2374 Likelihood ratio test of slopes 62.4 163.2 *Significant at 10 percent level. **Significant at 5 percent level. ***Significant at 1 percent level. Source: Authors' calculations using the 1993 and 1998 VLSSs. 114 Economic Growth, Poverty, and Household Welfare in Vietnam Involvement in several activities diversifies risk, however. The simplest approach, and the one that is followed here, is to stay with the maintained hy- pothesis that the observations on enterprises are independent of one another. Table 4.9 presents the results of estimating a logistic model, where the bi- nary dependent variable is set equal to one if the enterprise survived from 1993 to 1998. The empirical specification parallels that of other studies on firm survival, such as Littunen (2000), McPherson (1995), and Storey and Wynarczyk (1996). There are two versions of the model: one that relies on the community characteristics from 1993 and one that uses the community characteristics from 1998. The estimates of the two models are similar in most respects, but there are some notable differences in the community and regional effects. Judging by the likelihood ratio, the model with the 1993 community characteristics fits marginally better than the 1998 model. Table 4.9. Enterprise Survival: A Logistic Model Using community characteristics from 1993 1998 Coeffi- t Coeffi- t Variable cient statistic cient statistic Dependent variable: "1993 enterprise is surveyed again in 1998" Intercept -2.590 4.73*** -2.963 7.03*** Regional variables: South -0.454 3.23*** -0.256 1.91* Urban Northern Uplands 0.040 0.08 -0.077 0.27 Urban Red River Delta -0.004 0.01 0.125 0.43 Urban North Central Coast 0.686 1.21 0.479 1.20 Urban Central Coast 0.846 1.71* 0.451 1.65* Urban Southeast -0.076 0.15 -0.139 0.45 Urban Mekong Delta 0.351 0.73 -0.025 0.11 In rural areas: Presence and quality of roads 0.021 0.07 0.237 0.90 Presence and quality of waterways -0.039 0.27 Availability of public transportation -0.006 1.59 -0.009 1.28 Presence and frequency of local market 0.944 2.85*** 0.128 0.86 Presence of market in nearby community 0.593 1.49 -0.181 0.63 Use of electricity and piped water -0.330 1.29 -0.123 0.78 Local wage index -0.023 1.04 0.001 0.13 Dummy, = 1 if local wage index unknown -0.210 0.71 0.207 1.04 Local producer price of rice -0.015 0.21 0.175 2.28** Dummy, = 1 if local price of rice unknown -0.237 0.49 0.232 1.24 Household Enterprises in Vietnam: Survival, Growth, and Living Standards 115 Entrepreneur's characteristics Female 0.294 2.28** 0.293 2.29** Age <16 years -0.100 0.30 -0.111 0.33 Age between 26 and 35 years 0.620 4.41*** 0.639 4.53*** Age between 36 and 45 years 0.461 2.87*** 0.473 2.94*** Age between 46 and 55 years 0.227 1.14 0.234 1.17 Age between 56 and 65 years 0.048 0.20 0.041 0.17 Age >65 years -0.304 0.81 -0.274 0.72 Years of schooling -0.019 1.37 -0.018 1.29 Years of apprenticeship -0.032 0.34 -0.034 0.36 Chinese ethnicity 0.189 0.70 0.108 0.39 Other ethnicity (non-Kinh, non-Chinese) -0.226 0.96 -0.404 1.70* Former enterprise characteristics Operating from a fixed location 0.433 3.74*** 0.469 4.03*** 1993 enterprise age between 1.42 and 3.00 years 0.331 2.24** 0.343 2.31** 1993 enterprise age between 3.00 and 5.00 years 0.458 3.00*** 0.462 3.02*** 1993 enterprise age between 5.00 and 11.00 years 0.436 2.78*** 0.448 2.86*** 1993 enterprise age >11.00 years 0.759 4.65*** 0.736 4.51*** Fishery -0.955 5.19*** -0.949 5.19*** Food manufacturing -1.089 5.77*** -1.087 5.76*** Textiles manufacturing -0.790 4.27*** -0.792 4.29*** Other manufacturing -0.916 4.79*** -0.923 4.80*** Food/hotel commerce -0.345 1.87* -0.342 1.84* Transportation/communication -0.510 2.13** -0.544 2.26** Services -1.170 5.59*** -1.160 5.54*** Other industries -1.208 5.72*** -1.239 5.87*** Former scale of operation: Log(1993 enterprise income + 1) 0.251 5.45*** 0.259 5.61*** Log(1993 value of capital stock + 1) 0.066 3.73*** 0.062 3.47*** Log(1993 value of inventories + 1) 0.043 2.26** 0.046 2.41** Number of observations 2,376 2,368 Proportion affirmative 0.392 0.393 Average log-likelihood value -0.5908 -0.5926 Likelihood ratio test of slopes 374.22 367.21 *Significant at 10 percent level. **Significant at 5 percent level. ***Significant at 1 percent level. Note: In this table, the omitted categories against which comparisons are made are urban Central Highlands, an entrepreneur between 16 and 25 years of age of Kinh heritage, and an en- terprise operating from a variable location that has been in existence less than 1.42 years in the retail trade sector. Source: Authors' calculations using the 1993 and 1998 VLSSs. 116 Economic Growth, Poverty, and Household Welfare in Vietnam NFHEs were less likely to survive in the south of Vietnam, particularly in the Southeast region, which is dominated by Ho Chi Minh City. This is sur- prising at first sight because Ho Chi Minh City is the richest and most eco- nomically dynamic part of the country. Presumably the area is so dynamic that it is pulling people into wage employment, leaving fewer of them to op- erate NFHEs. Dynamism does not always have this effect, because firms are more likely to survive in rural areas where there is a nearby market (pre- sumably a sign of vigor, or at least of high population density). Of the firms surveyed in 1993, 39 percent survived in the sense that they were surveyed again in 1998. For enterprises operated by women, the esti- mated survival probability rises by a further 9 percentage points. This higher survival probability is not due to women being disproportionately concentrated in certain fields, because the equation holds other factors con- stant, including the activity in which the business operates (for example, food manufacturing, transportation, and so on). Enterprises operated by prime-age entrepreneurs were also more likely to survive, but it is surpris- ing that the survival rate was not influenced by the educational levels of the owners or by the owners' ethnicity. As is found in many other studies (Agarwal and Audretsch 2001; Goreski 1995), there is an important size effect. This is clear from table 4.10, which uses the estimated parameters from table 4.9 to compute the proba- bility that a firm survived from 1993 to 1998. Larger businesses, whether measured by the size of income or capital stock, were also more likely to still be in operation in 1998. If there is a lesson here, it might be that firms must grow to survive. The strongest predictor of future success is past success. Firms that had survived for three years or more by the start of the period were more likely to survive, a clear case of duration dependence. When combined with size, the effect is striking: A firm that was small and young in 1993 had a Table 4.10. Probability That a 1993 Enterprise Survived until 1998 Size of enterprise Enterprise age in 1993 Small Medium Large Between 0 and 1.42 years 0.21 0.28 0.38 Between 1.42 and 3.00 years 0.27 0.35 0.46 Between 3.00 and 5.00 years 0.30 0.38 0.49 Between 5.00 and 11.00 years 0.30 0.38 0.49 >11.00 years 0.36 0.45 0.56 Note: A "small" enterprise had a monthly enterprise income of Vietnam dong (D) 83,400 (US$6.78, using an exchange rate of US$1.00 = D 12,300), used D 10,000 (US$0.81) worth of cap- ital, and had no inventories. The income and capital stock of a "medium" enterprise were D 178,700 (US$14.53) and D 143,200 (US$11.64), but again there are no inventories. A "large" en- terprise had an income of D 376,100 (US$30.57) and a capital and inventory stock of D 771,100 (US$62.69) and D 40,200 (US$3.27), respectively. These values are chosen on the basis of the quartile values of the variables among the 1993 enterprises in panel households. Source: Based on calculations from the first column of table 4.9. Household Enterprises in Vietnam: Survival, Growth, and Living Standards 117 21 percent chance of surviving to 1998 (see table 4.10), and a large and old firm had a 56 percent probability of staying in business. The magnitude of this age effect is similar to the estimates reported by many other studies. Of course, this comparison assumes that other factors are held constant. How- ever, these other factors do matter. For example, compared with the retail sector (the excluded category among the industry dummy variables), enter- prises in the manufacturing and service sectors are more likely to be termi- nated, and enterprises near local markets or operating from a fixed location are more likely to survive. What Explains Startups? Between 1993 and 1998, households started 1,428 new NFHEs, which means that it is now possible to address the second key question, "What factors led households to start an NFHE between 1993 and 1998?" Conceptually, there are two distinct groups involved--those that oper- ated an enterprise in 1993 and started another business between 1993 and 1998 (box 3b in figure 4.1) and those that did not operate an NFHE in 1993 but had started one by 1998 (box 3c in figure 4.1). For households without an enterprise in 1993, the motives for starting a business may not be the same as for those that already had experience operating a business. To allow for this possible difference, separate logistic models are estimated for the two groups, as shown in table 4.11. The subsamples are statistically distinct, as witnessed by the p value of 0.0087 on the log-likelihood ratio test of para- meter equality. A familiar pattern emerges. Startup is less likely in the south, particu- larly the Mekong Delta region, and rural areas throughout Vietnam. If there is a secondary school nearby, fewer enterprises are expected to set up operations--presumably because the school reduces the availability of fam- ily labor. For new startups in inexperienced households, it greatly helps if the parents of the head were skilled manual workers or, perhaps, managers during their working lives. The same is no help in explaining whether households with established firms initiate another enterprise; but recall from table 4.5 that a history of proprietorship in the head's parental background was a strong determining factor in whether the household already operated an enterprise in 1993. Startups are also more likely if the household members are at least moderately well educated or have completed apprenticeships. There is a policy implication here, perhaps. Efforts to boost the level of worker skills appear to have an unexpected side effect of leading to the es- tablishment of new firms. Although a useful result, it is hardly surprising, because skilled and semiskilled workers such as carpenters and masons are often the ones who decide to go into business on their own. Performance of Nonfarm Household Enterprises over Time Survival is a minimalist measure of performance. It is at least as important to ask whether those firms that survived between 1993 and 1998 also a 5.44*** 3.63*** 0.99 0.90 2.20** 2.97*** 1.82* 1.97** 2.80*** 0.86 0.55 2.06** 1.33 1.50 2.84*** 0.03 2.00** 0.24 1.61 0.44 statistict without enterprise 10 1993 Households 2.140- 0.541- 0.477 0.312 1.038 1.109 0.787 0.676 0.696- 0.160- 0.154- 0.320 0.010- 0.228 1.123 0.006 0.022 0.064 0.1 0.305- Coefficient a with statistict 4.20*** 3.83*** 0.69 0.34 1.24 1.14 2.14** 1.28 1.91* 0.29 1.14 1.59 1.85* 1.97** 0.34 0.02 2.52** 1.84* 0.81 0.88 enterprise 1993 Households 1.475- 0.604- 0.233- 0.106- 0.785- 0.333 0.721 0.329 0.462- 0.059- 0.341 0.258 0.015 0.344 0.103- 0.004- 0.020 0.420 0.052- 0.521- Coefficient 1 statistict 7.55*** 5.10*** 1.01 1.25 1.39 3.13*** 3.78*** 2.53** 2.83*** 0.87 0.32 2.93*** 0.1 2.73*** 1.89* 0.01 3.40*** 1.70* 0.82 1.03 households All 16 1.922- 0.536- 0.265 0.279 0.492 0.695 0.967 0.505 0.475- 0.1- 0.064 0.321 0.001 0.302 0.442 0.002- 0.021 0.281 0.038 0.457- Model Coefficient Logistic A 1998" missing market is unknown started and community water rice Startup: 1993 local upper index of Coast index oadsr waterways of rice and of of transportation piped nearby of "Household Uplands Delta in wage price between Coast Delta and Central Enterprise River lower local price local schools extension quality quality public equencyfr .1 of of if if market index variable: 1 1 variables: eas enterprise Northern Red North Central Southeast Mekong and and and of 4.1 ar =, oducer =, electricity wage pr ableT ariableV newa cept of South Urban Urban Urban Urban Urban Urban rural secondary esence esence esence esence Dependent Inter Regional In vailabilityA Agricultural Pr Pr vailabilityA Use Pr Pr Local Dummy Local Dummy 118 1 0.19 0.81 1.84* 0.79 0.66 1.54 2.49** 1.88* 2.73*** 2.83*** 0.91 0.37 0.83 2.34** 0.37 0.00 0.1 1.83* 4.32*** 0.07 0.018 0.092- 0.235 0.107 0.097- 0.224- 0.364- 0.199 0.280 0.271 0.102 0.062 0.087 0.230 0.006 0.001 0.025 1.669 2.404 0.024 2,370 0.252 0.5261- 181.67 0.57 0.47 2.10** 1.24 0.18 0.53 0.28 1.19 2.65*** 0.87 0.12 1.87* 1.41 1.36 0.60 0.07 0.51 0.22 0.97 1.68* 11 0.051 0.057 0.276 0.173 0.028- 0.076 0.042- 0.143 0.306 0.095 0.014 0.272- 0.126 0.1 0.010 0.014- 0.099- 0.227- 0.354 0.597- 1,919 0.328 0.5994- 127.45 0.22 0.46 2.77*** 1.41 0.67 0.77 2.23** 2.38** 4.42*** 3.10*** 1.14 1.25 1.66* 2.58*** 0.73 0.16 0.75 1.05 3.19*** 1.17 11 0.014 0.037- 0.246 0.133 0.070- 0.077- 0.228- 0.183 0.326 0.214 0.089 0.135- 0.1 0.160 0.008 0.023 0.109- 0.707 0.950 0.288- 4,289 0.286 0.5661- 276.67 61.79 0.0087 VLSSs. 1998 and ence 1993 fer the older dif and schooling schooling enticeships using schooling schooling schooling 16 of head manual appr level. level. of of of training of level. value slopes sample ages years years years years years years of of cent ents cent cent 25­ 35­ 45­ 55­ 65­ years unknown farmer manager skilled years years years 12­ par schooling unknown per per per 16 26 36 46 56 >65 3­1 5­4 9­6 test test 5 1 calculations' 10 postsecondary technical completed of of 10 firmative at at at women observations af ratio ratio characteristics of ages ages ages ages ages ages with with with with with with with years schooling of log-likelihood Authors of occupation: occupation: occupation: occupation ce: Household Number Persons Persons Persons Persons Persons Persons Persons Persons Persons Persons Persons Persons Persons Characteristics verageA earsY oportion Major Major Major Major Number Pr verageA value *Significant **Significant ***Significant Sour Likelihood Likelihood p 119 120 Economic Growth, Poverty, and Household Welfare in Vietnam thrived. Are the most profitable NFHEs in 1993 still among the high- performing firms in 1998, or was 1993 just their lucky year? The simplest way to address this question is with the transition matrices that are presented in table 4.12. The columns of table 4.12, panel A, split the 1993 enterprises into quintiles according to their reported adjusted net rev- enue (that is, sales minus operating costs plus purchases of durable goods). The rows reflect where enterprises ended up--either in various income quintiles, as a terminated case, or as an enterprise that disappeared when the household attrited. Thus, each column adds up to 100 percent and con- tains one-fifth of the 1993 enterprise sample. Three conclusions follow from this table. First, there is clearly some sta- bility in the distribution of enterprise income. The best-performing enter- prises in 1993 are much more likely to be near the top in 1998, the middle- class remains in the middle, and the poor have difficulty rising from the bottom, although it is not impossible for them to do so. For most house- holds, the probability of building up a highly profitable enterprise in just a few years is very low. The second important finding is that enterprise termination is clearly re- lated to past enterprise performance, with the low performers being the most likely to go out of business. However, even in the highest quintiles, 40 percent or more of the enterprises do not survive until the fifth year. As noted, attrition is again seen as not strongly related to the recent perfor- mance of the enterprise. Part B of table 4.12 expands on this analysis by asking where the 1998 enterprises--distinguished by their quintile of 1998 performance--were in 1993. Here, the rows add up to 100 percent, and the columns describe the origin. The first five columns (with quintile headings) once again contain the panel enterprises and again demonstrate the stability in income that was seen in panel A. The next column provides evidence that start-up enter- prises are more likely to be among the poor performers, which is to be ex- pected given that they have not yet been winnowed out to the same degree as the more established firms. The "enterprise in new sample" column describes the position of the enterprises in households that were not part of the 1993 VLSS sample but were added in 1998 (see also table 4.6). Enter- prises in this subsample tended to perform relatively well. Table 4.13 goes a step further and asks what the sources of growth in en- terprise net revenue (that is, sales less expenses) might be. Regressions that explain the level of enterprise income have appeared elsewhere, both for 1993 (Vijverberg 1998b) and for 1998 (Tran 2000).9 The average value of the proportional difference in income is 0.418, which means that the average enterprise collected 41.8 percent more income in 1998 than in 1993. The independent variables refer to conditions in 1993, so the regression attempts to find determinants of future income growth. The middle two columns ("without selectivity correction") of table 4.13 show the results of estimating an ordinary least squares regression on enterprises that were in- cluded in the panel; it is thus conditional on the enterprise surviving from of 683 684 684 684 683 No. obser- vations (%) otalT 100.00 100.00 100.00 100.00 100.00 in sample 19.62 22.51 29.97 37.57 41.43 Enterprises new up 60.18 51.46 39.18 31.87 24.45 Enterprises started Upper 1.25 3.04 6.62 559 10.91 25.40 40.97 1.091 0.72 100.00 Upper 1.02 2.49 5.41 8.92 20.79 income 559 upper 3.58 7.33 9.84 12.88 7.87 income 44.90 1.631 1.97 6.44 Middle- 100.00 Middle- upper 2.93 5.99 8.04 10.53 enterprise 559 enterprise 6.62 8.77 VLSSs. Middle 10.73 7.33 5.01 52.06 8.05 1.43 5.42 7.16 8.77 5.99 4.10 1998? 100.00 Middle 1993 in 1993 1998 Income of of 1993? and 559 in Low- middle 6.80 7.87 6.08 3.58 2.50 63.86 8.23 1.07 100.00 Low- 5.56 6.43 4.97 2.92 2.05 1993 Quintile enterprises middle Quintile the Enterprise 1993 enterprises using in Low the 6.44 4.83 4.47 2.68 0.89 69.23 8.05 3.22 559 Low 5.27 3.95 3.65 2.19 0.73 to 100.00 1998 thee calculations' Dynamics happened opped were 1998 income: attrited dr 1998 income: of What of Wher Authors 4.12. A: observations (%) B: ce: of bleaT terminated Sour Quintile enterprise Panel Low Low-middle Middle Middle-upper Upper Enterprise Household Household otalT No. Quintile enterprise Panel Low Low-middle Middle Middle-upper Upper 121 0.13 statistict 0.64- 0.12- 0.04 1.05- 1.03- 1.88*- 0.74- 1.03- 0.36- 0.34- 0.84- 1.45- 1.49- 0.70- 1.18- 0.88 0.55 selectivity ection ith corr W 0.235 0.009 0.015 0.204 Parameter estimate 0.027- 0.004- 0.238- 0.287- 0.521- 0.302- 0.464- 0.143- 0.164- 0.447- 0.306- 0.474- 0.427- 0.654- 1)" + income statistict 2.77*** 3.06***- 0.96- 1.00- 2.16**- 2.68***- 4.02***- 3.20***- 1.00- 0.45 0.61 0.62- 1.32- 1.39- 0.26- 1.36- 1.41 0.68 enterprise selectivity ection corr 1993 ithout W Log(Annual 1.332 0.123 0.021 0.248 Income Parameter estimate - 0.053- 0.018- 0.121- 0.343- 0.433- 0.664- 0.540- 0.203- 0.085 0.127- 0.237- 0.352- 0.063- 0.331- 1) + Enterprise income in enterprise years years years Growth 3.00 5.00 1.001 of 1998 and and and location 1.42 3.00 5.00 years ce "Log(Annual 1) fixed 1.00 Determinants 1) + a between between between >1 + characteristics variable: inputs: characteristics omfr age age age age commer 4.13. manufacturing schooling years worker cept manufacturing manufacturing enterprises of <15 bleaT ariableV ln(Capital ln(Inventory Dependent Inter Enterprise Enterprise Operating Enterprise Enterprise Enterprise Enterprise Fishery Food extilesT Other Food/hotel ansportation/communicationrT Services Other Family earsY Age 122 1.31 0.90 0.78 0.91- 1.19- 0.38 0.90- 0.89- 0.40 0.36 0.30- 1.15 0.37 0.05- 0.37- 0.73 0.93 1.28- 1.58- 0.65 931 0.397 0.240 0.176 0.239- 0.569- 0.063 0.245- 0.252- 0.083 0.193 0.165- 0.868 0.264 0.027- 0.217- 0.384 0.465 0.457- 0.435- 0.698 0.088 1.49 0.64 0.63 0.90- 1.02- 0.08 1.10- 0.70- 1.46 0.12 0.70- 0.98 0.17- 0.37- 0.95- 0.37 0.69 1.42- 1.46- 10 1 931 0.228 0.1 0.134 0.232- 0.438- 0.01 0.286- 0.186- 0.187 0.061 0.332- 0.551 0.081- 0.176- 0.445- 0.127 0.294 0.494- 0.357- n.a. 0.087 VLSSs. 1998 and 1993 the market community water using local years years years years of oadsr level. level. level. 35 45 55 65 Coast nearby of piped term cent in cent cent and and and and Uplands Delta and per per per Coast Delta 1 25 35 45 55 Central equencyfr 10 5 calculations' quality ection at at at non-Chinese River market years characteristics and of and corr lambdas' observations of applicable. Authors Northern Red North Central Southeast Mekong electricity Not ce: between between between between >65 of esence esence esence n.a. *Significant **Significant ***Significant Sour 2 Age Age Age Age Age Female Chinese Non-Kinh, Regional South Urban Urban Urban Urban Urban Urban Pr Pr Pr Use Selectivity Heckman R Number 123 124 Economic Growth, Poverty, and Household Welfare in Vietnam 1993 to 1998. This does not, however, reflect the experience of all firms, be- cause more than 60 percent of firms that existed in 1993 were no longer in ex- istence in 1998 (that is, they had no profit in 1998). The two right-hand columns ("with selectivity correction") show estimates that, in principle, apply to all firms, using a Heckman adjustment (which means, first, esti- mate a probit regression of a model that tries to explain which enterprises survive, then use the conditional mean of the disturbance term, also called "Heckman's lambda," as an additional explanatory variable in the initial regression).10 The regression models do not have much explanatory power: The R2 val- ues are around 0.088. Thus, less than 10 percent of the variation in enterprise growth is explained by the model. This is in line with previous research showing that regression models of enterprise earnings leave most of the variation unexplained (for example, Tran [2000], Vijverberg [1998a, 1998b]). Because the dependent variable here refers to the difference in income be- tween two periods, the noise that one typically must deal with in enterprise earnings models is essentially doubled. Furthermore, whereas there are around 3,000 enterprises in each annual sample (see table 4.6), the require- ment that income be observed in both 1993 and 1998 lowers the sample size to only 931 enterprises. This further reduces the precision of the parameter estimates. All this suggests that a more adequate answer about the determi- nants of enterprise income growth can be derived only from much larger datasets. A number of interesting conclusions emerge from these estimates, al- though they are tentative, given the low levels of statistical significance. First, the size of the enterprise, as measured by the capital and inventory stocks, has little impact on enterprise income growth. Second, the youngest enterprises seem to grow the fastest, although this should be seen more as a learning effect than as an inherent, long-term productivity determinant. Third, the highest income growth rates are in retail trade (the excluded cat- egory among the market sectors). Fourth, there is a hint that educated and prime-age workers generate more growth. Differences across regions are minor, and the presence of markets appears to help. Conclusions Almost one-quarter of all adults worked in NFHEs in 1998, typically in com- bination with farming or another occupation. About 1 worker in 10 relied on NFHEs as his or her sole source of earnings. These averages hide more than they reveal, because participation in an NFHE is strongly related to living standards: Just 35 percent of chronically poor households operated such an enterprise in 1998, compared with 55 percent of solidly affluent households. It is difficult to identify the direction of causality, but it is probably bi- directional. Some evidence can be found to support the finding that operating an enterprise leads to affluence: Those households that jumped at least two expenditure quintiles between 1993 and 1998 ("shooting stars") began poor Household Enterprises in Vietnam: Survival, Growth, and Living Standards 125 and ended up relatively rich; they also were more likely to be operating an enterprise in 1998 than they were in 1993. Conversely, households whose relative expenditure level fell sharply ("sinking stones") were less likely to operate a business in 1998 than in 1993. To the extent that operating a busi- ness boosts a household's standard of living, it makes sense to encourage the establishment of such enterprises if the goal is faster economic growth. But what determines who operates a business? A formal analysis shows that geography matters, although perhaps not in the way that would be expected. Households in urban areas are more likely to engage in self- employment, but this effect is relatively weak in Ho Chi Minh City. Family history is also important, and the children of proprietors are much more likely to be proprietors themselves. Education helps, but only up to a point, and university graduates are less likely to operate a family enterprise than are those with just a high school diploma.11 Perhaps more interesting is the information on enterprise survival and formation. There is little published work on this subject, primarily because household survey data do not usually allow for the construction of the req- uisite panel of enterprises. NFHEs were found to be less likely to survive be- tween 1993 and 1998 in southern Vietnam, particularly in and around Ho Chi Minh City, than were NFHEs in the north. Older and larger firms were more than twice as likely to survive during this period as their smaller, younger peers. Startups were less common in the south of Vietnam, but were more common in households in which there was a skilled manual worker. An interesting pattern emerges from the analysis. As one moves from poor rural areas, through middle-income cities, to the most affluent part of the country (Ho Chi Minh City), the importance of NFHEs first rises and then falls. In poor areas, there is often a lack of education, credit, and effec- tive demand for the products of household enterprises. In rich areas, there are better alternatives to family business, typically in the form of wage labor. NFHEs thus play an important role in the period of transition, when agri- culture is declining in importance but before the formal industrial sector and service sector are large enough to take up all of the slack. For lack of data, this analysis of household enterprises in Vietnam has not been able to determine the impact of investment climate factors such as credit availability, regulations, permit requirements, health and safety in- spections, harassment by government officials, crime, and so forth. The VLSS did include (but only in rural areas) a number of relevant factors such as road and waterways infrastructure, community-level use rates of electric- ity and piped water, availability of public transportation, agricultural exten- sion services, and access to marketplaces. For enterprise startup, infrastruc- ture, availability of utilities, and market access matter. Survival is clearly related only to market access. Income growth does not clearly depend on any of these variables, but then again, income growth is such a noisy vari- able that the empirical model in this study was not able to explain much of its variation at all. Overall, therefore, the impact of investment climate vari- ables is still an open question. 126 Economic Growth, Poverty, and Household Welfare in Vietnam As Vietnam seeks to double GDP over the decade ahead, what role will NFHEs play? This chapter's findings are not particularly encouraging. The number of enterprise terminations is high, at 60 percent between 1993 and 1998. During the same period, the proportion of adults working in NFHEs fell, as did the proportion of households with such an enterprise. The growth in NFHE sales, expenditures, and income lagged behind GDP growth. This is not to argue that NFHEs should be neglected, but rather that, based on recent history, NFHEs play only a modest supporting role in fos- tering rapid economic growth in Vietnam. These findings should be qualified by noting that the economic environ- ment surrounding the private sector enterprises changed after the VLSS data were collected. Household enterprises can register quite easily now: They are required to file only the name and address of the business owner, the location of the business, the line of operation of the business, and the amount of business capital (Phan 2000a). Rural enterprises receive more support than before in access to credit, assistance with marketing, and fa- vorable tax treatment (Nguyen 2000). It is quite possible that these policies induce capable entrepreneurs to enter the private sector, but it still appears that private (household) enterprises start up with only one-third of the cap- ital that typical enterprises (such as limited companies, joint stock compa- nies, partnerships, or state-owned enterprises) begin with.12 Of course, it may be that, for purely financial reasons, successful NFHEs reregister under a more protected organizational form (Phan 2000b), which the VLSS does not capture. But this was not yet an issue when the VLSS data were col- lected. If the only thrust of the new policy direction lies in the facilitation of enterprise registration, it is quite likely that the main conclusions of this study are still valid under the new economic conditions, because such poli- cies do not address the long-term survival and success of small enterprises. However, given the rural policy initiative, it is certainly worth reexamining the issues with new data in the future. Appendix 4A Constructing Community Variables A number of rural infrastructure indexes are constructed from the rural community questionnaires. Here each index is defined; comments follow each. Road index = 1 - Km. to nearest road 10 × 1 - Months with impassable road . 12 This measures the availability of a viable road system. Distances greater than 10 are truncated at 10. The index declines if the nearest road that a car can travel on is farther away or impassable for longer periods.13 The Household Enterprises in Vietnam: Survival, Growth, and Living Standards 127 expectation is that a higher value of the road index, by opening up opportu- nities for business, will be associated with more involvement in NFHEs. Waterways index = 1 - Km. to nearest waterway 10 This measure is available only for the 1998 sample and is computed only if the respondent indicated that waterways were an important means of transportation for the community. Whenever the distance exceeds 10 kilo- meters or waterways are not deemed important, the index takes on a value of zero. Public transport index = 1 - Km. to nearest train, bus, or water transport 50 × Daily frequency The distance is truncated at 50 kilometers. The index is an indicator of connectedness. Daily market index = 1 - Distance to nearest daily market 72 Periodic market index = 1 - Distance to nearest periodic market 50 × Daily frequency For both market indexes, the truncation points are chosen according to values indicated in the survey. The daily frequency is a proportion, so a mar- ket that operates once a week has a daily frequency of 1/7. It is hypothesized that the presence of frequently operating markets enhances the viability of NFHEs. School index = 2 - Km. to lower secondary school - Km. to upper secondary school 10 10 Again, the distances are truncated at 10. A commune with both a lower and an upper secondary school in its center would have a school index value of 2. A larger school index may reflect higher levels of educational attain- ment locally, which should enhance enterprise performance. A larger school index also means that children stay at school longer, reducing the supply of labor. Agricultural extension index = 1 - Distance to nearest extension center + Number of extension visits per year . 50 40 128 Economic Growth, Poverty, and Household Welfare in Vietnam The truncation points are once again selected on the basis of values in the sample. The visits component on the second line contributes a maximum value of one to the index. Thus, the maximum possible value for this index equals two. Utilities index (1998) = (Proportion of households using electricity × Usability factor) The usability factor measures the proportion of the day that there is no utility outage. It is assumed that if the respondent indicates that outages occur, a daily outage lasts two hours on average or that weekly (or monthly) outages happen twice per week (or month) for two hours at a time. Utilities index (1993) = (Proportion of households with electricity + Proportion of households with piped water) The 1993 questionnaire did not provide information on electricity out- ages, so this simpler measure is used. The proportions take on values of zero, one-third, and two-thirds, depending on whether no, a few, or most households have the specified access. The objective of this index is to measure productive opportunities offered to the enterprise through access to electricity and water. Higher values of this index should benefit the enterprise. Wage index = (Average male and female wages for agricultural tasks) For a substantial number of communities in the 1998 survey, this average could not be computed with a reasonable level of confidence; in these cases, a dummy variable is added. The effect of this index on NFHE employment is ambiguous; higher wages imply that there are good alternative sources of income, making it less attractive to operate a business and more expensive to hire workers. Higher wages indicate greater affluence, however, and a higher demand for small-business services such as shops and restaurants. Notes The authors thank Tran Quoc Trung for his assistance in providing information about the enterprise laws that were instituted in 2000, and Dwayne Benjamin for sharing important data that he developed with Loren Brandt. Furthermore, the au- thors appreciate the comments provided by Paul Glewwe, as well as the many sug- gestions offered by the participants of the conference on "Economic Growth and Household Welfare: Policy Lessons from Vietnam," held in Hanoi, May 16­18, 2001. Each has contributed significantly to this chapter. 1. The 1992­93 survey spanned a full year, starting in October 1992; the 1997­98 survey began in December 1997 and also lasted a year. For brevity's sake, reference is made to the surveys as the 1993 VLSS and 1998 VLSS, respectively. 2. The figures in table 4.1 come from section 4A of the VLSS questionnaires, which asks whether someone was working in an NFHE. It would have been prefer- able to provide a breakdown of the hours worked, but unfortunately the relevant sections of the 1993 and 1998 questionnaires are not strictly comparable on this Household Enterprises in Vietnam: Survival, Growth, and Living Standards 129 matter. However, in 1993, the two breakdowns--by hours and by participation-- give broadly similar results (see Vijverberg [1998a]). 3. Here, "ethnic minority" is taken to refer to ethnic groups other than Kinh or Hoa (Chinese). 4. The official headcount poverty rate was 55 percent in 1993 and 37 percent in 1998 (General Statistical Office 2000). 5. The coefficients on the urban or region dummy variables compare these areas with a baseline rural region with zero values for all the rural indexes (includ- ing the wage dummy). Using the average values for rural areas, one would find that the baseline parameter for a "typical" rural area would be ­0.031. This is the number with which (for instance) the urban Red River Delta figure of 0.552 should be compared. 6. There have been several living standards surveys with a rolling panel design, most notably in Côte d'Ivoire and Ghana (Glewwe and Jacoby 2000). That is, one- half of the households in one year were visited again in the following year. There has been no attempt to create a panel of enterprises from the household panel informa- tion, as far as is known as of this writing. 7. Because the distribution of the financial performance variables is so highly skewed, the mean values are extremely sensitive to outliers and difficult to compare over time. Therefore, the table also reports median values, which are known to be less sensitive. 8. Before taking the logarithm, a value of one is added to the household's total enterprise income because some households report zero incomes. This transformation has little impact on the measurement of the effect of enterprise income on attrition. 9. The dependent variable is the difference in the natural logarithm of enter- prise income, which gives the proportional difference in income. However, because of the zero-valued incomes that a few enterprises report, a value of one has been added to the argument under the log function, so the dependent variable measures the proportional change relative to (enterprise income + one). Income values are ex- pressed in thousands of dong, measured in 1998 prices, and deflated for differences in prices across regions and survey months. 10. As is well known, it is highly recommended that the first-stage probit analysis incorporate some variables that are unique to the selection process and are not part of the explanatory variable set that is used in the second stage. This helps identify the explanatory influence of the added Heckman's lambda. In this case, the first-stage probit equation is the survival model reported in table 4.9 (estimated with probit instead of logit in line with the standard selectivity correction protocol). The identifying first-stage variables are availability of public transportation, local wage index, local producer price of rice, and the dummy variables indicating whether the latter two variables are missing (all pertaining to 1993 community characteristics). None of these are theorized to have a direct impact on the growth in enterprise income. Unfortunately, as was shown in table 4.9, they also lack a strong impact on enterprise survival. As a result, adding the Heckman's lambda variable to the model raises the degree of multicollinearity among the explanatory variables in a regression equation that already has low explanatory power. This is one more reason why the two right-hand columns show low t values. 11. As discussed in Edmonds and Turk (2004), the presence of NFHEs affects the likelihood of child labor in the household. However, our analysis shows that the impact of the presence of children on the incidence and performance of household enterprises is, to the degree analyzed, negligible. 130 Economic Growth, Poverty, and Household Welfare in Vietnam 12. Ministry of Planning and Investment. 2001. "Data on Registered Enterprises in 2000; Data on Registered Enterprises in the First Half of 2001." Unpublished tables. Hanoi. 13. The 1993 community survey did not specify that the road must be accessible by car. Bibliography Agarwal, R., and D. B. Audretsch. 2001. "Does Entry Size Matter? The Impact of the Life Cycle and Technology on Firm Survival." Journal of Industrial Economics 49(1): 21­43. Asia Pulse. 2001. "Vietnam's Private Sector Urged to Bridge Investment Gap." June 13. Hanoi. Audretsch, David B. 1995. "Innovation, Growth and Survival." International Journal of Industrial Organization 13(4): 441­59. Audretsch, David B., Patrick Houweling, and A. Roy Thurik. 2000. "Firm Survival in the Netherlands." Review of Industrial Organization 16(1): 1­11. Baldwin, John R., and Paul R. Gorecki. 1991. "Firm Entry and Exit in the Canadian Manufacturing Sector, 1970­1982." Canadian Journal of Economics 24(2): 300­23. Benjamin, Dwayne, and Loren Brandt. 2004. "Agriculture and Income Dis- tribution in Rural Vietnam under Economic Reforms: A Tale of Two Regions." In P. Glewwe, N. Agrawal, and D. Dollar, eds., Economic Growth, Poverty, and Household Welfare in Vietnam. Washington, D.C.: World Bank. Cabal, M. 1995. "Growth, Appearances and Disappearances of Micro and Small Enterprises in the Dominican Republic." Ph.D. diss. Michigan State University, Department of Agricultural Economics. Daniels, Lisa. 1995. "Entry, Exit and Growth among Small-Scale Enterprises in Zimbabwe." Ph.D. diss. Michigan State University Department of Agricultural Economics. Economist Intelligence Unit. 2002. Vietnam: Country Report. London. Edmonds, Eric, and Carrie Turk. 2004. "Child Labor in Transition in Vietnam." In P. Glewwe, N. Agrawal, and D. Dollar, eds., Economic Growth, Poverty, and Household Welfare in Vietnam. Washington, D.C.: World Bank. General Statistical Office. 2000. Vietnam Living Standards Survey 1997­98. Hanoi: Statistical Publishing House. Glewwe, Paul, and Hanan Jacoby. 2000. "Recommendations for Collecting Panel Data." In Margaret Grosh and Paul Glewwe, eds., Designing Household Enterprises in Vietnam: Survival, Growth, and Living Standards 131 Household Survey Questionnaires for Developing Countries: Lessons from 15 Years of the Living Standards Measurement Study. Washington, D.C.: World Bank. Goreski, P. A. 1995. "What Do We Know about Entry?" International Journal of Industrial Organization 13(4): 421­40. Haughton, Dominique, Jonathan Haughton, Le Thi Thanh Loan, and Nguyen Phong. 2001. "Shooting Stars and Sinking Stones." In Dominique Haughton, Jonathan Haughton, and Phong Nguyen, eds., Living Standards During an Economic Boom: The Case of Vietnam. Hanoi: United Nations Development Programme/Statistical Publishing House. Haughton, Jonathan. 2000. "Ten Puzzles and Surprises: Economic and Social Change in Vietnam, 1993­1998." Comparative Economic Studies (winter): 67­92. Littunen, H. 2000. "Networks and Local Environmental Characteristics in the Survival of New Firms." Small Business Economics 15(1): 59­71. McKinley, Catherine. 2001. "Government Must Help Private Companies Raise Competitiveness." Dow Jones Newswires. June 8. McPherson, Michael A. 1995. "The Hazards of Small Firms in Southern Africa." Journal of Development Studies 32(1): 31­54. Mead, Donald C., and Carl Liedholm. 1998. "The Dynamics of Micro and Small Enterprises in Developing Countries." World Development 26(1): 61­74. Nguyen Cong Tan. 2000. "Decision No. 132/2000/QD-TTg of November 24, 2000 on a Number of Policies to Encourage the Development of Rural Trades." Official Gazette 48 (December 31): 4­8. Perkins, Dwight. 1994. "Industrialization." In David Dapice, Jonathan Haughton, and Dwight Perkins, eds., In Search of the Dragons' Trail: Economic Reform in Vietnam [Viet Nam cai cach kinh te cheo huong rong bay]. Hanoi: Political Publishing House [Nha Xuat Ban Chinh Tri Quoc Gia]. Phan Van Khai. 2000a. "Decree No. 02/2000/ND-CP of February 3, 2000 on Business Registration." Official Gazette 9 (March 8): 3­13. _____. 2000b. "Decree No. 03/2000/ND-CP of February 3, 2000 Guiding the Implementation of a Number of Articles of the Law on Enterprises." Official Gazette 9 (March 8): 13­29. Storey, D. J., and P. Wynarczyk. 1996. "The Survival and Nonsurvival of Micro Firms in the U.K." Review of Industrial Organization 11(2): 211­29. 132 Economic Growth, Poverty, and Household Welfare in Vietnam Tran Quoc Trung. 2000. "Determinants of Income from Rural Nonfarm Business Activities in Vietnam." Master's thesis. National Economics University, Masters Programme in Development Economics, Hanoi. Viet Nam Living Standards Survey (VNLSS). 1992­1993. Web site: www. worldbank.org/lsms/country/vn93/vn93bid.pdf. Vietnam Living Standards Survey (VLSS). 1997­98. Web site: www. worldbank.org/lsms/country/vn98/vn98bif.pdf. Vijverberg, Wim P. M. 1998a. "Schooling, Skills and Income from Nonfarm Self-Employment in Ghana." In Paul Glewwe, ed., The Economics of School Quality Investments in Developing Countries: An Empirical Study of Ghana. London: Macmillan. _____. 1998b. "Nonfarm Household Enterprises in Vietnam." In David Dollar, Paul Glewwe, and Jennie Litvack, eds., Household Welfare and Vietnam's Transition. Washington, D.C.: World Bank. World Bank. 2000. "Vietnam 2020: Entering the 21st Century." Report 21411-VN. Hanoi. _____. 2003. Poverty: Vietnam Development Report 2004. Hanoi. 5 Agriculture and Income Distribution in Rural Vietnam under Economic Reforms: A Tale of Two Regions Dwayne Benjamin and Loren Brandt The period between 1993 and 1998 marked a continuation of the reforms of Vietnam's rural sector that began in earnest in 1988 with the implementation of Resolution 10, Vietnam's own version of the Chinese Household Respon- sibility System. Over this five-year period, new policies were implemented that provided households with better and more secure land use rights, ex- panded domestic and international marketing opportunities, and relaxed input supply constraints. These reforms, especially the relaxation of trade restrictions on rice and fertilizer, reinforced the incentive-enhancing effects of the earlier decentralization of decisionmaking in 1988 to farm households and, in the process, influenced supply and marketing decisions in agricul- ture, incomes, and welfare. This chapter aims to provide a description of the main changes occurring in agriculture in Vietnam during this five-year window, especially as they pertain to the distribution of incomes in the countryside. In particular, the panel dimension of the Vietnam Living Standard Survey (VLSS)1 is exploited to track households as they adjusted to the changes in the agricul- tural sector precipitated by ongoing economic reforms. This includes an assortment of adjustments in agricultural production and marketing, as well as in consumption. The chapter's chief concern is the analysis of the regional dimensions of these changes and the differential effect of these reforms on growth and dis- tribution in northern and southern Vietnam.2 The focus will be on the impact of two main policy changes: first, the increase in the rice export quota and the significant increase in the price of rice, especially in the south; sec- ond, liberalization of the fertilizer market and the sharp drop in the price of fertilizer. Most of the changes in agriculture are expected to center on rice 133 134 Economic Growth, Poverty, and Household Welfare in Vietnam production, and the south is expected to gain more than the north. Differ- ences between the north and the south in terms of their historical compara- tive advantage in rice production suggest that these marketing reforms may have accelerated a return to earlier production patterns. In addition to the differences across regions, the focus here will be on the possibly uneven impact of the reforms on households within regions. In short, this chapter investigates the linkages between the price policy changes and the associ- ated (possible) efficiency and equity consequences in the countryside. In the first section, the major institutional reforms affecting the rural sec- tor over this period are highlighted, and then their potential implications for agriculture and incomes are discussed. This is followed by a section giving a brief look at the changing structure of household incomes and changes in income inequality over this period. The discussion documents the continu- ing important role of agriculture in rural incomes, important changes in income distribution, and contrasts between the north and the south. The next section examines the empirically observable institutional envi- ronment and explores changes in rice and other crop prices, as well as in fer- tilizer prices. The authors believe that these changes are at the center of the changes in Vietnam's rural sector. With this as background, the next section explores changes in rice production, consumption, and marketing. Ques- tions addressed in this section are: Which regions saw increases in the pro- duction and consumption of rice? Were there declines? Moreover, do these changes line up with the patterns of price adjustment? This section also pro- vides a detailed look at the impact of rising household incomes on the level and pattern of food demand. Next, these changes in rice farming are placed in the broader context of agricultural production, and changes in cropping patterns across Vietnam are described. Given the significant increase in rice production and farm output, more generally, this section also investigates the extent to which these increases can be explained by increases in inputs such as fertilizer, cropping intensity, and increased yields. Finally, in the section titled "Agri- culture and Inequality," the focus is on the distributional impacts of these changes, including a detailed examination of the linkages between rice mar- keting and income distribution, and on a summary of the role of agricultural incomes more generally as they pertain to income inequality. Throughout the chapter, north-south regional breakdowns are provided. Differences between urban and rural Vietnamese households are also pre- sented. Urban households are of some interest, not only because a small frac- tion of them is made up of farmers, but also because the welfare consequences of changes in agriculture may have opposite implications for urban and rural households. That said, to keep the dimension of the tables to a reasonable level, the focus is primarily on those households classified as rural. Institutional Changes and Potential Implications Several important institutional changes occurred between 1993 and 1998 that might be expected to affect agricultural incomes.3 Some of these Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 135 changes represented a continuation of the move toward market-based production, and others concerned (implicit) taxes and rice marketing policy: · Between 1992 and 1998, key agricultural markets were gradually liberalized. Relaxation of restrictions on rice exports was the most im- portant. The export quota was increased from less than 1 million met- ric tons in 1992 to 4.5 million metric tons by 1998. Similarly, there was a relaxation of internal barriers to trade in rice that had restricted the flow of rice from the south to the north. Especially important in this regard was Decree Number 140/TTg, implemented in March 1997, which lifted internal trade restrictions on rice and eliminated some licenses and controls on transports. The combination of export quotas and internal trade restrictions, which restricted the movement of rice from south to north, had severely depressed the price of rice in the south. For example, in 1995, the export value of a ton of rice was US$269, but farmers in the Mekong Delta received only US$205 per ton, yielding an implicit tax of more than 30 percent.4 Clearly, an in- crease in the rice export quota would be expected to increase the price of rice, especially in the south. · Fertilizer supply constraints were reduced, with new freedoms to im- port fertilizer. In 1991, central and provincial state-owned enterprises thatearnedforeignexchangewereallowedtoimportfertilizerdirectly. Vietnam does not have much of a domestic fertilizer industry and therefore is heavily dependent on imports. Over the 1993­98 period fertilizer imports tripled in quantity terms. Although there are no tar- iffs on imports, there may still be a wedge between the domestic and international prices of fertilizer. Still, a sharp drop in fertilizer prices because of the liberalization of this key input market is anticipated. · Resolution 5, passed in 1993, aimed to enhance households' land use rights. Tenure security was extended to 20 years for annuals cropland and 50 years for perennials. Households were extended rights to ex- change, transfer, lease, inherit, and mortgage land. A land titling process was also begun. In the long run, these changes can be ex- pected to affect investment incentives in agriculture, including irriga- tion, drainage, perennials, and so on, and they are also expected to af- fect the efficiency with which land is allocated across households. Although titling programs encountered a variety of administrative difficulties, by 1997 half of all land had been titled, affecting two- thirds of all households. Most of the changes occurred toward the end of the period in question; thus, this increase in property rights security will probably not have had enough time to be reflected in production behavior and output. · Improved development of market infrastructure and the continued integration of Vietnam with world agricultural markets would possi- bly have made it easier for households to start growing exportable cash crops. 136 Economic Growth, Poverty, and Household Welfare in Vietnam In summary, the most important changes in farmers' economic environ- ment concerned prices: increases in crop prices and reductions in fertilizer prices. Changes in these markets are likely to have affected households in a variety of ways. First, ignoring production adjustments, changes in the price of rice (and fertilizer) will have direct effects on the well-being of rice pro- ducers and consumers through their farm balance sheets. This impact is quite likely to vary across the different regions of Vietnam (urban versus rural, north versus south, rice surplus versus rice deficit) and across differ- ent parts of the income distribution. Second, households are expected to make production adjustments. It may no longer be worthwhile for some households to grow rice, or it may be more profitable for them to move into the production of cash crops. For other households, rice may be a more lu- crative crop and may merit even deeper investments and specialization. In- deed, one of the most interesting questions pertaining to rice market liberal- ization concerns the extent to which crop production adjusts across regions. Through trade, rice-deficit regions move into other crops in which farmers have a comparative advantage, while rice-surplus areas cultivate rice more intensively. The reduced cost of fertilizer may induce farmers to substitute chemical for organic fertilizer, possibly increasing yields. At the household level, increased commercialization means that households are more en- gaged in--and affected by--markets. An important question is how this increase in market development affects household welfare and efficiency. In addition to responses to the specific price changes that resulted from government policy in the 1993­98 period, other factors are expected to affect agriculture. First, a continued adjustment to the market reforms that were begun in the late 1980s is expected. The decentralization of household deci- sionmaking, which was at the center of the reforms, is anticipated to im- prove farm efficiencies and possibly permit greater specialization within farming. An important question concerns whether some of the initial suc- cess of the market reforms was maintained through the 1990s.5 Second, increases in urban and rural incomes can be expected to increase the demand for food, especially for crops other than rice. This may have en- couraged farmers to diversify their crop portfolios, especially encouraging them to move into more lucrative perennials. The movement of farmers out of rice also suggests the importance of the codevelopment of rice markets and commercialization, which permits some farmers to stop growing their own food. Finally, combined with economic development, economic reforms may have permitted a diminished role for agriculture, because households in- creasingly participate in nonagricultural pursuits.6 In assessing the impact of changes in agriculture, it is important to see how nonagricultural pursuits may offset the impact on household welfare of changes in agricultural income. This chapter's focus is on agricultural income--crop revenue minus expenses--and its place in the context of overall household income. Prelim- inary evidence on the bottom line of the possible impact of these changes is provided by Glewwe, Gragnolati, and Zaman (2002), who look at changes in consumption levels between 1993 and 1998. They document a sharp drop in Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 137 poverty, from 58.2 percent to 38.4 percent, which suggests that incomes rose significantly. Most of the decline in poverty that they find is concentrated in urban areas. However, among rural households, they find the most pro- nounced drops in poverty for households that live in the south, had more irrigated land, and experienced greater increases in rice productivity. This suggests that changes in rice and fertilizer prices, or at least increases in the returns to rice farming, may have been responsible for improved living stan- dards. Direct examination of agricultural incomes is necessary to draw the link between agriculture and living standards. Agriculture and the Structure of Income This section documents changes in the structure of household income and consumption from 1993 to 1998 using the 1993 and 1998 VLSSs.7 This serves two purposes. First, it enables any changes in the role of agriculture in the portfolio of household income to become apparent; second, it provides use- ful background information on patterns of changes in living standards. As households become wealthier, their demand for agricultural goods is ex- pected to change, with implications for agricultural prices and, possibly, cropping patterns. Throughout, this chapter focuses only on the "panel households"--those households that can be tracked across the two survey periods. The VLSS in 1998 was expected to include a total of 6,000 households, including a resur- vey of the 4,800 households originally surveyed in 1993. As typically hap- pens, it was not possible to recontact some of the original households, so only 4,306 households can be precisely followed between the two surveys. The analysis is therefore based on the 3,496 rural and 810 urban households panel (or longitudinal) households that could be accurately matched across the 1993 and 1998 surveys. To the extent that these households are not representative of the entire population, there may be some limitations on applying conclusions from this analysis to the entire country. However, similar tables over the entire set of nonpanel households have also been calculated, and basically the same results were obtained. By focusing on the panel, the changes in outcomes for a specific group of households can be tracked with some confidence that ini- tial conditions are being held constant. This will be particularly helpful when changes (as opposed to levels) of outcomes are studied. One caveat, however, is that the members of any panel dataset age in the panel, so some of the observed changes may be a function of aging, in addition to any changes in aggregate economic conditions. Levels of Income Table 5.1 shows mean incomes for households by various sources, partitioned into a variety of subsamples.8 A few conventions are worth noting. First, all values are expressed in terms of 1998 prices, using the otalT 10.8 5.1 0.41 0.42 0.27 0.25 850.9 568.3 948.9 826.1 8,1 7,284.7 1,280.4 3,636.0 1,616.3 1,454.3 1,490.3 1,328.4 5.5 0.46 0.48 0.29 0.27 Rural South 246.8 851.8 997.2 8,779.3 7,782.1 1,355.7 1,548.2 3,779.7 1,629.7 1,449.7 1,612.0 1,432.0 4.8 0.37 0.37 0.24 0.22 475.1 807.8 698.7 North 7,613.0 6,914.3 1,081.1 3,529.1 1,021.2 1,604.9 1,458.3 1,387.2 1,240.6 1993 otalT 546.2 74.3 5.1 0.50 0.52 0.34 0.29 15,852.7 12,621.2 3,517.4 4,170.3 4,313.1 3,231.5 3,174.8 2,538.4 2,690.2 2,053.8 5.7 0.50 0.52 0.34 0.29 414.8 Urban South 17,993.6 14,021.8 4,261.8 4,876.4 29.2- 4,497.9 3,971.8 3,234.0 2,533.4 2,789.0 2,088.4 4.2 0.49 0.51 0.32 0.29 733.5 221.7 North Source 12,801.5 10,625.1 2,456.5 3,163.8 4,049.6 2,176.3 3,062.9 2,547.9 2,503.5 1,988.6 by (1) (2) Incomes (1) (2) (1) (2) (1) (2) income income consumption consumption Household (1) (2) dong) size income income consumption consumption durables capita capita capita capita 5.1. business income income per per per per components capita capita capita capita ableT ages (thousand Indicator otalT otalT By W Family Farming Livestock Other Services, Household Per Per Per Per Gini, Gini, Gini, Gini, 138 4.8 0.41 0.41 0.28 0.26 the otalT in 14,057.6 12,353.8 1,794.3 2,679.7 4,991.8 1,262.1 1,625.9 1,703.8 2,922.0 2,569.5 2,361.6 2,009.1 consump- listed weightede for ar ponents holds income) 5.2 0.42 0.42 0.29 0.27 ion com and Rural South 17,092.1 15,186.6 2,784.3 3,453.7 6,425.1 1,032.4 1,491.1 1,905.5 3,305.7 2,939.1 2,530.0 2,163.4 all on ze.si omfr distinct same income The (consumpti household 4.6 0.39 0.39 0.27 0.25 by total North 1,798.41 10,244.7 1,057.2 2,103.4 3,924.6 1,433.2 1,726.4 1,553.6 2,597.1 2,256.5 2,219.1 1,878.5 is housing. variables (1) weighted e capita 1998 ar Income -occupied per and 4.5 0.43 0.44 0.34 0.30 the otalT 818.7 267.0 owner for 27,086.1 21,977.0 6,058.1 8,808.8 6,024.5 5,109.0 6,026.1 4,892.5 4,566.2 3,432.5 indexes. income and price means The positive durables 4.9 0.44 0.43 0.35 0.29 monthly with 654.3 136.3 15.5 omfr Urban South 29,598.6 23,845.0 6,759.5 1,055.11 5,239.9 5,753.6 6,1 4,931.5 4,685.6 3,501.5 and households. income egionalr households panel by the those VLSSs. 4.0 0.42 0.44 0.32 0.30 imputed 453.3 over over 1998 North 23,505.3 19,314.9 5,058.5 5,607.5 1,052.9 7,142.6 4,190.4 5,871.9 4,825.1 4,360.3 3,313.5 deflated e and ar excluding and e, calculatede 1993 calculated e ar the ar (1) (2) ency measur omfr curr Means (1) (2) (1) (2) same ficients 1998 (2). in the coef (1) (2) e is income income consumption consumption ar (2) Gini calculations (1) (2) size size. income income consumption consumption capita capita capita capita values consumption durables income business Authors' All income income and per per per per and ce: components capita capita capita capita (1) household Indicator otalT otalT ages Note: Sour By W Family Farming Livestock Other Services, Household Per Per Per Per Gini, Gini, Gini, Gini, table, tion by 139 140 Economic Growth, Poverty, and Household Welfare in Vietnam recommended 1.456 deflator. Second, when discussing incomes for use in welfare analysis (as in table 5.1), the monthly and regional consumer price deflators are also used, so that regional incomes are adjusted for regional differences in purchasing power.9 Third, two measures of income (and con- sumption) are presented, with and without the imputed value of capital ser- vices. Imputed capital services, which measure the flow of services from durables and owner-occupied housing, are very difficult to estimate, and the authors would like to evaluate the robustness of their conclusions to this "made-up," but important, component of income. Finally, the per capita household-level variables are also weighted by household size to reflect individual-level averages. Two main conclusions follow from table 5.1. First, average incomes increased dramatically over this period, and, second, in- come growth was highest in the rural south and lowest in the rural north. The first category in table 5.1 to assess is overall household incomes from all sources (including capital services). Although urban households are not the focus of this chapter, it is still informative to examine urban income levels and growth rates, if only as a benchmark for the rural households. In 1993, urban households in the north had average income of dong (D) 12,802,000 compared with D 17,994,000 in the south, implying a ratio of 0.71 for north to south.10 Urban households in the north experienced income growth of 84 percent between 1993 and 1998, corresponding to average an- nual (compounded) growth of 13 percent. Urban household income growth was lower in the south, at 65 percent (10.5 percent a year) for this period. By comparison, rural households in the north had average income of D 7,613,000 in 1993. This was 59 percent as high as urban households in the north and 87 percent as high as rural dwellers in the south. Rural household incomes grew by 55 percent (9.2 percent a year) in the north and by an as- tounding 95 percent (14.3 percent a year) in the south. By 1998, the north- south rural income ratio had fallen to 0.69. This pattern of growth is similar to using "Total income (2)" in table 5.1, which excludes housing and durables services. The main difference that emerges in using this alternative income measure is that the rural-urban gap is slightly attenuated, reflecting the higher value of housing services in cities. These numbers do not take into account the differences in household sizes across regions and periods. In addition, household size fell in all re- gions, but not uniformly. For example, household size fell from 4.2 to 4.0 in the urban north, but from 5.7 to 4.9 in the urban south, from 4.8 to 4.6 in the rural north, and from 5.5 to 5.2 in the rural south. These differences and changes in household size are reflected in the per capita income figures, and they generate important nuances to the conclusions based on household in- come. First, per capita income differences are much smaller across regions. The ratio of northern to southern urban incomes is 0.95 in 1993 and 0.96 in 1998. Per capita income growth in the north and south is 92 percent (13.9 per- cent a year) and 89 percent (13.6 percent a year), respectively. At the same time, however, the gap between north and south is more pronounced in rural Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 141 areas: growth was 62 percent (10.1 percent a year) in the north and 103 per- cent (15.2 percent a year) in the south. The ratio of northern to southern rural per capita incomes fell from 0.99 in 1993 to 0.79 in 1998. Finally, adjustments for household size slightly widen the urban-rural gap. In addition to changes in levels, there were changes in the composition of incomes over this period. This can be seen directly in the numbers reported in table 5.1, as well as in the implied income shares illustrated in figure 5.1. Clearly, the levels of all types of income rose in all regions. In urban areas, wages and family-run businesses contribute the most income; in 1993 they together accounted for almost 45 percent of incomes in the north and 50 per- cent of incomes in the south. In 1998, in the south, this share rises to 60 per- cent. Agriculture is not a very important income source in urban areas, though the category "other income" is especially important. This category comprises remittances, government transfers, interest, and rental income. Government transfers, such as military pensions, are higher in the north. Turning to rural areas, agriculture is clearly the most important category. In the north, the share of agriculture falls from 46 percent in 1993 to 33 percent in 1998, while in the south, the share of agriculture falls from 43 percent to 38 percent. Animal husbandry grew slightly in importance and is a higher share of income in the north. Finally, income from wages and family-run businesses increased in importance in both the south and north, where they now make up about one-third of household income. In summary, broadly defined agricultural sources of income remain very important in the Vietnamese countryside, though agriculture was a slightly smaller share of income in 1998 than in 1993. Finally, mean per capita consumption levels are shown for each region. The consumption numbers paint a slightly different picture from income. Certainly, they indicate significant improvements in living standards, espe- cially in the cities, and suggest a sharp drop in poverty.11 The ratios of 1998 to 1993 consumption range from 1.57 in the rural south to 1.74 in the urban north. But unlike the income figures, the consumption numbers do not sug- gest a widening gap between regions: Consumption in urban areas in the north increased from 90 percent to 93 percent of the levels in the south (similar to the per capita income figures), while consumption in the rural north was 83 percent of that in the south in both years (in contrast to the in- come results). One possible explanation for this finding is that the income numbers are "noisier" and are more sensitive to year-to-year variation. In fact, evidence is presented below that suggests that incomes were unusu- ally high in the north in 1993, which would yield lower growth rates in subsequent years (mean reversion). However, it is also possible that con- sumption responds more slowly than income to changes in economic fun- damentals and increased saving, not consumption, is the response to higher incomes. If this is the case, then the income numbers may point ac- curately to long-run trends in living standards that are not yet reflected in annual consumption.12 142 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 5.1. Composition of Income, Vietnam a. North Share of income 100 80 60 40 20 0 Urban-93 Urban-98 Rural-93 Rural-98 Capital services Other Animal husbandry Farming Family non-ag businesses Wages b. South Share of income 100 80 60 40 20 0 Urban-93 Urban-98 Rural-93 Rural-98 Capital services Other Animal husbandry Farming Family non-ag businesses Wages Note: Based on data reported in table 5.1. Source: Authors' calculations from 1992­93 and 1997­98 VLSSs. Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 143 Income Inequality The Gini coefficients for the same subsamples are also presented in table 5.1. The income-based Gini coefficients suggest that income inequality dropped significantly in urban areas, declining from 0.50 to 0.43. This decline arises primarily from reductions in income inequality within the two regions, be- cause there was only a small closing of the gap between regions (which was small to begin with). The consumption-based Gini coefficients paint a slightly different picture: They suggest that there has been no change in urban in- equality. The income- and consumption-based Gini coefficients could both be correct if savings differences between rich and poor households are narrow- ing. Alternatively, the different picture could arise from some differential degree-of-measurement error in the income and consumption data. The rural Gini coefficients also show differences in the income- and consumption-based measures. The income-based Gini coefficients show no change in overall inequality, with a Gini of 0.41 in both years, but this hides a steep drop in income inequality in the south (0.46 to 0.42) and a slight in- crease in the north (from 0.37 to 0.39), as well as the higher between-region gap in 1998. Clearly, pooling of rural households from the north and south hides considerable variation in the evolution of the income distribution. The consumption-based inequality measures, however, suggest slight increases in inequality, both within the regions and between the north and south. Agricultural Prices One of the primary avenues by which changes in agricultural policy are ex- pected to affect household behavior and welfare is through changes in agri- cultural prices. Probably the most important set of policy changes concerns rice pricing. During most of the early reform period (including 1992­93), rice prices were not directly set by the government, but there were many mar- keting restrictions. In particular, as part of a broader policy of promoting rice self-sufficiency and possibly protecting northern farmers, there were strongly binding restrictions on the export of rice, and the movement of rice from south to north was impeded. These restrictions had the effect of se- verely distorting producer and consumer rice prices. More recently (includ- ing before 1997), the export quota was increased; thus, the implicit export tax was reduced. Furthermore, internal trade restrictions have been reduced, es- pecially since 1996, and private traders have become a more important part of the rice marketing system. Nonrice prices are also expected to have changed over this period, as a possible consequence of changing food de- mand patterns (see "Food Demand Patterns" section) and increased inte- gration of the Vietnamese economy into world agricultural markets. The VLSSs can be used--albeit indirectly--to explore the impact of lib- eralization on prices. This chapter explores two dimensions. First, evidence on the broad changes in crop prices is presented, looking separately at the north and south. For this exercise, the household and community parts of the VLSS are used. In the second exercise, the focus is on rice prices (unit 144 Economic Growth, Poverty, and Household Welfare in Vietnam values) and the extent to which rice prices have converged across the re- gions of Vietnam. Finally, the chapter looks at how rice producer and rice consumer prices changed over 1993­98. An Overview of Price Indexes for 1993 and 1998 Table 5.2 shows the results of the "big picture" of crop prices. In the first sec- tion of the table, crop prices (unit values) estimated from the cropping section of the household survey are used. Prices have been calculated on the basis of the revenue and quantity of each crop sold by households, and a Table 5.2. Selected Price Indexes for 1998 Relative to 1993, Rural Price index North South Crop producer prices Rice 1.62 1.83 Nonrice 1.97 2.26 All crops 1.83 2.05 Other staples 2.07 1.73 Vegetables 2.36 1.86 Annual industrial crops 1.45 1.74 Perennial industrial 2.63 3.19 Fruit 1.93 1.88 Selected foods (consumer) Rice 1.85 2.13 Pork 1.54 1.54 Beef 2.06 1.73 Chicken 1.61 1.60 Tofu 1.46 1.35 Cabbage 2.33 1.76 Tomatoes 2.94 1.73 Oranges 1.80 1.63 Bananas 1.74 1.66 Fertilizers Urea 1.13 1.14 Potassium sulfate 1.62 1.12 Phosphates 1.41 1.13 NPK 0.48 0.51 Note: NPK = nitrogen-potassium-phosphate compound fertilizer. Each index represents the price in the 1998 sample relative to the 1993 sample. The crop prices are calculated from unit values in the crop sales part of the household surveys. The indexes are based on weighted av- erages of price changes for the commodity group, where the base year weights are calculated in 1993. The crop unit values are calculated for all panel households. The constituent price ratios are calculated on the basis of sales-weighted average prices. The selected foods and fertilizer prices are taken from the community price surveys and are the simple ratio of average prices in the two VLSSs. Source: Authors' calculations from the 1993 and 1998 VLSSs. Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 145 price index for 1998 relative to 1993 has been created using the crop shares (of sales) from 1993 as base weights. In reading the numbers, recall that the overall consumer price index (CPI) for this period is 1.46, which serves as a useful benchmark. The two columns look separately at the north and south, because the degree of market integration between the regions is pertinent, as are the possibly different price signals being sent to farmers. Looking first at rice, it is clear that producer prices rose by factors of 1.62 and 1.83 in the north and south, respectively. Thus, the real price of rice (relative to the CPI) rose in both areas, with prices increasing more in the south. Nonrice crop prices increased even more, especially in the south. Almost all types of crop prices increased, but the increase was most pronounced for the ex- portable perennial industrial crops, especially in the south. The next section of table 5.2 corroborates these numbers with a selection of food prices drawn from the community questionnaires. Note that these are theoretical consumer prices, not realized producer prices (unit values), as in the first section.13 Here it is clear that consumer rice prices rose faster than the CPI and--at least with these numbers--also faster than producer prices. Consistent with the results for producer prices, rice prices also in- creased more in the south. The prices across the various goods confirm that food prices, and thus probably agricultural prices, have generally risen faster than the CPI. In the last section of the table, fertilizer prices are shown, again drawn from the community questionnaire. Fertilizer represents the largest compo- nent of farm input cash expenses, and this is especially true for rice. The fer- tilizer market, especially as it relates to imports, was increasingly liberalized over this period, and declines in fertilizer prices should occur concurrently with the increase in supply. Even if farmers had no response to fertilizer price changes, lower fertilizer costs would directly increase farm incomes. To the extent that lower prices also encouraged more use of fertilizer and en- abled higher yields, the lower prices are also expected to lead to higher crop output and revenues. The results emphatically demonstrate declining real prices of fertilizer. For urea, potassium sulfate, and phosphate-based fertil- izers, prices rose by less than the CPI (that is, real prices fell), especially in the south. Most dramatically, the nominal price of Nitrogen-Potassium- Phosphate (NPK) compound fertilizer fell by one-half in both the north and the south. The possible impact of these price changes on fertilizer use is explored in "Cropping Patterns" below. Rice Prices Table 5.3 examines rice prices more specifically, looking at the degree of market integration between north and south. First, the unit values of rice, both purchased and sold, are calculated, as estimates of the prices paid to farmers and by consumers. In table 5.3, consumer and producer prices of rice by region for 1993 and 1998 are documented. The prices are adjusted for overall changes in the CPI (that is, the 1993 prices are converted to 1998 146 Economic Growth, Poverty, and Household Welfare in Vietnam Table 5.3. Rice Prices across Time and over Space Consumer prices Producer prices Indicator 1993 1998 1993 1998 Prices in 1998 (thousand dong) Rural north 2.71 3.35 2.55 2.83 Rural south 2.58 3.45 1.92 2.43 Rural Vietnam 2.63 3.42 2.03 2.50 Urban north 2.91 3.29 n.a. n.a. Urban south 2.81 3.65 n.a. n.a. Urban Vietnam 2.85 3.50 n.a. n.a. ANOVA Variation of logs 48.40 78.42 108.83 52.29 Explained by region 7.09 4.43 28.78 5.19 (F statistic) (80.5) (30.2) (105.4) (41.01) R2 0.22 0.25 0.39 0.36 Unit value income elasticities (t values) Urban north 0.03 (3.5) 0.10 (6.0) n.a. n.a. Urban south 0.08 (8.8) 0.14 (11.1) n.a. n.a. Rural north 0.02 (2.0) 0.04 (3.4) n.a. n.a. Rural south 0.05 (6.0) 0.10 (10.9) n.a. n.a. Rural Vietnam n.a. Not applicable. Note: ANOVA = analysis of variance. All "prices" are really unit values, taken as the ratio of the value to the quantity of rice purchased or sold. Consumer prices are expressed as 1,000 dong per kilogram of purchased rice. The average unit values (prices) are weighted by the value of purchases. Prices for 1993 are expressed in 1998 values, inflated by 1.456. Producer prices are expressed as 1,000 dong per kilogram of sold rice (paddy converted to rice by a fac- tor of 0.66). The average unit values (prices) are weighted by the value of sales. The variation of logs is the total variation of log prices. The variation explained by region is the total explained variation (by region) in an ANOVA decomposition, including monthly dummies. The F statis- tic is that for the significance of the region effects, and the R2 is the fraction of total variation ex- plained by month and region. Income elasticities are the estimated coefficients from a regres- sion of log unit value on log per capita total expenditures and month dummies for each year and region. The t values are in parentheses. Source: Authors' calculations from the 1993 and 1998 VLSSs. currency by the factor 1.456). The paddy-to-rice prices are also converted, by a factor of 0.66. The first column shows mean consumer prices by region in 1993. Urban prices are (not surprisingly) higher than rural prices and are slightly higher in the north. An analysis of price variances (ANOVA) has been performed to see how much of the total variation can be accounted for by region. The total variation in log consumer prices is 48.4. Of this variation, 7.09 is explained by region (controlling for the month of survey). The F statistic on region is 80.5, so re- gion is an undeniably important predictor of rice consumer prices in 1993. The fraction of total variation in log prices explained by region--and Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 147 month--in 1993 is 22 percent. If rice markets become more integrated across regions, a reduction in price variability across regions in 1998 would be ex- pected, and this is what has occurred. First, real rice prices were significantly higher in 1998 than they were in 1993--30 percent higher in rural areas and 23 percent higher in urban areas. There is only a small gap between north and south (the south being higher), and urban prices remain slightly higher.As far as explaining price variation, while the total variation in log prices increased, both the absolute and relative variation explained by region declined, as would be expected as rice markets become more integrated across regions. Changes in the regional price structure are more dramatic for producer prices. First, note the north-south price divide in 1993: Prices were D 2,550 per kilogram in the north versus D 1,920 in the south. Producer prices are es- pecially low in the Mekong Delta, at D 1,860. This is exactly what would be expected, given the trading restrictions, which primarily affected Mekong Delta rice farmers. In 1993, they were receiving only 70 percent of the price for the same kilogram of rice as a farmer in the Red River Delta (in the north). In the ANOVA exercise, it is clear that 39 percent of the total variation in log producer prices is explained by region and month. Thus, producer prices vary much more across regions than do consumer prices. By 1998, most of this regional dimension disappears. Rice prices are still higher in the north (D 2,830 versus D 2,430), but the percentage gap is much smaller. Fur- thermore, the real price of rice rose across Vietnam, especially in the south, and in particular in the Mekong Delta. As the ANOVA shows, total variation in rice producer prices declined (from 108.83 to 52.29), and the portion of that variation that can be explained by region falls from 28.78 to 5.19. Thus, producer price dispersion has decreased, and region is a poorer predictor of rice prices--both of which are expected with improving market integration. It thus appears that the marketing reforms, which allowed increased exports internationally as well as domestically, have led to increases in rice prices, especially in the main rice-producing region of the Mekong Delta. An obvi- ous set of questions then follows as to who benefited or was hurt most by these price changes. Finally, a note of caution in interpreting unit values as "prices," espe- cially for consumers, is warranted. Consumers can choose to some extent how much to pay for their food, and, as incomes rise, they tend to substitute toward higher qualities, even for goods as standardized as rice. As incomes rise, some of the extra income is spent on more expensive rice, which ex- plains some of the increased dispersion in rice consumer prices. In the lower section of table 5.3, estimates of the price-expenditure elasticities for ordi- nary rice by region and survey year are given. These are estimated by the coefficients from a regression of the log unit values on log per capita house- hold expenditures, with controls for the month of the survey. As can be seen in table 5.3, the higher the average income levels, the more income elastic is the unit value of rice. For example, in 1993, southern urban dwellers had the highest elasticity of 0.08. By 1998, the urban elasticities had risen to 0.14 in the south and 0.1 in the north. Even in rural areas in the south, the 148 Economic Growth, Poverty, and Household Welfare in Vietnam expenditure elasticity is 0.1. This correlation between consumer rice prices and household income must be kept in mind when the welfare conse- quences of higher consumer prices are reviewed in the section titled "Changing Rice Prices and Inequality." Changes in Agriculture The various economic reforms and increased liberalization would be ex- pected to impact directly agricultural activity through the prices described in tables 5.2 and 5.3. Similarly, increased integration of the Vietnamese econ- omy into world agricultural markets, and the corresponding evolution of market infrastructure, should make it easier for farmers to move into exportable cash crops and away from rice. Food Demand Patterns The most important catalyst for change in agricultural production probably comes from an increased and changing domestic food demand. For exam- ple, an increase in incomes of 50­80 percent (as reported in table 5.1) will typically lead to increased food demand and movement along Engel curves toward more "luxurious" foods.14 However, it is also true that relative food prices changed over the five-year period--perhaps as a consequence of shifting demand in the face of relatively inelastic supply--and certainly some of these price changes also affected the food demand patterns. In this section, rather than focus on links from agriculture to income, the reverse causality is explored, describing the changing demand patterns and evalu- ating to what extent they are driven by increased affluence. To the extent that income growth is driving food demand changes, the impact of rising incomes on future demand for agricultural output may be extrapolated.15 Tables 5.4 and 5.5 report the estimated food expenditures and food shares for 10 types of food, for north and south and divided by urban and rural. As is true elsewhere in this chapter, the calculations are only over the panel households. The first two columns list the levels of household per capita expenditures for each food group, in addition to total food and non- durables consumption, expressed in 1998 prices. The next two columns show the corresponding budget shares. For total food, the share of the total budget that is devoted to food is used; for each individual food group, its share of the food budget is reported. Of interest here is the extent to which the observed changes lie on a simple Engel curve--that is, whether the change in demand from 1993 to 1998 can be explained by the increase in incomes from 1993 to 1998. To formally determine the extent to which the changes lie on a simple Engel curve, the panel structure of the data is exploited, and a standard Engel curve of the form is estimated: D-1 wi = j + j ln(pcxit) + j ln nit + ndit jt 0 1 2 j 2+d + j Cit + i 3 jt d=1 nit Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 149 Table 5.4. Urban Food Demand Patterns and Expenditure Elasticities Unexplained Per capita real (change from expenditures Shares Expenditure 1993 to Indicator 1993 1998 1993 1998 elasticity 1998) Urban north Per capita consumption 2,279.1 3,587.3 n.a. n.a. n.a. Food expenditures 1,490.9 2,050.5 0.67 0.62 0.82 Rice 380.6 433.3 0.32 0.25 0.49 Other grains 69.5 98.1 0.04 0.05 0.96 Meat 352.9 473.5 0.22 0.23 1.27 - Oils 20.7 47.3 0.01 0.03 0.83 + Fish 100.5 141.6 0.07 0.07 1.00 Other protein 72.7 88.3 0.05 0.05 0.85 Vegetables 70.0 96.1 0.05 0.05 0.56 + Fruit 68.5 97.6 0.04 0.04 1.32 Other foods 208.2 314.9 0.13 0.14 1.28 - Food consumed away from home 144.9 254.5 0.06 0.10 1.86 Urban south Per capita consumption 2,605.7 4,248.0 n.a. n.a. n.a. Food expenditures 1,503.0 2,264.6 0.59 0.57 0.82 + Rice 294.8 421.9 0.25 0.23 0.41 + Other grains 93.9 83.7 0.06 0.04 1.11 - Meat 286.1 409.3 0.17 0.17 1.41 - Oils 36.5 46.3 0.03 0.02 0.75 - Fish 166.0 248.9 0.11 0.11 0.94 Other protein 58.2 60.2 0.04 0.03 0.64 - Vegetables 75.3 121.0 0.05 0.06 0.63 + Fruit 68.8 101.4 0.05 0.04 1.16 - Other foods 202.3 265.5 0.13 0.12 1.15 - Food consumed away from home 221.1 504.7 0.10 0.19 1.59 + n.a. Not applicable. Note: All values are reported in thousands of dong, in 1998 values (deflated by the CPI). The first two columns show the average household per capita expenditures on a specific food group, in addition to total food and nondurables consumption, expressed in 1998 prices. The third and fourth columns give the budget shares of each food group as a share of total food expenditures. Total per capita expenditure is total nondurable expenditures. The fifth column reports the estimated expenditure elasticity, estimated at the average budget share, from a regression of the budget share on the log per capita expenditure, demographic controls, and commune fixed effects (FEs). The regression is estimated over all panel households, is pooled over the sample years, and includes a year dummy. The last column reports the result of a test of whether the food share in 1998 is statistically significantly higher than predicted from the Engel curve (that is, the t statistic on a year dummy in the Engel curve regression). If the food share in 1998 is higher than predicted by income and demographics, the cell contains "+," and if the food share in 1998 is less than predicted, the cell contains "-." Source: Authors' calculations from the 1993 and 1998 VLSSs. 150 Economic Growth, Poverty, and Household Welfare in Vietnam Table 5.5. Rural Food Demand Patterns and Expenditure Elasticities Unexplained Per capita real (change from expenditures Shares Expenditure 1993 to Indicator 1993 1998 1993 1998 elasticity 1998) Rural north Per capita consumption 1,225.6 2,001.1 n.a. n.a. n.a. Food expenditures 872.8 1,277.3 0.73 0.68 0.78 + Rice 421.9 518.1 0.51 0.44 0.64 Other grains 42.7 50.6 0.05 0.04 1.00 - Meat 139.8 245.0 0.15 0.18 1.48 - Oils 8.5 32.1 0.01 0.03 1.12 + Fish 56.8 92.1 0.06 0.07 1.25 Other protein 24.4 43.6 0.02 0.03 1.48 + Vegetables 43.0 58.1 0.05 0.05 0.87 Fruit 17.9 37.5 0.02 0.03 1.59 + Other foods 104.0 157.5 0.12 0.12 1.24 - Food consumed away from home 9.1 36.0 0.01 0.02 2.71 Rural south Per capita consumption 1,605.0 2,396.8 n.a. n.a. Food expenditures 971.0 1,405.5 0.63 0.62 0.80 + Rice 374.6 500.9 0.43 0.40 0.63 + Other grains 45.0 40.5 0.04 0.03 1.35 - Meat 143.1 240.5 0.13 0.16 1.50 Oils 23.8 34.2 0.03 0.02 0.80 Fish 113.9 161.4 0.11 0.11 0.99 Other protein 20.1 31.3 0.02 0.02 1.19 Vegetables 44.3 80.1 0.05 0.06 0.85 + Fruit 35.3 48.9 0.03 0.03 1.40 - Other foods 137.9 172.4 0.13 0.12 1.15 - Food consumed away from home 32.0 94.4 0.02 0.05 2.35 + n.a. Not applicable. Note: All values are reported in thousands of dong in 1998 values (deflated by the CPI). The first two columns show the average household per capita expenditures on a specific food group, in addition to total food and nondurables consumption, expressed in 1998 prices. The third and fourth columns give the budget shares of each food group as a share of total food expenditures. Total per capita expenditure is total nondurable expenditures. The fifth column reports the estimated expenditure elasticity, estimated at the average budget share, from a regression of the budget share on the log per capita expenditure, demographic controls, and commune FEs. The regression is estimated over all panel households, is pooled over the sam- ple years, and includes a year dummy. The last column reports the result of a test of whether the food share in 1998 is statistically significantly higher than predicted from the Engel curve (that is, the t statistic on a year dummy in the Engel curve regression). If the food share in 1998 is higher than predicted by income and demographics, the cell contains "+," and if the food share in 1998 is less than predicted, the cell contains "-." Source: Authors' calculations from the 1993 and 1998 VLSSs. Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 151 where wijt is the food share of food type j, for household i, in year t (1993 or 1998). The control variables are ln(pcx), or log per capita expenditures, household demographic variables (household size, nit, plus the ratios of each of D demographic groups), and commune indicators, Cit, which should account for spatial differences in preferences and long-run prices. To test whether the food share in 1998 is statistically higher or lower than ex- pected, given the control variables [especially ln(pcx)], a dummy variable for 1998 has been included. In the fifth column of the tables, the estimated expenditure elasticity (j ) from this exercise is reported, and in the sixth 1 column, the result of the hypothesis test of whether the 1998 budget share is out of line with the estimated Engel curve is reported. Tables 5.4 and 5.5 use a "+" if the 1998 share is statistically significantly higher than predicted and a "-" if the share is significantly lower. The results for urban areas are reported in table 5.4. Expenditures on vir- tually every food group increased, and total per capita food expenditures in- creased in the north by 38 percent and in the south by 51 percent. In terms of shares, the food share declined from 67 percent to 62 percent in the south and from 59 percent to 57 percent in the richer south. The food expenditure elasticities were estimated at 0.82 for both north and south (consistent with Engel's Law). For the north, the increased level of food expenditures and corresponding declines in the food budget share are estimated to be fully in line with the estimated Engel curve and the given income changes: This is one of those rare instances where the cross-section parameters explain changes over time. These results confirm that rising urban incomes alone have important ramifications for agricultural income. In the north, the food budget itself shifted away from rice (0.32 to 0.25) and toward other foods, notably toward meat, oils, and food consumed away from home. Expendi- ture on oils and vegetables increased more than predicted, given the increases in income, while demand for meat and other foods increased less. In the south, the rice share declined from 0.25 to 0.23, but the decline is sig- nificantly smaller than predicted, given the increase in incomes. This is pos- sibly due to the higher relative increase in the price of rice combined with a relatively low price elasticity of demand. Alternatively, some of this may represent a substitution toward higher-quality, more expensive rice, as sug- gested in table 5.3. Whatever the explanation, these households are spend- ing more on rice than predicted. The tilt toward rice spills over to other food demands: A smaller than anticipated shift toward meat, other grains, and fruit was observed. Vegetable demand, however, is higher than expected. Table 5.5 reports the corresponding rural results. In the north, real per capita food expenditures increased by 46.4 percent, while they increased a similar amount (44.6 percent) in the south. The matching food shares de- clined from 0.73 to 0.68 in the north and 0.63 to 0.62 in the south. Both de- clines are less than had been predicted, given the expenditure elasticities, which were estimated to be 0.78 and 0.80 for north and south, respectively. In fact, while the demand patterns shift in the direction expected, given the income increases, most are out of line with the Engel curve. As before, much 152 Economic Growth, Poverty, and Household Welfare in Vietnam of this is driven by rice. The rice share of food declined from 0.51 to 0.44 in the north and from 0.43 to 0.40 in the south. Given the increase in income, both shares should have been lower in 1998 (assuming that the estimated rice expenditure elasticity of 0.6 is not too high). As was the case in the urban south, the explanation is not obvious, but increasing rice prices may have played a role in tilting expenditures toward rice. That said, demand shifted toward meat, oils, fish, fruit, and vegetables, even if less than expected, given the increase in incomes. In summary, increased incomes have generated a significant increase in the demand for all types of food, including rice. There has been a significant shift toward nonrice foods, however, which should be reflected in cropping patterns. Cropping Patterns The broad changes occurring in agriculture discussed above are more di- rectly illustrated in tables 5.6 and 5.7, which focus on production. Table 5.6 reports summary data for all of Vietnam and for the north and south sepa- rately on the composition of output (value of production, deflated by a crop price index), sales, and acreage for the two panel years, 1993 and 1998. For Vietnam as a whole, fairly rapid growth has been observed in agriculture, averaging more than 6 percent a year. Rice production grew at a rate of slightly more than 5 percent a year, and nonrice crops (including annual in- dustrial crops and perennials) grew in excess of 8 percent. There are stark re- gional differences in the performance of the agricultural sector. Household production in the south grew faster than in the north for both rice and non- rice production, with growth rates almost three times that observed in the north.16 Overall, the more rapid growth in cash crop production, combined with a decline in the price of rice relative to cash crops, contributed to a marked decline in the role of rice, as rice production fell from 64.3 percent to 53.3 percent of crop output. This is not to diminish the still important role of rice, because it represents more than half of farm output and 40 percent of net farm income for rural households. Note that the reduction in rice's share of total output is much steeper than the reduction in acreage, which fell only from 67.5 percent to 64.1 per- cent. The much sharper decline in output reflects differences in the value of output per unit of land, which is much higher for cash crops, and the decline in the price of rice relative to cash crops between 1993 and 1998. Coffee, a key component of nonfood crops, provides a useful illustration. Coffee is grown almost entirely in the south, representing less than 1 percent of the value of output in the north. For the south, it accounted for 5.7 percent of the value of output and 10.4 percent of sales in 1993. By 1998, these figures had risen to 16.2 percent and 21.1 percent. Over the five-year period, coffee's share of agricultural production in the south nearly tripled. The increase in share was driven by the combination of a threefold increase in output and a sharp rise in the world price of coffee. Remarkably, the nominal value of Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 153 Table 5.6. Crop Output, Acreage, and Sales: Shares Output Acreage Sales Crop 1993 1998 1993 1998 1993 1998 Vietnam Rice 0.643 0.533 0.665 0.641 0.464 0.409 Other staples 0.055 0.044 0.098 0.068 0.039 0.019 Vegetables 0.066 0.064 0.045 0.039 0.095 0.067 Nonfood crops 0.149 0.239 0.099 0.145 0.260 0.342 Fruits and perennials 0.087 0.120 0.093 0.107 0.142 0.163 Total (thousand dong) 3,143.9 4,421.1 9,622.9 10,951.4 1,275.3 2,654.5 Percentage of output sold n.a. n.a. n.a. n.a. 40.6 59.6 North Rice 0.653 0.576 0.591 0.576 0.383 0.379 Other staples 0.097 0.093 0.184 0.132 0.106 0.054 Vegetables 0.080 0.101 0.053 0.048 0.155 0.153 Nonfood crops 0.107 0.135 0.088 0.106 0.211 0.232 Fruits and perennials 0.063 0.095 0.084 0.138 0.145 0.182 Total (thousand dong) 2,418.0 2,857.6 7,025.0 7,499.6 544.9 949.9 Percentage of output sold n.a. n.a. n.a. n.a. 22.5 33.2 South Rice 0.635 0.508 0.722 0.684 0.492 0.417 Other staples 0.020 0.016 0.030 0.026 0.016 0.011 Vegetables 0.053 0.043 0.040 0.033 0.074 0.045 Nonfood crops 0.185 0.300 0.108 0.171 0.276 0.369 Fruits and perennials 0.108 0.134 0.100 0.086 0.142 0.158 Total (thousand dong) 4,229.2 6,625.8 13,507.2 15,885.5 2,367.2 4,979.1 Percentage of output sold n.a. n.a. n.a. n.a. 56.0 75.1 n.a. Not applicable. Note: The value of crop output is expressed in both years in terms of 1998 dong, using re- gional price deflators based on prices received by farming households as reported in the VLSSs. The value of crop sales in both years is expressed in 1998 dong, using the same regional price deflators used to deflate the value of crop output. Source: Authors' calculations from the 1993 and 1998 VLSSs. coffee output increased by a factor of eight. Virtually all of this is exported, so the increased involvement of Vietnam in world coffee markets has had a significant impact on cropping patterns and incomes in the south.17 The high growth in output for all crops was accompanied by even more rapid commercialization of the farm sector. The percentage of output that owth cent) 8.90 6.99 1.42 3.46 1.96 Annual gr 14.60 1.41 (per 1.73 South 1998 6,912.6 4,108.4 9,543.6 5,786.3 5,213.8 16,466.4 1,263.61 1.60 1993 4,512.4 2,931.0 8,896.5 3,033.5 14,202.6 10,221.5 owth cent) 2.66 2.71 2.44 Annual gr 0.89- 3.46 0.59 0.16- (per 2.00 North 1998 2,871.4 1,646.4 3,710.3 2,540.3 1,960.7 7,440.1 4,345.6 Household 1.86 1993 Per 2,518.1 1,436.3 3,878.5 1,654.1 7,224.40 4,381.50 Use owth cent) 6.29 5.28 8.16 0.53 8.15 1.99 1.15 Annual gr (per Fertilizer 1.83 and ietnamV 1998 4,502.0 2,633.4 6,048.7 3,841.5 3,264.8 1,058.91 7,126.0 Use, 1.70 Land 1993 3,317.9 2,033.5 5,890.0 n.a. 2,207.1 10,021.8 6,722.6 Output, dong) dong) Crop in (thousand ) 2 (thousand (m Changes ) output oduction ea 2 pr ar (m (kg) 5.7. oduction ea pr ar ableT op Paddy Nonrice otalT Paddy Irrigated otalT Rice otalT Indicator Agricultural Cr Land Cultivated Sown MCI 154 el " 7.06- 10.54 1.12 5.12 12.58 17.60 1.411 34.08 13.64 31.67 5.97 pan n.a." 58.99 on egionalr implied the based using marked as is 514.7 200.1 53.0 25.5 12.01 300.9 858.3 14.31 55.4 27.8 146.3 1,043.6 ableT dong, and 1998 land). 1993, calculatede of ar 742.1 632.3 189.3 41.3 14.1 49.8 175.3 198.1 60.3 14.0 20.8 14.4 of for tes unit terms ra per in available owth ops is Gr 1.96- 7.10 2.56 6.20 39.22 22.05 1.35 cr essed 18.27 12.96 19.94 52.32 56.87 of expr is ietnam.V 310.8 86.0 107.8 20.4 13.0 744.7 121.7 34.4 34.0 4.1 9.5 number years information 1,808.7 both paddy southern average in No and 220.6 75.8 79.8 3.9 4.8 696.5 52.6 18.7 13.7 0.5 1.0 the 1,996.4 as output op VLSSs. cr northern of the for in 2.99- 9.47 1.83 5.84 23.49 18.71 2.85 29.87 13.66 25.17 9.95 (calculated 59.30 value indexes index The eportedr 597.5 131.2 86.2 22.4 52.1 569.1 402.7 66.0 42.4 13.5 63.6 as price 1,297.9 opping years. 1998. cr both fertilizer and VLSSs. in 380.1 19.81 64.9 7.8 22.1 494.6 109.0 34.8 13.8 8.4 6.2 households multiple 1993 1998 1,510.4 = farmed separate and MCI farming that between by using 1993 rate the meters.e surveys eceivedr deflated owth omfr dong) dong) is gr squar 1998 = 2 prices 1998 m and on ine (thousand 1993 calculations' e (kg) (thousand (kg) oduction: e (kg) (kg) the based compounded (kg) (kg) input pr (kg) (kg) kilogram. oduction: = in pr (kg) (kg) expenditur Authors ogen ogen kg average ce: ganic ganic deflators rice Or Expenditur Nitr Phosphates Potassium NPK nonrice Or Expenditur Nitr Phosphates Potassium NPK Note: Sour Fertilizer In In households price Fertilizer annual 155 156 Economic Growth, Poverty, and Household Welfare in Vietnam was sold increased from 41 percent to 60 percent over the period. This increase is not only a product of the increase in cash cropping, for which marketing ratios are higher, but also an increase in the marketing of rice, which increased significantly over this period. Although the percentage of total farm sales of rice declined, it still accounted for 41 percent of sales rev- enue in 1998. Furthermore, although the south was more commercialized at the outset, this period still saw greater gains in commercialization in the south. By 1998, more than three-fourths of all farm output in the south was being marketed; by comparison, in the north, only one-third was marketed. How do these results, based on VLSS household survey data, compare with aggregate administrative statistics? Actually, they compare quite well. Figures 5.2 and 5.3 plot indexes of the real value of rice production and the real value of agricultural output, respectively, based on data reported in the agricultural yearbooks. Superimposed on the figures are the corresponding indexes based on the VLSS data. A few methodological points are worth em- phasizing before turning to the comparison, because the data sources are not perfectly comparable. First, the reported VLSS indexes are recalculated on Figure 5.2. Trends in Rice Production Index of rice output (1993 100) 140 120 100 80 60 40 20 0 86 87 88 89 90 91 92 93 94 95 96 97 98 South (Yearbook) North (Yearbook) South (VLSS Panel) North (VLSS Panel) Note: The figure shows an index for agricultural output in the given crop year, rela- tive to the base year of 1993. Each crop year's output is based on a moving average of output in that year and the previous one (analogous to the VLSS). These aggregate data are drawn from the yearbook (GSO 2000). Superimposed on each graph is relative out- put in 1998 versus 1993, based on the VLSS panel households. Source: Authors' calculations from 1992­93 and 1997­98 VLSSs. Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 157 Figure 5.3. Trends in Total Agricultural Output Index of real value of agricultural output (1993 100) 160 140 120 100 80 60 40 20 0 86 87 88 89 90 91 92 93 94 95 96 97 98 South (Yearbook) North (Yearbook) South (VLSS Panel) North (VLSS Panel) Note: The figure shows an index for agricultural output in the given crop year, rela- tive to the base year of 1993. Each crop year's output is based on a moving average of output in that year and the previous one (analogous to the VLSS). These aggregate data are drawn from the yearbook (GSO 2000). Superimposed on each graph is relative out- put in 1998 versus 1993, based on the VLSS panel households. Source: Authors' calculations from 1992­93 and 1997­98 VLSSs. the basis of all households in the north and south (including urban and rural, as opposed to rural only, as was reported in table 5.6). Second, the rice output index (from both data sources) is based on physical output of paddy, as opposed to the value of rice deflated by the rice price index (as was done in table 5.6). Third, the real value of agricultural output reported in the year- book includes animal husbandry, whereas this index is based only on crop output. This does not make much difference, given the trends in animal husbandry income reported in table 5.1, in addition to its small share in out- put.18 Finally, the most difficult issue in comparing the data concerns the timing of the VLSS and the corresponding crop year in the yearbook. House- holds report output over the preceding 365 days (that is, the prior year). Given that survey interviews were spread evenly, more or less, over the sur- vey year, most households report output from crops harvested in two calen- dar years. As an approximation, the yearbook data are thus converted to a typical VLSS household "reporting horizon," by averaging output between the current and past years. For example, 1993 output equals the average of 1993 and 1992 output. The plotted yearbook data are thus a moving average. 158 Economic Growth, Poverty, and Household Welfare in Vietnam As shown in figures 5.2 and 5.3, the two data sources line up remarkably well, especially if north and south are pooled. The two regions' growth rates are also similar, although the VLSSs show slightly more divergence between north and south. More specifically, for rice the yearbook data imply a rice index for 1998 relative to 1993 of 122 for the north and 131 for the south. This compares with 114 for the north and 135 for the south, based on the VLSS data. The indexes for total output, which is dominated by rice, are 122 for the north and 140 for the south, based on the yearbook, compared with 113 and 147 in the VLSS, for the respective regions. Given especially the differ- ent sampling frames for these data, the series paint similar pictures: Growth is significantly higher in the south, and growth in total output is higher than rice alone, especially in the south. Although figures 5.2 and 5.3 provide strong corroboration for the agri- cultural data in the VLSSs, they also provide a strong source of caution about the representativeness of the two sample years, 1993 and 1998. The year 1993 was an especially good year for agriculture in the north and a slightly off year in the south. This means that mean reversion alone would generate higher growth rates in the south than in the north. The yearbook data do imply a divergence of north-south agricultural output, especially in nonrice; focusing on 1993­98 alone will exaggerate the trends.19 Of course, at this time, only VLSS data from 1993 and 1998 are available, so a full evalua- tion of the robustness of the conclusions to the choice of survey years is not possible. That said, the conclusions that follow, based on the VLSSs, must be placed in the broader context of the trends shown in figures 5.2 and 5.3. Although causal relationships must be approached with caution, the growth in output and farm sales appears to be correlated with the reform and liberalization of both input and output markets. As reported in table 5.2, over this short period, a marked increase (decrease) in the relative price of farm outputs (farm inputs) was observed. Overall, agricultural prices rose by 80­90 percent, compared with an increase of 45.6 percent in the CPI; fer- tilizer prices, on the other hand, rose only 10 percent. The behavior of these prices increased the returns to farming and provided farmers with powerful incentives for increasing output and sales. The more rapid growth in output in the south appears partially tied to more favorable movement in prices, as farm prices rose more and fertilizer prices rose less than in the north. Rice Output, Land, and Fertilizer Inputs Table 5.7 provides a breakdown for all of Vietnam, and by north and south, of the growth in crop output and inputs, most notably land and fertilizers. The eventual objective is to see to what extent changes in output can be ex- plained by changes in inputs. The estimates in this table are based on the panel of households, restricted to those who farmed in both 1993 and 1998. Some of these numbers are not exactly comparable to those in table 5.6 be- cause of a reduction between 1993 and 1998 in the number of households that were farming. Altogether, a reduction of nearly 10 percent in the Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 159 number of households farming can be seen over this period.20 Differences in samples give rise to modest differences between the two tables. The focus here is on rice production and total crop production, as well as on highlighting the differences between the north and the south. In the north, rice production measured in physical terms grew 2.8 percent a year. This is less than half the rate of growth of rice production in the south, which was nearly 7 percent. The overall rate of growth of crop production was 8.9 percent a year in the south, compared with 2.7 percent in the north. A likely explanation for these patterns is the differential effect of liberalization of rice marketing on the two regions, with the low-cost south taking advan- tage of the expanded export and domestic marketing opportunities for rice. The increase in rice production in the south did not seriously handicap the production of cash crops, however, which grew nearly 75 percent faster than rice output. In the north, relative price shifts also affected cropping decisions (see table 5.6), but output growth of nonrice production was actually slightly lower than that for rice. One potential explanation is that in the north, much of the shift was into perennials and fruits, both of which have three- to five-year lags in revenue generation. Data on acreage and fertilizer input provide important initial clues as to the margins on which output in agriculture was able to expand. Over this five-year period, cultivated area increased slightly, averaging less than one- half of 1 percent a year. This occurred largely in the south. In the north, cul- tivated area actually declined, largely because of the decision of households to take swidden land (land where the vegetation has been cut back or burned off) out of production. Sown area also increased slightly--2 percent per year nationwide--again largely because of increases in the south. In both regions, there was a modest increase in the cropping intensity as mea- sured by the multiple cropping index. In the north, however, this increase occurred largely because of a slight increase in sown area and a larger re- duction in cultivated area; in the south, it occurred because of growth in sown area that was more than double growth in cultivated area. With output in the aggregate growing at a rate in excess of 6 percent, and with sown area increasing only 2 percent nationwide, it would appear that much of the increase in output is coming from increases in cropping intensity and higher yields. Here the panel provides some insight into the role of increases in fertilizer inputs. The general tendency in both the north and south was a reduction in the role of organic fertilizers in rice and an increase in the use of commercial chemical fertilizers, largely consisting of urea, potassium sulfate, phosphates, and NPK. In real terms, commercial fertilizer use by the panel households increased slightly--less than 10 percent a year. For nonrice crops, there was some increase in the use of organic fertilizers, especially in the south, and even larger increases in the growth rate in chemical fertilizer use, as would be expected, given the liberalization of the fertilizer market. For the en- tire sample, the application of commercial fertilizers on nonrice production grew nearly 30 percent a year, with the annual growth in the south almost twice as high as in the north (34.1 percent versus 18.3 percent). As long as the 160 Economic Growth, Poverty, and Household Welfare in Vietnam increases in chemical fertilizer use are not being offset by the reduction in the application of organic fertilizers, real increases in the north and south in chem- ical fertilizer use are an important source of sown area yields. Decomposition of Output Growth To analyze the sources of output growth between 1993 and 1998, table 5.8 re- ports the results of a more formal decomposition exercise. The primary ob- jective is to estimate the contribution to the output growth of increases in input use versus increases in total factor productivity. Estimating produc- tion functions for agriculture and dealing with a host of econometric issues, including the endogeneity and measurement error of inputs, would be a chapter in themselves. This chapter stops short of carrying out this full- blown exercise, but the results presented here are still highly informative. The basic idea of the exercise is straightforward. Assume that the production function for agricultural output, Yit, is known for household i in year t: ln Yit yit = Xit + uit where Xit is a vector of inputs, is a vector of production function parame- ters, and uit captures the impact of unobservables. The change in output between two periods can be decomposed as: g¯ y¯it = Xit + u¯it ¯ where g¯ is the growth rate, defined as the difference in log output between the two years, yit = ln Y1998 - ln Y1993. To execute this decomposition, esti- mates of the change in input use are needed, Xit, as well as the parameters of the production function. It is also necessary to make some assumptions regarding the unobservables, uit. For example, it could be assumed that the unobservables are the same (on average) each year and so force the u¯it term to equal zero. More plausibly, it could be imagined that uit has the following structure: uit = t + it where t is a time effect and it represents a mean-zero error term. The time effect t will capture improvements in productivity that allow all farmers (on average) to obtain more output from their inputs in period t. This is com- monly labeled "total factor productivity." Of course, the source of the time effect is not directly observable, and it could as easily reflect differences in "luck" between periods. For example, figures 5.2 and 5.3 prove that output in the north was unusually high in 1993, and this "blip" was unlikely due entirely to productivity differences in 1993. The following functional form demonstrates the production function: yit = 0 + 1SAit + 2IRRit + 3Lit + 4DKit + 5Kit + 6DOFit 4 4 + 7OFit + 8f DCFERTf + it 9f CFERTf + 10Y98it + it. it f =1 f =1 Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 161 Table 5.8. Decompositions of Output Growth Vietnam North South Indicator OLS OLS-FEs OLS OLS-FEs OLS OLS-FEs Rice production Rice Output growth 21.00 21.00 15.70 15.70 30.90 30.90 Contribution to output growth (percent) Sown area 8.86 8.38 -0.83 -.89 17.57 15.53 Labor -- -- -- -- -- -- Capital 0.09 2.10 -0.47 2.80 3.53 0.97 Fertilizers 40.15 17.33 65.41 32.17 22.04 12.56 Chemical 43.67 21.62 70.70 36.31 22.20 15.21 Organic -3.52 -4.29 -5.29 -4.14 -0.16 -2.65 Land quality 12.27 1.40 16.37 -1.53 6.95 1.60 Residual 38.63 70.79 19.52 67.45 49.94 69.34 Total crop output Gross value of crops Output growth 17.10 17.10 8.00 8.00 32.40 32.40 Contribution to output growth (percent) Sown area 16.37 16.78 20.13 17.25 14.78 16.98 Labor 1.11 0.47 6.50 8.75 -5.03 -1.26 Capital 7.60 5.56 4.75 4.54 10.49 3.86 Fertilizers 92.08 49.82 155.25 120.88 58.74 34.65 Chemical 93.57 52.39 163.38 103.63 57.72 34.57 Organic -1.49 -2.57 -8.13 -9.75 1.02 0.08 Land quality 3.00 -0.78 4.75 -6.25 -1.85 -0.20 Residual -20.16 28.15 -98.64 -45.17 22.87 45.97 -- Not available. Note: FEs = fixed effects. OLS = ordinary least squares. These decompositions estimate the fraction of output growth explained or accounted for by changes in input use, based on pro- duction functions estimated for each subsample, pooled across years, and based on the panel households in 1993 and 1998 that farmed in both years. The residual is the coefficient on an in- dicator variable for 1998. The OLS-FEs decompositions are based on production functions esti- mated with household FEs. For rice, output is measured in physical units (kilograms), and total crop output is the value of crop output expressed in terms of 1998 dong, using regional price deflators based on prices received by farming households as reported in the VLSSs. The output growth rates used in the decomposition and reported here are computed by taking the differ- ence mean (logY98) - mean (logY93), where Y is either rice output, measured in physical terms, or the deflated value of crop output. Labor input is not provided separately for rice production and thus is not included in the rice decomposition. The capital stock in agriculture is measured by the deflated current market value of farm machinery and draft animals. The capital stock is deflated by a national deflator constructed on the basis of prices paid for capital machinery and draft animals in 1993 and 1998 as reported in the VLSSs. The contribution of fertilizer is the sum of the contribution of urea, phosphates, potassium sulfate, and NPK. Land quality is measured by the percentage of land irrigated. Source: Authors' calculations from the 1993 and 1998 VLSSs. 162 Economic Growth, Poverty, and Household Welfare in Vietnam Essentially, this functional form specifies (log) output (yit) as a function of log inputs, dummy variables (D) of whether the farmer uses the input, a time dummy to capture t (Y98it), and the error term, it.21 The inputs that have been included are sown acreage (SAit), the percentage of land that is irrigated (IRRit), labor (Lit), the real value of farm capital (Kit), and the quantities of various types of fertilizer (organic [OFit] and the four chemical fertilizers-- CFERTf --urea, potassium sulfate, phosphorus, and NPK). This particular it functional form accommodates the fact that there are households for whom either farm capital (machinery or draft animal) or fertilizer use is zero. To estimate the parameters of this production function, the data for the panel of households that farmed in both years have been pooled. The pro- duction functions for rice and for the gross value of crop output are esti- mated separately, measured in constant 1998 dong. The separate production functions have also been estimated for northern and southern Vietnam to allow for potential differences in technology and the productivity of the in- dividual factors of production. Finally, two sets of results are reported: ordi- nary least squares (OLS) and fixed effects (FEs). The motivation for using FEs is standard: Farmer's unobserved managerial ability, i, may be corre- lated with input use.22 In this case, the error term it = i + vit and OLS will yield inconsistent estimates of . This is a distinct possibility in this applica- tion, and it may affect the attribution of increases in output to increased fer- tilizer use. For example, if only the best farmers use chemical fertilizers, then the estimate of the coefficient on chemical fertilizer will be overstated. In that case, it will appear in the decompositions that the increased use of fertilizer explains most of the increase in output. The FEs specification has its own possible problems, especially in exaggerating measurement error in inputs. For this reason, both the OLS and OLS-FEs results are reported here. The results for rice are reported in the first half of table 5.8. On the basis of the OLS parameters for the pooled sample, sown area, land quality, and chem- ical fertilizer use explain nearly 60 percent of the growth in rice output; the residual (time effect) represents the remaining 40 percent. The latter includes the effect of new seed varieties, unmeasured labor effort, improvements in productivity, or better luck (for instance, rainfall). Decompositions based on separate parameter estimates for the north and south reveal a much larger role of the unexplained component for the south than for the north. In the north, increased chemical fertilizer use explains more than two-thirds of the growth in paddy production, with the residual the source of 20 percent of the growth. By contrast, in the south, increased fertilizer use explains only 20 percent, and the residual (time effect) is almost one-half. This sharp contrast between north and south disappears with the use of household FEs: In both regions, more than two-thirds of the increase in output can now be attributed to the residual. As suspected, the use of household FEs especially affects the estimated coeffi- cients on chemical fertilizer, and thus the role of fertilizer, in explaining growth. In general, the parameter estimates for fertilizer are smaller using OLS-FEs, with the OLS parameter differences greatest in the north.23 A similar exercise for the gross value of crop output is shown in the bot- tom half of table 5.8. The only difference with the decomposition for rice is Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 163 that labor is now included as an input (which turns out to be a minor fac- tor).24 The OLS estimates suggest that almost all of the growth in output is coming from the tremendous increase in chemical fertilizer use. The increase in total input use actually overexplains output growth, generating a negative residual. The contrast between the north and south here is particularly stark. There is a huge negative residual in the north, and in the south, slightly more than one-fifth of the increase can still be attributed to the residual. As in the case of rice, using parameters from OLS-FEs reduces the contribution of in- creases in input use (largely fertilizer) in explaining growth and increases the size of the residual. However, there remains a significant negative residual in the north, and in the south, the residual is equal to nearly one-half of output growth over the period. For the north, some of this negative residual may be coming from the shift into perennials and tree crops observed in table 5.6; these crops will not generate income for several more years. That would ex- plain why inputs went up proportionately more than output. Some of the difference in relative performance in the north and south may be due to mean reversion, as was suggested by figures 5.2 and 5.3. The main conclusions from this exercise are: · Most of the increase in rice output cannot be explained by increased inputs. Of the part that can be explained, increased use of chemical fertilizer is an important component of the explanation. · More of the increase in total output can be explained by increased in- puts. Nonrice output increased significantly because of increased acreage (sown area) and especially because of the increase in chemi- cal fertilizers. The unexplained component (perhaps productivity?) is positive in the south and negative in the north. This pattern is sug- gestive, and it could reflect differences in incentives generated by lib- eralization and the returns to increasing specialization, but there is no way these data alone can be used to confirm this hypothesis. Agriculture and Inequality In this section, some of the changes in income distribution over the period from 1993 to 1998 are explored, with a focus on the potential impact of changing rice prices. Clearly, increased rice prices will benefit rice producers while hurting consumers. On a regional level, rice-surplus regions will gain while rice-deficit regions lose. Moreover, to the extent that there are supply and demand responses to the price changes, there may be changes in the pattern of rice marketing. Rice Marketing This section begins with an exploration of the impact of rice price changes on regional rice marketing and cropping patterns. Various indicators of rice pro- duction, sales, and consumption indicators are presented in tables 5.9 and 5.10. Table 5.9 shows the breakdown by urban and rural and by north and south. Table 5.10 provides a more detailed regional breakdown for rural areas. 1 0.10 0.06 0.98 0.22 0.14 0.05 0.06 1998 471.85 200.32 300.12 537.21 485.51 168.1 51.69 1,868.98 1,700.87 ietnamV 1,400.75- 1,397.13- 0.13 0.06 0.98 0.17 0.18 0.05 0.06 Urban 1993 293.64 138.02 99.69 568.27 495.67 188.76 72.60 1,602.05 1,413.29 1,313.60- 1,308.42- 0.12 0.09 0.97 0.20 0.13 0.08 0.08 1998 734.82 312.47 508.15 549.58 489.62 210.90 59.96 1,998.98 1,788.07 south 1,279.93- 1,264.16- 1 Urban 0.13 0.08 0.97 0.16 0.15 0.06 0.07 1993 406.61 194.1 161.68 579.59 489.33 234.94 90.27 1,609.15 1,374.21 1,212.53- 1,202.54- 97.07 0.07 40.49 3.65 0.01 0.99 0.25 0.16 0.01 0.04 1998 519.57 479.66 107.12 39.91 1,683.72 1,576.60 north 1,572.95- 1,586.65- Urban 0.12 1.361 0.03 0.99 0.17 0.22 0.02 0.04 1993 132.63 58.08 552.13 504.71 122.94 47.42 1,591.94 1,469.00 1,457.64- 1,459.31- prices) prices) prices) prices) 1998 prices) 1998 prices) 1998 1998 dong, 1998 Household dong, 1998 prices) dong, dong, per (kg) 1998 dong, dong, (thousand rice (kg) (kg) (kg) rice? rice (thousand (thousand (thousand dong, (thousand Marketing, chased oduced oduced oduced (thousand oduced pr consumed chased pur oduced e Rice pr rice sold consumed rice pur (thousand home-pr shar surplus rice? rice? rice home-pr rice of rice rice of rice of home-pr of sales rice 5.9. of of of of of of rice? budget chased rice seller? ableT Indicator alueV oduced Pr Quantity alueV Sold alueV Quantity alueV Pur Quantity alueV Consumed Quantity Rice Net Net alueV Surplus? 164 1 . or is 0.79 0.46 0.70 0.81 0.28 0.42 0.47 way usual 1998 763.97 751.76 220.01 543.96 924.1 3,722.35 1,504.76 1,675.87 2,441.82 1,690.05 1,280.53 eportedr egion).r ietnamV (cluster same or surplus the that by the rice in rice Rural 0.85 0.42 0.74 0.86 0.33 0.36 0.51 estimate either (cluster The to 1993 758.05 763.04 567.51 215.43 547.61 190.54 727.09 2,722.43 1,198.53 1,995.34 1,427.83 ster values nea prices, verted unit estimate umption. the con estr is 0.64 0.47 0.88 0.66 0.25 0.44 0.45 or cons oducer sumer nea 1998 793.23 333.80 459.43 pr 5,171.44 2,154.24 3,162.05 2,576.40 1,152.97 1,423.43 2,009.08 2,595.04 sales at con the Paddy south at its or of nondurable valued Rural sales of prices). 0.75 0.42 0.85 0.76 0.28 0.36 0.47 is valued e basis is its 1993 789.20 733.44 284.64 504.56 632.35 3,389.70 1,605.90 1,365.79 1,991.67 1,258.23 1,398.03 the of shar rice on oduced the pr basis consumer is e and rice the 0.89 0.45 0.56 0.91 0.30 569.40 742.19 453.06 135.29 606.90 16.341 0.40 0.48 of consumed on shar 1998 301.87 2,643.49 1,021.22 2,341.61 1,888.56 household of north the alueV. oducer alueV budget pr by Rural way household rice 0.93 0.42 0.65 0.93 0.38 0.36 0.54 the 1993 895.24 305.58 743.55 443.97 163.89 579.66 2,225.64 1,998.07 1,554.10 138.39- 227.57 eportedr same egion).r by The. espectiver or the way the that in at (cluster eportedr same either values the 1.456. prices) that (valued in by unit VLSSs. 1998 prices, estimate prices) prices) prices) either est values prices 1998 prices) consumed 1998 prices) 1998 1998 dong, oducer near 1998 consumer prices, unit and 1998 pr at the and to prices) at 1993 dong, 1998 or dong, dong, (kg) oducer 1998 dong, sales pr consumer the oduced valued valued dong, (thousand rice? rice pr converted is is its at at (kg) (kg) (kg) e omfr of rice rice ar (thousand (thousand (thousand dong, rice valued valued of basis chased oduced oduced (thousand oduced is is prices oduced (thousand pr the value oduced pr consumed chased pur oduced e rice rice consumed on 1993 the calculations' pr rice sold consumed rice pur (thousand rice home-pr shar surplus of of oduced in and rice? rice? home-pr pr rice of rice rice of rice of home-pr of sales rice of of of of of of alueV alueV rice consumed ence 0.66, Authors rice? household budget of of fer of ce: chased rice seller? Note: the dif Indicator alueV oduced Pr Quantity alueV Sold alueV Quantity alueV Pur Quantity alueV Consumed Quantity Rice Net Net alueV Surplus? egion).r by alueV alueV Sour the factor 165 0.8 0.37 0.66 0.85 0.29 0.34 1998 Coast 921.01 499.36 714.35 624.78 183.30 531.05 2,207.29 2,284.83 1,660.06 125.41- 77.54- 0.38 Central 0.87 0.36 0.76 0.87 0.35 0.28 0.48 North 1993 830.58 244.66 716.01 628.43 237.38 478.63 1,900.69 1,810.42 1,181.99 383.77- 90.27 0.93 0.56 0.47 809.70 672.28 274.90 85.24 0.95 0.27 0.54 0.68 1998 587.05 534.80 3,137.01 1,124.00 2,047.14 1,772.24 1,089.87 Delta River Red 0.96 0.56 0.56 0.96 0.40 0.50 0.69 1993 425.17 730.28 287.42 104.66 625.61 137.75 678.55 2,648.45 1,037.63 1,969.90 1,682.49 Only 1 0.91 0.36 0.59 0.93 0.35 0.28 0.30 Rural 1998 975.78 320.61 856.39 533.56 158.1 698.28 2,386.22 2,770.22 2,236.66 212.94- 383.99- Uplands Region, by Northern 0.93 0.31 0.67 1.61 0.94 0.37 0.25 0.39 1993 767.91 204.38 784.56 485.57 176.43 608.13 1,962.93 2,197.18 1,71 281.19- 234.25- Household, Per (kg) rice (kg) (kg) (kg) rice? prices) prices) prices) prices) rice prices) prices) prices) Marketing oduced 1998 oduced 1998 1998 1998 chased 1998 oduced 1998 1998 chased oduced e Rice oduced pr consumed pur pr dong, rice sold dong, consumed dong, rice pur dong, dong, home-pr shar dong, surplus dong, rice? rice? rice home-pr rice of rice rice of rice of home-pr of sales rice 5.10. of of of of of of rice? budget rice seller? bleaT Indicator alueV (thousand oduced Pr Quantity lueaV chased (thousand Sold alueV (thousand Quantity alueV (thousand Pur Quantity alueV (thousand (thousand Consumed Quantity Rice Net Net alueV (thousand Surplus? 166 -er 0.64 0.56 0.89 0.65 0.25 0.53 0.55 or Delta 1998 814.18 329.53 484.65 prices 7,731.70 3,338.49 5,305.18 2,553.06 1,109.27 1,443.79 4,195.91 5,178.64 eported consumed 1998 (cluster self-r to 0.76 0.58 0.88 0.78 0.26 0.51 1.69 0.60 as and Mekong 1993 854.04 740.95 304.17 549.87 5,035.53 2,444.10 2,399.26 2,023.84 1,282.90 1,658.31 3,01 estimate est valuedsi oduced pr converted e 0.46 0.34 0.87 0.52 0.20 0.31 0.34 1998 766.40 433.28 333.12 525.59 853.64 earn ar rice rice 3,569.37 1,347.19 2,163.35 2,715.72 1,637.76 1,077.96 of ces the pri or value Southeast 0.59 0.35 0.84 0.59 0.24 0.28 0.37 1993 1993 928.98 877.24 723.49 864.15 328.04 395.45 13.09 308.21 sales -consumed the 2,199.23 1,891.02 1,026.87 its or in while of ence 15.41 0.43 0.12 0.90 1.07 0.49 0.30 0.07 0.09 0.66, basis oduced fer 1998 396.43 107.33 994.29 675.79 318.50 dif of 1,1 3,185.89 2,174.82 1,01 2,067.49- 2,070.48- the the Highlands on factor 1 home-pr is of 0.83 525.18 56.40 0.12 0.85 0.83 0.35 0.10 0.10 usual Central 1993 841.83 981.10 355.1 486.72 1,030.89 2,337.84 1,356.75 924.70- surplus 1,306.95- household value the rice by the The. by The rice 0.84 0.50 0.85 0.85 0.27 0.48 0.47 way to Coast 1998 848.84 713.14 536.29 157.68 555.47 312.56 535.41 2,857.84 1,100.12 2,322.43 1,786.14 eportedr same the 0.83 0.28 0.81 0.84 1.18 0.31 0.22 0.42 that consumption. converted Central 1993 893.47 205.51 702.83 539.34 191.65 51 in 1,953.54 1,904.44 1,365.10 333.83- 49.10 is either values Paddy VLSSs. prices, unit nondurable prices) prices) ofe prices). 1998 1998 prices) oducer shar and 1998 pr consumer the 1993 dong, 1998 prices) at at ise consumer (kg) 1998 dong, the shar and dong, rice? rice valued valued (kg) (kg) (kg) is is omfr prices) prices) rice prices) dong, (thousand rice budget oducer pr chased oduced (thousand oduced 1998 1998 1998 oduced oduced rice (thousand pr oduced pr consumed chased pur oduced e The calculations' rice pr rice sold consumed dong, rice pur dong, rice dong, (thousand home-pr shar surplus of consumed espectiver rice? rice? home-pr of rice of rice rice of rice of home-pr of sales rice the of of of of of of alueV: Authors rice? budget alueV at household. ce: chased rice seller? the 1.456. Indicator alueV oduced Pr Quantity alueV Sold alueV (thousand Quantity alueV (thousand Pur Quantity alueV (thousand Consumed Quantity Rice Net Net alueV Note Sour Surplus? gion). by (valued by 167 168 Economic Growth, Poverty, and Household Welfare in Vietnam The urban patterns can be quickly summarized. First, there are very few producers in urban areas. In 1998, 10 percent of urban households grew rice, a decline of 3 percentage points from 1993. Thus, slightly less than 25 per- cent of the urban rice growers in 1993 stopped farming rice (this was even more true in the north). Deflating by the CPI, real consumption expendi- tures on rice increased from D 1,602,000 to D 1,869,000 per household. Most of the increased expenditure comes from the higher relative rice prices, be- cause household physical consumption of rice declined slightly, from 568 to 537 kilograms per household. Recall, however, that household size fell, so per capita rice consumption actually increased slightly in urban areas. Finally, the share of rice in nondurable expenditures declined from 0.18 to 0.14. This decline is in line with what is expected from the Engel curve for rice, given that incomes almost doubled in urban areas (see tables 5.4 and 5.5). Looking at the combined rural areas for north and south, the value of rice produced increased by almost one-third. This is due partially to the increase in relative rice prices, but it also reflects an increase in rice production from 1,119 to 1,505 kilograms per household. Given that the percentage of house- holds growing rice declined from 85 to 79, this means that rice farmers were producing about 35 percent more rice per producing household. Compari- son of north and south shows that both areas experienced increases in out- put, but the largest increase by far was in the south, where prices increased the most. Most of the extra rice produced was sold to the market; house- hold rice consumption stayed the same in both north and south. In the two regions combined, 46 percent of farmers sold rice in 1998, compared with 42 percent in 1993. At the same time, farmers in the south also purchased more rice in 1998 than in 1993, so that commercial involvement in rice mar- kets was more important on both the production and consumption sides of the market, at least in the south. Rice expenditure shares remained much higher than in urban areas, but they did decline from 0.33 to 0.28 over the five-year period. Interesting regional patterns emerge in evaluating the changes in the rice surpluses. Looking at the rural north and south alone, the northern surplus rose only slightly, from D 228,000 to D 302,000, where surplus is defined as the per household difference in value of production and consumption. Al- though households produced more rice in the north, they increased their purchases just as much. The rice surplus position of southern households in- creased dramatically, from D 1,398,000 in 1993 to D 2,595,000 in 1998. The extra production went to exports, as well as to increased purchases by households in the north and in urban areas. Thus, at least at this coarse level, changes in marketing patterns match expectations, given the changes in rel- ative producer prices.25 Table 5.10 shows detail by subregions. Even stronger patterns of special- ization in rice can be seen here. The Northern Uplands go from a rice deficit of D 234,000 in 1993 to a deficit of D 384,000 (per household) in 1998. The Red River Delta, the main rice-growing region of the north, saw its surplus Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 169 increase from D 679,000 to D 1,090,000, while the North Central Coast region went from small surplus to small deficit. It is also true, however, that the apparent lack of change in this region hides largely offsetting increases in production and consumption. More pronounced specialization can be seen in the southern regions. In the Central Highlands, the rice deficit increased from D 1,307,000 to D 2,070,000 per household. There are sizable increases, however, in the surplus produced in the Central Coast (D 49,000 to D 535,000) and Southeast (D 308,000 to D 853,000) regions. Certainly the most dramatic change is in the Mekong Delta, though, where the household surplus goes from D 3,012,000 to D 5,179,000, despite an increase in rice consumption of almost D 500,000. It appears that at the household level, more households are rely- ing on the market for their rice, while at the national level, more regions are becoming "rice importers," with the Mekong Delta producing a growing share of national rice output. Changing Rice Prices and Inequality It is clear that, in total, rural income inequality declined in the south, while income differences widened between north and south. Is this fact linked to rice? To address this question, the position in the income distribution of winners and losers from rice price changes must be identified within each region. If rice farmers were concentrated at the bottom or middle of the income distribution in the south, reported price changes could lead to changes in line with those just described. Alternatively, the changes in in- come inequality may have nothing to do with rice. Many factors in addition to increases in rice prices changed over the period. In this section, the methodology outlined by Deaton (1989, 1997) is used to explore the association between benefits of rice price changes and a household's position in the income distribution. A brief review of the theo- retical motivation underlying the empirical analysis is helpful, even though the intuition is straightforward. Following Deaton's notation, household welfare can be summarized by the indirect utility function: uh = (wT + b + , p¯) where household "full income" is the sum of the value of the labor endow- ment, wT, unearned income, b, and farm profits, ; and p¯ is a vector of con- sumption prices. Assuming that the producer and consumer prices of rice are the same (which is not the case), then the change in household welfare associated with change in rice prices is: uh p = b × p + p = b × (y - q) that is, the effect of a change in income on household welfare, scaled by the difference between production (y) and consumption (q) of rice. Not surpris- ingly, welfare increases for those households that are in a surplus position 170 Economic Growth, Poverty, and Household Welfare in Vietnam and decreases for net consumers of rice. Note that the assumption in this formula (as simplified here) is that there is no supply or demand response. This gives a first-order approximation of the welfare change associated with a small price change. Clearly, households adjust their behavior to larger price changes, and this will add additional terms to the formula as produc- ers produce more rice and consumers reduce rice consumption. These changes will attenuate the potentially adverse impact of an increase in pro- ducer prices while compounding the benefits. Using the equation above, the amount of income that should be given to households to restore them to the original level of welfare is: dB = (q - y)dp = p(q - y)d ln p so that the amount of compensation depends on the net consumption posi- tion of the household, scaled by the price change. The marginal compensa- tion, dB, is expressed as a share of household expenditures: dB = pq - py d ln p x x which Deaton calls the "net consumption ratio." The focus is on the negative of this expression, the net benefit ratio: the value of production minus the value of consumption, relative to household expenditures. Clearly, house- holds should be relatively better off with an increase in rice prices if they are net sellers of rice. Because of the divergence between producer and con- sumer rice prices, however, it also makes sense to look at the components of this expression separately. The production ratio can be defined as the ratio of the value of rice production to total expenditures, and the consumption ratio can be defined as the ratio of the value of rice consumption to total expenditures (which is just the rice budget share). This chapter is specifically interested in the correlation of the benefit ratio with the position of the household in the income distribution, as sum- marized by log of household per capita expenditures (lnpcx). A useful start- ing point for this discussion is the results reported by Minot and Goletti (1998). They use the 1993 VLSS to calibrate a structural spatial model of rice markets in Vietnam to simulate the distributional impact of a relaxation of the rice export quota. Their focus, like ours, is on the impact of the change in rice prices on household welfare as summarized by the net benefit ratio. Their main conclusions are: · Higher rice prices will exacerbate regional income inequality. · Higher rice prices will worsen the within-region inequality, because the rural poor would be hurt more than the urban poor. · Higher rice prices will still yield net benefits to the poor, through higher incomes, and therefore reduce poverty. This chapter does not directly address their second point, because urban and rural households have not been pooled in the welfare analysis. How- ever, these results using the 1998 data directly address Minot and Goletti's other predictions. Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 171 In addition to the benefit of having more data, which allows for estima- tion rather than simulation of the impact of the price increase, it is clear that some of the assumptions in their analysis do not ultimately hold true. For ex- ample, consumer prices actually rose more than producer prices. (Minot and Goletti assume the reverse.) The divergence of consumer and producer prices provides an interpretation challenge to both their conclusions and this chapter's approach. Most important, and perhaps not surprisingly, many other factors that Minot and Goletti hold constant changed, notably the in- crease in incomes and the decrease in fertilizer prices. Both of these factors lead to rice supply and demand responses that are not easily anticipated. Following Deaton, nonparametric regressions of the association between rice benefit ratios and income lnpcx have been estimated.26 The resulting graphs provide a clear picture of who are the relative winners and losers from price changes. Because there are very few rice producers in urban areas, these households' net benefit ratios are essentially their rice budget shares. The welfare losses associated with rice price increases will thus be in direct proportion to rice consumption. Because rice expenditure shares are higher for lower-income households (as the Engel curve clearly shows), then in relative terms the poor will be most adversely affected by the price in- crease, and the price increase will worsen inequality of welfare (all else being equal). This chapter does not show the results of this exercise, but fo- cuses instead on the rural households, where the story is more complicated. The results are presented in figure 5.4 (rural north) and figure 5.5 (rural south). Three sets of graphs are shown: the net benefit ratio, the production ratio, and the consumption ratio. These variables are shown separately for 1993 and 1998, because it is unrealistic to imagine that the results from a single time point will apply over the entire five-year period, given all of the other changes in rice production, consumption, and household incomes. In- deed, the theory outlined above refers to the impact of marginal changes in prices on household welfare, holding everything constant. Over the five- year period, the rice price changes were far from marginal, and little was constant. The figures also show changes in the rice variables over the period and how these changes relate to a household's position in the original 1993 income distribution. This set of graphs allows review of relative improve- ments in living standards associated with changes in rice marketing. Two reference bars are presented in each figure, corresponding to the 25th and 75th percentiles of the lnpcx distribution. Note that although the domain of lnpcx includes all of the lnpcx-axis, most of the observations are concentrated in the middle of the figures; thus, most inferences should be drawn from the shapes of the estimated functions in the middle of the graphs. Finally, the dashed lines represent the bootstrapped 95 percent confidence intervals for the regression line. First consider results from the north. Panel A shows the net benefit ratio for 1993, and panel B presents the corresponding figure for 1998. Panel A shows that the net benefit of a rice price increase (given the values of vari- ables in 1993) was positive, in the 0.10 range. This means (approximately) that the difference in value between rice production and consumption is 8.8 8.8 1993 1993 to to 8 8 1998 1998 ratio, pcx ratio, pcx ln ln 7 7 benefit oduction in pr in 6 Change 6 C: Change .5 0 .25 .25 .5 1 .5 0 .5 F: Panel oitar tifeneb ni egnahC oitar noitcudorp ni egnahC Panel 8.8 8.8 North 8 8 1998 Rural 1998 ratio, pcx ratio, pcx ln ln Prices, 7 7 benefit oduction pr Rice Net B: Rice in E: 6 6 Panel 1 Panel .75 .5 0 .25 .25 .5 1 .75 .5 0 .25 Changes interval) oitar tifeneb teN oitar noitcudorp eciR of 8.8 8.8 Impact confidence cent 8 8 1993 1993 per ratio, 95 pcx ratio, pcx ln ln and 7 7 Distributional benefit oduction pr The ession Net A: Rice egrr D: 6 6 5.4. Panel 1 .75 .5 0 .25 .25 .5 1 .75 .5 0 Panel .25 oitar tifeneb teN oitar noitcudorp eciR Figure (estimated 172 8.8 1993 sur- to in oduc- pr ors. 8 1998 ence err rice d fer of ratio, ar dif pcx ln the ratio stand 7 as the is consumption in defined ratio is 6 bootstrapped on ratio 1 0 Change oduction oita.75r noitpmusnoc .5 .25 I: pr ni egnahC based benefit the Panel in 8.8 interval, change example, the For 1998 8 confidence distribution. while ratio, ), defined. cent pcx pcx ln pcx ln per 7 95 the lns' consumption. of 1992 consumption plus year to analogously e Rice centiles each ar 6 H: ession, to per elativer 1 egrr .75 .5 0 .25 Panel 75th changes, VLSSs. elative(r 1993, oitar noitpmusnoc eciR and and their 98­ 1998 as 1997 25th 1998 8.8 nonparametric the and well and at as consumption. 1993 93­ 1993 of between 8 ence, ratios, 1993 1992 estimated to ratio, eferr each omfr the pcx for for ln elativer 7 consumption) e ar ovided plotted consumption " consumption illustrates pr is e and minus calculations' Rice ar graph ratio 6 G: consumption. changes" bars Each oduction Authors 1 0 benefit oduction the .75 .5 .25 Panel pr total of ce: oitar noitpmusnoc eciR Note: erticalV (pr to The The All Sour plus tion 173 8.8 8.8 1993 1993 to to 8 8 1998 1998 ratio, pcx ratio, pcx ln ln benefit in production in 67 67 Change C: Change .5 0 .25 .25 .5 1 .5 0 .5 F: Panel oitar tifeneb ni egnahC oitar noitcudorp ni egnahC Panel 8.8 8.8 South 8 1998 1998 Rural ratio, pcx ratio, pcx ln ln Prices, 7 benefit production Rice Net B: Rice in 6 678 E: Panel 1 Panel .75 .5 0 .25 .25 .5 1 .75 .5 0 .25 Changes interval) oitar tifeneb teN oitar noitcudorp eciR of 8.8 8.8 Impact confidence cent 1993 1993 per ratio, 95 pcx ratio, pcx ln ln and Distributional benefit production The ession Net A: Rice egrr 678 678 D: 5.5. Panel 1 .75 .5 0 .25 .25 .5 1 .75 .5 0 Panel .25 oitar tifeneb teN oitar noitcudorp eciR Figure (estimated 174 8.8 1993 oduc- surplus to pr in ors. 8 rice 1998 err ence of d ratio, ar fer dif ratio pcx ln stand the 7 the as is consumption ratio in defined 6 bootstrapped is oduction 1 .5 0 Change on ratio oita.75r noitpmusn.25 oc I: pr the ni egnahC based Panel benefit in 8.8 interval, example, change For the 8 1998 confidence distribution. ratio, while defined. cent pcx ), pcx ln ln per pcx 7 95 the lns' of consumption. consumption analogously plus year e 1992 ar Rice centiles each to 6 H: ession, per to 1 egrr .75 .5 0 .25 Panel 75th elativer changes, VLSSs. oitar noitpmusnoc eciR elative(r and their 98­ 1993, as 25th 1998 1997 and 8.8 nonparametric the and well and at 1998 as consumption. 93­ 1993 of 8 1993 ence, ratios, 1993 1992 estimated to ratio, eferr each between omfr the pcx for for ln elativer 7 e consumption ar ovided plotted " consumption illustrates pr is consumption) and e calculations' Rice ar 6 graph ratio G: minus consumption. changes" bars Authors 1 0 Each oduction the .75 .5 .25 Panel benefit pr total of ce: oitar noitpmusnoc eciR Note: erticalV to The oduction The All Sour (pr tion 175 176 Economic Growth, Poverty, and Household Welfare in Vietnam 10 percent of income. The net benefits declined slightly with lnpcx, suggest- ing that the benefits were proportionately concentrated among lower- income households. Panel B suggests that, by 1998, the net benefits were lower, especially at the lower end of the income distribution. In fact, the net benefits were negative for the poorest households. This can occur, for exam- ple, because consumption prices have risen faster than producer prices, and rice consumption remains especially important in the consumption basket of the poor. Panel C looks at the change in the net benefit ratio. This is the change in the household rice surplus as a percentage of 1993 income (con- sumption). Clearly, households in the bottom 25 percent saw declines in their rice surplus position. For the majority of households in the north, the change in rice surplus was barely positive and was slightly more so for richer households. A slightly clearer understanding of what was happening can be obtained by splitting the net benefit ratio into its production and consumption com- ponents. Panels D, E, and F demonstrate that revenue from rice production represents a significant fraction of household income, and the importance of rice declines with household income (lnpcx). On the income side, then, in- creases in rice prices are strongly pro-poor. This relationship shifts down somewhat between 1993 and 1998. Still, as can be seen in panel F, the com- bined impact of higher rice prices and greater rice output yielded benefits that were concentrated among lower-income households. Whatever else was going on in terms of income generation, the liberalization of rice mar- kets seems to have unambiguously served to reduce income inequality. In- come does not translate directly into welfare, however (as the net benefit ra- tios show). In panels G, H, and I, higher rice prices for consumers offset most of the gains to income. The consumption ratios are essentially Engel curves, and they have the expected negative slope (as seen in the regressions reported in tables 5.4 and 5.5). Panel I demonstrates the very steep relation- ship between changes in rice expenditures and a household's lnpcx. Poor households increased their rice expenditures significantly more (as a per- centage of consumption) than rich households. As a result, the burden of the consumer rice price increases fell disproportionately on the poor. As shown in panel C, for the poorest households, the increase in consumer prices far outweighed the gains from higher producer prices. Figure 5.5 reports the results for the south. The patterns are similar to those in figure 5.4 with one important difference: The net benefits are much higher in the south. Across panels A to C, the net benefits in the south are higher each year than they are in the north, and they are also positive for most households in the income distribution. In fact, panel C shows that there were many more winners from liberalization in the south, because most households saw an increase in their rice surpluses. Furthermore, if anything, these net benefits were concentrated in the middle of the distribu- tion. Whether this reduced the inequality of welfare depends on the social welfare function used to weigh the social benefits of income to rich and poor Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 177 households, as well as the associated inequality index. Still, it does not ap- pear that benefits accrued disproportionately to the better-off, which would have unambiguously increased inequality. The disaggregated figures are also informative. The consumption pic- tures are similar to those for the north, but the production figures are strik- ing and tell most of the story. Although the benefits to rice revenue of higher rice prices are more evenly spread out in 1993, the distributional impact of higher prices in 1998 may still have served to equalize incomes, especially increasing incomes in the middle of the lnpcx distribution. In general, the change in the rice production ratio shows that households in the south ben- efited proportionately more than those in the north, and the increases in income from higher rice output and higher prices were shared relatively equally. This goes some way toward explaining why income inequality fell in the rural south. Liberalization of rice prices has been a double-edged sword, increasing the incomes of the poor, especially in the south, but also increasing the cost of food, which falls most heavily on the poor. On balance, except for the very poorest farmers, southern farmers across the income distribution benefited from the changes, while most northern households were slightly better off, except at the bottom of the income distribution. Note, however, that these conclusions ignore the fact that overall incomes in both the north and south have increased, and rice price liberalization may have facilitated the move- ment of northern farmers out of rice and into other crops. It is also not obvi- ous why such a gap has emerged between the producer and consumer prices. If people are voluntarily choosing to buy better-quality rice, for example, some of the adverse impact of the increase in rice consumer prices will be overstated. The results suggest, however, that in terms of rural wel- fare, there remains room for further liberalization in rice marketing. The Contribution of Agriculture to Income Inequality In this final exercise, the degree to which agriculture (and possibly increas- ing inequality of agricultural income) contributes to overall inequality will be examined by decomposing total income inequality by income source, using a method developed by Shorrocks (1982, 1983) and applied by Benjamin and others (2002). Shorrocks shows that, under reasonable as- sumptions, a decomposition of total inequality by source of income can be calculated whereby the decomposition applies to any inequality index. Basi- cally, the decomposition allows an answer to this question: What fraction of total income inequality is generated by inequality of income from income source k? Assuming that there are k sources of income--for instance, income from farming, wages, and other sources--the proportion of total inequality in Sk, deriving from yk, is given by: Sk = cov(yk, y) var(y) 178 Economic Growth, Poverty, and Household Welfare in Vietnam where yk is the income derived from source k, and y is total household income. Sk can easily be estimated by the following regression: ykh = 0 + k yh + h. The coefficient k yields the estimate of Sk. The key question is how to interpret Sk (that is, when is it "big" or "small"?). One benchmark is to compare Sk to zero. If Sk is negative, in- creases in the inequality of income source k will actually reduce inequality, reflecting that yk is an income source earned primarily by the poor. Alterna- tively, Sk can be compared to Wk, where Wk is the share of income derived from source k. Because the rich tend to earn more income from all sources, increases in inequality of any type of income will increase overall income in- equality. However, some sources will be relatively less disequalizing, and Wk is a useful benchmark. If Sk is greater than Wk, increases in the inequality of the distribution of yk can be viewed as disproportionately increasing income inequality, whereas if Sk is less than Wk, the income source is rela- tively less disequalizing. Two statistical issues may affect the interpretation of the decomposition. First, it may be desired to net out the spatial contributions to inequality. For example, it may be that nonfarm income generates most inequality. How- ever, this may simply reflect the possibility that areas with greater nonfarm incomes are richer. The decomposition would correctly attribute total in- equality to nonfarm incomes, but the results would not imply that increases in nonfarm income in poorer areas would increase inequality. The sensitiv- ity of the conclusions to spatial variation in the composition of income can be evaluated by allowing for cluster FEs and identifying k from the within- cluster variation in incomes. A second issue in the decomposition is mea- surement error. Any mismeasurement of an income source will lead to a spurious positive relationship between yk and y. Again, the OLS decompo- sition is mechanically correct, but interpretation is difficult. Essentially, the aim of this exercise is to estimate the correlation between a given income source and a household's position in the income distribution. The measure- ment error in this case could lead to either an overstatement of this relation- ship or an understatement (through conventional attenuation bias). A standard fix for this type of measurement error would be to use an alterna- tive estimate of the household's position in the income distribution as an instrument. In this case, household per capita consumption is used as an instrument for per capita income. Table 5.11 reports the results of the decompositions separately for 1993 and 1998. The sample here, as before, is the panel of households for which data for both 1993 and 1998 are available. Three alternative estimates are re- ported here: OLS, OLS with cluster FEs, and two-stage least square (2SLS) with cluster FEs, where total income is instrumented by consumption to adjust for possible measurement error in living standards. The focus here is on the changing role of agriculture in inequality, begin- ning with rural Vietnam (north and south combined). As seen before, Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 179 Table 5.11. Decompositions of Rural Income Inequality, by Source of Income 1993 1998 Coefficient on per Coefficient on per capita income capita income Cluster effects Cluster effects Source of income Share OLS OLS 2SLS Share OLS OLS 2SLS Rural Vietnam Wages 0.110 0.041 0.022 0.002 0.120 0.048 0.023 0.024 Family business 0.152 0.534 0.552 0.252 0.177 0.404 0.429 0.307 Farming 0.423 0.141 0.138 0.290 0.347 0.268 0.262 0.229 Animal husbandry 0.069 0.034 0.040 0.048 0.087 0.029 0.041 0.045 Services, durables 0.109 0.061 0.054 0.165 0.131 0.104 0.089 0.155 Other 0.137 0.187 0.194 0.243 0.137 0.147 0.157 0.240 Rural north Wages 0.062 0.027 0.013 0.046 0.087 0.062 0.044 -0.072 Family business 0.133 0.519 0.530 0.125 0.165 0.397 0.409 0.308 Farming 0.466 0.075 0.071 0.219 0.316 0.141 0.167 0.089 Animal husbandry 0.105 0.042 0.046 0.131 0.116 0.058 0.041 0.065 Services, durables 0.097 0.054 0.045 0.120 0.146 0.127 0.094 0.146 Other 0.167 0.282 0.295 0.358 0.170 0.215 0.218 0.320 Rural south Wages 0.162 0.044 0.027 - 0.013 0.153 0.028 0.011 -0.007 Family business 0.174 0.545 0.563 0.296 0.189 0.414 0.440 0.307 Farming 0.411 0.176 0.174 0.314 0.379 0.332 0.315 0.319 Animal husbandry 0.029 0.034 0.037 0.020 0.058 0.017 0.041 0.031 Services, durables 0.121 0.064 0.058 0.180 0.117 0.092 0.085 0.161 Other 0.104 0.137 0.140 0.203 0.105 0.116 0.123 0.190 Note: 2SLS = two-stage least squares. OLS = ordinary least squares. Value of the services ob- tained from household durables, including housing. FEs specification includes controls for cluster FEs. The 2SLS specification instruments total income by household total consumption. Source: Authors' calculations from the 1993 and 1998 VLSSs. farming represents a significant portion of rural household income. In 1993, its share was 42.3 percent, and in 1998, it was only slightly lower, at 34.7 per- cent. All but a small percentage of this is income from crop production. (The rest is the value of crop by-products.) The OLS estimate for 1993 for Sk is 0.141, which is considerably less than that of farming's share. Adding clus- ter effects has no impact on this estimate, and the 2SLS estimate is 0.290, which is larger than the OLS. This arises largely because of the spillovers from correcting the measurement error in income from family-run busi- nesses. Family business income is notoriously difficult to measure, and this almost certainly contaminates the OLS estimates of the decomposition. Even 180 Economic Growth, Poverty, and Household Welfare in Vietnam after instrumenting, the contribution of family-run business income to total inequality significantly exceeds its income share. However, the OLS esti- mates are even higher, which (because of adding up) leads to an underesti- mate of the impact of agricultural income on overall inequality. Throughout the remaining discussion, the focus will be on the 2SLS estimates. Turning to the 1998 figures, the contribution of agriculture to inequality has risen to 0.229, but it is still lower than agriculture's share of income (0.347). The next two panels of table 5.11 provide the breakdowns by north and south. One finding worth noting is the strongly equalizing impact of wage income: Clearly, it appears that the development of off-farm labor markets will improve income inequality. Another finding common to both regions is that inequality of income from family-run businesses is the largest contributor to overall income inequality. In the north, the impact of farming on income inequality fell from 1993 to 1998 (as did the share of income from farming) while it rose slightly in the south (with only a slight decline in its importance). Given the results from the previous section, it appears that in- equality of nonrice farm income may be generating some income inequality. However, the overall conclusion that can be drawn from this exercise is that inequality of farm income is not the primary source of inequality in rural areas. Concern that agricultural reforms may have adverse distributional consequences, because of increasing inequality in farm incomes, seems unwarranted. Conclusions While there have undoubtedly been continuous changes--both subtle and not so subtle--in the institutional environment in which farmers make deci- sions, the most dramatic and observable changes over the 1993­98 period in- volved liberalization of the rice market. The panel dimension of the rich VLSSs has been exploited in this chapter to document changes in prices and explore the possible impact of these changes on efficiency and equity in rural Vietnam. Using these data, the following conclusions can be drawn: Prices in the VLSS data reflect the main policy changes: (a) rice producer prices rose in all regions, but especially in the south, where the implicit tax of the export quota had most depressed prices; (b) fertilizer prices declined dramatically in all regions. It also appears that the rice market became more integrated as the variation in rice producer prices fell between 1993 and 1998 and the correlation of price with region also declined. Rural households experienced rapidly rising incomes over this period, and some of this increase can be directly--and indirectly--linked to rice market liberalization. First, even if there had been no behavioral response by farmers, the higher rice prices that arose from the reforms significantly in- creased the value of rice output while the sharp drop in fertilizer prices cut farm costs. Second, the increased incentives in rice farming line up with the anticipated behavioral response: Rice output increased--especially in the south, where prices rose the most--and farmers made considerable in- creases in their use of fertilizer, further improving yields. Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 181 Despite the direct linkages between rice market liberalization and in- come, however, the greatest increases in agricultural output occurred in nonrice crops. Cheaper fertilizer may have facilitated this expansion, but the primary driving forces appear to have been changes in local food demand patterns, combined with broadening export opportunities, that provided in- centives for farmers to expand their nonrice production, especially produc- tion of perennials. The changing patterns of returns to various agricultural activities also appear to have shifted geographic patterns of production and marketing, with rice production shifting to the south and the north moving toward non- rice crops. At the same time, there were significant increases in the amount of domestic trade in rice at both the national and household levels: Vietnamese households are increasingly reliant on markets for obtaining rice. Not surprisingly, given that the burden of the export quota fell most heavily on the south, and especially on farmers in the Mekong Delta, a re- laxation of the quota yielded disproportionate benefits for southern farmers. This served to widen regional income differences as relative rice production and incomes returned to historical patterns of specialization. However, total rural income inequality did not increase, despite the in- creased regional inequality. This occurred because of significant reductions of inequality among southern households. This chapter has shown that in both the north and south, increased producer prices for rice primarily bene- fit poorer and middle-income farmers. The increases in incomes from rice thus tended to equalize household incomes. To some extent, the corre- sponding increases in consumer prices undid these benefits, especially for the urban poor. However, taking into account the increases in rice consumer prices, this chapter did not find that the overall (net) impact of the increase in rice prices had an adverse impact on inequality among rural households. Evidence of increasing income inequality within regions was not found by this study. Moreover, compared with other sources of income (such as family-run businesses), inequality of agricultural incomes was found to con- tribute less than its share to total inequality. This suggests that policymakers need not be concerned about adverse linkages between further agricultural liberalization and income inequality. This chapter finds that the agricultural reforms had a largely beneficial impact on the well-being of rural households throughout Vietnam. An im- portant follow-up question is whether these benefits can be expected to con- tinue. There are certainly reasons to be cautious in extrapolating results from the 1993 to 1998 period to the next five years. First, to the extent that in- creased use of fertilizer is driving improved yields, there are technological limits as to the degree that additional fertilizer can further increase yields. Second, if the incentives to cultivate rice more intensively or more efficiently are driven by increased rice prices--or a continuation of other institutional reforms--a limit to the opportunities for this type of growth can be antici- pated as prices converge to international levels. Again, technological con- straints eventually dominate improvements in incentives. Finally, care must be taken in extrapolating time trends from any two years of data. As was 182 Economic Growth, Poverty, and Household Welfare in Vietnam seen from comparing the VLSS data to aggregate data, the overall trends ob- served from 1993 to 1998 are consistent with aggregate trends. However, the relative performance of the south and the north observed may in part reflect mean reversion in that 1993 was a relatively good year for the north, com- pared with 1998 for the south. There may also be lags in the realization of output of perennials, especially in the north, so that the observed divergence is transitory. That said, the broad picture provided by the VLSSs suggests that even over the brief span of five years, farmers responded to the changes in their economic environment generated by the reforms, and these responses are not likely to be reversed in the years ahead. Notes The authors thank Daniel Lee for exceptional research assistance and the Social Sci- ences and Humanities Research Council of Canada and the World Bank for financial support. The authors are also grateful to Paul Glewwe for detailed suggestions, and for comments from Chris Gibbs and participants at the World Bank conference in Hanoi, May 2001. 1. The 1992­93 survey spanned a full year, starting in October 1992; the 1997­98 survey began in December 1997 and also lasted a year. For brevity's sake, reference is made to the surveys as the 1993 VLSS and 1998 VLSS, respectively. 2. In the 1993 VLSS, the country was divided into seven administrative regions for the purpose of sampling. The north is defined in this chapter to include the Northern Uplands, the Red River Delta, and the North Central Coast. The south con- sists of the Central Coast, the Central Highlands, the Southeast, and the Mekong Delta. In the 1998 VLSS, eight administrative regions were used, with the Northern Uplands subdivided into the Northeast and Northwest. 3. See the reports completed for the Asian Development Bank (1996, 2000) and prepared by Goletti (1998); Goletti and Minot (1997); and Goletti, Minot, and Berry (1997) for a more detailed overview of changes in rice policy, as well as a compre- hensive review of the performance of the agricultural sector in Vietnam. Timmer (1996) discusses some of the general issues concerning economic transition and the role of agriculture in economic development as they may pertain to Vietnam. 4. See Minot and Goletti (1998, p. 739). 5. See Bautista (1999) and Tran (1998) for preliminary discussion of the impact of Vietnamese agricultural policy changes and farmer productivity. 6. If Engel's Law holds--that is, if the income elasticity for food is less than one--the overall increase in the demand for food is expected to be less than propor- tional to the increase in incomes, and the food share of expenditures is expected to fall. This would correspond to a diminished share of agriculture in economic activ- ity. Of course, the demand for individual agricultural commodities can be consider- ably more income elastic, and the composition of agricultural output can be expected to shift toward these "luxuries." 7. See World Bank (2000) for more details concerning the VLSS, in particular the sampling frame of 1998 compared with 1993. 8. See Wiens (1998) for additional discussion of the role of agriculture in house- hold incomes, based on the 1993 survey. 9. The monthly price changes are from the Vietnamese consumer price index (CPI), which is constructed on the basis of a basket of food and nonfood items. The Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 183 regional price indexes were constructed by the General Statistical Office and are for 1998 (World Bank 2000). These indexes are discussed in more detail in World Bank (2000). 10. This is a useful point at which to provide a comparison of the panel house- holds to the two cross-sections (1993 and 1998). The panel households are virtually in- distinguishable from the full cross-section in 1993: The ratio of cross-section to panel household incomes is 1.02 in urban areas and 1.00 in rural areas. Some differences emerge for the urban households in 1998. The ratio of cross-section to panel house- hold incomes is 1.11 in the south (urban) and 0.95 in the north (urban). Thus, the cross-section would yield a slightly greater widening of the north-south differential for urban areas. The rural samples are closer: The ratio of cross-section to panel house- hold income in 1998 is 0.98 in the north and 0.94 in the south. The cross-section data thus imply slightly less widening of rural incomes between north and south (by 4 per- centage points). However, these differences are very small. The overall story is not sensitive to whether the analysis is restricted to the panel households. Also note that some of the divergence may be driven by the change in the national sampling frame in 1998, as opposed to sample attrition from the 1993 survey. That this may be the case is underlined by the similarity between the panel and nonpanel households in 1993. 11. These results broadly corroborate those in Glewwe, Gragnolati, and Zaman (2002). 12. A similar possibility in data drawn from a household survey conducted in north China in 1995 has been identified. In Benjamin and others (2002), evidence is found that higher-income households save higher fractions of their incomes; thus, consumption-based measures of inequality may understate true differences in income-earning potential. 13. "Theoretical" here means that, in contrast to unit values, these prices are based on a price survey with "unconsummated" transactions, as opposed to prices based on realized transactions. 14. Of course, this ignores general equilibrium considerations, such as price changes, and other factors. 15. In rural China, increasing incomes have helped spur rapid investment in greenhouses and other more capital-intensive (and lucrative) forms of agriculture. 16. In 1993, land productivity, measured in terms of either rice yields or the gross value of agricultural output per unit of land, was significantly lower in the south than the north. As a result of the more rapid growth in the south between 1993 and 1998, the gap largely disappeared. Professor Jean-Pascal Bassimo (Paul Valery University, and Centre for International Economics and Finance, Chateau Lafarge, Aix-en Provence) has informed the authors, in personal correspondence (January 2002), that in the 1920s, rice yields in the north and south were actually similar. 17. An important avenue of future research will be to evaluate the distributional consequences of this striking increase in income from perennials and evaluate the possible role that land security played in encouraging investment in trees. 18. The national data suggest that income from animal husbandry grew at roughly the same rate as that from crop production. 19. Note that the 1986­98 period shows a marked divergence between north and south: The index of 1998 total agricultural output relative to 1986 is 1.96 for the south, compared with 1.56 for the north. For rice, the corresponding 1998 to 1986 out- put indexes are 1.87 for the south and 1.64 for the north. 20. Although some households that were not farming in 1993 were farming in 1998, this entry into farming is more than offset by the decision of a significant number of households to exit agriculture. 184 Economic Growth, Poverty, and Household Welfare in Vietnam 21. To get around the log of zero problem for input use, the log of input use is specified as ln(1 + Xit). In general, this would not be the right approach, but the in- clusion of the dummy variables for whether input use is or is not nonzero tidies up the impact of this otherwise arbitrary transformation. 22. This corresponds to Mundlak's original motivation for using an FEs specifi- cation for the estimation of farm production functions. 23. Although the contribution of productivity growth measured in percentage terms is the same in the north and the south, the much higher growth in the south implies a much larger role of increases in productivity. 24. Labor input is not broken down by crop in the survey. 25. In physical terms, the north goes from a slight per household deficit of 138.4 kilograms to a surplus of 116.3 kilograms; in the south, the surplus per household goes from 632.4 kilograms to 2,009.1 kilograms. Differences in producer and con- sumer prices explain the fact that, in value terms, the north was in surplus in 1993. 26. Specifically, the Fan nonparametric regression estimator, as described in Deaton (1997, chapter 4), has been used. Bibliography The word processed describes informally reproduced works that may not be commonly available through libraries. Asian Development Bank. 1996. "Rice Market Monitoring and Policy Op- tions Study." Final Report, TA 2224-VIE. Prepared by the Interna- tional Food Policy Research Institute, Washington, D.C. _____. 2000. "Vietnam Agricultural Sector Program: Phase I Technical Report." ADB TA 3223-VIE. Hanoi. Bautista, Romeo M. 1999. "The Price Competitiveness of Domestic Rice Production in Vietnam: Effects of Domestic Policies and External Factors." ASEAN Economic Bulletin 16(1): 80­94. Benjamin, Dwayne, Loren Brandt, Paul Glewwe, and Li Guo. 2002. "Markets, Human Capital, and Inequality: Evidence from China." In Richard Freeman, ed., Global Inequality: Where Are We and Where Are We Headed? New York: Macmillan. Deaton, Angus. 1989. "Rice Prices and Income Distribution in Thailand: A Non-Parametric Analysis." The Economic Journal 99(395): 1­37. _____. 1997. The Analysis of Household Surveys. Baltimore, Md.: Johns Hopkins University Press and World Bank. General Statistical Office of Vietnam; Department of Agriculture, Forestry and Fishery. 2000. Statistical Data of Vietnam Agriculture, Forestry and Fishery, 1975­2000. Hanoi: Statistical Publishing House. Glewwe, Paul, Michele Gragnolati, and Hassan Zaman. 2002. "Who Gained from Vietnam's Boom in the 1990s?" Economic Development and Cultural Change 50(4): 773­92. Agriculture and Income Distribution in Rural Vietnam under Economic Reforms 185 Goletti, Francesco. 1998. "Trade Distortions and Incentives in Agricultural Trade: The Case of Rice, Sugar, Fertilizer, and Livestock-Meat- Feed Sub-Sectors in Vietnam." Background paper prepared for Vietnam Rural Development Strategy, World Bank, Washington, D.C. Processed. Goletti, Francesco, and Nicholas Minot. 1997. "Rice Markets, Agricultural Growth, and Policy Options in Vietnam." Market and Structural Studies Division Discussion Paper 14. International Food Policy Research Institute, Washington, D.C. Goletti, Francesco, Nicholas Minot, and Philippe Berry. 1997. "Marketing Constraints on Rice Exports from Vietnam." IFPRI Working Paper. International Food Policy Research Institute, Washington, D.C. Minot, Nicholas, and Francesco Goletti. 1998. "Export Liberalization and Household Welfare: The Case of Rice in Vietnam." American Journal of Agricultural Economics (November): 738­49. Shorrocks, Anthony F. 1982. "Inequality Decomposition by Factor Compo- nents." Econometrica 50(1): 193­211. _____. 1983. "The Impact of Income Components on the Distribution of Family Incomes." Quarterly Journal of Economics 98(2): 311­26. Timmer, C. Peter. 1996. "Agriculture and Economic Growth in Vietnam." In R. A. Goldberg, ed., Research in Domestic and International Agribusiness Management, vol. 12. Middlesex, U.K.: JAI Press Ltd. Tran, Thi Que. 1998. "Economic Reforms and Their Impact on Agricultural Development in Vietnam." ASEAN Economic Bulletin 15(1): 30­46. Viet Nam Living Standards Survey (VNLSS). 1992­1993. Web site: www. worldbank.org/lsms/country/vn93/vn93bid.pdf. Vietnam Living Standards Survey (VLSS). 1997­98. Web site: www. worldbank.org/lsms/country/vn98/vn98bif.pdf. Wiens, Thomas B. 1998. "Agriculture and Rural Poverty in Vietnam." In David Dollar, Paul Glewwe, and Jennie Litvack, eds., Household Welfare and Vietnam's Transition. Washington, D.C.: World Bank. World Bank. 2000. "Vietnam Living Standards Survey (VLSS), 1997­98: Basic Information." World Bank, Poverty and Human Resources Division, Washington, D.C. Processed. Part II Poverty Reduction in Vietnam in the 1990s 6 The Static and Dynamic Incidence of Vietnam's Public Safety Net Dominique van de Walle Vietnam has a system of centrally determined and mandated poverty and social welfare programs that are implemented by local authorities according to local norms, local poverty standards, and--in large part--local financing. Resources are scarce. Although they may be intended to cover the mandates, insufficient central and provincial allocations may never even reach the communes. These central allocations must inevitably be supplemented by means of local resource mobilization. There is evidence that the rural popu- lation, and the poor among them, are heavily taxed, including through numerous locally levied "fees, charges, and other contributions" (Govern- ment of Vietnam­Donor Working Group 2000). In addition, standards of "poverty" used by different authorities vary across locations; these various standards often simply mirror local resources. For these reasons, there is thought to be uneven coverage and leakage of poverty and social welfare programs. The poorest in Vietnam often need to rely on charity from within their communities, but the communities in which they live are often poor, so other households have little to spare. In this context, too, it has been argued that coverage among Vietnam's poor may be quite uneven spatially, with poor people who live in poor areas faring much worse than poor people who live in well-off areas (Rao, Bird, and Litvack 1999; van de Walle 1999). The decentralized nature of Vietnam's public safety net also raises wider concerns from recent literature (Bardhan and Mookherjee 2000; Conning and Kevane 1999; Galasso and Ravallion 2004). A popular argument in recent years is that decentralized programs are better at reaching the poor. The contention is essentially that local authorities are better placed to accurately identify and target poor people and their problems. Against that, 189 190 Economic Growth, Poverty, and Household Welfare in Vietnam counterarguments can be made that local entities may not share the objec- tives of the central government and may be more liable to political capture. In the light of these concerns--both specific to Vietnam and more general--this chapter examines the extent to which existing programs and expenditures on poverty reduction in Vietnam are well targeted to poor communes and poor people. Surprisingly little is known about this issue. Cross-province regressions of budgetary allocations for health- and education-related national programs strongly suggest that transfers from the center are progressive, in that they result in higher per capita spending in poor and middle-income provinces (Fritzen 1999). Fritzen also finds that central health transfers are well targeted based on health needs. However, little is known about the within-province allocations to districts and com- munes. Others have noted the lack of cross-commune redistribution of re- sources and the consequent disparities between communes in their abilities to provide basic services and assistance to the local poor (Litvack 1999). Moreover, nationally representative data on household-specific program in- cidence have not been available for more than one or two programs. Fortu- nately, new data from the 1997­98 Vietnam Living Standards Survey (VLSS) enable an analysis of the incidence across households and communes of some social welfare- and poverty-related initiatives and provide an oppor- tunity to explore these concerns more rigorously. The availability of an earlier dataset for 1992­93 also allows some comparisons over time, includ- ing longitudinal comparisons for the same households. Total spending on certain transfers more than doubled between 1992­93 and 1997­98. This situation permits an interesting study in who benefited from the changes in outlays. The main question this chapter addresses is whether current public so- cial welfare programs are targeted to the poor.1 In answering this question, the chapter explores sensitivity to the definition of poverty and what is as- sumed about household behavioral responses to the programs. The chapter examines whether programs perform a safety net function, recognizing that this involves both protection from poverty and promotion from poverty (Dreze and Sen 1989). The chapter also examines the role of non- income factors, including whether equally poor communes in different provinces are treated equally and, if not, what accounts for any differences in treatment. The chapter begins with a discussion of the background, setting, and overall system of poverty alleviation and safety net programs and their fi- nancing. The next two sections discuss the data and welfare measurement. Implications for the incidence of program spending are next addressed, fol- lowed by a section that looks at how much the system protects versus pro- motes the poor. This is followed by an examination of the importance of fac- tors other than welfare to incidence, including where one lives, and then conclusions. Despite experiencing a large reduction in poverty since embracing the market economy in the late 1980s, Vietnam remains a poor country, with The Static and Dynamic Incidence of Vietnam's Public Safety Net 191 more than one-third of its population in poverty. Vietnam's population is primarily rural; they are engaged in small-scale agricultural activities and subject to seasonality in incomes, recurring natural disasters, and other sources of vulnerability and impoverishment. Geographic differences and the existence of disadvantaged ethnic minority (non-Kinh) groups add to the complexity of the poverty picture (van de Walle and Gunewardena 2001). The country also faces severe budget constraints. Yet on paper, Vietnam has--by the standards of low-income countries-- an extensive social security and safety net system. This reflects a strong historical commitment to combating inequality and raising the living stan- dards of all its regions and people. The surviving concern for and frequently expressed political commitment to ensuring a minimum level of welfare for all and maintaining a low variance in incomes also do much to preserve the regime's political legitimacy, but the government's aspirations in this area are often overshadowed by its lack of resources. Doi Moi ("renovation") profoundly changed the way social services were delivered, leaving peasants more vulnerable (Glewwe and Litvack 1998; Kolko 1997). Cooperatives that had financed and supported health and edu- cation services for their members, as well as insurance against shocks, were disbanded in 1988. The social protection system that has evolved since decol- lectivization is composed of a number of different initiatives that are centrally mandated but locally implemented and often rely heavily on local resources.2 · The Social Security System provides pensions and other employment- related social insurance payments to formal sector workers. Public servants and armed forces personnel have been covered since 1947. In 1995, the scheme was expanded to private sector employees working in firms with 10 or more employees (Ministry of Labor, Invalids and Social Affairs [MOLISA] 1999). Although these social insurance payments are employment related and eventually meant to be fully funded from payroll taxes and employee contributions, they continue to be heavily subsidized by the central budget. · The Social Guarantee Fund for Veterans and War Invalids extends compensation and assistance in the form of social subsidy transfers to those who contributed and suffered from the war efforts--such as disabled veterans, relatives of deceased soldiers, and others who contributed to the revolution. · The Social Guarantee Fund for Regular Relief targets assistance to those unable to support themselves, including the disabled, orphans, and the elderly. Here especially, however, scarce public resources imply that implementation and coverage ultimately depend in large part on local-level governments and resources. · The central government also runs a Contingency Fund for Preharvest Starvation and Natural Disasters, whose role is to minimize the con- sequences of natural calamities and other emergencies by dispensing disaster relief to regions and households. 192 Economic Growth, Poverty, and Household Welfare in Vietnam · Finally, the government has devised a number of National Develop- ment Programs that aim to reduce poverty and are often targeted to "poor and remote" communes. These include interventions such as employment generation, reforestation, school and health fee exemp- tions, microcredit schemes, and physical infrastructure investments. Their focus is generally more on promoting growth than on provid- ing protection. In 1996, the government also proposed a national Hunger Eradication and Poverty Reduction (HEPR) program to coordinate existing and new ef- forts, as well as the resources for combating poverty. Since then, many pub- lic programs have been consolidated under the HEPR national program to better mobilize and coordinate antipoverty resources. Under the HEPR um- brella, the government implemented the National Target Program on Poverty Alleviation between 1998 and 2000 and has recently prepared a Poverty Alleviation Strategy for 2001­10 (MOLISA 2001). The HEPR and these other efforts do not appear, however, to have entailed much change in policy focus. The policy areas have all been emphasized in the past and have been addressed by past programs and a variety of ad hoc schemes. In addi- tion, there is little new funding for HEPR from the center. New poverty mandates and targets are imposed on ministries by HEPR without the ben- efit of additional funding or reductions in other mandated responsibilities (Nguyen 1999; van de Walle 1999). Throughout these programs, eligibility criteria, guidelines, and norms are largely dictated by the center, and implementation is chiefly the re- sponsibility of the communes. Poverty and needs are locally determined following national norms, but heavily influenced by available local means and resources. Clearly, there are incentives for the local-level agencies to overstate their needs and understate their resources. Communes initially draw up lists of eligible candidates (people or households, depending on the program) for the different social protection programs to reflect their needs. These are gathered, altered, and eventually approved and passed on to the center by the districts and the provinces. After a process of review and negotiation among Vietnam's Ministry of Finance, Ministry of Plan- ning and Investment, and MOLISA in Hanoi, transfers are made to the provinces. Although transfers from the central budget appear to be insufficient to cover local needs or even centrally mandated spending, there is evidence that they are quite redistributive, aiming to equalize resources across provinces (Rao, Bird, and Litvack 1999). However, use of the funds and in- traprovincial distribution are largely at the discretion of the provincial au- thorities. The evidence suggests that the redistributive process often breaks down at this level (Litvack 1999). Provinces distribute resources to districts based on criteria that vary widely from one province to another. Similarly, districts distribute to communes in disparate ways, and there is great dis- parity in the resources available to communes. Expenditure mandates are The Static and Dynamic Incidence of Vietnam's Public Safety Net 193 sometimes ignored and sometimes funded from other recurrent transfers or locally mobilized resources ("contributions"). There is often pressure on the communes to raise the resources to implement central programs through charging various fees and levying "voluntary contributions" from their pop- ulations. Communes are likely to contribute their own additional resources depending on several factors, including the economic status of households in the commune and local leadership. It is nevertheless likely that the need- iest communes are often the ones that are least able to mobilize local funds. Existing fiscal arrangements that ensure progressive redistribution to poor provinces, at least for some programs, are nevertheless likely to lead to low and uneven coverage and horizontal inequity as a result of the lack of cen- tral incentives or mandates for targeting the poor within provinces. Statistics published by MOLISA (1999) show the large gap between the numbers of people who are eligible for each of the social welfare programs and the actual numbers of beneficiaries. The probability of participation is likely to depend on local budgets and leadership--hence on where the potential beneficiaries live. In exploring the implications for the poor of the existing safety net in Vietnam, this chapter emphasizes a number of concerns. One issue relates to defining "the poor." This chapter uses per capita consumption expenditures as its general welfare measure but recognizes that some components of the observed household consumption data reflect public transfers. This has im- plications for drawing conclusions about the counterfactual situation-- what welfare would have been without transfers--and hence about the incidence of transfers. The chapter describes a method for dealing with this concern. A second issue addressed by this chapter concerns how the safety net performed over time. In principle, a safety net can reduce poverty either by protecting nonpoor people from becoming poor or by promoting poor people out of poverty. How does Vietnam's existing safety net perform in both functions? With panel data, methods exist to address this question (Ravallion, van de Walle, and Gautam 1995). These methods are applied to Vietnam's safety net. Afinal issue addressed by this chapter concerns possible determinants of program incidence other than consumption expenditures. One possibility is that interventions are aimed at nonincome dimensions of welfare, so that the incidence picture based on consumption gives a skewed view of targeting. Another possibility is that, given public institutional arrangements for de- livering social welfare programs, nonwelfare factors--in particular, political and geographic factors--may be found to matter a great deal in determining whether transfers reach the poor. In this respect, it may not be poverty that attracts benefits but rather the characteristics of the commune where the po- tential beneficiary lives. Communes in better-off provinces will generally have more resources for helping the poor. To what degree is the interaction of geography and low living standards the determining factor in whether the poor are assisted nationally? 194 Economic Growth, Poverty, and Household Welfare in Vietnam Data The analysis is based on the nationally representative 1993 and 1998 VLSSs.3 These are multitopic household consumption expenditure surveys with modules covering numerous aspects of living standards.4 The surveys cov- ered 4,800 households spread across 150 communes in 1993 and 6,000 households living in 194 communes in 1998. In both years, a community questionnaire was administered in rural and small-town communes--120 communes in 1993 and 156 communes in 1998. A panel of 4,308 households was also contained in the surveys. The welfare indicator is annual per capita consumption. This includes the value of consumption from own production and the use value of con- sumer durables, including imputed housing expenditures (World Bank 1995, 2000). Consumption expenditures and other monetary amounts are expressed in real January 1998 national prices and therefore take into ac- count both inflation through the survey year and spatial price differences. The 1998 survey sought to improve the measurement of consumption in certain ways. For example, it records the consumption of own-produced nonfood items such as coal, wood, and flowers and strives for a better ac- counting of tobacco consumption. Although some questionnaire changes were introduced in the 1998 VLSS, care was taken to ensure comparability across the two surveys. Two total consumption expenditure measures--one that is the best possible measure for 1998 and another made comparable to the 1993 expenditure totals--are therefore available. For all comparisons over time, this chapter uses the temporally comparable measures of con- sumption but sticks with the best 1998 measure otherwise. The questionnaires changed between the two surveys in certain other re- spects as well. In particular, the 1998 VLSS contains considerably more infor- mation on government programs and policies than the 1993 survey. This puts certain limitations on the types of issues that can be examined with re- spect to public interventions. The only transfer receipts recorded in 1993 for which a comparison can be made over time are education scholarships, so- cial insurance, and social subsidy funds. Details are also available for 1998 on whether the household received transfers from the poverty alleviation fund or nongovernmental organizations (NGOs). In addition, there is information on the existence of programs and numbers of beneficiaries of various inter- ventions at the commune level for 1998. It should be noted that there are a number of other ways in which the government intervenes to increase social welfare--for example, through subsidizing microcredit and various goods, and disaster relief (MOLISA 1999). Although information at the commune level for some such schemes (for example, disaster relief) permits analysis of geographic reach, an analysis of household-level benefits is not feasible. Behavioral Responses to Transfers In assessing whether programs reach the poor, a first step involves accu- rately identifying them. To accurately distinguish the poor, it is necessary to The Static and Dynamic Incidence of Vietnam's Public Safety Net 195 determine what welfare would have been without the government interventions. Program outcome assessments may depend on that choice-- the appearance of weak targeting may just be due to deficient welfare measurement. Typically, studies of the incidence of public spending subtract the entire amount of government transfer receipts from household income or con- sumption to approximate preintervention welfare and to rank the popula- tion into quintiles, for instance. Netting transfers out fully assumes that there is no replacement through savings, labor effort, schooling decisions, interhousehold transfers, and other potential changes in household behav- ior. That assumption is implausible. Yet treating posttransfer consumption as the welfare indicator instead is just as problematic. Ideally, the interven- tion amount would be subtracted, but the replacement income households would have had if they had not benefited from the intervention would be added. Van de Walle (2003) addresses these concerns by estimating the mar- ginal propensity to consume out of social income (PCSI) (see also Ravallion, van de Walle, and Gautam [1995]). The estimated PCSI is then used to de- termine the net gain to consumption from social transfers and construct the counterfactual consumption level without intervention. This section sum- marizes the key results from van de Walle (2003). The estimate is then used for the chapter's incidence analysis. In the following analysis, transfers com- prise social insurance, social subsidies, and education scholarship receipts-- the only components of social income that can be identified in both surveys. Consumption of household i at time t (t = 1993, 1998) (Cit) is assumed to be represented by an additive function of public transfers (Tit), observed household characteristics (Xit), and time-varying (t) and time-invariant (i) latent factors: (6.1) Cit = + Tit + Xit + i + t + it. There are a number of potential problems with estimating directly with equation 6.1. For example, transfers are likely to be correlated with time- invariant household characteristics [cov(Tit i) = 0], such as if there is pur- posive targeting to the long-term poor. Another possible source of endo- geneity arises if transfers are correlated with time-varying determinants of consumption [cov(Tit t) = 0 or cov(Tit it) = 0]. This would occur if transfers target those who suffered a shock. Alternatively, transfer eligibility may have changed because of the death of an elderly household member who had received a pension. Furthermore, such changes may not all be observed in the data. Finally, the behavioral response, and hence the PCSI, may well vary across households with different household characteristics. A number of alternative specifications are run to test for these possibili- ties. A double differencing model where all variables are expressed in first differences is used to purge the estimate of fixed effects and deal with the first source of endogeneity. Equation 6.1 is then: (6.2) Cit = Tit + Xit + t + it. 196 Economic Growth, Poverty, and Household Welfare in Vietnam Because there are only two rounds of data, in equation 6.2 the term t becomes an ordinary intercept term in a regression of the change in consumption on the change in transfers. This regression was initially run as- suming that Xit = 0 (characteristics do not change or do not have any effect), giving the standard double difference estimate of the consumption impact of transfers. This gives a estimate of 0.45 with a heteroskedasticity and clustering-corrected t statistic of 4.3 (van de Walle 2003). To deal with potential remaining contamination through dependence of the change in transfers on time-varying characteristics, a regression is run that controls for changes in observable household characteristics in the double differencing model of consumption as a function of transfers. A number of variables are found to be significant--changes in household size and in the language of interview have a negative impact, and an older head of household and a higher educational level influence consumption positively.5 The estimate is 0.37 (t = 3.6) and is not significantly different from the initial simple dou- ble difference estimate. To deal with possible omitted variables that alter over time and affect transfers, the last ordinary least squares regression (OLS) is reestimated with the change in transfers instrumented by transfer receipts in the first pe- riod.6 This gives an estimated of 0.72 (t = 3.7). This is higher, but still not statistically significantly different from the first, naive estimate. Van de Walle (2003) also tests for heterogeneity in impacts by adding in- teractions between the change in transfers and household characteristics to the OLS regression with controls for time-varying changes in characteristics. The results suggest that the impact of transfers on consumption is higher in households with more education. However, a test of the joint significance of the interaction terms shows them to be not significantly different from zero. The analysis in van de Walle (2003) suggests a range of estimates of the PCSI, none of which are significantly different from the simple double dif- ference estimate of 0.5.7 So, in the following analysis, consumption expendi- tures are net of half of the value of transfer receipts that can be identified, unless otherwise noted.8 Incidence of Poverty-Related Programs In exploring the evidence from the 1998 VLSS on the incidence of programs and policies aimed at raising living standards, the focus is squarely on the distributional impacts and who is getting how much. In reality, these pro- grams serve other objectives--such as assisting the elderly or those who contributed to and suffered from the war effort--that one may want to take into account when assessing whether to expand or contract them. At the same time, it is often argued that there is a coincidence of objectives and that some of the larger funds--particularly social subsidies--are quite pro-poor. Substantial public resources are spent on these programs--although poverty may not be their sole objective, it is important to ask how much is reaching the poor. The Static and Dynamic Incidence of Vietnam's Public Safety Net 197 In table 6.1, individuals are ranked into national population quintiles on the basis of their household per capita expenditures--net of half of current transfer receipts as discussed in "Behavioral Responses to Transfers." (These are referred to as net quintiles.) The table presents real monetary amounts per capita of various types of public transfers received by households dur- ing the 12 months before the survey. Amounts are expressed averaged over each quintile's population--recipient and nonrecipient. Percentages of the population living in households where at least one member benefited from these transfers are given in table 6.2. In general, outlays are small and there is weak coverage.9 The largest payments are from the Social Insurance Fund, covering pen- sion and disability benefits for civil servants and employees of state-owned enterprises. As a result, one would expect these payments to be more wide- spread in urban areas and to be not particularly pro-poor. In fact, they are predominant in urban areas--where 18.3 percent of the population lives in households where someone received these payments in 1998 (table 6.2). Yet per capita amounts from this source are by far the largest for the poor in urban areas. In rural areas, by contrast, the amounts received from the Social Insurance Fund rise steadily with standards of living. The Social Insurance Fund also touches the greatest number of people of any program (11.2 per- cent of the population, nationally). Social subsidies, which include payments to veterans and the families of those who died during the war, as well as to those who are unable to support themselves, are much smaller in absolute amounts. These programs are often claimed by the government to be reaching the poor in Vietnam. Per capita amounts of these social subsidies are largest for those in the poorest quintile in urban areas. In rural areas, the poorest quintile follows the top quintile with the second largest per capita amounts. In general, receipts are much more even across expenditure levels than for social insurance benefits. Interestingly, mean payments are larger in rural areas, although coverage is relatively similar across the sectors. Actual individual social insurance and subsidy payments are found to vary widely across recipient households. For example, social insurance out- lays range from D (dong) 49,252 to D 21,500,000 a year and social subsidy outlays from D 14,264 to D 8,645,464 a year. It should be noted that some of this variance is expected. For instance, the VLSS does not allow identifica- tion of recipients, and some households may have more than one benefi- ciary. Furthermore, social insurance payments consist of pensions, but they also consist of disability payments that are likely to be lower than pensions. Government-set minimum regular relief transfers also vary across the dif- ferent types of potential beneficiaries (MOLISA 1999). The VLSSs also asked about transfers received under policies or pro- grams supported by the government's scholarship program, its poverty al- leviation efforts, and transfers received from NGOs. Few scholarships are awarded (141 were reported in the sample). Their incidence is regressive: The top quintile has the largest share of recipients as well as the highest of es 8.3 5.2 5.4 4.4 3.1 5.3 7.3 4.7 4.2 3.5 2.5 4.8 centage household Per expenditur e per social otalT welfar income capita) 97,130 87,785 (dong 18,9011 130,764 167,785 120,474 84,431 79,032 93,199 103,195 122,638 92,776 per income 0 508 338 286 443 521 546 384 373 546 capita) 1,030 1,071 (dong NGO per 607 829 654 398 444 161 Poverty income capita) 2,652 1,600 1,268 2,707 1,721 1,249 alleviation (dong per 1 1 772 817 capita) 1,158 1,856 2,806 7,912 2,901 1,058 1,41 2,823 6,01 1,944 Education (dong scholarships per Social subsidies capita) 22,785 17,021 17,556 17,503 18,337 18,639 21,649 17,237 17,437 18,862 23,625 19,340 (dong Income fund per elfare W Social capita) 69,506 67,883 98,543 (dong 109,339 140,439 97,145 57,947 58,712 73,569 80,694 92,885 69,697 insurance Social of of 937 887 917 997 973 607 1,001 1,165 1,319 1,576 5,998 4,381 Number households Incidence 6.1. quintile ableT otalT quintile otalT National population Net 1 2 3 4 5 Rural Net 1 2 3 4 5 198 ex- the for e consist associ- 34.6 12.4 14.1 7.4 3.4 7.3 capita befor Fund per ograms subsidies months pr 12 Guarantee 1 Social omfr household the on Social 413,747 204,200 307,41 221,858 200,010 225,369 the during eceivedr based payments. ty omrf funds NGOs. 0 0 0 0 come all quintiles 760 425 government disabili the and These esents population omfr eprr international pensions facilities. and 0 to income national eceivedr 1,296 2,140 2,104 1,037 1,341 into efersr private oduction pr income alleviation or omfr ranked e insurance ar eported poverty eceivedr 185 3,642 5,126 2,752 9,269 6,527 self-r Social ganizations The or Individuals Relief. assistance quintile. social is household each omfr 51,106 14,148 18,431 13,015 14,562 15,981 of Regular definition. esent income and for eprr VLSS Fund NGO. population veterans 1993 policy amounts the the The oss Guarantee 357,704 189,868 281,715 203,987 174,381 201,095 disabled acr follows and alleviation Social eceipts.r basis the dead 05 48 eakdown capita and poverty 168 346 969 br transfer war 1,617 of per of a half on Invalids ral-urbanur of families ar W government's VLSS. net essed to The es and the 1998 expr ce: quintile otalT with Note: transfers Sour Urban Net 1 2 3 4 5 penditur survey of eteransV ated 199 200 Economic Growth, Poverty, and Household Welfare in Vietnam Table 6.2. Population Receiving Social Welfare Income Percent of population living in households who received National Social Social Education Poverty NGO population insurance subsidies scholarships alleviation income Net quintile 1 9.5 11.6 1.1 6.4 0.5 2 9.1 9.4 0.8 2.1 1.2 3 11.6 9.6 1.9 1.3 0.3 4 12.1 10.0 2.7 0.9 0.2 5 13.9 7.3 5.6 0.2 0.1 Total 11.2 9.6 2.4 2.2 0.5 Rural Net quintile 1 8.7 11.5 0.9 6.6 0.5 2 8.5 9.4 0.7 2.3 1.3 3 9.5 9.6 1.8 1.1 0.3 4 10.1 10.8 2.3 0.8 0.3 5 11.1 9.4 5.1 0.1 0.0 Total 9.4 10.2 1.8 2.6 0.6 Urban Net quintile 1 27.8 13.0 4.6 2.5 0.0 2 17.3 8.9 2.6 0.0 0.0 3 27.0 9.2 2.7 2.8 0.0 4 18.6 7.4 4.1 1.2 0.0 5 15.8 5.9 6.0 0.3 0.2 Total 18.3 7.1 4.9 0.8 0.1 Note: The rural-urban breakdown follows the 1993 VLSS definition. Individuals are ranked into national population quintiles based on household per capita expenditures net of half of transfer receipts. Source: 1998 VLSS. per capita amounts in both rural and urban areas. However, the urban pop- ulation in the bottom quintile is also notable for having the second highest incidence of beneficiaries. In general, scholarships benefit a larger share of the urban than of the rural population. Per capita amounts are also higher in urban than in rural areas. The amounts involved in the poverty alleviation and NGO funds are neg- ligible and are equivalent to approximately US$0.22 per person a year (1998 official exchange rate) in the case of poverty alleviation funds and US$0.08 from NGOs for the quintile with the largest receipts. The little money there is appears to be moderately well targeted in rural areas, in that per capita amounts fall with higher quintiles. However, there is also evidence of capture by the well-off, because all quintiles receive something. This is more pronounced in urban areas for both poverty and NGO transfers. The Static and Dynamic Incidence of Vietnam's Public Safety Net 201 Finally, expressing all transfers together as a share of household per capita expenditures indicates progressive overall incidence in both rural and urban areas.10 Transfers to the urban poor in the bottom national quin- tile account for 35 percent of their consumption, which is quite a contrast with the poorest in rural areas, for whom transfers account for 7.3 percent. Nonetheless, it is clear that income from social welfare programs accounts for only a small percentage of consumption expenditures overall. The low average amounts received from social welfare in table 6.1 could reflect either low coverage or low monetary amounts among those covered. Table 6.2 provides information on percentages of the population in each subgroup whose household received social welfare transfers (as discussed in table 6.1). The patterns across quintiles are what one would expect from the discussion of table 6.1. Only 2.2 percent of the population (2.6 and 0.8 percent of the rural and urban populations, respectively) belongs to house- holds that received assistance under a poverty program. This rises to a maximum of 6.6 percent for the poorest rural quintile. These figures may well underestimate the coverage of poverty programs if households do not know the source of assistance. Nevertheless, the data suggest very limited coverage. Table 6.3 further shows the urban bias of spending on these pro- grams. Although only 22 percent of the population and 6 percent of the poor lived in urban areas in 1998, 46 percent of total spending goes to urban areas. One important initiative under the education-related national programs has been targeted exemptions from paying school fees and other contribu- tions. Such exemptions appear to be received by children at all levels of ed- ucation but most commonly for primary, followed by secondary, schooling. Because primary school fees were abolished in 1993 (Behrman and Knowles 1999), the exemptions picked up by the 1998 VLSS and received by primary school children must cover other school expenditures. Table 6.4 presents percentages of the population living in households with at least one child benefiting from exemptions across quintiles, as well as the reasons given for exemptions. Unfortunately, the data do not allow a calculation of the pecu- niary benefit of the fee discharges. Exemptions can be partial or total. In the 1998 VLSS sample, there were only 862 households that had at least one recipient child, though many had more than one. One thousand children benefited from partial exemptions, and 571 had total exemptions. In both urban and rural areas, more partial than total exemptions were bestowed--3.7 percent versus 2.1 percent of the rural population and 1.8 percent versus 0.7 percent of the urban. There are clear indications that total exemptions are better targeted than partial ones. This can also be seen in the reasons given for receiving the exemption. Of the reasons listed in the questionnaire, unspecified "other" is the most common for partial exemptions in both urban and rural areas (see endnote 11 for an explanation of "other"). This is followed by living in a remote or mountain- ous region and having a parent who is a disabled soldier or cadre in rural 202 Economic Growth, Poverty, and Household Welfare in Vietnam Table 6.3. Total Spending on Social Welfare in 1998 as Reported in the Vietnam Living Standards Survey Program Rural Urban Total Social insurance Total amount 1,458,655.0 1,443,274.0 2,901,929.0 Percent of total 50.3 49.7 100 Social subsidies Total amount 404,762.6 117,436.0 522,198.6 Percent of total 77.5 22.5 100 Education scholarships Total amount 40,680.61 46,779.39 87,460.0 Percent of total 46.5 53.5 100 Poverty alleviation funds Total amount 26,137.04 9,613.08 35,750.12 Percent of total 73.1 26.9 100 NGO funds Total amount 11,431.44 3,044.012 14,475.452 Percent of total 79.0 21.0 100 Total social income Total amount 1,941,667.0 1,620,147.0 3,561,814.0 Percent of total 54.5 45.5 100 Percent of poor 94 6 100 Percent of population 78 22 100 Sample observations 4,381 1,618 5,999 Note: Amounts are in thousands of 1998 dong and equal the weighted sums of monetary amounts received by households as reported in the 1998 VLSS. Source: 1998 VLSS. areas, and the latter reason and being poor in urban areas. In contrast, living in a remote or mountainous region is the most commonly given reason for receiving the total exemption in rural areas, followed by being from an eth- nic minority group and poor. In urban areas, poverty is given as the main reason, and it is given as a reason across all quintiles. For example, 35 per- cent of exemptions received in the fourth quintile give poverty as the reason. Targeting exemptions to the children of disabled soldiers or cadres primar- ily benefits the most well-off groups in both sectors. However, 33 percent of all reasons in rural areas and 43 percent in urban areas were given as "other" (which is omitted from the table).11 Table 6.4 shows the incidence of school fee exemptions to be mildly pro- poor. Similar conclusions are reached when the incidence is instead ex- pressed across the percentage of children ages 6 to 14 across consumption quintiles (Government of Vietnam­Donor Working Group 2000). However, as noted by Behrman and Knowles (1999), school fees account for only a is e otalT 3.6 4.0 12.6 6.0 15.8 6.3 3.4 4.3 14.9 7.2 0 5.7 page) soldier ent cadr Par or 8.8 1.91 19.5 28.2 26.4 18.1 9.1 12.6 20.5 28.6 19.1 17.1 following disabled Partial on or otalT 31.3 36.1 27.2 37.7 15.1 31.3 31.9 38.8 32.2 45.0 36.4 35.2 continues ea ar (table Remote 6.0 6.7 cent) mountainous 24.8 22.5 24.3 16.2 20.0 25.6 23.7 28.1 19.2 22.8 Partial (per otalT 17.4 27.9 34.6 1.21 10.7 21.8 15.9 22.5 27.7 6.6 16.3 18.6 exemption Poverty fee 12.7 1.01 12.6 6.0 6.9 10.3 1.21 10.1 9.8 3.4 4.8 8.6 Partial for Reason otalT 37.9 15.9 14.8 26.8 0 24.7 38.7 17.1 15.0 28.5 0 27.1 Ethnic minority 8.6 10.7 5.2 4.4 2.3 6.6 8.9 1.31 6.1 4.0 2.9 7.3 Partial or otalT 1.6 2.3 0 6.4 3.9 2.1 1.6 2.4 0 3.6 0 1.7 orphan Disabled 0.6 2.1 0.4 0.9 4.4 1.5 0 2.2 0.4 1.1 0.5 0.9 Partial Exemptions 7.0 5.9 5.7 4.2 3.0 5.1 7.1 6.1 5.8 4.7 4.4 5.8 Fee fee Either of with School 3.8 2.2 1.4 1.1 0.6 1.8 3.8 2.3 1.4 1.2 0.6 2.1 of centage otalT exemption Per population 3.2 3.7 4.3 3.1 2.4 3.3 3.2 3.8 4.4 3.5 3.8 3.7 Partial Incidence 6.4. quintile ableT otalT quintile otalT National population Net 1 2 3 4 5 Rural Net 1 2 3 4 5 203 is e otalT 10.7 0 0 0 26.9 1.21 capita ethnic soldier shown) ent cadr per (not Par or 0 0 orphan; 13.6 26.0 38.9 24.2 disabled Partial and household emainderr on or otalT 0 0 0 0 0 0 the disabled e; ea based n: ar adrc Remote 0 0 0 0 4.9 1.9 cent) mountainous Partial quintiles exemptio (per fee a government otalT 89.3 100 72.5 34.9 6.7 47.1 population eceivingr disabled exemption Poverty for or, fee 52.4 27.2 30.0 20.2 10.6 21.4 national Partial for into given soldier Reason otalT 0 0 13.3 18.3 0 5.8 ranked easonsr e wounded ar oss Ethnic minority acr 0 0 0 6.5 1.3 1.8 Partial made seriously, Individuals been soldier or otalT 0 0 0 20.9 6.7 5.7 has definition. orphan deceased Disabled 17.5 0 0 0 1.01 5.3 egation is VLSS Partial aggr ent 1993 par the Some ea; 4.2 2.6 4.9 2.4 2.1 2.6 ar fee Either of ." follows with eceipts.r minority 2.2 1.0 1.1 0.6 0.6 0.7 in "other ) centage otalT exemption eakdown transfer and Per br of student population 2.1 1.6 3.8 1.9 1.5 1.8 half Partial of ding farmer" continued( is ral-urbanur net VLSS. boar ent es The and 6.4. "par 1998 ce: quintile ableT otalT Note: Sour National population Urban Net 1 2 3 4 5 expenditur minority includes 204 The Static and Dynamic Incidence of Vietnam's Public Safety Net 205 small share of total school-related expenditures and have a negligible impact on poverty outcomes. Households in Vietnam are expected to make cash or in-kind contribu- tions to a myriad of funds, associations, and national causes. Table 6.5 pro- vides some information about average household per capita annual contri- butions to their commune's labor and local security and police funds, as well as to mass associations. These are the funds for which the household survey collected information, but they represent just a few among the many payments made by households. Such funds collect fees that are earmarked for particular services. For example, contributions to the labor fund can be made in labor time, cash, or in-kind contributions and are intended to fi- nance road maintenance and small construction works in the commune. With the exception of the labor fund in rural areas, absolute amounts gener- ally rise with standards of living for all categories. As a share of household expenditures, they are still moderately regressive for the rural population, but they are income neutral for the urban population at a consistent 0.4 per- cent of expenditures across quintiles. Strikingly, more is paid per capita by all but the top quintile in rural areas. This is driven by much higher contri- butions to the labor fund by the rural population. Amuch larger percentage of the population contributes to one of the three funds (for which there is self-reported information) than benefits from social welfare income. In rural areas, this varies from 70 percent of the population to 54 and 49 percent for the labor fund, security fund, and associations, re- spectively.A compulsory contribution of 10 labor days a year for able-bodied adults within a certain age range is an established tradition in Vietnam. With the introduction of the market economy, the labor contribution has been partly or fully replaced by a cash or in-kind contribution in some regions. A national ordinance specifies the money amounts to be paid for each workday and details a number of characteristics that exempt individuals either tem- porarily or permanently. The 1998 VLSS asked the household about both the amount of time given in labor and the cash and in-kind payments made by family members during the past year. The data, as well as other sources, sug- gest that there is liberal interpretation of the national ordinance at the local level. For example, a study of six communes in three provinces found the time obligation to vary between 10 and 15 days and the cash alternative to be between D 3,400 and D 10,000 per day (Government of Vietnam­Donor Working Group 2000). The evidence thus suggests that the cash amounts paid in lieu of labor are considerably lower on average than daily wage rates. Imputing a cash value for labor time by using mean, commune-specific, level agricultural and nonagricultural unskilled wages will tend to overestimate the labor contributions.12 Short of going to every commune, it is impossible to know how the policy is enforced for each household. The discussion below uses what appears to be reasonable, if upper-bound, estimates of D 10,000 and D 15,000 per day for rural and urban areas, respectively. Imputed labor time is added to the cash and in-kind contributions to give the total payments to the labor fund presented in table 6.5. Participation in es centage 1.3 1.1 0.8 0.6 0.4 0.9 1.3 1.1 0.9 0.7 0.4 1.0 per household As of expenditur payments otalT per 11 capita 15,005 18,496 18,835 19,263 22,773 18,875 15,396 19,312 19,991 20,903 22,066 29,1 Dong with cent) 38.9 44.3 49.3 56.9 66.5 51.2 38.8 44.9 49.3 57.0 65.9 49.0 payments (per Population Associations per 902 870 capita 1,206 2,030 2,538 7,987 2,933 1,222 1,979 2,551 5,737 2,038 Dong with fund cent) 53.6 52.8 56.7 57.6 73.1 58.8 53.4 52.2 54.5 52.5 59.0 53.8 payments (per Population security Local per 852 827 capita 1,156 1,450 1,999 5,239 2,140 1,130 1,321 1,576 2,437 1,323 Dong Contributions with cent) 72.3 70.1 65.3 61.4 49.8 63.8 74.2 72.9 69.3 65.9 61.6 69.9 fund payments (per Household Population of Labor per capita 13,251 16,134 15,355 14,726 9,546 13,803 13,699 16,960 16,691 16,776 13,892 15,750 Dong Incidence 6.5. quintile ableT otalT quintile otalT National population Net 1 2 3 4 5 Rural Net 1 2 3 4 5 206 time capita (mass fund. 0.3 0.4 0.4 0.4 0.3 0.4 per labor labor in the associations to household ous on contributions 5,294 7,648 10,357 13,842 23,277 17,981 numer of based contributions the of valuee cash quintiles any Th to or 40.3 36.9 49.7 56.5 67.0 59.7 added population population. and, government national local quintilee to espectivelyr 985 1,694 2,407 2,494 9,593 6,320 into entir eas, the ar ranked payments oss e acr urban ar and basis 59.6 61.0 73.1 74.5 83.1 77.7 household uralr capita in Individuals eported per a on worked self-r definition. e ar day essed 1,480 1,508 2,396 3,396 7,240 5,231 per VLSS expr amounts 1993 15,000 survey D the Dong thee and 24.6 33.2 35.8 46.5 41.4 40.7 follows eceipts.r befor 10,000 D of months eakdown transfer br 12 of values the 2,120 5,155 5,553 7,952 6,444 6,431 half of using during VLSS. ural-urbanr net es The imputed 1998 ce: quintile otalT been Note: ganizations) Sour Urban Net 1 2 3 4 5 expenditur or has 207 208 Economic Growth, Poverty, and Household Welfare in Vietnam the labor fund decreases with increasing living standards in rural areas. The picture is quite different in urban areas: In all quintiles, a smaller percentage contributes to the labor fund than in rural areas, and participation rises with expenditures--from 25 percent of the poorest to 42 percent of the top quin- tile. A large percentage contributes to local security (59 percent overall), and the more so the higher the quintile. Close to 60 percent of the urban popula- tion also contributed to associations over the last year. For these contributory "funds," coverage appears reasonably wide, though average amounts paid among those contributing are clearly low. As noted, however, the reviewed charges account for just part of the amounts levied from households. A re- cent study suggests that in aggregate they can be quite burdensome as a share of household expenditures. Conversely, they clearly play a crucial role in commune-level budgets (Government of Vietnam­Donor Working Group 2000). Tables 6.6 and 6.7 combine data from the household and commune surveys to present percentages of the rural and small-town populations classified into poor and nonpoor groups, by whether (a) they live in com- munes where any of seven public programs are currently active (poverty Table 6.6. Rural Population Living in a Commune with Poverty Programs and Other Programs (percent) Program Total Poor Nonpoor Poverty alleviation 79.1 83.6 76.2 Employment generation 21.1 19.1 22.5 Environmental or clean water 15.3 13.7 16.4 Public health 25.0 28.7 22.5 Infrastructure development 49.5 52.7 47.3 Education and culture 18.9 18.7 19.1 Other project 7.6 7.8 7.5 Disaster relief 66.1 71.5 62.5 Recent infrastructure investments 92.4 93.4 91.9 Roads 50.5 45.8 53.6 Electricity 28.1 26.9 28.8 Irrigation 36.7 40.6 34.1 Schools 58.9 52.8 63.0 Health center 36.2 33.3 38.1 Water sources 18.1 18.9 17.5 Other 0.9 0.5 1.1 Observations 4,269 1,439 2,830 Note: The table combines information from the household and commune datasets. "Rural" is defined according to the 1998 VLSS definition. The questionnaire asked for the first, second, and third kinds of government or other projects or programs currently existing in the commune. The table reports the percentage of population living in communes where a kind of project was listed either first, second, or third. Source: 1998 VLSS. The Static and Dynamic Incidence of Vietnam's Public Safety Net 209 Table 6.7. Small-Town Population Living in a Commune with Poverty Programs and Other Programs (percent) Program Total Poor Nonpoor Poverty alleviation 83.1 86.2 82.7 Employment generation 38.4 45.0 37.4 Environmental and clean water 20.5 23.3 20.1 Public health 6.6 14.1 5.5 Infrastructure development 22.5 24.2 22.3 Education and culture 26.5 20.0 27.5 Disaster relief 50.7 61.4 49.1 Recent infrastructure investments 78.0 83.3 77.2 Roads 67.6 78.5 65.9 Electricity 21.3 17.2 21.9 Irrigation 12.7 18.2 11.8 Schools 57.7 62.8 57.0 Health center 23.3 20.6 23.7 Water sources 27.2 26.9 27.2 Observations 581 59 522 Note: The table combines information from the household and commune datasets. The ques- tionnaire asked for the first, second, and third kinds of government or other projects or programs currently existing in the commune. The table reports the percentage of population living in communes where a kind of project was listed either first, second, or third. Source: 1998 VLSS. alleviation, employment generation, environmental or clean water, public health, infrastructure development, education or culture, or other); (b) whether the commune received disaster relief in the last year; and (c) whether any physical infrastructure was built or improved during the past three years, and if so what type it was.13 Poverty programs are the most common of all among these public programs. These were active at the time of the 1998 survey in communes covering 80 percent of the rural population and 84 percent of the rural poor. However, they were slightly more common in small towns, where 83 per- cent of the entire population and 86 percent of the poor were covered. Em- ployment generation, sanitation and clean water, and education and culture projects also reached a larger proportion of small-town residents than rural residents. By contrast, public health and infrastructure development pro- grams covered more of the rural population. Disaster relief was also re- ceived in communes covering 65 percent of the nonurban population (rural and small-town). Finally, infrastructure investments are extremely wide- spread, covering communes containing 92 percent of the rural and 78 per- cent of the small-town populations. In both sectors, roads and schools are the most common investments. In rural communes, both roads and schools tend to benefit larger percentages of the better-off than of the poor. 210 Economic Growth, Poverty, and Household Welfare in Vietnam In the programs reviewed in tables 6.6 and 6.7, there is some evidence of targeting of the poorer population groups. Disaster relief, for example, is re- ceived by the communes with a greater percentage of poor than nonpoor households. However, based on these data, it is not possible to judge whether, relative to needs, disaster relief would still appear well targeted. Many of the other programs are thought to be geographically targeted to government- identified "poor and remote" communes. Yet on the whole, the impression is one of programs being spread widely across expenditure groups and the rural population generally. This may reflect problems in identifying the poor through the current "poor and remote" commune classification, corroborat- ing the results of Minot and Baulch (2004). It could also indicate that com- munes are heterogeneous in terms of standards of living, and geographic targeting may be an inefficient way to help the poor. Of course, these tabula- tions tell nothing about the magnitude or impact of the programs. Careful evaluation of Vietnam's various poverty program disbursements must be made to better understand what does and does not work. However, the data reviewed at both household and commune levels suggest a gov- ernment preference for programs that are community based rather than tar- geted to households. Transfers to households are negligible, and coverage is weak. By contrast, the data indicate substantial community-based programs and investments. Again, how much is being spent is unclear, as is the impact of the latter programs. However, as assessed by incidence across per capita expenditure quintiles, such interventions appear to be only weakly targeted to Vietnam's poor. The data suggest that transfers are redistributive, but not particularly well targeted, in that the poor receive less in absolute amounts than the nonpoor, in general. Protection versus Promotion As can be seen in table 6.8, there was a clear expansion in the total outlays going to social welfare programs between 1993 and 1998.14 As reported in the survey, mean overall real per capita amounts rose from D 51,443 to D 116,641 in 1998 prices, a 127 percent proportionate increase. Was this expansion pro-poor? A comparison of panel households over time can help answer this and other pertinent questions concerning the per- formance of the safety net. An important role for the public sector in a poor rural economy such as Vietnam is to provide protection for those who are vul- nerable to poverty as a result of uninsured shocks. The preceding incidence picture is uninformative about whether transfers perform such a safety net function. The static incidence may not seem particularly well targeted, but it may be deceptive about the degree to which outlays, coverage, and changes over time were perhaps correlated to poverty-related shocks and changes in exogenousvariables.Theconsiderablevariabilityinpaymentamountsacross recipients has already been seen. There is also much instability over time in whogetstransfers.Forexample,outofatotalof744and769panelhouseholds who respectively received social insurance or social subsidy outlays in one of cent ease social 122.0 131.0 134.2 139.9 15.01 126.7 129.6 139.4 133.5 137.6 16.21 126.7 page) Per incr in transfers following on cent) (775) (829) (850) (891) (958) (4,303) (740) (809) (872) (924) (958) (4,303) Population (per 16.3 17.0 21.2 21.6 23.2 19.8 16.5 17.9 22.2 20.5 22.0 19.8 continues (table es transfers cent) 5.8 5.0 5.5 5.4 0.6 4.5 7.1 5.1 6.0 5.5 1.5 4.5 Social Household (per expenditur 1998 per capita 76,197 90,452 Dong 101,858 130,822 184,128 16,6411 80,468 78,878 17,4421 139,395 166,996 16,6411 meiT cent) (775) (830) (850) (895) (958) (4,305) (740) (809) (872) (924) (960) (4,308) over Population (per 22.1 19.7 21.7 23.4 24.2 22.2 24.2 19.4 21.3 23.8 22.5 22.2 es ransfersT transfers cent) 4.8 3.4 2.9 2.8 2.5 3.3 4.6 2.8 3.6 3.0 2.5 3.3 Social Social Household (per expenditur of 1993 per Incidence capita 34,330 39,166 43,492 54,532 85,654 51,443 35,041 32,952 50,290 58,657 77,257 51,443 in Dong Changes quintile quintile 6.8. Net net ableT otalT otalT National population 1993 1 2 3 4 5 Mean 1 2 3 4 5 211 cent ease social 136.8 154.9 19.91 132.0 122.7 Per 126.7 incr in transfers households sample of cent) (735) (797) (879) (929) (963) (4,303) Population (per 17.6 18.1 22.3 19.3 21.8 19.8 umbern The es transfers transfers. cent) 3.2 5.8 5.6 4.3 3.4 4.5 Social Household (per social expenditur of populations. 1998 half of quintile per net the es capita 91,545 89,965 14,2181 16,3251 171,121 16,6411 oss Dong acr basis expenditur capita capita cent) (735) (797) (879) (929) (965) (4,305) per per a Population (per 23.0 21.8 22.7 21.0 22.6 22.2 on on based essed es expr ucted e transfers ar VLSSs. cent) 4.1 3.1 3.5 3.0 2.9 3.3 constr 1998 Social Household (per expenditur amounts and 1993 quintiles Dong 1993 the per ) capita 38,652 35,299 51,934 50,131 76,857 51,443 population entheses. using Dong par in (2003), national e ar given alle continued( W is de quintile van 6.8. Quintiles quintile Net ce: ableT otalT Note: each Sour National population 1998 1 2 3 4 5 in 212 The Static and Dynamic Incidence of Vietnam's Public Safety Net 213 the two years, only 402 and 111 received them in both years. Does this reflect a response to changing household circumstances on the part of the system? This section examines social welfare incomes from this perspective. When using the panel to study the incidence of the changes in social in- come, there is a question of how households should be ranked in deciding who is "poor." Table 6.8 ranks households by three different definitions of welfare, which can be loosely referred to as denoting the initial, new, and long-term poor--namely, per capita expenditures (net of half of transfers) in the initial period, the same in the later period, and by the mean over both years--and presents a comparison of mean per capita social income receipts in both survey years. The proportional gains from expansion were fairly uniform across groups. However, among the "poor" in each of the three senses, the "initial poor" clearly had the lowest gains, with a 122 percent proportionate increase in benefits for the bottom quintile and a 131 percent increase for the second lowest. The "new poor" had the highest proportion- ate gains (137 percent and 155 percent increase, respectively), and the "long- term poor" fell somewhere in between (130 percent and 139 percent). Per capita amounts increased for all groups, but the share of the population re- ceiving transfers declined slightly overall (22 to 20 percent), as did the pro- portion of the poor receiving them by all three definitions. The evidence does not suggest that the poor were targeted by the program expansion. Were changes in transfers responsive to poverty-related shocks? Ta- ble 6.9 presents information on mean changes in transfers received by panel Table 6.9. Incidence of Changes in Transfers by Initial Consumption and Changes in Consumption over Time Fall in Consumption Large rise in Indicator consumption stayed the same consumption Low initial 34% 27% 27% consumption 111,901 246,476 241,658 80 506 848 Middle initial 32% 30% 30% consumption 408,469 251,619 296,513 240 422 772 High initial 33% 36% 32% consumption 481,618 343,329 367,991 496 221 720 Note: The population is ranked into three equal groups based on 1993 per capita expendi- tures net of half of transfers and cross-tabbed against the level of their change in consumption over time net of half the change in transfers. The first number gives the percentage of house- holds in the cell that received transfers in 1998. The second number gives the per capita amount (in dong) of the change in transfers received by those with positive receipts only. The final num- ber gives the number of households in the cell. Changes in transfers refer to changes in amounts received from social insurance, social subsidies, and school scholarships. Source: van de Walle (2003), using the 1993 and 1998 VLSSs. 214 Economic Growth, Poverty, and Household Welfare in Vietnam households classified into a three-by-three matrix. Households ranked into terciles of their initial 1993 level of per capita consumption (low, middle, or high) are cross-tabbed against the change in their consumption between the two dates categorized into whether it underwent a fall, stayed more or less the same, or rose significantly.15 So, for example, 34 percent of those who were in the bottom third of the distribution in 1993 and experienced a fall in consumption over time received transfers equal to about D 111,901 per per- son in recipient households. There is little sign that the system responded to consumption shocks. In- deed, the percentage of households who benefited from social incomes is relatively uniform across cells. Neither starting out poor, nor experiencing negative consumption shocks, appears to have elicited a response from social welfare programs. Thirty-two percent of those who enjoyed the highest initial consumption and the highest gains to consumption were ben- eficiaries, compared with 34 percent of the worst-off in both respects. Fur- thermore, if anything, the per capita transfers to participants increase with initial and rising welfare. The smallest amount went to the neediest. These specific programs appear unresponsive to shocks.16 As discussed in the beginning of this chapter, and to be further discussed in the section on geographic targeting, location may be an important factor in the determination of program participation. Possibly the absence of a pattern in table 6.9 arises from variation across geographic areas that is obscuring patterns within them. To test this, a dummy variable indicating whether transfers were received in 1998 was regressed against initial (1993) per capita consumption and the change in per capita consumption (1993 to 1998). A linear probability model was used, and it was run with and without commune effects. With commune effects, there is no sign of transfers re- sponding to either initial consumption or changes in consumption. Without commune effects, the results suggest that transfers respond perversely to initial consumption ( = 1.12e­8, t = 2.52) and not to shocks (similar to table 6.9). This suggests that it is households in better-off communes that primarily benefit from these transfers. It is of further interest to examine what role transfers played in the im- pressive reduction in poverty that occurred over this period. The panel structure is now exploited to evaluate how well the safety net performed dynamically, including how well it protected against poverty distinguished from how well it promoted out of poverty, following the approach proposed in Ravallion, van de Walle, and Gautam (1995). In comparing joint distribu- tions of consumption expenditures, such as with and without policy changes, the approach tests a policy's ability to protect the poor (PROT) and its ability to promote the poor (PROM).17 It indicates which distribution offered more protection and which offered more promotion and allows a calculation of the statistical significance of the difference. Table 6.10 presents the baseline joint distribution of consumption in the two survey years. Households are classified into four groups according to whether they were poor or nonpoor in both years, as well as whether they escaped or fell into poverty over the five-year period between the surveys. The Static and Dynamic Incidence of Vietnam's Public Safety Net 215 There is evidence of a large fall in poverty: 27 percent of the population es- caped poverty, 5 percent fell into poverty, 34 percent were persistently poor, and 35 percent were never poor. There is considerable persistent poverty. What is the effect of transfers on poverty? To answer this question, it is necessary to simulate the counterfactual joint distribution without transfers. As in the static incidence calculations, this is done by subtracting half the transfers received in each respective year from consumption in that year. The simulated joint distribution is given in table 6.11. Transfers are found to Table 6.10. Baseline Discrete Joint Distribution 1998 1993 Poor Nonpoor Total Poor 33.54% 26.58% 60.12 (55.78) (44.22) 100 Nonpoor 4.84% 35.04% 39.88 (12.14) (87.86) 100 Total 38.38 61.62 100 Note: The population is ranked into poor and nonpoor groups based on actual per capita expenditures at each date and cross-tabbed. The first number in each cell gives the percentage of total population that was in that row's poverty group in 1993 and that column's group in 1998. The numbers in parentheses give the proportion of each row's population that is in each column's group in 1998 or the transition probability. Source: van de Walle (2003), using the 1993 and 1998 VLSSs. Table 6.11. Joint Distribution without Transfers 1998 1993 Poor Nonpoor Total Poor 35.21% 25.88% 61.09 (57.63) (42.37) 100 Nonpoor 5.15% 33.76% 38.91 (13.24) (86.76) 100 Total 40.36 59.64 100 [PROT = 0.31(0.66); PROM = 0.70(0.74)] Note: PROT = Test of policy's protection of the poor. PROM = Test of policy's promotion of the poor. The population is ranked into poor and nonpoor groups based on their simulated "without transfer" per capita expenditures (net of half of the transfers) at each date and cross- tabbed. The first number in each cell gives the percentage of total population that was in that row's poverty group in 1993 and that column's group in 1998. The numbers in parentheses give the proportion of each row's population that was in each column's group in 1998. z scores for the PROM and PROT tests are given below the table. Critical values: 1.96 (2.58) at the 5 percent (1 percent) level. Source: van de Walle (2003), using the 1993 and 1998 VLSSs. 216 Economic Growth, Poverty, and Household Welfare in Vietnam Table 6.12. No Change in Transfers between 1993 and 1998 1998 1993 Poor Nonpoor Total Poor 34.23% 25.89% 60.12 (56.94) (43.06) 100 Nonpoor 5.19% 34.69% 39.88 (13.02) (86.98) 100 Total 39.43 60.57 100 [PROT = 0.36(0.76); PROM = 0.69(0.73)] Note: PROT = Test of policy's protection of the poor. PROM = Test of policy's promotion of the poor. The population is ranked into poor and nonpoor groups based on actual per capita ex- penditures for 1993 and the simulated 1998 distribution had there been no change in transfers (per capita expenditures in 1998 net of half of the change in transfers) and cross-tabbed. The first number in each cell gives the percentage of total population that was in that row's poverty group in 1993 and that column's group in 1998. The numbers in parentheses give the proportion of each row's population that was in each column's group in 1998. z scores for the PROM and PROT tests are given below the table. Critical values: 1.96 (2.58) at the 5 percent (1 percent) level. Source: van de Walle (2003), using the 1993 and 1998 VLSSs. have a negligible impact on poverty. Without them, 1 and 2 additional per- cent of the population would have been poor in 1993 and 1998, respectively. The measures of promotion and protection are not statistically significantly different from zero. Table 6.12 simulates the joint distribution had there been no changes in transfers between the two years. The change in the proportion who fell into poverty identifies the degree of protection offered, and the change in the proportion who escaped poverty indicates promotion. Changes enabled just over 1 percent of the population to escape poverty, and they protected about 1 percent from falling into poverty. Again, these are not statistically different from zero effect. Low spending, low coverage, and poor targeting together explain the negligible impact of transfers and changes in transfers on poverty. How much could better targeting improve impacts on poverty inci- dence? Table 6.13 compares the current distribution relative to a simulated uniform allocation of actual 1998 social income across the entire population. This would have a small, but statistically significant, further impact on poverty: An additional 3 percent of the population (7 percent of the poor) under the actual allocation would escape poverty (standard error of 0.4 per- cent). Just over 2 percent of the nonpoor would have fallen into poverty (standard error of 0.3 percent). What if 1998 transfers were instead targeted based on an equal allocation to those below the poverty line only? The re- sults in table 6.14 show that outlays would be sufficient to bring 17 percent of the poor (7 percent of the population, with a standard error of estimate of 0.4 percent) out of poverty. Only 3 percent of the nonpoor would have fallen into poverty (2 percent of the population, standard error of 0.2 percent). The Static and Dynamic Incidence of Vietnam's Public Safety Net 217 Table 6.13. Actual 1998 Distribution versus Uniform Allocation of 1998 Transfers 1998 Simulated 1998 Actual Poor Nonpoor Total actual Poor 35.54% 2.83% 38.38 (92.61) (7.39) 100 Nonpoor 1.54% 60.09% 61.62 (2.49) (97.51) 100 Total simulated 37.08 62.92 100 Note: The population is ranked into poor and nonpoor groups based on actual per capita expenditures for 1998 and the simulated 1998 distribution, if the five transfers identifiable in 1998 had been distributed uniformly across individuals and cross-tabbed. The first number in each cell gives the percentage of total population that was in that row's poverty group in 1993 and that column's group in 1998. The numbers in parentheses give the proportion of each row's population that was in each column's group in 1998. Source: van de Walle (2003), using the 1998 VLSS. Table 6.14. Actual 1998 Distribution versus 1998 Transfers Targeted on Equal Per Capita Basis to the Poor 1998 Simulated 1998 Actual Poor Nonpoor Total actual Poor 31.72% 6.66% 38.38 (82.66) (17.34) 100 Nonpoor 1.98% 59.64% 61.62 (3.21) (96.79) 100 Total simulated 33.70 66.30 100 Note: The population is ranked into poor and nonpoor groups based on actual per capita ex- penditures for 1998 and the simulated 1998 distribution, if the five transfers identifiable in 1998 had been distributed on a per capita basis only to the poor and cross-tabbed. The first number in each cell gives the percentage of total population that was in that row's poverty group in 1993 and that column's group in 1998. The numbers in parentheses give the proportion of each row's population that was in each column's group in 1998. Source: van de Walle (2003), using the 1998 VLSS. Finally, going back to the concerns of table 6.8, table 6.15 presents the joint distribution of the incidence of proportionate gains in social incomes. When ranked by their 1998 welfare, large gains are again apparent for the nonpoor. The new information here is that within the nonpoor population, the largest gains went to the "initial poor." Once again, the evidence sug- gests very poor performance on protection. Poverty fell quite dramatically in Vietnam between 1993 and 1998, but social insurance, social subsidy, and scholarship income transfers appear to 218 Economic Growth, Poverty, and Household Welfare in Vietnam Table 6.15. Incidence of Proportionate Changes in Social Incomes (percent) 1998 1993 Poor Nonpoor Poor 102 189 Nonpoor 54 125 Note: The population is ranked into poor and nonpoor groups based on their actual per capita expenditures at each date and cross-tabbed. The numbers give the percentage change in the three transfers between the dates. Source: van de Walle (2003), using the 1993 and 1998 VLSSs. have had negligible bearing on that outcome. They did not fulfill a safety net role, either, in protecting those who faced falling living standards dur- ing this period. Part of the reason for this failure of promotion and protec- tion is low overall spending on these programs. However, the simulations above suggest that poor targeting is a fundamental problem, as are low total outlays. Geographic Targeting One possible explanation for the picture that has emerged so far may be the narrowness of the welfare indicator that has been used. Consumption ex- penditures per capita may simply be too narrow a welfare metric to reveal the underlying pro-poor targeting. Programs may well respond to in-the- field definitions of welfare that are considerably more complex than per capita consumption. Another possible explanation is that, given Vietnam's institutional arrangements for delivering social welfare programs, nonwelfare dimen- sions, such as politico-geographic dimensions, may largely determine whether transfers reach the disadvantaged. This section explores these pos- sibilities. Poor communes have greater needs, but better-off communes can better afford poverty-related programs. Better-off communes may also be better at implementing programs and reaching their poor residents. One means of equalizing resources is through the central government's national pro- grams. Two obvious questions to ask at this point are: To what degree does redistribution occur through these programs? Are the limited resources that are transferred from the national programs to the local level targeted to poorer communes? It is not possible to answer these directly, because there is no way to use the VLSSs to determine whether a sampled household ben- efits from a national program (with the exception of school fee exemptions, for which a benefit amount is not identifiable). However, incidence at the The Static and Dynamic Incidence of Vietnam's Public Safety Net 219 commune level is observed in the commune-level data for employment gen- eration, poverty alleviation, education and culture, infrastructure develop- ment, public health, environment, and other programs. Similarly, household participation at the commune level is observed for microcredit, school and health fee exemptions, tax exemptions, and training and disaster relief pro- grams. Most of these programs are probably centrally mandated "national programs," although they cannot be identified specifically. Table 6.16 links the household- and commune-level data to show the incidence of programs and of beneficiary households across communes classified into three equal groups--poor, middle, and rich--by the mean per capita consumption ex- penditures of their population as sampled in the household survey. A com- parison of commune mean per capita expenditures gives an indication of the income disparities across communes. The mean for the poorest 10 per- cent of communes is 70.3 percent lower than that for the 10 percent richest communes. Are poor communes more likely to have poverty programs? Table 6.16 suggests that the answer is yes. In general, poorer communes appear to have both more poverty-related programs and a greater share of their popula- tions participating--but the exceptions to this generality are interesting. The percentage of households benefiting from occupational training is highest in better-off communes. Education, employment generation, and environmen- tal programs are all most common in the most well-off communes. There ap- pears to be capture of skills and employment-related schemes in better-off communes, perhaps because they are already well endowed with the bene- fits offered by the other programs. Overall, the incidence across communes is redistributive in that there is a greater concentration of programs in the poorest communes. However, it is also true that programs are geographi- cally spread around quite widely.18 The above results tell nothing about the benefits to households from liv- ing in a commune with a program or from being among the beneficiaries of a program. To obtain this information, it is necessary to turn to household- level data. Linking up household and commune information further allows an exploration of the importance of location to participation in programs. For example, how do the poor in poorer communes fare compared with the poor in better-off communes? To what degree does a poor household's loca- tion determine whether and how much it benefits from assistance pro- grams? Do poor households in the most well-off communes do better than those in poor communes? Are there signs of better targeting when more is spent overall in a location? (See Ravallion [1999].) Tables 6.17 and 6.18 ex- amine these issues by looking at the distribution of beneficiaries and of social income payments (as reported at the household level) across the populations of poor, middle, and rich communes ranked into national terciles of per capita consumption net of transfers. Table 6.18 clearly shows that, not only is more being spent per capita overall in better-off communes, but much more is also going to the poor. Total mean per capita payments in the richest communes are more than the 7.7 3.9 7.7 6.4 eliefr 8.2 4.0 1.8 4.7 Other All Disaster pulation. b water po Communes onment 9.6 19.2 21.2 16.7 their 1.1 0.9 4.0 1.9 clean of Rich rainingT Envir and and incomes a social of Middle,, eceivedr xaT 19.2 19.2 32.7 23.7 13.4 5.4 7.0 8.8 half Poor who Employment generation of exemptions ograms net pr es Rural by with households and health fee 30.8 23.1 9.6 21.2 expenditur Health public commune 13.6 2.1 4.0 6.7 communes capita of Hospital exemptions of per Beneficiaries e mean and centage centage cultur 25.0 15.4 26.9 22.4 the Per Per Education and on fee based Programs 13.1 6.8 4.8 8.3 taxes. c School exemptions oups data. gr business training. investments. 53.9 42.3 36.5 44.2 or equal Development investments ee thr technology Poverty-Related commune-level oduction development into pr e of edit cr 19.4 12.6 1.41 the 14.6 of on uctur Subsidized ranked 88.5 76.9 71.2 78.8 e agricultural Poverty ar based alleviation is eductionr and infrastr Incidence or and VLSS. 1998 6.16. Communes information ce: bleaT otalT otalT Exemption Occupational Economic Note: a. b. c. Sour Communes Poor Middle Rich Communes Poor Middle Rich other 220 The Static and Dynamic Incidence of Vietnam's Public Safety Net 221 Table 6.17. Incidence of Social Transfers across the Rural Population by Terciles and Poor, Middle, or Rich Communes Percentage of population benefiting from the following household level funds Population Social Social Poverty Education tercile insurance subsidies alleviation NGO scholarships Total Poorest communes 1 (968) 7.6 10.2 6.7 0.7 0.9 23.4 2 (542) 10.7 8.5 2.4 0.7 2.3 21.6 3 (150) 12.2 4.3 3.0 0 3.4 22.3 Total (1,660) 8.8 9.5 5.1 0.7 1.5 22.8 Middle communes 1 (405) 11.9 12.3 1.3 1.5 0.7 25.3 2 (741) 9.0 12.3 0.5 0.5 1.5 21.8 3 (489) 7.8 13.4 0 0 2.6 21.6 Total (1,635) 9.5 12.6 0.6 0.7 1.5 22.8 Richest communes 1 (149) 8.7 6.9 3.1 0 2.7 18.4 2 (479) 14.9 7.4 0.7 0.2 1.7 21.2 3 (927) 12.3 6.5 0.2 0 4.2 19.2 Total (1,555) 12.8 6.8 0.7 0.1 3.2 19.8 Note: Communes are ranked into three equal groups based on the mean per capita expendi- tures net of half of social incomes of their population. The rural population is ranked into population terciles. The number of sample households in each tercile is given in parentheses. Source: 1998 VLSS. double that in the poorer communes. Mean per capita amounts going to the poor are 136 percent higher. There are signs of better targeting in better-off communes. Social insurance and social subsidies largely drive these results. Although table 6.17 indicates that more of the poor live in households that participate in programs in poor communes than in rich communes, the per capita amounts received by the poor in the latter dwarf the former. They ac- count for 7.1 percent of household expenditures compared with 4.3 percent for the bottom tercile in the poorest communes. Although small, outlays from the poverty alleviation fund tend to be concentrated in poor com- munes and on the poorest. The targeting differentials, given by the differ- ence between the mean expenditures going to the poorest 50 percent of the population to that going to the top 50 percent, are 1,202, 1,210, and 161 for poor, middle, and rich communes, respectively. Conclusions and Policy Implications This chapter's results reveal little overall sign of targeting to poor people or poorer communes in terms of their standards of living measured by of es uralr cent 4.3 3.5 2.7 3.9 6.2 3.5 2.5 4.0 7.1 5.6 3.1 4.4 The Rich Per household or expenditur pulation. po Middle,, their otalT of 68,257 89,915 107,924 77,296 15,5481 93,572 105,337 102,948 160,927 166,038 148,672 155,701 Poor incomes and funds omfr social of entheses. ercilesT 447 788 1,339 1,646 4,735 1,643 1,369 2,859 1,475 1,839 6,745 4,480 par half in by Education scholarships of net population given es is rural 0 0 0 0 cile Population by 366 874 384 447 316 634 207 ter NGO 1,204 expenditur each in Rural eceivedr capita the per dong 0 210 746 890 491 83 306 mean households 2,132 1,009 1,672 2,979 1,210 across Poverty capita alleviation the on sample Per of based Amounts oups number Social 18,461 21,537 8,802 gr 18,797 22,820 17,495 27,084 21,462 17,322 17,052 15,486 16,202 subsidies The ransferT equal ciles. ee ter thr Social into of Social 45,122 65,356 94,177 54,310 88,919 73,515 75,394 78,485 141,927 146,022 126,358 134,507 population insurance ranked e into ar Incidence ranked VLSS. is 6.18. communes (1,660) communes (1,635) communes (1,555) 1998 Communes ce: est bleaT cile (968) (542) (150) otalT (405) (741) (489) otalT (149) (479) (927) otalT Note: Sour Communes Population ter Poor 1 2 3 Middle 1 2 3 Richest 1 2 3 population 222 The Static and Dynamic Incidence of Vietnam's Public Safety Net 223 consumption. If anything, transfer receipts rise with consumption per person, though there are signs that the share of social incomes in consump- tion falls with consumption, implying that transfers reduce inequality. At the same time, the existing system is ineffective in protecting households that are vulnerable to falling living standards. Differences are apparent when specific programs are compared, though these are not marked. Social insurance payments are clearly urban biased where they are also quite pro- poor. In rural areas, they accrue primarily to better-off households. Though they benefit the urban poor less, education scholarships show a similar pat- tern. Social subsidy benefits are distributed more evenly across deciles and are larger in rural Vietnam. Finally, poverty alleviation and NGO transfers, though negligible in size, tend to be well targeted to the poor and to favor rural areas. Household payments and contributions also appear to be regressive. The current system suffers from the lack of national norms for identify- ing the poor consistently across regions; the lack of survey and other instru- ments with which to consistently measure and monitor local needs and program performance; a lack of integration and coordination among sub- programs with well-defined and universal rules for implementation at the local level; insufficient welfare-maximizing redistribution of resources across space so that everyone is treated equally regardless of where they reside; and a lack of resources and attention to helping households and com- munities deal with covariate risk. Progress in these areas could lead to sig- nificant improvement in social protection for Vietnam's poor and vulnerable households. In terms of funding and priorities, it is clear that the primary focus of HEPR continues to be microfinance and infrastructure development. The potential immediate importance of HEPR lies in the possibility of greater consistency in priorities and norms, better monitoring of outcomes, much needed integration and coordination between programs, better coverage of the poor, and redistribution toward poorer and less administratively capa- ble provinces. Here too, though, there has been little discernible progress since HEPR's inception. While the HEPR concept offers the potential for considerable improve- ments in the safety net, the government of Vietnam faces a number of difficult challenges. The very principles on which the current highly decen- tralized, community-based assistance and safety net system is built are threatened by the emerging market economy. In particular, increasing mobility--important to a well-functioning market system--dictates a thor- ough rethinking of the safety net's foundations. Household mobility renders community-level identification and targeting of the poor less effective and is likely to make the mobilization of community resources for helping the poor more difficult. The high level of decentralization inhibits the country's ability to provide adequate protection from covariate risks, which, in turn, appear to be on the rise because of environmental destruction. Adequately addressing this challenge, and the consequent widening urban-rural and 224 Economic Growth, Poverty, and Household Welfare in Vietnam regional inequalities, will require a greater level of risk pooling nationally through greater reliance on state-contingent redistribution mediated through the center. Important political hurdles can also be expected in ef- forts aimed at reallocating resources to better protect Vietnam's poor and vulnerable. Geographic targeting is a widespread practice and is generally assumed to work well when there are geographic concentrations of poverty and iden- tification of the poor is possible at a sufficiently disaggregated level. How- ever, it may well be that poorer areas are less capable of reaching their poor well or implementing poverty programs, or both, than are their better-off counterparts. This chapter finds that across Vietnam's communes, more is spent relatively and absolutely on the poor in better-off communes. This is likely to reflect the large differences in resources across regions. More re- search is needed to understand whether it also reflects a weaker capacity for reaching the poor. However, in the absence of a reform of the fiscal redis- tributive system--whereby the center's redistributive process promotes an equalization of resources all the way to the commune level--if the question is where resources will have the greatest impact, perhaps better-off com- munes rather than poorer ones should be targeted. The data do not allow identification of whether funding comes from the national or local level. Past evidence seems to indicate that existing national resources are relatively well targeted spatially at the provincial level, but the redistributive effect is mitigated by the distribution that then occurs within provinces. Although this chapter cannot answer this question, it does show that the combination of funding and implementation mechanisms results in poor areas and people getting less than better-off areas and people. This sug- gests the need for more compensatory mechanisms from the center, which could take the form of more money, better incentives for fiscal redistribution at the local level, more monitoring of central norms, or administrative con- straints on local discretion in the implementation of centrally mandated social welfare programs. Notes Special thanks go to Dorothyjean Cratty. Helpful comments were received from Paul Glewwe, Jennie Litvack, Martin Ravallion, Kalanidhi Subbarao, and participants at the May 2001 research workshop, "Economic Growth and Household Welfare," in Hanoi. The support of the World Bank's Research Committee is gratefully acknowl- edged. 1. This chapter's focus is on public transfers only. For a discussion of private in- terhousehold transfers, see Cox (2004). 2. For more details, see van de Walle (1999). 3. The 1992­93 survey spanned a full year starting in October 1992; the 1997­98 survey began in December 1997 and also lasted a year. For brevity's sake, reference is made to the surveys as the 1993 VLSS and 1998 VLSS, respectively. 4. The World Bank (1995, 2000) provides detailed information on the surveys, accessible at www.worldbank.org/lsms/. The Static and Dynamic Incidence of Vietnam's Public Safety Net 225 5. The regression controls for changes in household size and composition--in particular, the number of members in the birth to 6 and 7 to 16 age groups, the number of women and men older than 55 and 60, respectively (the formal sector legal retirement age)--and for a change in the highest grade completed by the most educated member of the household, the change in the age and gender of the household head, and finally a change in the language of interview. (Households had the option of being interviewed in a language other than the Kinh language in both survey years.) See van de Walle (2003) for full regression results and explanation. 6. Ahigh correlation is found between these variables (0.50). The key untestable exclusion restriction is that transfers in 1993 do not appear on the right-hand side of the equation [such as cov(it, Tit ) = 0]. This appears plausible but would not hold if, -1 say, the initial level of transfers helps prevent households from falling into destitu- tion or succeeds in putting them on a different growth path. There is no other obvi- ous instrument with which to do an overidentification test. 7. The lower the PCSI, the more targeted transfers appear to be to the poor. See van de Walle (2003) for a discussion. 8. Note that this means half of the total of scholarships, social insurance, and subsidy funds for 1993 and half that same total plus poverty alleviation and NGO funds for 1998. 9. The official January 1998 exchange rate was about D 12,290 to the U.S. dollar. 10. Note that throughout this chapter, "progressive" is defined as meaning that, as a proportion of expenditures, transfers decline as expenditures increase. 11. Other (not individually recorded) reasons for receiving exemptions in- cluded being a student at a pedagogic college; being an excellent student, a class monitor, the children of teachers, and/or the children of officers and workers for whom tuition is paid by the parent's work; and households with two or more chil- dren attending school (General Statistical Office communication of May 2001). 12. For example, commune mean daily unskilled agricultural wages in real 1998 prices are D 19,421 and D 16,609 for men and women, respectively. 13. Here and elsewhere, this chapter uses the national poverty lines described in Glewwe, Gragnolati, and Zaman (2002). 14. Note that this refers only to programs--scholarships, social insurance, and social subsidies--covered in both VLSSs. Although these do not account for all pro- grams, they cover the bulk of social income receipts. 15. Consumption in 1993 is net of half of transfers, and changes in consumption are net of half the change in transfers. 16. One may not expect a pension scheme to respond to changing household circumstances. However, as noted, there is huge instability in who gets social insur- ance transfers over time. These consist of disability, maternity, and death as well as pension payments and could thus respond to consumption shocks. 17. Details on the tests are given in Ravallion, van de Walle, and Gautam (1995) and van de Walle (2003). 18. Again, the empirical evidence does not support the claim that truly poor communes are being targeted much more than others. This could reflect deficiencies in the government's identification of poor communes (see Minot and Baulch [2004]) or point to the inefficiency of geographic targeting due either to fundamental heterogeneity among communes or, alternatively, to targeting not actually being implemented. 226 Economic Growth, Poverty, and Household Welfare in Vietnam Bibliography The word processed describes informally reproduced works that may not be commonly available through libraries. Bardhan, Pranab, and Dilip Mookherjee. 2000. "Capture and Governance at Local and National Levels." American Economic Review, Papers and Proceedings 90(2): 135­39. Behrman, Jere, and James Knowles. 1999. "Household Income and Child Schooling in Vietnam." World Bank Economic Review 13(2): 211­56. Conning, Jonathan, and Michael Kevane. 1999. "Community Based Target- ing Mechanisms for Social Safety Nets." Williams College, Williamstown, Mass. Processed. Cox, Donald. 2004. "Private Interhousehold Transfers in Vietnam in the Early and Late 1990s." In Paul Glewwe, Nisha Agrawal, and David Dollar, eds., Economic Growth, Poverty, and Household Welfare in Vietnam. Washington, D.C.: World Bank. Dreze, Jean, and Amartya Sen. 1989. Hunger and Public Action. Oxford: Oxford University Press. Fritzen, Scott. 1999. "Decentralization, Disparities, and Innovation in Vietnam's Health Sector." In Jennie Litvack and Dennis Rondinelli, eds., Market Reform in Vietnam: Building Institutions for Development. Westport, Conn.: Quorum Books. Galasso, Emanuela, and Martin Ravallion. 2004. "Decentralized Targeting of an Anti-Poverty Program." Journal of Public Economics. forthcoming. Glewwe, Paul, Michele Gragnolati, and Hassan Zaman. 2002. "Who Gained from Vietnam's Boom in the 1990s?" Economic Development and Cultural Change 50(4): 773­92. Glewwe, Paul, and Jennie Litvack. 1998. "Provision of Health Care and Education in Transitional Asia: Key Issues and Lessons from Vietnam." Working Paper 147. World Institute for Development Economics Research, Helsinki. GovernmentofVietnam­DonorWorkingGroup.2000."AnnexC:Commune- Level Fees and Contributions." In Vietnam, Managing Resources Better: Public Expenditure Review 2000. Vol. 2. Joint Report on Public Expenditure Review. Hanoi. Kolko, Gabriel. 1997. Vietnam: Anatomy of a Peace. London: Routledge. Litvack, Jennie. 1999. "The Dilemmas of Change: Revitalizing Social Ser- vices in a Period of Transition." In Jennie Litvack and Dennis Rondinelli, eds., Market Reform in Vietnam: Building Institutions for Development. Westport, Conn.: Quorum Books. Minot,Nicholas,andBobBaulch.2004."TheSpatialDistributionofPovertyin Vietnam and the Potential for Targeting." In Paul Glewwe, Nisha The Static and Dynamic Incidence of Vietnam's Public Safety Net 227 Agrawal, and David Dollar, eds., Economic Growth, Poverty, and House- hold Welfare in Vietnam. Washington, D.C.: World Bank. (Originally published for the International Food Policy Research Institute in 2001.) MOLISA (Ministry of Labor, Invalids and Social Affairs). 1999. Statistics on Social Safety Net in Vietnam. Hanoi: Labor and Social Affairs Publishing House. ------. 2001. "Poverty Alleviation Strategy: 2001­2010." Hanoi. Nguyen, The Dzung. 1999. "Capacity Assessment for Vietnam's National Target Program for Hunger Eradication and Poverty Reduction: Pro- gram Assessment." United Nations Development Programme, Hanoi. Rao, Govinda, Richard Bird, and Jennie Litvack. 1999. "The Changing Re- quirements of Fiscal Relations: Fiscal Decentralization in a Unified State." In Jennie Litvack and Dennis Rondinelli, eds., Market Reform in Vietnam: Building Institutions for Development. Westport, Conn.: Quorum Books. Ravallion, Martin. 1999. "Are Poorer States Worse at Targeting Their Poor?" Economics Letters 65: 373­77. Ravallion, Martin, Dominique van de Walle, and Madhur Gautam. 1995. "Testing a Social Safety Net." Journal of Public Economics 57(2): 175­99. van de Walle, Dominique. 1999. "Safety Nets in an Emerging Market Econ- omy." In Jennie Litvack and Dennis Rondinelli, eds., Market Reform in Vietnam: Building Institutions for Development. Westport, Conn.: Quorum Books. ------. 2003. "Testing Vietnam's Public Safety Net." Social Protection Dis- cussion Paper 0319, World Bank, Washington, D.C. van de Walle, Dominique, and Dileni Gunewardena. 2001. "Sources of Ethnic Inequality in Vietnam." Journal of Development Economics 65(1): 177­207. Viet Nam Living Standards Survey (VNLSS). 1992­1993. Web site: www.worldbank.org/lsms/country/vn93/vn93bid.pdf. Vietnam Living Standards Survey (VLSS). 1997­98. Web site: www. worldbank.org/lsms/country/vn98/vn98bif.pdf. World Bank. 1995. "Vietnam Living Standards Survey (VLSS). 1992­93: Basic Information." World Bank, Research Development Group, Washington, D.C. Processed. ------. 2000. "Vietnam Living Standards Survey (VLSS). 1997­98: Basic Information." World Bank, Poverty and Human Resources Division, Washington, D.C. 7 The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting Nicholas Minot and Bob Baulch This chapter combines household survey and census data to construct a provincial poverty map of Vietnam and evaluate the accuracy of geograph- ically targeted antipoverty programs. Per capita expenditure is estimated as a function of selected household and geographic characteristics using the 1998 Vietnam Living Standards Survey (VLSS).1 Next, these results are com- bined with data on the same household characteristics from the 1999 census to estimate the incidence of poverty in each province. The results indicate that rural poverty is concentrated in 10 provinces in the Northern Uplands, two provinces in the Central Highlands, and two provinces in the Central Coast. Finally, receiver operating characteristics (ROC) curves are used to evaluate the effectiveness of geographic targeting. The results show that the existing poor communes system excludes large numbers of poor people, but there is the potential to sharpen poverty targeting by using a small number of easy-to-measure household characteristics. In most countries, poverty is spatially concentrated. Extreme poverty in inaccessible areas with unfavorable terrain often coexists with relative afflu- ence in more favorable locations close to major cities and markets. Informa- tion on the spatial distribution of poverty is of interest to policymakers and researchers for a number of reasons. First, it can be used to quantify sus- pected regional disparities in living standards and identify which areas are falling behind in the process of economic development. Second, it facilitates the targeting of programs--such as education, health, credit, and food aid-- whose purpose is, at least in part, to alleviate poverty. Third, it may shed light on the geographic factors associated with poverty, such as mountain- ous terrain or distance from major cities. Traditionally, information on poverty has come from household income and expenditure surveys. These surveys generally have sample sizes of 2,000 to 8,000 households, which allow estimates of poverty for only 3 to 12 regions 229 230 Economic Growth, Poverty, and Household Welfare in Vietnam within a country. Previous research has shown, however, that geographic tar- geting is most effective when the geographic units are quite small, such as a village or district (Baker and Grosh 1994; Bigman and Fofack 2000). The only household information usually available at this level of disaggregation is census data, but census questionnaires are generally limited to household characteristics and rarely include questions on income or expenditure. In recent years, new techniques have been developed that combine house- hold and census data to estimate poverty for more disaggregated geographic units. Although various approaches have been used, they all involve two steps. First, household survey data are used to estimate poverty or expendi- ture as a function of such household characteristics as household composi- tion, education, occupation, housing characteristics, and asset ownership. Second, census data on those same household characteristics are inserted into the equation to generate estimates of poverty for small geographic areas. For example, Minot (1998, 2000) used the 1993 VLSS and a probit model to estimate the likelihood of poverty for rural households as a function of a series of household and farm characteristics. District-level means of these same characteristics were then obtained from the 1994 Agricultural Census and inserted into this equation, generating estimates of rural poverty for each of the 543 districts in Vietnam. Hentschel and others (2000) developed a similar method using survey and censusdatafromEcuador.Usinglog-linearregressionmodelsandhousehold- level data from a census, they demonstrate that their estimator generates un- biasedestimatesofthepovertyheadcount,andtheyshowhowtocalculatethe standard error of the poverty headcount.2 This approach has been applied in a number of other countries, including Panama and South Africa (see Statistics South Africa and the World Bank [2000]; World Bank [2000]). The earlier Vietnam study (Minot 1998, 2000) has several limitations. First, because it relied on an agricultural census, it generated poverty esti- mates only for the rural areas. Second, the use of a probit regression and district-level means, although intuitively plausible, does not necessarily generate consistent estimates of district-level poverty.3 Third, in the absence of household-level census data, it was not possible to estimate the standard errors of the estimates to evaluate their accuracy. Accordingly, this chapter has three objectives. First, it explores the house- hold factors associated with poverty in Vietnam, using the 1998 VLSS. In this task, it builds on an earlier report describing the characteristics of poor households and individuals in Vietnam (Poverty Working Group 1999). Second, it examines the spatial distribution of poverty in Vietnam. It is well known in Vietnam that "poverty has marked regional characteristics" with the Northern Uplands, Central Highlands, and North Central Coast re- gions being the poorest (Socialist Republic of Vietnam 2001). However, be- cause of sampling considerations, the VLSS can be disaggregated only down to the regional level. Our analysis uses the 1998 VLSS and a 3 percent sam- ple of the 1999 Population and Housing Census. It therefore represents an improvement on the earlier Vietnam studies: (a) the data are more recent, an important consideration in a rapidly growing country such as Vietnam; The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting 231 (b) the analysis covers the spatial distribution of poverty at the provincial level, providing a more disaggregated view of poverty than is normal in Vietnam; and (c) the standard error of the poverty headcount is calculated. The standard errors are based on the methods suggested by Hentschel and others (2000), with extensions to incorporate the sampling error associated with using a 3 percent sample of the census, rather than the full census. Third, this study examines the efficacy of Vietnam's existing geographi- cally targeted antipoverty programs and investigates the potential for improving the targeting of the poor by using the type of additional household-level variables that could be collected in a "quick and dirty" enumeration of households. The section that follows this brief introduction describes the data and methods used to generate poverty maps for Vietnam from household sur- vey data and census data. The next section, "Factors Associated with Poverty in Vietnam," describes the results of the regression analysis. Al- though these are an input in the poverty mapping procedure, they also yield insights on the factors associated with poverty and how they vary between urban and rural areas. "Poverty Maps of Vietnam" presents the provincial estimates of urban and rural poverty in Vietnam, along with the standard er- rors of these estimates. The next section examines the efficacy of Vietnam's poor and disadvantaged communes program and investigates whether use of additional household variables might improve poverty targeting. Finally, the last section summarizes the results, discusses some of their policy impli- cations, and suggests areas for future research. Data and Methods This study makes use of two datasets: the 1998 VLSS and the 1999 Popula- tion and Housing Census. The VLSS was implemented by the General Statistical Office (GSO) of Vietnam with funding from the Swedish International Development Agency and the United Nations Development Programme (UNDP) and technical assistance from the World Bank. The sample included 6,000 households (4,270 in rural areas and 1,730 in urban areas) in Vietnam, selected using a stratified random sample. Data The 1999 census was carried out by the GSO and refers to the situation as of April 1, 1999. It was conducted with the financial and technical support of the UnitedNationsFamilyPlanningAssociationandUNDP.Thefullresultsofthe census have not yet been released; thus, this analysis is based on a 3 percent sample of the census. The 3 percent sample was selected by the GSO using a stratified random sample of 5,287 enumeration units and 534,139 house- holds. It was designed to be representative at the provincial level. A number of variables are common to both the VLSS and the census, which allows household-level expenditures to be predicted and disaggre- gated poverty estimates produced. Table 7.1 summarizes the 17 variables that were selected for inclusion in this poverty-mapping exercise. 232 Economic Growth, Poverty, and Household Welfare in Vietnam Table 7.1. Household Characteristics Common to the Census and the Vietnam Living Standards Survey Question number Variable 1999 1998 name(s) Description of variable Census VLSS hhsize Household size (number of people) Pt I,Q4 S1A pelderly Proportion of elderly people Pt I, Q4 S1A,Q2 (age over 60) in household pchild Proportion of children (aged under 15) Pt I, Q4 S1A,Q6 in household pfemale Proportion of females in household Pt I, Q3 S1A,Q6 Iedchd_1 to 6 Highest level of education completed by Pt I,Q11-13 S2A head (less than primary school, primary school, lower secondary school, upper secondary school technical or vocation training, college diploma or university degree) Iedcsp_0 Dummy for no spouse Pt I, Q2 S1B,Q3 Iedcsp1 to 6 Highest level of education completed by Pt I,Q11-13 S2A spouse (less than primary, primary school, lower secondary school, upper secondary school technical or vocation training, college diploma or university degree) ethnic Dummy for ethnic minority head Pt I, Q4 S0A (not Kinh or Chinese) Ioccup_1 to 7 Occupation of head over last 12 months Pt I, Q16 S4D (political leader or manager, professional or technical worker, clerk or service worker, agriculture nonfarm enterprises, unskilled worker, not-working) Ihouse_1 to 3 Type of house (permanent; semipermanent Pt III, Q3 S6A,Q1 or wooden frame, "simple") htypla1 to 2 House type interacted with living area (m2) Pt III, 4 S6C,Q1a electric Household with electricity Pt III, Q7 S6B,Q33 Iwater_1 to 3 Main source of drinking water (private Pt III, 8 S6B,Q25 or public tap, rainwater and wells, rivers and lakes) Itoilet_1 to 3 Type of toilet (flush, latrine/other, none) Pt III, Q9 S6B,Q31 tv Dummy for TV ownership Pt III, Q10 S12C radio Dummy for radio ownership Pt III, Q11 S12C reg7_1 to 7 Regional dummies (7 regions) page 1 S0A Source: Questionnaires for 1998 VLSS and 1999 Population and Housing Census. The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting 233 To estimate the poverty headcount, expenditures were predicted using these common variables and then the food and overall poverty lines devel- oped by the GSO and the World Bank for use with the VLSSs (Poverty Working Group 1999) were applied. The lower of these two lines, the food poverty line, corresponds to the expenditure (including the value of home production and adjusted regional and seasonal price differences) required to purchase 2,100 kilocalories per person a day. The upper overall poverty line also incorporates a modest allowance for nonfood expenditures.4 Vietnam's Ministry of Labor, Invalids and Social Affairs (MOLISA) esti- mates provincial poverty rates based on a system of administrative report- ing that uses different welfare indicators (rice equivalent income), different poverty lines, and a different unit of analysis (households).5 Nonetheless, the results are fairly similar to those obtained in this study. Estimating Poverty with a Household Survey As mentioned above, the first step in implementing this approach is to esti- mate poverty or household welfare as a function of household characteris- tics. In this study, per capita consumption expenditure is used as the measure of household welfare. The explanatory variables must be useful in "predicting" household welfare, and they must exist in both the household survey and the census. Economic theory provides no guidance on the func- tional form, but often a log-linear function is used: (7.1) ln(yi) = Xi + i where yi is the per capita consumption expenditure of household i, Xi is a 1 × k vector of household characteristics of household i, is a k × 1 vector of coefficients, and i is a random disturbance term distributed as N(0, ). Because the main interest here is predicting the value of ln(y) rather than assessing the impact of each explanatory variable, the possible endogeneity of some of the explanatory variables is not a concern. Hentschel and others (2000) show that the probability that household i with characteristics Xi is poor can be estimated as: ln z - Xi^ (7.2) E[Pi | Xi, ^, ^ ] = ^ where Pi is a variable taking a value of one if the household is poor and zero otherwise, z is the poverty line expressed in terms of consumption expendi- ture per capita, ^ and ^ are the estimated coefficients from the regression in equation 7.1, and is the cumulative standard normal function. Applying Regression Results to the Census Data In the second step, the estimated regression coefficients from equation 7.1 are combined with census data on the same household characteristics to predict the probability that each household in the census is poor. This is 234 Economic Growth, Poverty, and Household Welfare in Vietnam accomplished by inserting the household characteristics for household i from the census, Xi , into equation 7.2, as shown in equation 7.3. C C (7.3) E Pi | Xi , ^, ^ = C ln z - Xi ^ ^ For a given area (such as a district or province), Hentschel and others (2000) show that the proportion of the population living in households that are below the poverty line is estimated as the mean of the probabilities that individual households are poor, as shown in equation 7.4. N mi ln z - Xi C (7.4) E P | XC, , 2 = i=1 M where mi is the size of household i, M is the total population of the area in question, N is the number of households, and XC is an N × k matrix of house- hold characteristics. The advantage of using the census data, of course, is that the large number of households allows estimation of poverty headcounts for geographic units much smaller than would be possible with the VLSS data. Provided that (a) the error term is homoskedastic, (b) there is no spatial autocorrelation, and (c) the full census data are used, the variance of the estimated poverty headcount can be calculated as shown in equation 7.5. (7.5) 2 N 2 var(P) = P P P 2^ 4 mi Pi (1 - Pi ) ^ var(^) ^ + ^ 2 n - k - 1 + i=1 M 2 where P* is the estimated poverty headcount and n is the sample size in the regression model. Thus, n, k, and 2 are from the regression analysis, and mi, M, and N are obtained from the census data. The partial derivatives of P* with respect to the estimated parameters can be calculated as shown in equations 7.6 and 7.7. P N mi -xi ln z - Xi^ (7.6) j ^j = i=1M ^ ^ P N mi ln z - Xi ^ C ln z - Xi ^ C (7.7) ^ 2 = -12 i=1M ^ 3 ^ The first two terms in equation 7.5 represent the "model error," which comes from the fact that there is some uncertainty regarding the true value of and in the regression analysis. This uncertainty is measured by the estimated covariance matrix of and the estimated variance of 2, as well as the effect of this variation on P*. The third term in equation 7.5 measures the "idiosyn- cratic error," which is related to the fact that, even if and are measured exactly, household-specific factors will cause the actual expenditure to differ The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting 235 from predicted expenditure. These equations are described in more detail in Elbers, Lanjouw, and Lanjouw (2003) and Hentschel and others (2000). As noted, equation 7.5 is valid only if the full census data are available for the second stage of the mapping procedure. When using a sample survey or a sample of the census data in the second stage, this expression must be modified as shown in equation 7.8. P P P 2 2^ 4 (7.8) var(P) = ^ var(^) ^ + ^ 2 n - k - 1 N + mi Pi (1 - Pi ) 2 + Vs f i=1 M2 where Vs represents the variance associated with the sampling error in the census, taking into account the design of the sample. In this study, the software package Stata is used to calculate the variance associated with the sampling error, taking into account the design of the survey.6 To compare poverty headcounts in different regions or provinces, it is convenient to calculate the variance of the difference between two estimates of poverty. Hentschel and others (2000, footnote 17) provide an expression for the case when full census data are used. The expression is extended to in- clude the variance associated with sampling error, as shown in equation 7.9. (7.9) 2 var(P1 - P2) = P1 - P2 P1 - P2 P1 - P2 2^ 4 ^ var(^) ^ + ^ 2 n - k - 1 + Vi(P1) + Vi(P2) + Vs(P1) + Vs(P2) - 2covs(P1, P2) where Vi(Pr) is the idiosyncratic variance of the poverty estimate for region r (the third term in equation 7.5), Vs(Pr) is the sampling variance of the poverty estimate for region r, and covs(P1, P2) is the covariance in the poverty estimates for regions 1 and 2 associated with sampling error. Two qualifications need to be made regarding the implementation of this method in the case of Vietnam. Researchers at the World Bank have recently been addressing the issue of spatial autocorrelation in the first-stage regres- sions (equation 7.1).Analytical solutions for the variance of the headcount are not possible in this case, and it becomes necessary to use complex simulation methods to calculate the estimators and their standard errors (Elbers, Lanjouw, and Lanjouw 2003). Although preliminary analysis indicates the presence of some spatial autocorrelation, in this case it could not be elimi- nated by including community-level variables in the regression analysis. This suggests that there may be some inefficiency in the results of the first-stage re- gression analysis, though the magnitude of these effects is difficult to assess. In addition, the estimate of the variance associated with sampling error produced by Stata is only an approximation. Exploratory analysis reveals that the sampling error is relatively small compared with the model error, suggest- ing that this approximation does not substantively influence the results. 236 Economic Growth, Poverty, and Household Welfare in Vietnam Factors Associated with Poverty in Vietnam The first step in constructing a poverty map is to estimate econometrically per capita consumption expenditure as a function of variables that are com- mon to the census and the VLSS. These household characteristics include household size and composition, ethnicity, education of the head of house- hold and his or her spouse, occupation of the head of household, housing size and type, access to basic services, and ownership of selected consumer durables. Table 7.1 lists the variables, and appendix 7A provides descriptive statistics for each of them. It is reasonable to expect that the factors that "predict" expenditure in rural areas may be different from those predicting expenditure in urban areas. Indeed, a Chow test strongly rejects the hypothesis that the coeffi- cients for the urban subsample are the same as those for the rural subsample (F = 6.16, p < .001). This implies that separate analyses should be carried out on rural and urban samples. The next level of disaggregation is the stratum used in the VLSS sample. The VLSS was designed to be representative for each of 10 strata, compris- ing 3 urban strata and 7 rural strata. For this analysis, it was necessary to col- lapse the three urban strata (Hanoi and Ho Chi Minh City, other cities, and towns) into two (Hanoi and Ho Chi Minh City, other urban areas) because the census data do not allow distinction between "other cities" and "towns." Within urban areas, a Chow test suggests that Hanoi and Ho Chi Minh City differ significantly from other urban areas (F = 2.20, p < .001). In addition, the seven rural regions differ significantly from each other (F = 12.61, p < .001). In other ways, however, the stratum-level regressions are not very satisfactory. Because of the small sample size in each stratum (ranging from 368 to 1,111 households), many of the coefficients are not statistically signif- icant at conventional levels or have counterintuitive signs. Furthermore, the goodness-of-fit of most of the stratum regressions is below 0.5, compared with 0.54 and 0.55 for the rural and urban regressions. One result of this is that the standard errors of the poverty estimates from the stratum-level re- gressions are higher than those obtained from the urban-rural regressions (see "Regional Poverty Estimates" below). This chapter presents the results of both the urban-rural regressions (see tables 7.2 and 7.3) and the stratum-level regressions (see appendixes 7B and 7C), as well as the poverty estimates derived from each (tables 7.4­7.6 and appendix 7D). However, greater prominence is given to the results from the urban-rural regression analysis. As will be shown in another section, the two methods yield similar poverty headcounts and rankings, particularly for the poorest provinces. The results of the regression analysis are summarized to "predict" per capita expenditures in the next sections. Household Size and Composition Large households are strongly associated with lower per capita expenditure in both urban and rural areas, as shown in table 7.2. The negative sign of the Table 7.2. Determinants of Per Capita Expenditure for Rural and Urban Areas Rural model Urban model N 4269 1730 R-squared 0.536 0.550 Variable Coefficient t Variable Coefficient t hhsize -0.0772 -19.5*** hhsize -0.0785 -8.1*** pelderly -0.0831 -2.4** pelderly -0.1026 -1.6 pchild -0.3353 -9.4*** pchild -0.2368 -3.6*** pfemale -0.1177 -3.5*** pfemale 0.0386 0.5 ethnic -0.0765 -1.9* ethnic 0.0142 0.2 Iedchd_2 0.0585 3.4*** Iedchd_2 0.0616 1.7 Iedchd_3 0.0883 4.5*** Iedchd_3 0.0338 1.3 Iedchd_4 0.0884 3.3*** Iedchd_4 0.1368 3.2*** Iedchd_5 0.1355 4.2*** Iedchd_5 0.1603 3.5*** Iedchd_6 0.2552 4.9*** Iedchd_6 0.1843 3.7*** Iedcsp_0 0.0173 1.0 Iedcsp_0 0.0344 0.8 Iedcsp_2 0.0049 0.3 Iedcsp_2 0.0642 1.9* Iedcsp_3 0.0132 0.6 Iedcsp_3 0.0987 2.6** Iedcsp_4 0.0107 0.3 Iedcsp_4 0.1912 2.7** Iedcsp_5 0.0921 2.3** Iedcsp_5 0.1285 3.2*** Iedcsp_6 0.1571 2.7*** Iedcsp_6 0.1752 3.1*** Ioccup_1 0.1414 3.5*** Ioccup_1 0.2312 3.0*** Ioccup_2 0.1350 3.3*** Ioccup_2 0.0576 1.2 Ioccup_3 0.1362 3.4*** Ioccup_3 0.0357 0.9 Ioccup_4 -0.0163 -0.6 Ioccup_4 -0.0093 -0.2 Ioccup_5 0.0701 1.9* Ioccup_5 0.0071 0.2 Ioccup_6 -0.0586 -1.7* Ioccup_6 -0.1599 -2.9*** Ihouse_1 -0.9228 -4.3*** Ihouse_1 -0.5194 -3.4*** Ihouse_2 -0.3120 -3.6*** Ihouse_2 -0.4001 -3.8*** htypla1 0.2958 5.7*** htypla1 0.2001 5.4*** htypla2 0.1180 5.2*** htypla2 0.1403 4.6*** electric 0.0765 2.7*** electric -0.0026 0.0 Inwate_1 0.0828 1.4 Inwate_1 0.2289 5.3*** Inwate_2 0.1157 4.4*** Inwate_2 0.0340 0.6 Itoile_1 0.2700 5.5*** Itoile_1 0.1311 2.2** Itoile_2 0.0556 2.6** Itoile_2 0.0049 0.1 tv 0.2124 15.1*** tv 0.2167 5.5*** radio 0.1009 7.0*** radio 0.1599 6.2*** Ireg7_2 0.0314 0.6 Ireg7_2 0.0693 0.7 Ireg7_3 0.0485 0.8 Ireg7_3 0.0445 0.6 Ireg7_4 0.1373 2.2** Ireg7_4 0.1460 1.9* Ireg7_5 0.1708 2.1** lreg7_5 omitted Ireg7_6 0.5424 9.4*** Ireg7_6 0.4151 5.5*** Ireg7_7 0.3011 5.1*** Ireg7_7 0.1895 2.1** _cons 7.5327 108.7*** _cons 7.7538 64.7*** Note: The dependent variable is log of per capita expenditure. *Coefficient is significant at the 10 percent level, **at the 5 percent level, and ***at the 1 percent level. Source: Regression analysis of 1998 VLSS. 237 238 Economic Growth, Poverty, and Household Welfare in Vietnam coefficient on household size implies that, other factors being equal, each additional household member is associated with a 7­8 percent reduction in per capita expenditure.7 The stratum-level regressions show similar results (see appendix 7B). In rural areas, a household with a large number of elderly members, chil- dren, and females is likely to have low per capita expenditure. In urban areas, however, only the number of children is statistically significant (see table 7.2). Household composition appears to matter less in urban areas than rural ones. It may be that the number of children, women, and elderly people has less effect on household welfare in urban areas because income-earning capacity in the cities and towns is less dependent on physical strength. Ethnicity8 is a predictor of per capita expenditure, but a surprisingly weak one, after controlling for other household characteristics (education, occupation, ownership of consumer durables, and so on). In rural areas, the coefficient on ethnicity was significant only at the 10 percent level, and in urban areas, it was not statistically significant (see table 7.2). The urban co- efficient is not surprising, given the very small sample of ethnic minority households in urban areas (just 19 households). The weakly significant, although appropriately signed, coefficient for rural areas is more surprising, given the strong correlation between poverty and ethnicity in Vietnam. Other research (Baulch and others 2004; van de Walle and Gunewardana 2001) suggests that ethnic minorities have both lower levels of endowments and lower returns to those endowments. The results in this chapter are con- sistent with these findings, showing that after controlling for differences in endowments (education, housing characteristics, and ownership of con- sumer durables), differences in per capita expenditure between ethnic minority households and others remain, but are much smaller. Education In both urban and rural areas, the level of schooling of the head of house- hold is a good predictor of a household's per capita expenditure.9 The five dummy variables that represent the education of the head of household are jointly significant at the 1 percent level in both rural and urban areas (see table 7.3). In rural areas, heads of household who completed primary school earn 6 percent more than those who did not complete primary school. In urban areas, households whose head has completed primary or lower sec- ondary school do not seem to be better off than those whose head has not completed primary school, but higher levels of education are associated with significantly higher earnings (see table 7.2). In general, the educational level of the spouse is less significant than that of the household head as a predictor of per capita expenditure.10 In the rural areas, only the highest two levels of education of the spouse (advanced tech- nical training and postsecondary education) show any significant effect rel- ative to the base level (not completing primary school). The education of the spouse is a better predictor in urban areas than in rural areas (see table 7.2). The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting 239 Table 7.3. Tests of Significance of Groups of Explanatory Variables in Urban-Rural Regressions Sector Variables df1 df2 F statistic Probability Rural Education of head of household 5 129 7.80 0.0000*** Education of spouse 6 129 1.97 0.0738* Occupation of head 6 129 12.65 0.0000*** Type of housing 2 129 14.00 0.0000*** Main source of water 2 129 9.69 0.0001*** Type of sanitary facility 2 129 15.64 0.0000*** Region 6 129 26.20 0.0000*** Urban Education of head of household 5 55 4.01 0.0036*** Education of spouse 6 55 3.10 0.0110** Occupation of head 6 55 2.90 0.0157** Type of housing 2 55 10.76 0.0001*** Main source of water 2 55 17.17 0.0000*** Type of sanitary facility 2 55 4.12 0.0216** Region 5 55 10.29 0.0000*** Note: The dependent variable is log of per capita expenditure. *Coefficient is significant at the 10 percent level, **at the 5 percent level, and ***at the 1 percent level. Source: Regression analysis of per capita expenditure using 1998 VLSS. Occupation The occupation of the head of household is a statistically significant predic- tor of per capita expenditure in rural and urban areas.11 In rural areas, the first three occupational categories (political leaders or managers, profes- sionals or technicians, and clerks or service workers) are significantly better off than households in which the head of household is not working. There is no statistically significant difference between the expenditure of farm households and households with nonworking heads, however (see table 7.2). This somewhat counterintuitive finding reflects the fact that nonworking heads include retirees as well as a disproportionate number of skilled work- ers who can "afford" to look for better-paid work. In urban areas, households whose head is a leader or manager are sig- nificantly better off than those with nonworking heads, and those whose head is an unskilled worker are significantly worse off (see table 7.2). This suggests that in urban areas, a nonworking head of household is not a reli- able indicator that the household is poor. Housing and Basic Services Various housing characteristics are good predictors of expenditures. Living in ahouseorotherdwellingmadeofpermanentratherthantemporarymaterials is associated with 19 percent (24 percent) higher per capita expenditure in 240 Economic Growth, Poverty, and Household Welfare in Vietnam rural (urban) areas.12 Similarly, having a house of semipermanent rather than temporary materials implies a significantly higher level of per capita expendi- ture. The living area of houses is also a useful predictor of household well- being. Houses in Vietnam have an average living area of about 45 square me- ters, and each 10 percent increase in area is associated with a 12­30 percent increase in per capita expenditure, depending on the place of residence (urban or rural) and the type of house (permanent or semipermanent).13 Electrification14 is a statistically significant predictor of household welfare in rural areas, where 71 percent of the households have access to electricity. By contrast, in urban areas, where 98 percent of the households are already elec- trified, electricity is not a significant predictor of expenditures (see table 7.2). The main source of water is also useful in distinguishing poor house- holds. In rural areas, households with access to well water have a higher level of per capita expenditures than households using river or lake water (the omitted category). Access to tap water is not a statistically significant predictor of expenditures in rural areas, presumably because just 2 percent of the rural households fall into this category. By contrast, in urban areas, more than half the sample households (58 percent) have access to tap water, and this variable is a good predictor of urban per capita expenditures. Finally, sanitation facilities can be used to separate poor from nonpoor households. In rural areas, flush toilets and latrines are statistically signifi- cant indicators of higher per capita expenditure at the 5 percent level. In urban areas, having a flush toilet is a significant predictor of expenditures at the 5 percent level but having a latrine is not (see table 7.2). Consumer Durables Television ownership is one of the strongest predictors of per capita expenditures--a statistically significant predictor in both urban and rural areas. Radio ownership is almost as good a predictor: statistically significant at the 1 percent level in both urban and rural areas. As expected, the coeffi- cient for radio ownership is smaller than that of television ownership (see table 7.2). Later in this chapter, the extent to which the addition of variables reflecting ownership of consumer durables or housing characteristics can improve the geographic targeting of the poor is examined. Region Regional dummy variables were included in the urban and rural regression models, with the Northern Uplands as the base region. Even after control- ling for other household characteristics, rural households in the four south- ern regions are shown to be better off than those in the Northern Uplands. The coefficient in the Southeast is the largest, implying that households in this region have expenditure levels 72 percent higher than similar house- holds in the Northern Uplands. A similar pattern holds for urban house- holds (see table 7.2). The regional dummy variables are jointly significant at the 1 percent level in both urban and rural areas (see table 7.3). The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting 241 Poverty Maps of Vietnam The second stage in constructing a poverty map is to combine the regression coefficients estimated from the VLSS in the first stage and the census data on the same household characteristics. This gives predicted expenditures for each household in the census, which are then used to estimate the incidence of poverty (the poverty headcount) for individual regions and provinces, as well as the standard errors associated with these estimates. The estimates of the incidence of poverty are presented first at the regional level and then at the provincial level. Regional Poverty Estimates Regional poverty headcounts and their standard errors, as estimated di- rectly from the 1998 VLSS, are shown in the first two columns of table 7.4. For the country as a whole, the incidence of poverty is 37.4 percent, with a 95 percent confidence interval of ±3.2 percentage points. The regional poverty headcounts range from 0.9 percent in urban Hanoi and Ho Chi Minh City to 65.2 percent in the rural Northern Uplands. The standard er- rors suggest that the degree of precision in the estimates of regional poverty using the VLSS is relatively low: Four of the nine regions have confidence limits of ±10 percentage points or more. By combining the urban-rural regression models and the census data (as described above), an alternative set of estimates of regional headcount poverty rates and standard errors, shown in the second pair of columns in table 7.4, can be seen. Seven of the nine regional estimates are within 3 per- centage points of the corresponding estimate from the VLSS. However, the census-based poverty estimates tend to be less extreme: They are higher than the VLSS estimates where the incidence of poverty is low (such as in the rural Southeast and urban areas) and lower where the incidence is high (such as in the rural Northern Uplands). In every region except one (Hanoi and Ho Chi Minh City), the standard errors of the census-based estimates are substantially smaller than those of the VLSS estimates. Apparently, the gains in accuracy from using a larger sample exceed the losses due to estimating expenditure based on household characteristics. According to the urban-rural regression results in table 7.4, the rural Northern Uplands is the poorest region. In fact, it is significantly poorer than the other eight regions at the 1 percent confidence level (see table 7.5). The rural Central Highlands and the rural North Central Coast are the next poor- est regions, although there is no statistically significant difference between the two. The rural South Central Coast, the rural Mekong Delta, and the rural Red River Delta follow, with the differences being statistically significant in each case. The rural Southeast and "other urban" areas are significantly less poor than the rural Red River Delta, but the difference between the former two areas (that is, rural Southeast and "other urban") is not statistically sig- nificant. The ninth region, Hanoi and Ho Chi Minh City, is significantly less poor than any of the other eight regions (see tables 7.4 and 7.5). or err 1 ession data 0.009 0.014 0.037 0.031 0.036 0.028 0.046 0.018 0.031 0.01 egrr Standard Census Stratum with 0.031 0.146 0.626 0.407 0.490 0.400 0.525 0.173 0.386 0.365 Poverty or err 1 1 essions data 0.007 0.012 0.01 0.006 0.01 0.010 0.016 0.004 0.007 0.012 egrr Standard Census with Census. Urban-rural Headcounts 0.037 0.145 0.598 0.379 0.513 0.460 0.533 0.234 0.397 0.365 Poverty Housing and Poverty or centages. err per 0.004 0.021 0.057 0.038 0.058 0.075 0.097 0.022 0.033 0.016 than Population 1998 Standard 1999 rather Census-Based of VLSS and fractions sample as 0.009 0.138 0.652 0.361 0.488 0.436 0.524 0.130 0.412 0.374 cent Poverty essed per Original 3 of expr e and City ar VLSS Coast Coast Minh 1998 Uplands Delta Chi Comparison Highlands Delta headcounts Central Central omfr Ho River Data 7.4. and urban Northern Red North South Central Southeast Mekong Poverty ce: ableT otalT Note: Sour Region Hanoi Other Rural Rural Rural Rural Rural Rural Rural 242 -- -- Rural Southeast 0.271***- (0.008) es. Rural -- -- enthes Highlands 0.392*** (0.017) 0.120*** (0.017) par C in ors err 1) d Coast -- -- Rural C S 0.042**- (0.019) 0.349*** (0.01 0.078*** (0.012) Standar n.s. Coast egion.r -- -- 17*** Significance Rural C 0.038*** N (0.014) 0.004- (0.020) 0.388*** (0.012) 0.1 (0.013) owr of level. cent Statistical Delta per Census. Rural R -- -- headcount 1 0.134***- (0.013) 0.096***- (0.012) 0.138***- (0.018) 0.254*** (0.007) 0.017**- (0.008) the Their Red poverty ***at Housing and and and minus -- -- Rural Uplands 0.218*** (0.012) 0.084*** (0.016) 0.123*** (0.016) 0.081*** (0.021) 0.472*** (0.012) 0.201*** level, (0.013) N egionr cent Population per Headcounts n.s. 5 column 1999 -- -- of the of Other urban 0.452***- (0.016) 0.234***- (0.013) 0.368***- (0.016) 0.330***- (0.015) 0.372***- (0.020) 0.020 (0.012) 0.252***- (0.013) level. **at Poverty cent sample & level, per Chi City headcount cent 10 cent per Regional Hanoi Ho Minh 0.109***- (0.012) 0.561***- (0.013) 0.343***- (0.009) 0.477***- (0.013) 0.438***- (0.012) 0.481***- (0.017) 0.089***- (0.008) 0.360***- (0.009) per the 3 poverty 10 at in and as the at VLSS Coast Coast essed significant ferences 1998 Uplands Delta expr Dif Highlands Delta Central significant Central omfr ences River fer statistically Data 7.5. urban Northern Red North South Central Southeast Mekong Dif Not ce: ableT Note: *Statistically n.s. Sour Region Other Rural Rural Rural Rural Rural Rural Rural 243 244 Economic Growth, Poverty, and Household Welfare in Vietnam Combining the stratum-level regression models with the census data yields results similar to those based on the urban-rural regression models, as shown in the last two columns of table 7.4. Again, the poverty estimates are less extreme than the VLSS estimates and the standard errors are some- what lower. One notable difference is that the standard errors of the poverty estimates based on the stratum-level regression models are higher, often two to three times higher, than those based on the urban-rural regression models. Provincial Poverty Estimates One of the main advantages of using census data is that they allow the gen- eration of reliable estimates of poverty for smaller geographic units, such as provinces or districts, which would be difficult or impossible to estimate with a household sample survey such as the VLSS.15 Table 7.6 shows the estimated provincial poverty rates, along with the standard errors of the estimates, based on the urban-rural regression models (the corresponding results from the stratum-level regressions are given in appendix 7B). Map 7.1 shows the geographic distribution of poverty at the provincial level, also based on the urban-rural regression models. The results indicate that Lai Chau, located at the extreme northwest cor- ner of Vietnam, is the poorest province, with more than three-quarters of its population living below the poverty line. The next five poorest provinces (Ha Giang, Son La, Cao Bang, Lao Cai, and Lang Son) are all provinces in the Northern Uplands on the northern border with China or the western border with the Lao People's Democratic Republic. In fact, the 10 poorest provinces are all in the Northern Uplands. This is probably a reflection of their moun- tainous topography, distance from major markets, and limited infrastruc- ture, all of which reduce the returns to agriculture in this region. Ethnic minorities also make up more than half of the population of these provinces. Poverty is not limited to the Northern Uplands, however. The North Central Coast composes six provinces, all of which are among the poorest 21 provinces in the country. The incidence of poverty in these provinces ranges from 44 percent to 52 percent. The Central Highlands region includes three provinces. Two of the three, Kon Tum and Gia Lai, are among the 15 poorest provinces in Vietnam, with poverty headcounts of more than 50 percent. The third province, Dak Lak, is more prosperous, with a poverty headcount similar to the national average. This is probably due to the importance of coffee production. Vietnam now exports US$500 million worth of coffee per year, most of which is grown in Dak Lak province. Poverty is less severe in the southern regions, although each region has at least one province with a poverty headcount over 40 percent. The South- east region is the least poor region, but it has two provinces, Ninh Thuan and Binh Thuan, with poverty headcounts of more than 40 percent. These provinces are farther from Ho Chi Minh City than the other provinces in the Southeast. In the South Central Coast, Quang Ngai has a poverty headcount The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting 245 Map 7.1. Incidence of Poverty by Province Poverty headcount 0­20% 20­40% 40­50% 50­60% 60­70% 70­100% 0 100 200 300 400 Kilometers Source: Estimated from urban-rural regression models of the 1998 VLSS and household characteristics in the 1999 Population and Housing Census. of 47 percent. In the Mekong Delta region, Soc Trang, Tra Vinh, and An Giang have rates of more than 40 percent. The lowest incidence of poverty is found in Ho Chi Minh City (less than 5 percent), followed by four provinces in the Southeast (Binh Duong, Ba Ria-Vung Tau, Dong Nai, and Tay Ninh), all of which have poverty head- counts under 15 percent. The headcounts for Hanoi and Da Nang are both close to 15 percent. otalT 0.034 0.036 0.034 0.033 0.036 0.032 0.034 0.036 0.038 0.036 0.047 0.033 0.043 0.034 0.040 0.041 0.038 0.033 0.047 0.041 0.040 0.047 0.033 0.042 0.029 ors err Urban 0.036 0.032 0.029 0.034 0.031 0.033 0.037 0.028 0.026 0.027 0.032 0.038 0.035 0.034 0.028 0.029 0.030 0.033 0.028 0.027 0.030 0.032 0.040 0.024 0.040 Standard Rural 0.038 0.039 0.039 0.037 0.043 0.038 0.039 0.041 0.043 0.044 0.062 0.041 0.061 0.043 0.044 0.046 0.043 0.043 0.050 0.045 0.044 0.052 0.041 0.049 0.034 Model otalT 0.777 0.722 0.714 0.675 0.652 0.617 0.609 0.586 0.583 0.550 0.538 0.525 0.522 0.520 0.491 0.477 0.474 0.472 0.470 0.460 0.445 0.442 0.435 0.431 0.424 Regression headcount Urban 0.221 0.195 0.153 0.142 0.197 0.141 0.189 0.155 0.161 0.165 0.194 0.214 0.221 0.192 0.132 0.140 0.153 0.185 0.164 0.135 0.151 0.199 0.235 0.132 0.244 Urban-Rural Poverty with Rural 0.857 0.770 0.795 0.739 0.747 0.724 0.676 0.655 0.635 0.644 0.650 0.618 0.670 0.618 0.532 0.515 0.513 0.579 0.494 0.492 0.474 0.470 0.498 0.482 0.463 Estimated NU NU NU NU NU NU NU NU NU NU CH SE HC NCC NCC NCC SCC NCC NU NCC NCC NU SE NU MRD Region Headcounts Poverty Hue- Son Quang irT Binh Ngai Thuan An Hoa Thien Phuc Thuan ovince Chau La Cai Binh Giang Bang Kan Bai Lai muT Giang inhT Tho rangT Pr Provincial Lai Ha Son Cao Lao Lang Bac Hoa uyenT enY Gia Ninh Kon Quang Quang Nghe Quang Thua Bac Thanh Ha inhV Binh Phu Soc 7.6. ableT 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Rank 246 page) 0.038 0.030 0.035 0.036 0.027 0.050 0.031 0.028 0.032 0.029 0.046 0.030 0.028 0.027 0.031 0.029 0.028 0.025 0.027 0.039 0.027 0.032 0.028 0.028 0.027 0.027 0.022 0.026 0.024 following on 0.023 0.032 0.036 0.035 0.036 0.029 0.028 0.036 0.033 0.026 0.028 0.037 0.036 0.037 0.031 0.026 0.026 0.031 0.030 0.024 0.030 0.021 0.029 0.027 0.024 0.031 0.019 0.025 0.017 continues (table 0.047 0.034 0.042 0.041 0.033 0.063 0.033 0.032 0.041 0.033 0.050 0.032 0.034 0.033 0.033 0.048 0.032 0.031 0.032 0.061 0.031 0.033 0.031 0.032 0.040 0.031 0.032 0.030 0.028 1 0.419 0.418 0.416 0.408 0.406 0.395 0.395 0.391 0.391 0.385 0.383 0.383 0.380 0.377 0.376 0.357 0.351 0.349 0.345 0.337 0.330 0.330 0.325 0.319 0.31 0.305 0.286 0.276 0.179 10 0.126 0.191 0.188 0.191 0.196 0.176 0.125 0.195 0.179 0.109 0.166 0.163 0.210 0.207 0.143 0.155 0.1 0.156 0.152 0.144 0.148 0.075 0.137 0.106 0.126 0.151 0.074 0.105 0.076 0.495 0.452 0.469 0.443 0.454 0.451 0.417 0.424 0.460 0.424 0.405 0.403 0.428 0.430 0.391 0.519 0.385 0.402 0.388 0.458 0.360 0.345 0.342 0.353 0.416 0.335 0.395 0.301 0.197 NU MRD CCS SCC MRD CH RDR MRD SCC RRD NU RDR MRD MRD RRD NU RRD MRD MRD SE MRD RRD MRD RRD CCS MRD RRD MRD SE Ninh Hoa Nguyen inhV neY Nam Binh neY An Giang Lac yaT Thap Dinh Giang Dinh Long Ninh Lieu Nam Tho Dong Binh Mau erT Duong Phong Giang Phuoc Thai raT Phu Quang An Dac Ha Dong Binh Ninh Bac Hung Kien Bac Ha Quang Nam Can Ca Lam inhV Thai Ben Hai Khanh Long Hai eniT Binh 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 247 1 es otalT 0.019 0.015 0.016 0.014 0.01 0.010 0.007 0.012 Highlands, expenditur ors err pita 1 1 ca Central = Urban 0.022 0.010 0.017 0.01 0.013 0.012 0.008 0.01 per CH Standard with Coast, Rural 0.038 0.031 0.019 0.020 0.016 0.014 0.014 0.015 households Central in South live = 11 Giang SCC otalT 0.156 0.152 0.124 0.1 0.090 0.079 0.044 0.365 An in Coast, Census. Central headcount 11 population Urban 0.106 0.037 0.081 0.048 0.062 0.051 0.036 0.1 North Housing the = of and Poverty cent NCC per Delta, Population Rural 0.346 0.306 0.130 0.137 0.109 0.092 0.082 0.441 40.6 River 1999 that of Red = implies sample SCC RDR SE SE ES RRD SE SE Region Giang cent Delta. per An 3 Uplands, for River and uaT line. 0.406 of Northern VLSS ) Minh Mekong = ung poverty = 1998 Nai Chi Duong NUe ovince Nang Noi Ninh Ria-V Ho headcount ar MRD omfr continued( Pr Da Ha ayT Dong Ba Binh TP otalT GSO/WB and codes poverty 1998 7.6. A the gioner Estimated ce: Southeast, ableT 55 56 57 58 59 60 61 Note: The = Sour Rank below SE 248 The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting 249 Map 7.2. Incidence of Rural Poverty by Province Rural poverty headcount 0­20% 20­40% 40­50% 50­60% 60­70% 70­100% 0 100 200 300 400 Kilometers Source: Estimated from urban-rural regression models of the 1998 VLSS and household characteristics in the 1999 Population and Housing Census. Poverty headcounts in rural areas are similar to the overall provincial poverty levels, which is not surprising, given the large proportion of the population living in rural areas in most provinces (see table 7.6 and map 7.2). Rural poverty is greatest in the border provinces of the Northern Uplands. The Central Highlands provinces of Gia Lai and Kon Tum are among the 10 poorest provinces in terms of rural poverty. 250 Economic Growth, Poverty, and Household Welfare in Vietnam As expected, the incidence of poverty in urban areas is consistently lower than that in rural areas. Even in the poorest provinces, where more than 70 percent of the rural population is poor, urban poverty is below 25 per- cent. In contrast, the difference between rural and urban poverty head- counts is relatively small in the more prosperous provinces in the Southeast (see table 7.6). To determine whether the poverty estimates for any two provinces are statistically different from one another, the standard error of the difference between their poverty headcounts must be calculated. This statistic can be computed using equation 7.9, which takes into account the modeling error, the idiosyncratic error, and the sampling error associated with the 3 percent sample census. The standard errors of these differences (based on the urban-rural regressions) have been calculated for the 1,830 possible pairs of provinces, and their rural and urban subsamples, with the 61 urban-rural pairs in the same province. The results can be summarized as follows: · About one-quarter (23 percent) of the provincial pairs with a 6 per- centage point gap16 in their poverty incidence are significantly differ- ent from each other at the 5 percent level of statistical significance. Forty-three percent of the provincial pairs with an 8 percentage point gap and 70 percent of those with a 10 percentage point gap have statistically different poverty levels. This implies that poverty headcounts are generally not statistically different from one another in provinces that are adjacent to each other in poverty rankings. Provinces that are four to five provinces away from each other in the ranking, however, will usually have statistically significant differ- ences in their poverty headcounts. · Poverty headcounts in 65 percent of the rural provincial pairs, but just 33 percent of the urban pairs, are significantly different from one another (at the 5 percent level). This is largely due to the fact that rural areas have higher and more diverse poverty headcounts, so the (absolute) differences are larger. · In every province except one, Tay Ninh, the incidence of rural poverty is significantly higher than that of urban poverty. Finally, the sensitivity of these results to the type of regression models esti- mated in the first step of the analysis was examined. In particular, how dif- ferent are the results when stratum-level regressions were used instead of urban-rural regressions? The average (absolute value) gap between provin- cial poverty headcounts obtained from these two regression models is 2.2 percentage points. Just eight provinces have differences of more than 5 percentage points, and none have differences of more than 10 percentage points. Furthermore, the ranking of the 10 poorest provinces is the same according to the two approaches. Figure 7.1 shows the similarity of the rural and urban poverty head- counts for each province (identical headcounts would be represented by points along the diagonal line). The two methods are most similar for the The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting 251 Figure 7.1. Provincial Poverty Headcounts Estimated Using Urban-Rural and Stratum-Level Regression Models Poverty headcount from rural-urban regressions 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 0.2 0.4 0.6 0.8 1.0 Poverty headcount from stratum-level regressions Urban Rural Source: Estimated from rural-urban regression models of 1998 VLSS and household characteristics in the 1999 Population and Housing Census. poorest rural regions, where the difference in estimates is typically just 1 percentage point. They are less similar for more prosperous rural areas and for urban areas. The urban poverty headcounts often differ by 4 to 8 per- centage points. The standard errors of the provincial poverty headcounts were also com- pared. Those based on the urban-rural regression models were often (72 per- cent of the time) lower than the corresponding standard errors based on the stratum-level regressions. For the poorest provinces, the standard errors of the headcount based on the stratum-level regressions are roughly twice as large as those based on the urban-rural regressions. The Potential of Geographic and Additional Targeting Variables Given knowledge about where the poor live, a natural question to ask is, "How effective are geographic variables in identifying the poor?" Experi- ence in other countries indicates that the ability to target poor households typically improves with greater geographic disaggregation (Baker and Grosh 1994; Bigman and Fofack 2000). Many of Vietnam's antipoverty pro- grams use highly disaggregated listings of "poor and remote communes"; 252 Economic Growth, Poverty, and Household Welfare in Vietnam thus, one would expect the efficiency of its geographic targeting programs to be quite high.17 In addition, the institutional impediments to internal migra- tion imposed by the ho khau (registration permits) system mean that the spatial distribution of poverty is likely to be more persistent than in coun- tries with free mobility of labor. Furthermore, spatial dependence means that sizable differences in living standards can persist even after one con- trols for observable household and individual characteristics (Ravallion and Wodon 1997). The ability to target poor households and individuals is there- fore an integral part of the government of Vietnam's desire to limit urban- rural disparities and inequality (Socialist Republic of Vietnam 2001). The government of Vietnam is strongly committed to poverty reduction and over the 1990s has developed a complex array of geographically targeted antipoverty programs and policies (Conway 2001). The most important of these are the Hunger Eradication and Poverty Reduction (HEPR) program, also known as Program 133, and Program 135, which assists communes in the most mountainous and remote areas. The HEPR program was established in 1996 with the objective of fighting poverty through the coordination and im- provement of existing targeted poverty alleviation programs in education, health, agricultural extension, irrigation, job creation, training, microfinance, and basic infrastructure. It is implemented in a number of poor communes (1,715 communes in 1998) identified by MOLISA, although program funding and implementation remain under the line ministries responsible for specific sectors. A commune is eligible to benefit from Program 133 grants if its poverty rate is above 40 percent, and poor households within these com- munes are eligible for targeted assistance (such as free or subsidized school- ing, health insurance cards, and sometimes exemption from local taxes).18 MOLISA conducts an annual exercise in which its district and commune rep- resentatives, along with party cadres, make lists of the poor households in each of the more than 10,000 communes in the country. In 1998, Program 135 was introduced to assist in the building of basic infrastructure in 1,000 com- munes in "especially difficult circumstances in mountainous and remote areas." This program is implemented by the Committee for Ethnic Minorities in Mountainous Areas in association with sector ministries. The mechanism used to target communes under Program 135 uses five criteria, which include the altitude of the commune, distance to the nearest town, adult illiteracy rate, and productive structure, in addition to the poverty rate. In addition to Programs 133 and 135 there are a number of antipoverty interventions, such as paying transportation subsidies and higher salaries to teachers working in remote areas, that are implemented at the local level. It is therefore important to know whether the poor can be identified more accurately if additional information other than place of residence is available. Implicitly, this is what the commune and district level staff of MOLISA do in their annual exercise to determine whether a household is classified as poor.19 Put differently, can the geographic targeting of the poor be improved by the use of the type of additional socioeconomic variables that can be collected easily in a "quick and dirty" enumeration of households? The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting 253 The efficacy of different targeting variables was assessed using a rela- tively novel technique: Receiver Operating Characteristic (ROC) curves, a graphic and nonparametric way of portraying the accuracy of a diagnostics test originally developed for use in electrical engineering and signal pro- cessing (Stata Corporation 2001a). An ROC curve shows the ability of a di- agnostic test to correctly distinguish between two states or conditions. In the context of poverty targeting, an ROC curve plots the probability of a test cor- rectly classifying a poor person as poor (the test's sensitivity) on the vertical axis against one minus the probability of the same test correctly classifying a nonpoor person as nonpoor on the horizontal access (the test's speci- ficity).20 When the diagnostic test (here, the values of a targeting variable) takes several discrete values, the ROC curves will consist of a series of linear segments corresponding to these discrete values. The greater the area under an ROC curve and the closer it is to the left and top axes, the greater the ef- ficacy of a diagnostic test. The closer an ROC curve is to the 45-degree line, the weaker its efficacy. To the best of the authors' knowledge, the only previous use of ROC analysis for analyzing the impact of poverty targeting is by Wodon (1997), who used household survey data from Bangladesh. As Wodon points out, unlike conventional statistical hypothesis tests, ROC analysis can take ac- count of continuous as well as categorical targeting variables. However, like conventional hypothesis tests, ROC analysis can be employed only for dichotomous outcome variables, so it cannot be used to examine the depth or severity of poverty. Figure 7.2 shows an example of two pairs of ROC curves drawn using data from the 1998 VLSS. Because the curve for the index of radio and tele- vision ownership in rural areas lies everywhere above and to the left of the curve for the education level completed by the household head, panel A shows that use of the television and radio ownership variables unambigu- ously dominates that for education of the household head as a poverty tar- geting variable. Note that the ROC for the index of radio and television own- ership has four linear segments corresponding to the four values of the index, and the ROC curve for the head of household's education has six seg- ments corresponding to the six educational levels a household head may complete. Panel B shows the contrasting situation in which the ROC for quintiles of land area and the number of children per household cross, in which case neither variable unambiguously dominates the other from a tar- geting perspective.21 As long as a potential targeting variable increases in value as the likeli- hood of poverty increases (that is, it is "monotonically increasing with the risk of failure") (Stata Corporation 2001a), the area under an ROC curve can be used for ranking the accuracy of different targeting variables. The more a test's ROC curve is bowed toward the upper left-hand corner of the graph, the greater the accuracy of the test. Because ROC curves are bounded by the interval [0, 1], the maximum value for the area under an ROC curve is one (in which case, the test would predict poverty perfectly and the ROC curve 254 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 7.2. Receiver Operating Characteristic Curves for Selected Targeting Variables a. Radio/TV ownership and education Sensitivity 1.00 0.75 0.50 0.25 0 0 0.25 0.50 0.75 1.00 1-Specificity Radio and TV ownership Education level of head 45-degree line b. Land ownership and children Sensitivity 1.00 0.75 0.50 0.25 0 0 0.25 0.50 0.75 1.00 1-Specificity Land quintiles Number of children 45-degree line The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting 255 would coincide with the left-hand vertical and top horizontal axes). In con- trast, a test with no predictive power would correspond to an area of 0.5 under the ROC curve (which would itself coincide with the 45-degree line in the ROC diagram). Table 7.7 shows the area under the ROC curves for a number of possible poverty targeting variables for which information could be obtained relatively easily in a "quick and dirty" survey. It can be seen that the current system for classifying poor and remote communes does not perform particularly well in identifying poor people, especially for the overall poverty line. The simple reason for this is that the vast majority of poor people in Vietnam do not live in an officially desig- nated poor or remote commune. With the exception of educational level of the spouse, land allocated, and livestock owned in rural areas, table 7.7 shows that household-level targeting variables are generally much better at identifying poor individuals than whether they live in a poor or remote commune. The four categories of provincial poverty headcounts identified in the national poverty maps (map 7.1) also do quite well according to this criterion. Nonetheless, as shown by this and the ranking of poor communes according to their mean expenditures, there is considerable potential for im- proving the targeting of Vietnam's poor and remote communes programs. Table 7.7 also shows that the most effective poverty targeting variables are those related to housing quality and ownership of durable assets. Floor type is generally a better predictor of both food poverty and overall poverty than roof or toilet type.22 The level of education completed by household heads and their spouses performs considerably better as a targeting indica- tor in urban than in rural areas. Demographics, as proxied by the number of children younger than 15 years of age (the age by which Vietnamese chil- dren should have completed lower secondary school), are a better indicator of food poverty than overall poverty in both rural and urban areas. Ethnic- ity of the household head is a reasonable predictor of both food and overall poverty in rural areas, but it performs poorly in urban areas with few ethnic minority households. An unexpected finding is that a simple index of radio and television ownership is a better targeting indicator than all other asset, demographic, or educational variables. Indeed, table 7.7 confirms that the radio and tele- vision ownership index dominates all other targeting variables, with the ex- ception of communes ranked by the level of their median per capita expen- ditures. Using a cutoff point corresponding to ownership of neither a radio nor a television, the index is able to correctly classify some 76 percent of poor people in the VLSS sample.23 It would be possible to further increase the accuracy of targeting by com- bining a few of the above variables into a composite targeting indicator. Ini- tial work on developing such an indicator using stepwise regressions shows that sex variables (the number of children younger than 15 and number of females in the household, fuel type, and the ownership of a television and motorcycles), with the choice of an appropriate poverty cutoff point, allow up to 94 percent of poor and nonpoor households to be correctly identified 256 Economic Growth, Poverty, and Household Welfare in Vietnam Table 7.7. Accuracy of Different Variables in Targeting Poor Households Area under ROC curve Rural Urban All Vietnam Food Overall Food Overall Food Overall Targeting variable poverty poverty poverty poverty poverty poverty Poor or remote commune 0.586 0.589 0.557 0.503 0.591 0.560 Categories in national poverty map 0.641 0.622 0.620 0.645 0.663 0.650 Communes ranked by median expenditure 0.829 0.790 0.726 0.808 0.849 0.827 Land allocated (quintiles) 0.529 0.542 n/a n/a 0.619 0.646 Livestock owned (animal eq. units) 0.474 0.448 0.591 0.541 0.467 0.441 Educational level of household head 0.601 0.579 0.715 0.685 0.625 0.609 Educational level of spouse* 0.570 0.554 0.739 0.727 0.602 0.597 Number of children under 15 0.733 0.690 0.753 0.789 0.742 0.714 Number of females 0.636 0.618 0.578 0.671 0.632 0.616 Ethnicity 0.650 0.615 0.515 0.533 0.660 0.625 Floor type 0.696 0.665 0.694 0.773 0.734 0.720 Roof type 0.630 0.585 0.687 0.658 0.637 0.594 Toilet type 0.597 0.577 0.773 0.730 0.650 0.648 Radio and TV ownership 0.736 0.711 0.876 0.792 0.771 0.751 Notes on targeting variables: Poor commune: 0 = commune not included in CEMMA's list of remote communes or the MOLISA list of poor communes; 1 = commune included in either CEMMA's list of remote com- munes or the MOLISA list of poor communes; 2 = commune included in MOLISA's list of poor communes; 3 = both a remote and a poor commune. Categories in national poverty map: 0 = provincial poverty headcount <25 percent; 1 = headcount 25­45 percent; 3 = headcount 45­60 percent; 4 = headcount >60 percent. Livestock owned: number of livestock multiplied by their livestock equivalents units: 0.7 = cow, horses, and water buffalo; 0.1 = goats, pig, and deer; 0.01 = ducks and chickens. Educational level completed (both heads and spouses): 0 = postsecondary; 1 = advanced technical; 2 = upper secondary; 3 = lower secondary; 4 = lower secondary; 5 = primary; 6 = less than primary (* Note: 1,284 households do not include a spouse). Ethnicity: 0 = Chinese; 1 = Kinh; 2 = Khmer; 3 = Northern Uplands minorities; 4 = Central Highlands minorities. Floor type: 0 = earth; 1 = other; 2 = bamboo/wood; 3 = lime and ash; 4 = cement; 5 = brick; 6 = marble or tile. Roof type: 0 = other; 1 = leaves/straw; 2 = bamboo/wood; 3 = canvas/tar paper; 4 = pan- els; 5 = galvanized iron; 6 = tile; 7 = cement or concrete. Toilet type: 0 = flush; 1 = other; 2 = none. Radio and TV ownership: 0 = color TV; 1 = black and white TV; 2 = radio; 3 = none. The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting 257 in urban areas. Developing a composite targeting indicator is more difficult in rural areas, although the addition of three more variables (ethnicity, floor type, and ownership of radios) allows up to 75 percent of households to be correctly classified as poor or nonpoor. A further advantage of this method is that it allows the tradeoff between coverage of the poor and exclusion of the nonpoor to be quantified in terms that are readily understandable by policymakers.24 Summary and Conclusions Vietnam's current antipoverty programs rely heavily on the geographic targeting of poor households. Yet, as in many developing countries, the rel- atively small numbers of households that are sampled in its national house- hold surveys do not allow poverty statistics below the regional level to be estimated accurately. Meanwhile, questions have been raised about the com- parability and reliability of the more disaggregated province, district, and commune poverty statistics that are collected through Vietnam's adminis- trative reporting system. This chapter has shown how the data collected by the 1998 VLSS may be combined with that of the 1999 Population and Housing Census to bridge this gap and allow disaggregated maps of poverty to be constructed. The procedure to construct these maps involves two steps. First, the VLSS is used to explore the factors associated with poverty at the household level and develop linear regression models for predicting per capita expenditures at the rural or urban and strata levels. Second, these regression models are applied to household data from the 3 percent enu- meration sample of the census to derive and map provincial-level estimates of the percentage of people living in households whose per capita expendi- tures fall below the GSO­World Bank poverty line (the poverty headcount). The national poverty map resulting from this two-step procedure shows that poverty is concentrated in Vietnam's Northern Uplands, in particular in the six provinces that border China and the Lao People's Democratic Re- public. Fourteen other provinces, most of which are located in the Northern Uplands, Central Highlands, and North Central Coast, have poverty head- counts above 45 percent. When rural areas are considered separately from urban areas, rural poverty is also found to be high in most of the remaining provinces of the Northern Uplands, together with Gia Lai and Kon Tum and the Central Highlands. A group of moderately poor rural provinces (with rural headcounts between 45 and 50 percent) can also be seen clustered in the North Central Coast and Red River Delta. However, even relatively prosperous regions have their own pockets of poverty, such as Ha Tay in the Red River Delta and Ninh Thuan in the Southeast. To consider the effectiveness of Vietnam's existing geographically tar- geted antipoverty programs, we applied the relatively novel technique of ROC curves to the VLSS data. The results confirm that a consistent ranking of communes has high potential to identify Vietnam's poor population. How- ever, the existing, officially designated list of poor and remote communes is 258 Economic Growth, Poverty, and Household Welfare in Vietnam less effective in targeting the poor, because it excludes a large number of poor people living in other areas. Among the additional household-level variables that might be used to help sharpen the focus of targeting, demographics (in particular, the number of children younger than 15 in a household), housing characteristics (especially floor type), and ownership of durable assets per- form well. A simple index of radio and television ownership dominates all other individual targeting variables, with the exception of communes ranked by their median per capita expenditures. Combining several household-level variables into a composite targeting indicator offers the potential to further improve the targeting of the poor, especially in urban areas. When household-level data from the full sample of the 1999 census be- come available, it should be possible to extend this poverty mapping to the district level. Because the determinants of expenditures and poverty are likely to remain relatively stable over time, the authors believe that this will be a useful exercise, even though the 1999 census and VLSS are now three to four years old. In addition, although censuses are conducted only every 10 years, the first step of the poverty mapping calculations (the expenditure regressions) can be reestimated and new poverty maps derived each time a nationally representative household sample survey is conducted. The com- plete provincial poverty map could also be redone every five years with in- formation from the interdecadal censuses.25 Furthermore, international experience (Baker and Grosh 1994; Bigman and Fofack 2000) indicates that greater geographic disaggregation is likely to improve the targeting of Vietnam's antipoverty programs. With more computational effort, it is also feasible to estimate poverty headcounts (and other poverty and inequality measures) at the commune and ward levels, although the confidence inter- vals around some of these estimates will be large. More regionally specific analysis of the use and combination of additional household-level targeting variables, such as housing characteristics and asset ownership, would also be useful at this time. Nonetheless, it is hoped that this chapter has demon- strated the feasibility and policy relevance of these tools to targeting an- tipoverty interventions in Vietnam. page) 10.732 19.000 1.000 0.750 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 Maximum following on eas ar 6.526 1.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Minimum continues Urban (table dev Std 0.602 2.196 0.191 0.201 0.177 0.099 0.433 0.406 0.437 0.280 0.318 0.281 0.405 0.413 0.369 0.408 0.229 17 14 1 Mean 8.293 5.221 0.1 0.244 0.526 0.010 0.249 0.208 0.256 0.086 0.1 0.086 0.207 0.218 0.163 0.21 0.056 Analysis 10.148 16.000 1.000 0.833 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 Maximum eas Regression ar 5.879 1.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Minimum in Rural dev Used Std 0.478 1.904 0.187 0.214 0.173 0.384 0.487 0.425 0.448 0.198 0.214 0.102 0.344 0.493 0.384 0.403 0.173 ariablesV Mean 7.56 5.55 0.10 0.35 0.51 0.18 0.39 0.24 0.28 0.04 0.05 0.01 0.14 0.42 0.18 0.20 0.03 for school school e (fraction) (fraction) minority primary primary education Statistics primary lower upper years primary lower upper advanced spouse a expenditur (members) years (fraction) ethnic 15 60 is ee have completed Description completed school school school school capita over under female head degr not not completed completed completed not (omitted) completed completed completed completed postsecondary (omitted) Descriptive per household has has has has has has has has has has does of of 7A oportion oportion oportion school secondary secondary technical school secondary secondary Log Size Pr Pr Pr Household Head Head Head Head Head Head Head Spouse Spouse Spouse Spouse Appendix ariableV lnrpce hhsize pelderly pchild pfemale ethnic Iedchd_1 Iedchd_2 Iedchd_3 Iedchd_4 Iedchd_5 Iedchd_6 Iedcsp_0 Iedcsp_1 Iedcsp_2 Iedcsp_3 Iedcsp_4 259 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 5.835 4.973 1.000 Maximum eas ar 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Minimum Urban dev Std 0.287 0.230 0.176 0.300 0.441 0.356 0.392 0.245 0.401 0.480 0.500 0.346 1.914 1.865 0.133 Mean 0.090 0.056 0.032 0.100 0.264 0.149 0.190 0.064 0.201 0.361 0.500 0.139 1.417 1.832 0.982 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 5.537 5.293 1.000 Maximum eas ar 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Minimum Rural dev Std 0.163 0.073 0.126 0.163 0.212 0.458 0.259 0.241 0.261 0.283 0.486 0.456 1.108 1.876 0.456 Mean 0.03 0.01 0.02 0.03 0.05 0.70 0.07 0.06 0.07 0.09 0.62 0.29 0.34 2.35 0.71 , manager education materials materials or worker estry ea) ea) ar ar advanced or for e, worker (omitted) ) leader service ee worker or permanent semipermanent temporary Description log(house log(house completed degr postsecondary ofessional of of of worker working political pr clerk agricultur unskilled of of skilled electricity continued( has has a a a in a an not is is is is fishing is is is made made made Ihouse_1 Ihouse_2 has 7A technical technical or materials (omitted) and and Spouse Spouse Head Head Head Head Head Head Head House House House Interaction Interaction House Appendix ariableV Iedcsp_5 Iedcsp_6 Ioccup_1 Ioccup_2 Ioccup_3 Ioccup_4 Ioccup_5 Ioccup_6 Ioccup_7 Ihouse_1 Ihouse_2 Ihouse_3 htypla1 htypla2 electric 260 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 0.000 1.000 1.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.494 0.465 0.307 0.487 0.437 0.333 0.382 0.490 0.290 0.417 0.225 0.355 0.000 0.459 0.385 0.578 0.316 0.106 0.615 0.257 0.127 0.822 0.599 0.092 0.224 0.053 0.148 0.000 0.301 0.181 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.136 0.467 0.459 0.188 0.439 0.416 0.500 0.499 0.403 0.391 0.369 0.294 0.212 0.267 0.418 weights. 0.02 0.68 0.30 0.04 0.74 0.22 0.51 0.47 0.20 0.19 0.16 0.10 0.05 0.08 0.22 sampling using Coast Coast Delta nor Delta public Highlands River calculated a toilet Central e Central River ar omfr lake flush water or Red North South Central Southeast Mekong toilet television radio Northern the the the the the the deviations water tap well river flush latrine neither has has in (omitted) in in in in in in d (omitted) (omitted) uses uses uses has has has private standar or water latrine Uplands and VLSS. House House House House House House Household Household Household Household Household Household Household Household Household 1998 Means ce: Note: Sour Inwate_1 Inwate_2 Inwate_3 Itoile_1 Itoile_2 Itoile_3 tv radio eg7_1r eg7_2r eg7_3r eg7_4r eg7_5r eg7_6r eg7_7r 261 t River 12.8***- 0.3 5.3***- 3.3***- 0.3- 0.3- 1.4 1.8* 1.0 2.2** 0.2- 0.8 0.7- 1.6- 0.1- Delta Mekong 830 0.508 Coefficient 0.0876- 0.0208 0.3240- 0.2205- 0.0192- 0.0072- 0.0512 0.1093 0.1614 0.3651 0.0046- 0.0299 0.0432- 0.0921- 0.0366- t 4.0***- 0.5- 1.3- 0.6 1.0- 1.8* 1.0 2.6** 3.2*** 0.1 0.4 1.2 1.5 2.8** 0.1- Southeast 514 0.482 Coefficient 0.0530- 0.1000- 0.1399- 0.0521 0.1268- 0.0801 0.0769 0.2199 0.2753 0.0057 0.0154 0.0839 0.1084 0.3571 0.0083- t 4.9***- 0.2 2.2**- 1.4- 2.3**- 0.2 1.7 0.7 1.0 0.3- 0.3 1.1- 0.5 1.3- 1.3 Highlands 368 0.671 Central Coefficient 0.0587- 0.0414 0.2424- 0.1041- 0.2360- 0.0092 0.1235 0.0989 0.1929 0.0982- 0.0177 0.0782- 0.0401 0.0851- 0.1595 t 5.4***- 0.0 2.7**- 1.4- 5.7***- 2.0* 1.7 0.4 0.9 1.7 0.1- 0.7 1.8* 2.3** 0.3- Central Coast South 502 0.712 Coefficient 0.0697- 0.0006- 0.2247- 0.1449- 0.4229- 0.0925 0.1045 0.0397 0.1071 0.4427 0.0034- 0.0310 0.1720 0.3033 0.0565- Stratum t 12.2***- 2.4**- 3.3***- 2.1*- 1.5- 0.4 0.5 0.3- 1.6 2.1* 0.2- 0.4- 0.9- 0.7- 0.1 Each Central Coast of 16 North 600 0.451 Coefficient 0.0758- 0.1491- 0.3163- 0.1993- 0.0940- 0.0152 0.0206 0.0173- 0.0932 0.1436 0.0123- 0.0170- 0.0508- 0.0648- 0.01 t 9.7***- 2.0*- 6.6***- 0.7- 1.0- 2.0* 3.3*** 3.0*** 3.6*** 4.2*** 0.2 0.4- 0.3 0.4 1.3 River Delta Red Expenditure 783 0.414 Coefficient 0.0996- 0.1435- 0.4184- 0.0559- 0.0471- 0.0972 0.1619 0.1628 0.1898 0.4954 0.0083 0.0149- 0.0090 0.0203 0.0948 t Capita 7.9***- 1.3- 3.7***- 1.4- 0.4 0.8 1.2 2.3** 1.6 0.9 0.5 1.1- 0.6 0.0 3.1*** Per Northern Uplands 178 101 672 0.539 of Coefficient 0.0835- 0.1- 0.3242- 0.1- 0.0220 0.0394 0.0495 0.1299 0.0837 0.1313 0.0232 0.0397- 0.0219 0.0029 0.1641 t 6.7***- 2.4**- 2.9***- 0.4 0.7 1.1 0.8- 2.7** 1.2 3.1*** 0.2- 2.0* 1.1 1.0 0.2 urban eas ar 11 Other 1,1 0.486 Coefficient 0.0806- 0.1849- 0.2641- 0.0387 0.0629 0.0454 0.0265- 0.1437 0.0725 0.1766 0.0087- 0.0764 0.0508 0.0838 0.0091 Determinants Chi t Ho City 4.4***- 0.4- 0.1- 0.6 1.5- 2.1* 2.7** 0.9 2.0* 2.5** 1.2 1.3 2.0* 3.2*** 2.1* 7B & 19 Minh 198 619 Hanoi 0.4330 Coefficient 0.0688- 0.0408- 0.01- 0.0877 0.2614- 0.1 0.1504 0.0864 0.1358 0.2101 0.0876 0.0996 0.1423 0.4751 0.1802 ed N Appendix R-squar ariableV hhsize pelderly pchild pfemale ethnic Iedchd_2 Iedchd_3 Iedchd_4 Iedchd_5 Iedchd_6 Iedcsp_0 Iedcsp_2 Iedcsp_3 Iedcsp_4 Iedcsp_5 262 0.6- 0.9 2.4** 1.5 0.5 0.7 1.5- 2.2**- 1.1- 2.6** 1.7 2.4** 3.2*** .8**2 4.9*** 2.0* 6.8*** 5.2*** 105.8*** 711 0.1293- 0.0727 0.2130 0.0917 0.0197 0.0579 0.0832- 1.6038- 0.1717- 0.4052 0.0628 0.0903 0.1542 0.1 0.3758 0.0567 0.1512 0.1492 7.9655 1.2- 1.9* 0.5 1.3 0.8- 0.7- 2.6**- 2.3**- 3.5***- 2.4** 4.0*** 4.4*** 0.1 2.5** 2.7** 1.3 4.5*** 1.3 45.2*** 0.1791- 0.3054 0.0619 0.1426 0.0829- 0.0818- 0.2348- 2.7300- 0.7977- 0.6983 0.2531 0.1725 0.0189 0.0649 0.1856 0.0824 0.2094 0.0537 7.7554 0.1 0.2- 0.7- 0.7- 1.2- 1.7- 0.7 0.8- 0.4 3.2*** 1.4 1.0- 4.4*** 4.1*** 0.1- 2.6** 3.1** 31.5*** . 120 15 511 level 0.0083 0.0281- 0.0879- 0.0588- 0.1286- 0.2498- 0.4357 0.1755- 0.0545 0.1604 0.1557 0.1713- 0.1 0.41 0.0057- 0.1 0.1415 7.4845 cent per 0.3- 5.1*** 1.9* 1.1 0.5 0.0 0.6 0.6- 2.4**- 1.1 2.8** 1.5 2.9** 0.6 0.8 1.9* 4.3*** 2.2** 69.5*** 1 the 103 0.0609- 0.2959 0.1778 0.1 0.0289 0.0028 0.1239 0.4968- 0.5064- 0.2399 0.1687 0.0899 0.2541 0.0254 0.0556 0.0741 0.1917 0.0809 7.6878 ***at and 1.8* 0.7 0.5 0.5 1.2- 0.9 0.7- 1.4- 1.3- 1.6 1.7* 0.0 5.3***- 2.8**- 1.3 1.3 5.9*** 3.9*** 44.2*** level, 126 cent 0.3326 0.0952 0.0502 0.0559 0.0687- 0.1033 0.0448- 1.4392- 0.2913- 0.4639 0.1 0.0015 0.4072- 0.1565- 0.2321 0.0444 0.2439 0.1533 8.0240 per 5 the 2.8** 1.7 1.9* 3.5*** 0.8 1.2 0.2 7.1***- 0.6- 8.1*** 1.6 3.3*** 0.2 2.3** 3.4*** 1.0 8.1*** 3.5*** 45.1*** **at e. level, 0.1520 0.1464 0.1393 0.2760 0.0436 0.0892 0.0204 1.1440- 0.0902- 0.3552 0.0669 0.1918 0.0200 0.1741 0.3322 0.0699 0.1907 0.0913 7.3747 cent per 1.6 1.2 1.5 0.4 0.8- 2.4** 4.7*** 0.2- 1.6- 0.9 2.2** 0.3 1.9* 6.5*** 1.7 10.4*** 0.9 75.2*** expenditur 10 the 188 0.1 0.1595 0.1408 0.0498 0.0591- 0.1772 0.3607 0.0977- 0.3355- 0.0918 0.1233 0.0217 0.0959 0.4844 0.0681 0.2624 0.0313 7.6097 capita at per of VLSS. 0.5 2.7** 1.9* 1.1 0.0 1.5 2.3**- 3.5***- 3.7***- 4.6*** 4.6*** 0.0 3.5*** 0.1 1.7* 0.3- 4.0*** 4.6*** 82.1*** log significant is is 1998 138 of 0.0353 0.2371 0.1284 0.0466 0.0012 0.0736 0.1292- 0.9722- 0.5709- 0.3095 0.1826 0.0019- 0.1782 0.0030 0.1 0.0152- 0.2056 0.1573 8.0018 ficient variable coef analysis 3.0*** 1.5 0.5- 0.3 2.8**- 0.9- 1.6- 4.3***- 3.6***- 4.7*** 3.6*** 3.6*** 1.5 0.1 3.3*** 0.8 3.6*** 5.5*** 29.8*** the ession dependent that 0.2505 0.1849 0.0377- 0.0192 0.1906- 0.0614- 0.1697- 0.8704- 0.7219- 0.2274 0.1850 0.6201 0.1200 0.0073 0.2932 0.1079 0.2363 0.2558 7.3886 Regr The ce: Note: *Indicates Sour Iedcsp_6 Ioccup_1 Ioccup_2 Ioccup_3 Ioccup_4 Ioccup_5 Ioccup_6 Ihouse_1 Ihouse_2 htypla1 htypla2 electric Inwate_1 Inwate_2 Itoile_1 Itoile_2 tv radio _cons 263 12** obability 0.0557* 0.01 0.0008*** 0.0004*** 0.1340 0.0090*** 0.0108** 0.2364 0.0054*** 0.0007*** 0.0006*** 0.0252** 0.3501 0.0275** 0.0009*** 0.2986 0.0743* 0.0000*** 0.0010*** 0.1306 0.0033*** 0.0000*** 0.0025*** 0.0074*** Model Pr Regression 2.65 3.84 6.45 statistic 12.29 2.24 6.09 3.52 1.41 3.74 8.88 9.24 4.08 1.19 3.05 6.13 1.28 3.55 21.33 5.99 1.85 4.54 25.39 7.78 6.06 F Stratum-Level in df2 91 19 91 91 91 91 63 36 63 63 63 63 02 20 02 02 02 02 42 24 42 42 42 42 ariablesV df1 5 6 6 2 2 2 5 6 6 2 2 2 5 6 6 2 2 2 5 6 6 2 2 2 Explanatory of household household household household Groups of of of of facility facility facility facility of head water head water head water head water head spouse of of head spouse of of head spouse of of head spouse of of of of ce of of ce of of ce of of ce housing sanitary housing sanitary housing sanitary housing sanitary of sour of of sour of of sour of of sour of ariablesV Education Education Occupation ypeT Main ypeT Education Education Occupation ypeT Main ypeT Education Education Occupation ypeT Main ypeT Education Education Occupation ypeT Main ypeT Significance of estsT 7C City eas ar River Minh urban Northern Red Chi Appendix Stratum Hanoi and Ho Other Rural Uplands Rural Delta 264 1 0.1071 0.5103 0.0219** 0.181 0.0001*** 0.2577 0.1882 0.1909 0.0014*** 0.0572* 0.0278** 0.1855 0.1031 0.0040*** 0.3623 0.5310 0.0026*** 0.0001*** 0.0302** 0.1848 0.0034*** 0.0007*** 0.0746* 0.0514* 0.1208 0.3374 0.0001*** 0.0767* 0.0058*** 0.0001*** 2.14 0.91 3.33 1.88 15.26 1.46 1.73 1.69 6.66 3.48 4.59 1.89 2.42 6.79 1.23 0.67 10.68 21.98 3.32 1.7 5.35 1.811 3.07 3.59 1.95 1.20 7.59 2.85 6.37 12.80 81 18 81 81 81 81 51 15 51 51 51 51 11 11 11 11 11 11 61 16 61 61 61 61 52 25 52 52 52 52 5 6 6 2 2 2 5 6 6 2 2 2 5 6 6 2 2 2 5 6 6 2 2 2 5 6 6 2 2 2 level. cent per 1 the ***at and household household household household household level, VLSS. of of of of of cent facility facility facility facility facility e. head water head water head water head water head water 1998 per head spouse of of head spouse of of head spouse of of head spouse of of head spouse of of 5 of of ce of of ce of of ce of of ce of of ce the using e housing sanitary housing sanitary housing sanitary housing sanitary housing sanitary expenditur **at of sour of of sour of of sour of of sour of of sour of Education Education Occupation ypeT Main ypeT Education Education Occupation ypeT Main ypeT Education Education Occupation ypeT Main ypeT Education Education Occupation ypeT Main ypeT Education Education Occupation ypeT Main ypeT capita level, expenditur per cent of per capita log 10 is per the of at variable analysis Central significant Central is ession dependent Regr The ficient Northern South Central ce: Note: *Coef Sour Rural Coast Rural Coast Rural Highlands Rural Southeast Rural Mekong River Delta 265 otalT 0.033 0.039 0.037 0.036 0.035 0.034 0.039 0.042 0.045 0.038 0.043 0.046 0.055 0.055 0.035 0.027 0.043 0.050 0.046 0.043 0.039 0.028 0.033 0.055 0.036 ors err Urban 0.022 0.020 0.016 0.020 0.018 0.020 0.022 0.019 0.016 0.017 0.029 0.025 0.018 0.021 0.028 0.021 0.017 0.015 0.014 0.018 0.017 0.021 0.017 0.018 0.023 Standard Rural 0.037 0.042 0.042 0.040 0.042 0.041 0.045 0.049 0.051 0.047 0.062 0.060 0.060 0.062 0.045 0.034 0.048 0.059 0.058 0.047 0.043 0.038 0.036 0.061 0.041 1 otalT 0.765 0.709 0.699 0.664 0.653 0.61 0.597 0.585 0.581 0.554 0.533 0.532 0.530 0.493 0.480 0.470 0.469 0.469 0.450 0.440 0.435 0.431 0.430 0.429 0.427 Regression 15 14 12 12 11 17 headcount Urban 0.150 0.130 0.103 0.094 0.140 0.090 0.121 0.120 0.1 0.1 0.217 0.193 0.121 0.140 0.208 0.160 0.1 0.092 0.092 0.122 0.1 0.150 0.1 0.1 0.155 Stratum-Level Poverty with Rural 0.853 0.763 0.785 0.732 0.760 0.728 0.673 0.659 0.638 0.661 0.689 0.642 0.561 0.533 0.561 0.562 0.509 0.532 0.542 0.471 0.467 0.536 0.456 0.460 0.460 Estimated Region NU NU NU NU NU NU NU NU NU NU HC CH NU NU SE NCC NCC NU NU NCC NCC NCC RRD NU SCC Headcounts code Poverty Son Quang irT Ngai Phuc Thuan An Hoa Thien-Hue Chau Giang La Bang Cai Kan Binh Bai muT Lai Giang Tho Nguyen Ninh 7D ovince inhT yaT Pr Lai Ha Son Cao Lao Lang Bac Hoa uyenT enY Kon Gia Bac inhV Ninh Quang Nghe Phu Thai Ha Thanh Thua Ha Bac Quang Appendix 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 266 0.044 0.031 0.031 0.033 0.031 0.034 0.033 0.049 0.034 0.030 0.028 0.032 0.030 0.034 0.032 0.035 0.028 0.031 0.029 0.031 0.032 0.025 0.030 0.033 0.022 0.032 0.025 0.033 0.035 page) following on 0.017 0.015 0.025 0.018 0.022 0.030 0.026 0.025 0.023 0.025 0.027 0.016 0.030 0.027 0.025 0.015 0.025 0.015 0.023 0.016 0.020 0.025 0.024 0.021 0.015 0.021 0.019 0.018 0.024 continues (table 0.049 0.036 0.038 0.035 0.033 0.041 0.038 0.061 0.039 0.037 0.036 0.036 0.038 0.042 0.038 0.038 0.036 0.054 0.035 0.036 0.051 0.032 0.035 0.036 0.032 0.038 0.038 0.037 0.040 0.425 0.408 0.406 0.403 0.401 0.392 0.389 0.387 0.386 0.377 0.374 0.373 0.369 0.368 0.355 0.353 0.350 0.348 0.347 0.343 0.338 0.336 0.332 0.323 0.297 0.296 0.286 0.263 0.235 0.105 0.095 0.208 0.120 0.142 0.250 0.205 0.184 0.200 0.220 0.221 0.102 0.232 0.192 0.190 0.073 0.171 0.107 0.167 0.096 0.153 0.183 0.162 0.152 0.073 0.166 0.130 0.123 0.177 1 0.462 0.453 0.452 0.421 0.424 0.423 0.419 0.439 0.414 0.421 0.422 0.41 0.412 0.410 0.382 0.370 0.399 0.540 0.387 0.382 0.454 0.385 0.360 0.339 0.412 0.321 0.375 0.283 0.245 NCC RRD MRD RRD RDR MRD MRD CH MRD MRD MRD RRD SE CCS SCC RRD MRD NU MRD RRD SE SCC MRD MRD RRD MRD SCC MRD SE Binh Binh neY Giang Nam rangT Thap An Lac inhV Giang Dinh Long Lieu Thuan neY Nam Ninh Binh Tho Mau Duong Dong Dinh erT Hoa Phong Giang Phuoc Quang Ninh An Ha Hung Soc Dong Dac raT Kien Bac Nam Binh Phu Quang Thai Can Quang Ca Hai Lam Binh inhV Ben Hai Long Khanh ienT Binh 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 267 1 1 otalT 0.018 0.017 0.014 0.017 0.013 0.013 0.01 0.01 below es Highlands, ors err expenditur Central = Urban 0.025 0.019 0.005 0.020 0.020 0.019 0.012 0.002 capita CH Standard per Coast, with Rural 0.021 0.033 0.032 0.022 0.017 0.016 0.022 0.010 Central South households = in 18 SCC otalT lives 0.159 0.151 0.149 0.145 0.129 0.1 0.048 0.365 ayT Coast, Ha Census. in Central 15 10 headcount Housing Urban 0.177 0.1 0.014 0.122 0.139 0.123 0.038 0.1 North population = and the Poverty of NCC cent 16 Delta, Population per Rural 0.156 0.290 0.331 0.155 0.122 0.1 0.096 0.441 43 River 1999 of that Red = sample implies RRD SE SCC RDR SE ES SE SE ayT cent Region Delta. per Ha 3 Uplands, for River and ) 0.430 uaT of Northern VLSS Mekong line. = Minh = 1998 continued( code ung Nai Chi NUe MRD omfr Ninh Duong headcount poverty ar 7D ovince Nang Noi Ria-V Ho and Pr ayT Da Ha Dong Ba Binh TP otalT codes poverty A GSO/WB gioner Estimated ce: Southeast, Note: 1998 The = Sour Appendix 55 56 57 58 59 60 61 the SE 268 The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting 269 Notes The authors thank Phan Xuan Cam and Nguyen Van Minh for their help in under- standing the Vietnam census data and Peter Lanjouw for helpful methodological dis- cussions. Paul Glewwe, participants at workshops in Hanoi, and two internal World Bank reviewers provided valuable comments on earlier versions of this chapter. The financial assistance of the Department for International Development (United King- dom) Poverty Analysis and Policy Support Trust Fund and World Bank Develop- ment Economics Research Group is acknowledged. 1. The 1992­93 survey spanned a full year, starting in October 1992; the 1997­98 survey began in December 1997 and also lasted a year. For brevity's sake, reference is made to the surveys as the 1993 VLSS and 1998 VLSS, respectively. 2. The poverty headcount is defined as the proportion of the population with per capita expenditures below the poverty line. 3. Minot and Baulch (2002) show that using aggregated census data under- estimates the incidence of poverty when it is below 50 percent and overestimates it when it is above 50 percent. The absolute size of the error, however, can be as low as 2 to 3 percentage points in some circumstances. 4. In 1998, the food poverty line was D (dong) 1,286,833 and the overall poverty line was D 1,789,871 per person a year. See annex 2 of Poverty Working Group (1999) for further details concerning the estimation of these poverty lines. 5. Between 1997 and 2000, MOLISA's area-specific poverty lines were D 45,000 per capita per month for rural mountainous regions and islands, D 70,000 for rural midland and delta regions, and D 90,000 for urban areas. Note that these poverty lines were developed for monthly rice equivalent income and cannot be applied to the income data collected by the VLSS. 6. This is accomplished with the "svymean" command. Stata calculates a linear approximation (a first-order Taylor expansion) of the sampling error variance based on information on the strata, the primary sampling unit, and the weighting factors. See Stata Corporation (2001b) for more information. 7. A coefficient of ­0.772 implies that a one-unit increase in the explanatory variable is associated with a 7.4 percent reduction in per capita expenditure, because (exp­0.772) = 0.926 = 1­7.4 percent. However, for two reasons, care must be taken before inferring that larger households are worse off than smaller ones. First, there may be economies of scale in household size, so that larger households do not need the same per capita expenditures as smaller households to reach an equivalent level of welfare. Second, the measure of welfare used here does not take into account household composition, so if larger households have more children than smaller households, they might still have equivalent levels of expenditure per adult equiva- lent. 8. In common with other studies of ethnic minority issues using the VLSS, the Hoa (Chinese) are grouped with the Kinh (ethnic Vietnamese) households. 9. This study also experimented with using the number of years of education for the head of household and spouse as explanatory variables, but found that the level of education completed gave better results. 10. Education of the spouse may have other benefits, such as improved health or nutrition, that are not captured by the measure of welfare used in this analysis, per capita expenditure. Note that 11.4 percent of spouses in the VLSS are male. 11. Although information on the employer of household heads is available in both the census and the VLSS, the categories they use to describe different categories 270 Economic Growth, Poverty, and Household Welfare in Vietnam of employers differ substantially and cannot be reconciled. For this reason, a set of dummies for employer of the household head was not included in the predictive regressions. 12. Because the permanent housing dummy enters both as a separate variable (Ihouse_1) and in the interaction term htypla1 (=Ihouse_1 × ln(area)), the marginal ef- fect is calculated as Ihouse _1 + htypla1 × ln (area). The marginal effect is evaluated at the mean values of ln(area), which are 3.72 in rural areas and 3.66 in urban areas. 13. The census did not collect information on the area of houses made of tem- porary materials, so housing area cannot be used to help predict expenditures for these households. 14. More specifically, this variable refers to whether the household said that electricity was the main source of lighting for the house. 15. There are three factors that complicate using the VLSS for estimating provincial poverty. First, three provinces are not included in the VLSS sample. Sec- ond, in the remaining provinces, the sample size is small: most provinces have fewer than 100 households, and some have as few as 32. Third, the sample (and hence the sampling weights) is not designed to produce precise estimates at the provincial level. For example, the proportion of urban households in each province is not accu- rate, even after applying sampling weights. 16. A 6 percentage point gap refers to a gap greater than 5.5 percent and less than or equal to 6.5 percent. 17. Because the geographic areas in which Programs 133 and 135 operate are reasonably distinct, we have combined them into one list of "poor and remote com- munes" for our analysis of the potential of targeting. This list was then matched to commune information in the VLSS to identify households living in areas identified as poor by MOLISA or the Committee for Ethnic Minorities in Mountainous Areas. 18. Note that this is the poverty estimate from applying MOLISA's poverty lines (see endnote 5) to household rice equivalent income, and it is not directly com- parable with the estimates of the poverty headcount in this chapter. 19. See Conway (2001) for a careful account of the often highly variable way in which this exercise is implemented at the commune level. 20. ROC curves can be linked to the occurrence of Type I and Type II errors fa- miliar from conventional statistical hypothesis testing (known as "false positives" and "false negatives" in epidemiology and medicine and as F and E errors in the tar- geting literature). Sensitivity is one minus the probability of a Type I error (incor- rectly classifying a poor household as nonpoor), and one minus the specificity of a test is the same as the probability of a Type II error (incorrectly classifying a nonpoor household as poor). In many respects, this is akin to describing whether a glass is half empty or half full, in that both are simply different methods of presenting the same data. 21. This is rather similar to the problems encountered in making unambiguous comparisons of inequality when the Lorenz curves cross or in making comparisons of inequality when cumulative income distribution curves cross. 22. Ownership of the dwelling in which a household lives was considered for inclusion in the list of asset-based targeting variables, but it was found to perform poorly because the vast majority of households in the 1998 VLSS sample (5,703 out of 5,999) own their own dwellings. 23. It may seem surprising that in a country with Vietnam's level of per capita income, radio and television ownership has such potential for targeting the poor. Radio and television ownership is, however, quite widespread throughout The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting 271 Vietnam--with 53 percent of households owning a television and 45 percent of households owning a radio, according to the 1999 Population and Housing Census. Many of the televisions owned, especially in rural areas, are relatively inexpensive 14-inch, battery-operated televisions produced in China. 24. See Baulch (2002) for further details. 25. The need for updated census data is greatest if changes in poverty are prin- cipally associated with changes in household characteristics. The need for new household survey data is greatest if poverty changes are linked to changes in the co- efficients of the expenditure regressions. Further research is needed into the relative importance of these two factors. Bibliography The word processed describes informally reproduced works that may not be commonly available through libraries. Baker, J., and M. Grosh. 1994. "Poverty Reduction through Geographic Targeting: How Well Does It Work?" World Development 22(7): 983­95. Baulch, B. 2002. "Poverty Monitoring and Targeting Using ROC Curves: Ex- amples from Vietnam." Working Paper 161. Institute of Development Studies, University of Sussex, Brighton, U.K. Baulch, B., T. K. C. Truong, D. Haughton, and J. Haughton. 2004. "Ethnic Minority Development in Vietnam: A Socioeconomic Perspective." In P. Glewwe, N. Agrawal, and D. Dollar, eds., Economic Growth, Poverty, and Household Welfare in Vietnam. Washington, D.C.: World Bank. Bigman, D., and H. Fofack. 2000. Geographic Targeting for Poverty Alleviation: Methodology and Applications. Washington, D.C.: World Bank. Conway, T. 2001. "Using Government Data to Target Activities to Poor Com- munes and Monitor Poverty Reduction: A Review of Options for the Cao Ban-Bac Kan Rural Development Project." Commission of the European Communities, Hanoi. Processed. Cornia, G., and F. Stewart. 1995. "Two Errors of Targeting." In D. van de Walle and K. Nead, eds., Public Spending and the Poor. Baltimore and London: Johns Hopkins University Press. Elbers, C., J. Lanjouw, and P. Lanjouw. 2003. "Micro-Level Estimation of Poverty and Inequality." Econometrica 71(1): 355­64. Hentschel, J., J. Lanjouw, P. Lanjouw, and J. Poggi. 2000. "Combining Census and Survey Data to Trace the Spatial Dimensions of Poverty: A Case Study of Ecuador." World Bank Economic Review 14(1): 147­65. Minot, N. 1998. "Generating Disaggregated Poverty Maps: An Application to Viet Nam." Discussion Paper 25. Markets and Structural Studies Division, International Food Policy Research Institute, Washington, D.C. 272 Economic Growth, Poverty, and Household Welfare in Vietnam _____. 2000. "Generating Disaggregated Poverty Maps: An Application to Vietnam, 2000." World Development 28(2): 319­31. Minot, N., and B. Baulch. 2002. "Poverty Mapping with Aggregate Census Data: What Is the Loss in Precision?" Discussion Paper 49. Markets and Structural Studies Division, International Food Policy Research Institute, Washington, D.C. Poverty Working Group. 1999. Vietnam: Attacking Poverty. A Joint Report of the Government of Vietnam­Donor­NGO Poverty Working Group, presented to the Consultative Group Meeting for Vietnam. Hanoi: The World Bank. Ravallion, M. 1992. "Poverty Comparisons." Living Standard Measurement Working Paper 88. World Bank, Washington, D.C. Ravallion, M., and Q. Wodon. 1997. "Poor Areas, or Only Poor People?" Policy Research Working Paper 1798. World Bank, Policy Research Department, Washington, D.C. Socialist Republic of Vietnam. 2001. "Comprehensive Poverty Reduction and Growth Strategy." Ministry of Planning and Investment, Hanoi. Stata Corporation. 2001a. "Receiver Operating Characteristics (ROC) Analy- sis." Stata 7 Reference Manual 3: 131­51. College Station, Tex.: Stata Press. _____. 2001b. "Svymean." Stata 7 Reference Manual 4: 52­74. College Station, Tex.: Stata Press. Statistics South Africa and the World Bank. 2000. "Is Census Income an Adequate Measure of Household Welfare? Combining Census and Survey Data to Construct a Poverty Map of South Africa." Pretoria. Processed. van de Walle, D., and D. Gunewardana. 2001. "Sources of Ethnic Inequality in Vietnam." Journal of Development Economics 65(1): 177­207. Viet Nam Living Standards Survey (VNLSS). 1992­1993. Web site: www. worldbank.org/lsms/country/vn93/vn93bid.pdf. Vietnam Living Standards Survey (VLSS). 1997­98. Web site: www. worldbank.org/lsms/country/vn98/vn98bif.pdf. Wodon, Q. 1997. "Targeting the Poor Using ROC Curves." World Develop- ment 25(12): 2083­92. World Bank. 2000. Panama Poverty Assessment: Priorities and Strategies for Poverty Reduction. Washington, D.C.: World Bank. 8 Ethnic Minority Development in Vietnam: A Socioeconomic Perspective Bob Baulch, Truong Thi Kim Chuyen, Dominique Haughton, and Jonathan Haughton Vietnam is an ethnically diverse society. The Kinh (ethnic Vietnamese) majority, which accounts for 84 percent of the population, coexists with 53 smaller ethnic minority groups, some of which have fewer than 1,000 members (Dang, Son, and Hung 2000). Previous research using the Vietnam Living Standards Surveys (VLSSs), in which the Kinh are usually grouped together with the Hoa (Chinese), has shown that the remaining 52 ethnic minorities constitute the poorest, least educated sections of Vietnamese society (Vietnam Poverty Working Group 1999).1 Furthermore, the gap in living standards between the Kinh and Hoa majority and the other ethnic minorities grew between 1992­93 and 1997­98 (the years when the closely comparable VLSSs were undertaken).2 Geography, in particular the fact that many ethnic minorities live in remote and mountainous areas, explains only a part of the difference in living standards between these two groups. There are systematic differences in endowments and the returns to those endowments for members of the Kinh-Hoa majority and the ethnic minorities, most of which are in favor of the majority group (van de Walle and Gunewardena 2001). These and other more detailed qualitative studies (see in particular Huy and Dai [1999]; Jamieson, Cuc, and Rambo [1998]; and Winrock International [1996]) have led to an emerging consensus among donors and nongovernmental organizations (NGOs) that a new, more differentiated approach to ethnic minority policy is required in Vietnam. This chapter seeks to contribute to this debate by examining and decom- posing the latest quantitative evidence on disparities in living standards between and among the different ethnic groups in Vietnam. We first use a range of socioeconomic variables to examine the differences in living stan- dards between the Kinh-Hoa majority and the other ethnic minorities, and 273 274 Economic Growth, Poverty, and Household Welfare in Vietnam how these changed between 1993 and 1998. This is followed by a more detailed examination, employing data from both the VLSSs and the 1999 Population and Housing Census, of socioeconomic differences among mi- nority groups. A more nuanced picture starts to emerge, in which the ethnic groups that have done best are shown to be those that have assimilated most with Kinh society and the less assimilated groups (particularly those in the Central Highlands and the Hmong in the Northern Uplands) have been left behind.3 After a brief examination of government policy toward ethnic minori- ties, we turn to a more detailed explanation of why many ethnic minority households are so poor. Distinguishing between endowments (comprising both physical and human capital) and returns to those endowments, we separate the effects of each of these using the VLSS data. Expenditure re- gressions are estimated and decomposed, which show that even if ethnic minority households had the same endowments as the Kinh and Hoa, this would close no more than one-third of the gap in their living standards. Such diversity in the socioeconomic development experiences of the differ- ent ethnic minority groups indicates the need for a similar diversity in the policy interventions designed to assist them. The Majority-Minority Gap in Living Standards The clearest evidence of the gap in living standards between the Kinh-Hoa majorityandtheethnicminoritiescomesfromtheVLSSsof1993and1998.The 1993 survey covered 4,234 Kinh and Hoa households and 566 ethnic minority households; the sample sizes for the 1998 survey were 5,151 and 848 house- holds, respectively. As can be seen from map 8.1, with the exception of the Chinese (Hoa), the ethnic minorities are concentrated in the more remote regions of Vietnam, especially the Northern Uplands and Central Highlands. Where 54 percent of Kinh-Hoa had expenditures below the General Statistical Office (GSO)­World Bank poverty line in 1993, this proportion had dropped to 31 percent by 1998. During the same period, the poverty headcount among the remaining minorities fell from 86 percent to 75 per- cent. So despite constituting just 14 percent of the total population, ethnic minorities now make up 29 percent of all the poor in Vietnam (Vietnam Poverty Working Group 1999). Provincial-level poverty maps constructed by merging data from the 1998 VLSS with the 1999 Census show that there are 14 provinces with rural poverty headcounts of more than 60 percent (Minot and Baulch 2004). Of these 14 provinces, 12 have populations in which ethnic minorities make up more than half of the total. A number of socioeconomic indicators related to the household are gath- ered together in table 8.1, which is based on the data from the 1993 and 1998 VLSSs. For 1993, the summary measures in table 8.1 are based on the full sample of 4,800 households. For 1998, data are presented for both the full sample of 5,999 households living in 194 communes and a subsample of 48 ethnically mixed communes.4 This subsample can be used to examine Ethnic Minority Development in Vietnam: A Socioeconomic Perspective 275 Map 8.1. Ethnicity and Expenditures Hanoi Haiphong Ethnicity Kinh Chinese Hue Expenditure per capita 1,000 dong Other Under 900 N 1993 1998 901­1,800 1,801­2,700 2,701­3,600 Over 3,600 Ho Chi Minh City Sources: 1993 and 1998 VLSSs. whether the living standards of ethnic minority households are worse than those of their Kinh and Hoa neighbors and so provides a crude way to con- trol for the otherwise pervasive effects of geography. To test whether the values of each of these variables are the same for majority and minority households, we have computed p values based on t tests (for continuous variables) and chi-squared tests (for binary variables); these are displayed in the "Test" columns. The data for 1998 have been weighted to correct for the sampling design of the second VLSS (in which different households have different probabilities of being enumerated). Table 8.1 shows that, with an annual per capita expenditure that aver- aged D (dong) 1.54 million (US$125) in 1998, minority households were far poorer than their Kinh and Hoa counterparts (D 3.0 million).5 Although spending by the majority groups rose by 38 percent in real terms between 1993 and 1998, the increase for minority households was much smaller, at 18 percent. The lower living standards of minority households are partly due to the fact that they tend to be larger than Kinh households (5.4 versus 4.6 members in 1998) and are more likely to include young children (15 per- cent versus 10 percent); they are also more likely to span three generations (27 percent versus 18 percent). The fertility rate for minority women is about 25 percent higher than for Kinh and Hoa women (Desai 2000). Ethnic a estT 0.00 0.00 0.01 0.05 0.06 0.00 0.00 0.02 0.04 0.15 0.00 0.68 0.00 0.01 0.00 0.00 only 575 ities 1,604 5.37 0.14 0.27 0.28 0.31 0.04 0.07 0.19 0.41 0.27 0.02 44.0 0.17 0.89 0.02 Minor- communes Mixed 1 1 931 Kinh- Hoa 2,742 4.71 0.1 0.24 0.29 0.36 0.10 0.1 0.25 0.35 0.17 0.02 46.8 0.26 1.00 0.25 1998 estT 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.01 0.00 0.00 0.78 0.00 0.00 0.00 0.00 esent. pr e ar 738 1998 sample ities Minor- 1,536 5.41 0.15 0.27 0.28 0.31 0.04 0.07 0.18 0.41 0.27 0.02 44.2 0.17 0.79 0.02 and Full households 1993 Kinh- Hoa 5,261 2,952 4.61 0.10 0.23 0.30 0.37 0.10 0.14 0.24 0.33 0.18 0.02 48.3 0.28 1.00 0.27 minority and Households, ities 565 Minor- 1,299 5.52 0.20 0.23 0.27 0.30 0.03 0.12 0.15 0.44 0.23 0.03 42.1 0.16 0.47 0.04 Kinh-Hoa e 1993 sample wher Full Minority Kinh- Hoa 4,234 2,142 4.89 0.16 0.22 0.28 0.35 0.07 0.15 0.21 0.38 0.17 0.02 45.8 0.28 1.00 0.22 and communes in those eas ar Majority prices) of en only of households 1998 childr urban is e years interviewed in includes capita that consisting ne mor that VLSSs. per e January or 16 child childr ee household 1998 16 household, Characteristics (weighted) dong, one two thr female-headed households households and size household age households of subsample of age adults 6 16 of of of of and and and on 1993 8.1. size expenditur to to over two 0 7 over head ces: or ent(s) ent(s) ent(s) ee-generation of Based ableT (thousand oportion oportion oportion oportion ietnameseV oportion a. Sour Indicator Sample Annual Household Pr Ages Ages Male, Female, Pr One Par Par Par Thr Other Age Pr Pr Pr 276 Ethnic Minority Development in Vietnam: A Socioeconomic Perspective 277 minority households are also less likely to be able to speak Vietnamese and are much less likely to live in urban areas (2 percent versus 27 percent). Ethnic minority households are less served by the health system (Desai 2000). Just 47 percent of ethnic minority mothers in the 1998 VLSS sample sought prenatal care, compared with 70 percent for Kinh mothers. Further- more only 30 percent of ethnic minority births were assisted by a physician or nurse-midwife, compared with 81 percent for the Kinh. Similarly, 75 per- cent of ethnic minority parents consulted a health care provider when a child (that is, ages 5­60 months) was sick, compared with 88 percent for Kinh households. Roughly 50 percent of minority children of ages one year or older have received the four main vaccinations, compared with about 60 percent for Kinh children.6 However, it is important not to overemphasize the contrasts, because an outside observer is more likely to be struck by the similarities between the sociodemographic characteristics of the two groups. For instance, Desai (2000) shows that contraceptive usage rates are broadly similar across ethnic groups: 55 percent of married women in ethnic minorities who were ages 15 to 44 reported that they use a modern method of contraception, compared to 59 percent among Kinh women and 35 percent among Chinese women. Although the expenditure level of minority households is much lower than that of Kinh-Hoa households, the mean consumption of calories is only slightly lower (2,068 calories per day per capita for minorities versus 2,115 for Kinh). If adult equivalents are used, the difference (2,681 versus 2,695 calo- ries) is negligible (Desai 2000, table 3.6). This helps explain the otherwise sur- prising finding that the mean body mass index of minority men is the same as that for Kinh men (19.9) and only slightly lower for minority women (19.6) than Kinh women (20.1). Indeed, Desai (2000, table 6.2) finds that a smaller proportion of minority men are severely malnourished (3.6 percent) than Kinh men (6.3 percent), although the gap is less evident for women (8.0 per- cent for minorities versus 9.4 percent for Kinh). Nonetheless, it remains the case that the children of ethnic minorities are more likely to be stunted, a measure of long-term malnutrition (Haughton and Haughton 1999). In short, by Vietnamese standards, ethnic minority households look sig- nificantly different from Kinh-Hoa households. However, both fit groups broadly within Vietnamese norms, and both groups have experienced simi- lar trends in living standards: rising expenditures, falling fertility and household size, and comparable levels of malnutrition. Differences among Minority Groups Expenditures Not all ethnic minority groups are equally disadvantaged. This is an impor- tant point, because if ethnicity is used to help target government interven- tions such as food subsidies or scholarships, then there will be less wastage if the relevant targets can be identified more precisely. The practical problem here is that the VLSSs did not sample enough ethnic minority households to 278 Economic Growth, Poverty, and Household Welfare in Vietnam allow for much disaggregation; moreover, the 1993 VLSS codes allowed for only 10 different ethnic groupings rather than the standard official list of 54 distinct groups. The VLSS questionnaires also collected information only on the ethnicity of the head of the household. This does not allow analysis of, for instance, minority issues at the individual (as distinct from household) level or exploration of the extent of intermarriage between ethnic groups. The best that one can do with the VLSS data under these circumstances is to separate households into a few relatively homogeneous categories based on the ethnicity of the head. We distinguish three of the main ethnic groups (Kinh, Hoa, and Khmer), together with a composite category for eth- nic minorities that traditionally live in the Central Highlands and another for those that originate in the Northern Uplands. The relevant details are summarized in table 8.2, along with a listing of ethnic groups by composite category. This disaggregation, crude as it may be, is helpful. The data in table 8.2 show clearly that the poorest group consists of the Central Highlands minorities. Their expenditure per capita was D 1.02 million in 1993, barely Table 8.2. Key Indicators for Major Minority Groups, 1993 and 1998 Expenditure Poverty per capita headcount (thousand Sample (percent of dong, 1998 Household size people) prices) size (weighted)a Percent Ethnic group 1993 1998 1993 1998 1993 1998 1993 1998 of pop. Vietnam overall 55 36 2,043 2,751 4.97 4.71 4,799 5,999 Kinh 52 30 2,105 2,899 4.86 4.60 4,145 5,030 83.9 Hoa (Chinese) 11 8 3,843 5,119 6.55 5.18 89 121 2.0 Khmer 70 57 1,521 1,882 5.44 5.33 89 122 2.0 Central Highlands minoritiesb 92 91 1,021 1,090 6.28 5.79 103 167 2.8 Northern Uplands minoritiesc 84 73 1,323 1,594 5.33 5.31 373 560 9.3 Note: One hundred thirty-two households coded as belonging to "Other" ethnic minorities in the 1993 VLSS and 39 households belonging to the "other" category in the 1998 VLSS have been subdivided between the last two groups in this table using the regional and religion variables. Details are available from the authors on request. The categories may not be strictly compara- ble between 1993 and 1998. a. Unweighted sample size: Kinh, 5,172. Hoa, 131. Khmer, 95. Central Highland minorities, 193. Northern Uplands minorities, 411. b. Central Highlands minorities: Ba-Na, Co-Ho, E-De, Gie-Tieng, Hre, Ma, Ra Glai, Xo-Dang. c. Northern Uplands minorities: Dao, Hmong, Muong, Nung, Tay, Thai, San Diu, Dan Chay, Tho. Sources: 1993 and 1998 VLSSs. Ethnic Minority Development in Vietnam: A Socioeconomic Perspective 279 rising to D 1.09 million by 1998; this stagnation meant that the Central High- lands minorities saw their relative position fall, with an expenditure level that was half the national average in 1993 but little more than one-third of the national average by 1998. The poverty headcount for this group went from 92 percent in 1993 to 91 percent in 1998. This group missed the eco- nomic boom of the 1990s; thus, it is not surprising that dissatisfaction, which was also related to land and religious conflicts, bubbled over into the signif- icant demonstrations by ethnic people in several places in the Central High- lands in February 2001 (Economist Intelligence Unit 2001). It is possible to get a more complete picture of the distribution of per capita expenditures by ethnic category from the kernel densities shown in figure 8.1. These may be thought of as histograms that have been smoothed to iron out minor irregularities in the data (Deaton 1997; Stata Corporation 1999) and draw the eye to the essential features of the distribu- tions. In figure 8.1a, the kernel densities for the Kinh, Hoa, and Khmer are shown. The density for the Hoa stands out: its peak is far to the right of the other distributions, and there is a wider variation in per capita expenditures than the other four categories. The slightly bimodal distribution is due to the heavy, if partial, concentration of Hoa households in large urban areas, par- ticularly in the Southeast. In contrast, the distribution of expenditures for Khmer households, which live primarily in the Mekong Delta, has a peak just below the GSO­World Bank poverty line, and most of the observations are highly concentrated in that vicinity. This indicates that as long as those regions continue to benefit from general economic growth, a large propor- tion of the Khmer should move out of poverty in the next five years or so. Figure 8.1b shows the kernel densities for the Central Highlands and Northern Uplands minorities, with that for the Kinh included for compari- son purposes. The distributions of expenditures for Northern Uplands minorities, and especially for Central Highlands minorities, are even more highly concentrated than for the Khmer. The mode of the density for the Northern Uplands minorities is, however, relatively close to the poverty line, indicating that they are also likely to benefit from equitable economic growth. In contrast, the Central Highlands minorities are considerably poorer in expenditure terms than the other four categories, as both their density in figure 8.1b and the poverty headcounts in table 8.2 confirm. Exceptionally rapid growth, or other special measures, or both will there- fore be needed if poverty is to be reduced significantly among the ethnic minorities indigenous to the Central Highlands. Schooling Although a finer breakdown by ethnic group is not possible using VLSS data, one can get greater precision using the 3 percent enumeration sample of the 1999 Population and Housing Census. Although the census data do not pro- vide information on incomes or expenditure, they do allow one to construct gross and net school enrollment rates for the 12 ethnic groups for which there 280 Economic Growth, Poverty, and Household Welfare in Vietnam Figures 8.1. Kernel Densities of Per Capita Expenditure for 1998 .0006 Poverty line .0004 .0002 0 0 5,000 10,000 Real per capita expenditure (thousands of dong) Kinh Hoa Khmer .0006 Poverty line .0004 .0002 0 0 5,000 10,000 Real per capita expenditure (thousands of dong) Kinh N. Upland Min. C. Highland Min. are at least 1,000 observations in the enumeration sample. School enrollment rates are usually highly correlated with income and may therefore be used as an approximate indication of the standard of living in a community. Table 8.3 shows primary school enrollment rates by sex for each of the 12 ethnic groups with more than 1,000 children of primary school age Ethnic Minority Development in Vietnam: A Socioeconomic Perspective 281 Table 8.3. Primary School Enrollment Rates by Ethnic Group and Sex, 1999 Ethnic group Gross Net Net (percentage Net (percentage Sample (percent) (percent) of boys) of girls) size Kinh 113.6 93.4 93.5 93.4 229,503 Hoa 122.6 93.7 94.5 92.9 2,361 Khmer 114.5 76.3 77.3 75.3 3,879 Central Highlands Gia-rai 126.3 66.4 67.6 65.1 1,695 Ba-na 108.9 57.8 55.0 60.4 1,335 Xo-dang 139.3 62.2 64.7 59.3 1,233 Northern Uplands Tay 135.4 94.7 94.9 94.4 11,079 Thai 135.5 83.9 87.2 80.5 5,004 Muong 133.4 94.5 94.9 94.0 3,851 Nung 136.6 89.3 89.7 88.9 5,010 Hmong 80.5 41.5 51.5 31.5 4,090 Dao 126.4 71.4 73.7 68.8 4,091 All Vietnam 115.4 91.4 91.7 91.0 280,262 Note: To be consistent with Vietnamese school enrollment procedures, these enrollment rates have been computed using calendar year of birth as stated in the census files to determine whether or not a child is of primary or lower secondary age. The net enrollment rate can fall by several percentage points if the child's actual age (for example, 6 to 10 years old for primary school) is used. Gross enrollment rate is total enrollment in level X per number of children eligible to attend level X. Net enrollment rate is total enrollment in level X of children eligible to attend level X per number of children of age eligible to attend level X. Source: Based on the 3 percent enumeration sample of the 1999 Census. included in the census 3 percent sample. By the standards of comparably poor countries, the primary school net enrollment rates (NERs) in Vietnam are quite high (91 percent).7 However, primary NERs are below 70 percent for five ethnic groups: the Ba-na, Gia-rai, and Xo-dang in the Central High- lands and the Dao and the Hmong in the Northern Uplands, as table 8.3 shows. In addition to poverty and remoteness, one of the factors discour- aging ethnic minority children in these groups from attending primary school is lack of instruction in ethnic minority languages (especially in the lowest grades).8 On average, primary school enrollments are relatively balanced between the sexes, with an overall primary NER of 91.7 percent for boys and 91.0 per- cent for girls. Again, this blurs differences at the level of individual ethnic minorities: for all groups except the Ba-na, primary NERs are slightly lower for girls than for boys, although in most cases the difference in the NER is small and not statistically significant. Three exceptions to this rule stand out: girls' primary NERs substantially lag those for boys among three ethnic groups in the Northern Uplands: the Dao (-4.9 percent), the Thai (-6.7 per- cent), and, in particular, the Hmong (-20.0 percent). 282 Economic Growth, Poverty, and Household Welfare in Vietnam Table 8.3 also shows that primary school gross enrollment rates (that is, the number of pupils enrolled in primary school divided by the number of children eligible to attend primary school) are in some cases very high in- deed. The implication is that a substantial proportion of Vietnamese children are starting primary school late and are repeating grades frequently--this is especially true of most ethnic minority children. Table 8.4 summarizes lower secondary school enrollment rates by eth- nicity and sex.9 As expected, both gross and net lower secondary school en- rollment rates are much lower than primary school enrollment rates. For Vietnam overall (in 1999), the net enrollment rate falls from 91 percent for primary school to 60 percent for lower secondary school. At the lower sec- ondary level, there is a clear gap between the Kinh (65 percent) and all other groups (52 percent or less). Five ethnic groups--the Gia-rai, Ba-na, and Xo-dang in the Central Highlands and the Hmong and Dao in the north-- have NERs at the lower secondary level of less than 20 percent, with that for the Hmong just under 5 percent. Overall, the lower secondary NER is es- sentially the same for boys and girls, but this hides some variation by ethnic group: among the Hmong and Xo-dang girls, lower secondary NERs are at Table 8.4. Lower Secondary School Enrollment Rates by Ethnic Group and Sex, 1999 Gross Net Net (percentage Net (percentage Sample Ethnic group (percent) (percent) of boys) of girls) size Kinh 80.6 64.8 65.5 64.0 185,772 Hoa 71.0 51.7 50.4 53.1 1,989 Khmer 35.9 22.5 23.8 21.2 3,041 Central Highlands Gia-rai 37.1 14.9 15.2 14.5 1,354 Ba-na 20.0 8.9 9.0 8.9 1,024 Xo-dang 35.2 10.1 12.7 7.1 1,071 Northern Uplands Tay 77.0 51.0 47.1 55.2 9,082 Thai 55.2 32.1 33.6 30.5 4,402 Muong 76.7 52.3 50.8 53.9 3,265 Nung 61.8 39.2 37.0 41.6 4,055 Hmong 9.8 4.5 7.5 1.6 3,092 Dao 20.3 11.8 11.9 11.8 3,026 All Vietnam 76.2 60.0 60.5 59.3 226,649 Note: To be consistent with Vietnamese school enrollment procedures, these enrollment rates have been computed using calendar year of birth as stated in the census files to determine whether or not a child is of primary or lower secondary age. The net enrollment rate can fall by several percentage points if the child's actual age (for example, 6 to 10 years old for primary school) is used. Gross enrollment rate is total enrollment in level X per number of children eli- gible to attend level X. Net enrollment rate is total enrollment in level X of children eligible to attend level X per number of children of age eligible to attend level X. Total sample size for All Vietnam also includes ethnic groups not listed in the table. Source: Based on the 3 percent enumeration sample of the 1999 Census. Ethnic Minority Development in Vietnam: A Socioeconomic Perspective 283 least 5 percent lower than for boys, and for the Tay and Nung the female enrollment rates are at least 5 percent higher than for boys. These findings on enrollment rates allow us to start to explore the extent to which different ethnic minorities are assimilated with the Kinh majority. If ethnic groups are classified according to the extent to which their school enrollment rates are similar to those of the Kinh, one might reasonably argue that the Hoa, Tay, Muong, Nung, and perhaps Thai are assimilating rela- tively fast, and the other minorities (the Dao and Hmong in the Northern Uplands, the Khmer in the South, and all the Central Highlands minorities) are assimilating much less rapidly. If this speculation is correct, then we might expect a relatively high degree of intermarriage among the first ("more assimilated") group than among the second ("less assimilated") group. We now examine this proposition. Intermarriage The 3 percent census enumeration sample, but not the VLSSs, allows us to measure the extent of intermarriage among the 12 main ethnic groups.10 The results are summarized in table 8.5. The most striking finding is that Chinese are the most likely to marry partners of a different ethnic group; one-third of Chinese heads of household are married to a member of another ethnic group, primarily Kinh. The Nung and the Tay are also likely to intermarry, Table 8.5. Intermarriage of Household Heads, 1999 Married To member of another To Kinh partner Sample size Ethnic group ethnic group (percent) (percent) (unweighted) Kinh 0.9 99.1 339,633 Hoa 33.3 30.1 3,283 Khmer 11.4 10.9 4,196 Central Highlands Gia-rai 1.2 0.6 1,872 Ba-na 1.4 0.3 1,440 Xo-dang 2.0 0.2 1,536 Northern Uplands Tay 19.1 12.0 15,161 Thai 6.4 2.6 5,816 Muong 10.2 7.6 4,957 Nung 25.0 12.1 6,562 Hmong 0.8 0.5 3,676 Dao 6.5 4.1 4,225 All Vietnam 2.5 1.1 399,573 Note: Among household heads, 134,566 (23.6 percent) are single, widowed, separated, or divorced. Among married household heads, 9.5 percent are female. Total sample size for All Vietnam also includes ethnic groups not listed in the table. Source: Authors' calculations based on the 3 percent enumeration sample of the 1999 Census. 284 Economic Growth, Poverty, and Household Welfare in Vietnam with one in four Nung and one in five Tay heads married to a partner from a different ethnic group. With the exception of the Thai, at least 10 percent of household heads in the more educated ethnic groups are married to someone from another eth- nic group, typically Kinh. This is an intermarriage rate comparable to or higher than that of second-generation Italian-Americans and Jews in the United States in the middle of the 20th century (The Economist 2001). This suggests that the cultural and perhaps economic distance between these groups and the Kinh majority is relatively modest; we might speculate that these groups have embarked on a path of economic development that will lead to "assimilation" with the dominant group. The Khmer may also fit into this mold, although less clearly. In contrast, the Thai appear to have chosen to keep their distance--a rel- atively low rate of intermarriage, particularly with Kinh partners--while emphasizing education. In this respect, they are following a similar path to the (ethnically similar) Tai in Xishuangbanna, a region of southern China that abuts Vietnam. The Tai's unwillingness to assimilate into mainstream Han culture has led to an increasing degree of economic marginalization (Hansen 1999). The remaining ethnic groups, particularly the Central Highlands minori- ties and the Hmong in the Northern Uplands, have very low rates of inter- marriage with members of other groups. It is perhaps surprising that the Hmong and the Dao, who live in overlapping mountainous areas and be- long to the same Kadai subgroup, intermarry very infrequently. When the Dao do intermarry, it is most often with a Tay partner. The low-intermarriage groups are also the groups for whom school en- rollment rates are the lowest. It is unclear whether these groups' apartness is a matter of choice or an unintended consequence of linguistic and geo- graphic barriers. We would, however, suggest that the most difficult chal- lenge of public policy toward ethnic groups is bringing the less assimilated groups into the economic mainstream; most of the more assimilated ethnic groups are already halfway there. Some anthropologists argue that it may be more socially acceptable for a Tay, Nung, or Dao man to marry a woman from one of the other Northern Uplands minorities than to marry a Kinh woman. However, the evidence from the 3 percent census enumeration survey does not bear this out: More than half of the Tay and Dao husbands, and almost half of the Nung hus- bands, who have married an outsider have Kinh wives.11 Religion Religion is the final aspect of the assimilation of different ethnic groups into Kinh society that can be examined using the 1999 Census. This is a sensitive issue in Vietnam. The protests in the Central Highlands by ethnic minority groups in early 2001 were partly in response to official efforts to restrict religious practice in the region, especially among the growing number of evangelical Protestants. Ethnic Minority Development in Vietnam: A Socioeconomic Perspective 285 Article 70 of the 1992 Constitution guarantees all Vietnamese citizens freedom of religion or nonbelief, but indirect controls and local restrictions often discourage particular religious groups (United Nations Economic and Social Council 1998). Furthermore, in the past, some religious groups (espe- cially Protestant Christians in the Central Highlands and Northern Up- lands) have been accused of being aligned with organizations whose aim is the overthrow of the state (Winrock International 1996) or were historically associated with opposition to the government (for instance, the Cao Dai). Although in recent years, the government's attitude toward religion has be- come noticeably more relaxed, many de facto regulations still exist; thus, the position of many religious communities is best described as one in which "circumscribed areas of freedom are emerging within a general framework of controls, limitations and even prohibitions" (United Nations Economic and Social Council 1998). For those minority groups that have large num- bers of religious practitioners, these restrictions are an important source of irritation and even alienation from the central authorities. Table 8.6 shows the percentage breakdown of professed religion at the time of the 1999 Census.12 More than three-quarters of people in Vietnam stated they had no religion, with Buddhism, Christianity (mainly Catholi- cism), Cao Daism, and Hoa Hao (two indigenous religions that blend a number of Eastern and Western beliefs and practices), and Islam accounting for the remainder. Some of the smaller ethnic minorities are known to have their own, often animist-based, religions, and it is unclear how well these were enumerated in the census. Table 8.6. Distribution of Ethnicity and Religion (percent) Ethnic group No religion Buddhist Christian Other religion Kinh 77.7 10.9 7.9 3.3 Tay 99.3 0.3 0.1 0.0 Thai 99.6 0.0 0.1 0.0 Hoa (Chinese) 74.7 22.7 2.4 0.2 Kho-me (Khmer) 37.4 62.3 0.2 0.1 Muong 98.4 0.1 1.4 0.0 Nung 98.0 1.6 0.2 0.0 Hmong 95.2 0.1 4.5 0.0 Dao 99.2 0.2 0.3 0.0 Gia-rai 80.3 0.1 19.6 0.0 Ba-na 52.2 0.0 47.8 0.0 Xo-dang 71.3 0.0 28.6 0.0 All Vietnam 78.8 10.5 7.7 3.0 Note: "Other religions" include Cao Dai, Hoa Hao, and Islam. Source: Authors' calculations from the 3 percent sample of the 1999 Census. 286 Economic Growth, Poverty, and Household Welfare in Vietnam About three-quarters of the Kinh and Hoa stated that they practiced no religion; among practitioners, Buddhism is the most common religion, fol- lowed closely by Christianity. A significant number of ethnic groups, partic- ularly in the Northern Uplands, profess essentially no religion, including the Tay, Thai, Muong, Nung, Dao, and Hmong. However, a number of the Central Highlands minorities count a high proportion of believers: almost half of the Ba-na are Christian (mainly Protestants), as are substantial percentages of the Xo-dang and Gia-rai. A majority of the Khmer are practicing Buddhists. Islam has a significant num- ber of adherents only among the Cham, and Cao Daoism and Hoa Hao are practiced mainly by the Kinh living in the Southeast and Mekong Delta. Just under 5 percent of the Hmong are Christian (most of whom are Protestants), though it seems likely that the Hmong's traditions of spirit worship have been overlooked in the census data. Government Policy toward Minority Groups To provide some context for the subsequent discussion, we now briefly sum- marize government policy toward ethnic minorities. The main vehicle for implementing government policies on ethnic minorities is the Committee for Ethnic Minorities in Mountainous Areas (CEMMA). This is a cabinet-level committee, established in 1993, that is charged with identifying, coordinat- ing, implementing, and monitoring projects geared toward ethnic minority development. CEMMA has a budget of D 7.9 trillion (US$546 million) to be spent on its main programs and projects over the five-year period through 2005; if realized, this would amount to a substantial US$50 per ethnic minor- ity household per year. However, since 1998, CEMMA has been criticized for various instances of corruption. In February 2001, 13 CEMMA officials were disciplined by the Communist Party for "violating regulations on manage- ment" (Cohen 2001). In March 2001, the 11th plenum of the Central Commit- tee of the Party disciplined CEMMA chairman Hoang Duc Nghi (Xinhua 2001). Cohen (2001) has written that "at the heart of CEMMA's failings is a top-down approach. . . . Ethnic minorities rarely participate in planning de- velopment projects, and rarely know what they are entitled to once projects are implemented." In addition, under Program 133, the Ministry of Labor, Invalids and Social Affairs (MOLISA) coordinates a Hunger Eradication and Poverty Reduction program that is designed to combat poverty by providing addi- tional resources to the poorest communes in the country. Given the high lev- els of poverty among ethnic minorities, this program necessarily helps ethnic minority households disproportionately, even though MOLISA's list of poor communes includes many in lowland and midland areas. The main weakness of this program is that it is not sufficiently targeted. By spreading its largesse--about D 410 billion (US$28 million) annually--so thinly, it pro- vides only limited help to the poorest households, which dissipates its effectiveness as an antipoverty program. The bluntness of the targeting is Ethnic Minority Development in Vietnam: A Socioeconomic Perspective 287 clear from the numbers compiled by van de Walle (2004) in this volume. She reports that in 1998, 71 percent of the richest rural communes had a poverty alleviation program, compared with 89 percent for the poorest rural com- munes (table 6.16), and poverty alleviation programs touched villages with 84 percent of the rural poor and 76 percent of the nonpoor (table 6.6). Mean- while, large numbers of poor people living in nonpoor communes are excluded from receiving many benefits (Minot and Baulch 2004). A wide range of government interventions designed to help the ethnic minorities have been introduced since 1993. These interventions include subsidies for the cost of transporting essential goods to remote areas; funds for resettlement and sedentarization; subsidies for salt, reforestation funds, the provision of potable water, road maintenance and upgrading, the provi- sion of livestock and seedlings to farmers; gifts of radios to remote house- holds; subsidies for connecting villages to the national grid; and the provi- sion of educational scholarships. Government policy is not, however, universally supportive of ethnic mi- norities. On the one hand, there is official interest in maintaining (bao ton) and developing (phat huy) cultural identity, particularly dances, folklore, and modes of dress. On the other hand, the standard textbooks tend to em- phasize, and even glorify, Kinh culture and history. Similarly, the expansion of education has at last led to a rapid rise in enrollment rates for ethnic mi- nority children. However, Vietnamese remains the dominant language of in- struction, and most officially sanctioned textbooks are in Vietnamese. There is an ongoing tension between the willingness to accept differences (cong nhan) and cultural assimilation or Vietnamization (dong hoa). The most important rural development policies have not helped, and may have hurt, many ethnic minority households. The government discour- ages drug production, which reduces the income of some growers in the Northern Mountains. Agricultural extension and research tend to favor low- land rice over upland crops (Huy and Dai 1999). The formalization of land rights has tended to squeeze slash-and-burn farmers, especially because tra- ditional land and forest use rights are poorly defined and frequently not rec- ognized by the formal legal system (Huy and Dai 1999). Government subsi- dies have encouraged people to move to the "New Economic Zones" in the Central Highlands. Even though only half of the (mostly Kinh) migrants to those zones have stayed there, the in-migration has contributed to tension with the indigenous ethnic minorities in the Central Highlands (especially over land). There is strong interest among donors and NGOs in projects that would alleviate poverty. These efforts have the effect of helping ethnic minority households, although not explicitly. There are also a number of projects, or components of projects, that are explicitly geared toward ethnic minorities.13 With NGO support, an Ethnic Minority Forum (and now working group) was established in 1993 and serves as a locus for sharing experiences and lessons learned from the many efforts that are geared toward ethnic minor- ity development. 288 Economic Growth, Poverty, and Household Welfare in Vietnam Explaining the Divergence between Majority and Minority Living Standards Why are Vietnam's ethnic minority households so poor? Following other studies using the VLSS, we measure material living standards using expen- diture per capita. So our question becomes: Why is expenditure per capita so low, and growing so slowly, among ethnic minority households? The standard economic explanations may be grouped into two. First, people may be poor if they lack endowments. The main factors of production are land, physical capital, and human capital (education). To the extent that a household lacks these endowments, it is likely to be relatively poor. Table 8.7 summarizes some of the main variables on household en- dowments. It shows that although ethnic minority households tend to have a relatively large quantity of land, this land is generally of poorer quality (re- flected in part by the relatively low holdings of irrigated land).14 Ethnic mi- nority households are likely to be poorly endowed with capital, as reflected by their lack of access to credit and lower receipts of remittances; in rural areas, the value of farm tools owned by Kinh households is twice as high as the value of those owned by ethnic minority households. As would be ex- pected from the school enrollment data, ethnic minority households also have lower levels of education than the Kinh-Hoa majority. For households that remain in farming, it may not make sense to acquire more education, but the modest level of education also serves to reduce the number of eco- nomic opportunities open to them elsewhere in the country. Second, people may be poor because of their knowledge, customs, or culture--meaning that they do not use the available factors of production as efficiently as possible or because they face discrimination, and so they would have more difficulty getting a good job than another, equivalently qualified individual. Either of these would lead to the same result, which is low "returns on characteristics." For instance, a poorly educated farmer from an ethnic minority may not be able to get a high return on land because he or she does not know how to cultivate high-yielding crop varieties, or be- cause the local agricultural extension agent cannot speak the local language or never visits. Ethnic minority people have low endowments, and poor returns to char- acteristics, in part because many of them live in remote areas and are dis- connected from the rest of the economy. Traditionally, remoteness is seen as a geographic concept. Households living in remote areas find it expensive to buy inputs or to bring their goods to market. If the density of population is low, it is harder to provide schooling and other amenities. But remoteness may also be thought of as a social concept, so that some households may be distant from their neighbors because of barriers of language or culture. The ethnic minority households in rural areas that do not speak Vietnamese have per capita expenditures (D 1.074 million) that are only three-fifths as high as those of their Vietnamese-speaking counterparts (D 1.641 million), according to the 1998 VLSS. Many minority groups also feel remote from the estT 0.00 0.01 0.00 0.00 0.65 0.61 0.00 0.00 0.01 0.00 only 575 1,604 0.01 0.10 12,035 2,886 1,284 5,027 216 6.04 7.55 Minorities communes Mixed 931 2,742 0.04 0.20 7,628 3,176 1,164 1,044 484 7.36 9.21 Kinh-Hoa oups. 1998 gr estT 0.00 0.00 0.00 0.00 0.52 0.39 0.00 0.00 0.00 0.00 minority 738 1,536 0.01 0.09 sample 1,7471 2,403 1,454 4,630 213 5.53 6.94 ethnic 1998 Minorities and Full and majority 1993 5,261 2,952 0.06 0.23 5,469 2,704 1,079 505 425 7.36 9.36 Kinh-Hoa Kinh-Hoa 1 the 565 12 1,299 0.02 0.1 569 959 216 8,002 1,1 4.72 6.57 both Households, Minorities sample omfr 1993 Full Minority 4,234 2,142 0.06 0.20 682 170 486 5,004 1,531 6.58 9.04 Kinh-Hoa households and a sampled size. Majority prices) and with of ) 1998 2 eceivingr eceivingr those capita e weights ar . per (meters head best-educated VLSSs. only e January b of of b is b 1998 b household Endowments (weighted) dong, households households ops tools/household dong) communes by and of emittancesr of emittancesr b cultivated which land cr households 1993 8.7. size expenditur ea of land farm education education eign ar household ennial est of Mixed of household of eighted ces: W Rural ableT (thousand oportion for oportion domestic per Irrigated Per For Indicator Sample Annual Pr Pr Land Portion alueV (thousand earsY of earsY member Note: a. b. Sour 289 290 Economic Growth, Poverty, and Household Welfare in Vietnam process of policymaking and decisionmaking; the recent (April 2001) eleva- tion of Nong Duc Manh, an ethnic Tay, to the position of General Secretary of the Communist Party is an exception to this rule. Remoteness is more likely to be a problem if there are additional barriers--administrative, social, or other--that prevent households from migrating in response to better op- portunities elsewhere. Several measures of remoteness are summarized in table 8.8. Children from ethnic minorities have to travel farther to school. Their parents have to travel farther to go to a market, hospital, post office, or factory. Their fami- lies are less likely to live in a village or commune that is served by public transport, electricity, or telephone. These effects are particularly dramatic for households living in minority-only villages, where on average the clos- est market is 13.5 kilometers away. Only 11 percent of minority-only villages have phones. Although they use a somewhat different vocabulary, Vietnamese social scientists typically point to similar causes of poverty among the ethnic mi- norities. Ethnologist Bui Van Dao (personal communication, March 2001) argues that ethnic minorities are persistently poor because of "objective rea- sons" (isolated villages, poor soils, inadequate water, unsuitable climate), "subjective reasons" (low educational levels, population pressure, shortage of capital, slow technical change), and "institutional reasons" (government pol- icy insufficiently targeted, overlapping programs, top-down administration). Pham and Tuan (1999) come up with a similar list, but they add that the sociopolitical institutions and customs of ethnic minorities are "still back- ward" and "subversive forces" have "abused" religion and ethnicity "to de- stroy national unity." Implicit in this diagnosis is that the solution is for eth- nic minorities to assimilate. This is the most widely held view in official circles. The reference to national unity is important, because a number of the ethnic minority groups worked closely with the Americans during the war in the 1960s and 1970s, and their political reliability is still considered to be suspect. Others have argued that ethnic minorities are poor because they have been trapped in a downward spiral: Population growth puts pressure on the natural carrying capacity of the uplands, which leads to environmental degradation and poverty (Jamieson, Cuc, and Rambo 1998). This in turn leads to social, cultural, and economic marginalization and increased de- pendence on nonlocal support systems (NGOs, government subsidies), which make it even harder for them to rise out of poverty. Jamieson, Cuc, and Rambo stress this last component. Decisionmaking, they argue, is too centralized and remote. It also occurs without adequate representation of local people, which in turn fuels distrust and misunderstanding. Much less has been written about how minority people characterize and explain their own poverty. As part of a participatory poverty assessment, a recent study in Lao Cai found that people place great emphasis on the lack of natural resources, particularly high-quality land and reliable water sup- plies, in explaining their own poverty (Vietnam-Sweden Mountain Rural 1 1 0.42 2.4 6.5 9.1 29.1 25.0 0.23 0.12 13.5 0.1 0.29 0.1 24.4 25.5 0.89 Nonmixed communes Minorities estT 0.66 0.70 0.83 n.a. 0.04 0.12 0.95 0.03 0.01 0.04 0.41 0.06 0.14 0.15 0.06 only 0.43 1.9 2.6 n.a. 16.5 6.7 0.54 0.21 4.0 0.83 0.31 0.33 4.5 1.21 0.53 Minorities communes 1998 Mixed 0.38 1.8 2.5 n.a. 9.1 5.2 0.55 0.38 2.4 0.95 0.41 0.54 2.7 8.5 0.33 Kinh-Hoa Households, estT 0.43 0.02 0.01 0.03 0.00 0.01 0.13 0.00 0.00 0.00 0.05 0.00 0.01 0.06 0.00 Minority and 0.43 2.0 3.0 8.0 18.9 10.1 0.48 0.19 5.8 0.70 0.31 0.29 8.2 13.6 0.60 sample Minorities Full Majority for 0.35 1.4 1.9 5.0 8.8 4.2 0.63 0.48 1.5 0.96 0.48 0.66 1.4 8.3 0.19 Kinh-Hoa ariablesV village km commune in available e school school 10 the Remoteness in wher school commune school within transport in home VLSSs. secondary secondary fice market of villages at in 1998 primary factory any electricity public phone Community primary lower upper center post market phone hospital and variable with est est est est with with with with with est living usually e applicable. ar 1993 8.8. near near near district near closest closest near Not ces: to to to to to to to to ableT oportion oportion oportion oportion oportion oportion oportion births n.a. Sour Remoteness Pr Km Km Km Km Km Pr Pr Km Pr Pr Pr Km Km Pr 291 292 Economic Growth, Poverty, and Household Welfare in Vietnam Development Project 2000). Bui Minh Dao also argues that many ethnic groups explain poverty on the basis of superstitions (tam linh). People become rich thanks to spiritual support, or they are poor because they are encountering a bad time (van han). Although a listing of the possible causes of poverty is certainly useful, such an exercise does not give a good sense of what the most influential fac- tors might be. In an important study based on the 1993 VLSS data, van de Walle and Gunewardena (2001) examine the relative contributions of charac- teristics, the return to characteristics, and geography in explaining why eth- nic minority households are poorer than the rest of society. They use the Blinder-Oaxaca decomposition (described below) to determine the extent to which the lower expenditure levels of minority households is due to weaker characteristics (that is, lower educational levels, poorer quality of land) and how much is due to lower returns on these characteristics. Using expenditure regressions estimated for households living in rural areas of northern and central Vietnam, they find that about half of the difference in expenditure per capita between the two groups is due to differences in their characteristics and endowments, with the remainder attributable to the lower return to characteristics obtained by minority households. Some writers interpret the portion of the expenditure differential due to return to characteristics as a measure of discrimination. However, this is not entirely satisfactory, because the differences in characteristics between majority and minority households may themselves be the result of unequal treatment in the past. Nor is dis- crimination the only possible explanation of the expenditure differential-- other unobserved factors, including cultural history, could play a role. Do the findings of van de Walle and Gunewardena still hold? They used data from 1993, when restrictions on in-country migration had only just been eased and were still of some importance. In the next section, we apply their model to the 1998 VLSS data using both the simple majority-minority split and the disaggregation into composite categories (Kinh, Hoa, Central Highlands minorities, and Northern Uplands minorities) developed above. We find that the differences in returns to characteristics by ethnicity are gen- erally stronger than they were in 1993; certainly they remain very important. Decomposition Analysis Updated To explain the gap between the living standards of majority and minority households, we begin by estimating regressions in which the dependent variable is the log of expenditure per capita (ln E) and the independent variables consist of household- and community-level endowments and characteristics (X). Formally, we regress ln(Eijk) = Xijkjk + jk + ijk where the observations are for the ith household in the jth ethnic group in the kth commune. Here the jk are fixed, commune-level effects and ijk is a random error with zero mean. Separate regressions may be estimated for Ethnic Minority Development in Vietnam: A Socioeconomic Perspective 293 each ethnic group. For instance, indexing the Kinh and Hoa majority with a and the ethnic minorities with b, it can be shown that ln E¯a - ln E¯b = (Xa - Xb)a + Xa(a - b) ¯ ¯ ¯ Total difference = Characteristics + structure where the ln E¯ terms represent the mean log of expenditure per capita and the Xi give the mean characteristics of each group. This is the Blinder- Oaxaca decomposition (Blinder 1973; Oaxaca 1973), which separates the dif- ferences in expenditure per capita into the part that is due to the different characteristics of the two groups and another component that reflects "structural" differences between the two groups. Note that the decomposi- tion shown here uses the parameters for group a, but this choice is arbitrary. One could equally well use the parameters from the equations estimated for group b, and this will generally give a different decomposition. When fixed effects are included (the jk terms) in the regressions, they drop out of the decomposition provided that the equations for each group are estimated for communes where there are both majority and minority households--in our terms, the "mixed commune sample." Our regression results are set out in appendix 8A for the full sample and appendix 8B for the mixed-commune sample (which includes only the 48 communes with both majority and minority households). The dependent variable is the log of per capita expenditure. Separate equations are esti- mated, using the Stata statistical package, for the Kinh and Hoa majority and for ethnic minority households.15 In each case, we estimate a version of the equation with commune-level fixed effects and another without these effects. The regressions are weighted by the inverse of the probability that a household is sampled, and they also account for clustering and stratification of the 1998 VLSS (see Stata Corporation [1999, Vol. 4, pp.18­30]). There is clear evidence that the minority and majority regressions are structurally different, in the sense that at least some of the coefficients are not the same in the two cases. For the full dataset, a Chow test of the equal- ity of coefficients is rejected at the 1 percent level both for the case of no fixed effects (F(20,164) = 14.09) and when there are fixed effects (F(20,164) = 2.75); in the latter case, we are testing for the equality of all the coefficients except for the commune fixed effects dummies. When the sample is re- duced to those communes that include both majority and minority house- holds, the Chow test rejects the null hypothesis of equal coefficients at the 1 percent level when there are no fixed effects (F(21,18) = 6.29), but when fixed effects are included, the equality of the noncommune coefficients is rejected only at the 5 percent level (F(21,18) = 2.64). This hints at the possi- bility that much of the explanation for the differences in per capita expen- diture level between majority and minority households is due to the fixed location effects. Further evidence that the factors that influence Kinh-Hoa households differ from those that affect ethnic minority households comes from 294 Economic Growth, Poverty, and Household Welfare in Vietnam estimates of multiple adaptive regression spline (MARS) models. These models allow for nonlinearities as well as interactions among the variables in the models, but they aim to identify parsimonious sets of basis functions (Friedman 1991). Separate MARS models were estimated for Kinh-Hoa and for ethnic minority households, and these yielded very different models (see appendix 8C for details). For Kinh-Hoa households, the MARS model shows (among other things) that education has the most dramatic effect on living standards for those who have little or no land. For the ethnic minority households, the MARS model shows that the profitability of land is closely associated with complementary family labor inputs--the ethnic minorities need large families to make their land productive. By and large, the regressions in appendixes 8A and 8B accord with our prior expectations. Larger households have lower per capita expenditure levels. For both minority and majority households, an extra household member is associated with a drop in per capita expenditure of about 7 per- cent. Having a higher proportion of adults in the household also raises per capita expenditures, an effect that is significantly stronger for majority than minority households (as may be seen from the "p, eq. coeff." Column in ap- pendix 8A, which gives the p values for a test of coefficient equality; where the coefficients differ between majority and minority households, they are shown in bold face). Education, as proxied by the number of years of education of the best- educated household member who is not in school, is also a significant pre- dictor of expenditures, but the results differ depending on whether the full sample or only the sample of households in mixed communes is used. Using the full sample, the relative return to education (as measured by the per- centage change in expenditure per capita relative to a change in the numbers of years of education achieved by the best-educated household member) is higher for minority than majority households, up to seven years of educa- tion. Beyond that point, the relative return to education is slightly higher for majority households. However, when one confines the sample to only those living in mixed communes, then the relative return to education is higher for majority households.16 A plausible interpretation is that education brings a high return to ethnic minority households when they also are free to mi- grate, an effect that is best seen when using the full sample. However, when migration is limited (for legal, linguistic, institutional, or cultural reasons), it is more difficult to find profitable outlets for additional education. Thus, the efficacy of education as a way to raise the living standards of ethnic minori- ties depends fundamentally on the degree to which they are geographically mobile and are willing to become assimilated. The quality of education received by children from ethnic minority groups may also be poorer. In 1998, their curriculum was shorter and their instruction was most often in Vietnamese (a foreign language for many mi- nority children). It is plausible that minority children need to have at least several years of schooling before they are able to acquire the language and other skills needed for inclusion into the economic mainstream. Ethnic Minority Development in Vietnam: A Socioeconomic Perspective 295 Finally, when the sample is confined to households in ethnically mixed communes, access to land appears to play a bigger role, especially for mi- nority households. Minority households, when asked, tend to emphasize the importance of land as a cause of poverty (Vietnam-Sweden 2000). The re- gression results in appendix 8B help one to understand why this might be so. Confining the sample to households in ethnically mixed communes and allowing for fixed effects, an extra hectare of irrigated land is associated with additional expenditure per capita of approximately D 2 million, both for majority and for minority households. An extra hectare of irrigated land would raise the per capita expenditure of a typical Kinh-Hoa household by 13 percent, but it would boost expenditures for a minority household by 25 percent on average. It is hardly surprising, then, that ethnic minority households put more emphasis on access to land as a way out of poverty. In table 8.9, we present the main results of our decomposition analysis. This decomposes the sources of differences in per capita expenditure levels between pairs of ethnic groups into a component that is due to different characteristics (age, education, land, gender, location, and so forth) and a component that may be interpreted as reflecting different returns to charac- teristics. To interpret the table, consider the first line: the difference in pre- dicted per capita expenditures between the Kinh-Hoa majority and minority groups is D 1,173,000 (in January 1998 prices). Of this difference, 44 percent is a result of minority households having less education, fewer remittances, and other characteristics than the Kinh-Hoa majority; the remaining 56 per- cent is attributable to differences in returns to those characteristics. So if the characteristics of minority households could be boosted up to the level of the majority, then almost half of the expenditure gap would disappear. However, there would still be a substantial gap because of the lower returns to characteristics of ethnic minorities: Even if minority households had the same characteristics as the Kinh-Hoa majority, they would still be substan- tially poorer. The magnitude of the components due to different characteristics and re- turns to characteristics is substantially different, depending on which group is used as the reference and which sample is used. If the sample is confined to those communes where there are both Kinh-Hoa and minority house- holds (the mixed communes), we again find that about 45 percent of the ex- penditure per capita differential is attributable to differences in characteris- tics. However, when the equation is estimated with commune fixed effects (section 3 of table 8.9), almost two-thirds of the difference in per capita ex- penditure is due to differences in characteristics. In other words, when we compare Kinh-Hoa with minority households within a given commune, much of the gap between the groups is due to such factors as differences in education. Thus, minority households are poor in part because they lack ed- ucation and other assets, but also because they are disproportionately lo- cated in poorer communes. Only 19 of the households surveyed by the 1998 VLSS consisted of ethnic minority households in urban areas (out of a total urban sample of 1,200 296 Economic Growth, Poverty, and Household Welfare in Vietnam Table 8.9. Decomposition of the Sources of Ethnic Inequality, 1998 Per capita expenditure Percentage of of reference Percentage of difference due group difference due to different Number Reference (thousand to different returns to of obser- Location equation 1998 dong) characteristics characteristics vations All Vietnam Kinh-Hoa 2,651 44 56 5,294 Other minorities 1,478 31 69 698 All Vietnam Kinh-Hoa 2,456 45 55 993 (mixed) Other minorities 1,563 29 72 510 All Vietnam Kinh-Hoa 2,456 66 34 993 (mixed, Other fixed) minorities 1,563 54 46 510 Rural areas Kinh-Hoa 2,254 29 71 4,377 Other minorities 1,460 38 62 679 Rural areas Kinh-Hoa 2,254 28 72 4,377 Central Highlands minority 1,012 34 66 191 Rural areas Kinh-Hoa 2,254 26 74 4,377 Northern Uplands minority 1,551 16 84 402 Urban areas Kinh 4,249 -80 180 1,484 Hoa 5,426 -61 161 112 Note: For each pairwise comparison, the decomposition based on the Kinh-Hoa (or, for urban areas, the Kinh) equation is reported first, and the results based on the minority equation follow on the next line. The per capita expenditures are geometric mean values. (Mixed) = regressions based on data from communes having both minority and nonminor- ity households. (Fixed) = regressions that include community fixed effects. Source: Based on the 1998 VLSS. urban households). Thus, it may make more sense to confine the sample to rural areas and compute the Blinder-Oaxaca decomposition for this subset. The results are shown in sections 4­6 of table 8.9. For minority households overall, and for the Central Highlands minorities, about one-third of the dif- ferences in per capita expenditure is attributable to differences in character- istics such as education or age. This proportion is closer to one-fifth for Northern Uplands minority groups. Even if this group had the same charac- teristics as the Kinh-Hoa majority, four-fifths of the per capita expenditure Ethnic Minority Development in Vietnam: A Socioeconomic Perspective 297 gap would remain.17 Here, as elsewhere, we are reluctant to ascribe the dif- ferences in returns to characteristics to labor market discrimination: so few people living in the generally somewhat remote villages where minority households live are engaged in the labor market, whether or not they belong to an ethnic minority. Table 8.9 also reveals an interesting result when the living standards of the urban Kinh and the urban Chinese are decomposed. The Chinese are more affluent, but they actually have lower levels of education and other ob- servable expenditure-raising characteristics than do the Kinh. Thus, the dif- ference in per capita expenditure between the two groups is entirely due to the higher returns to characteristics that Chinese households enjoy. For- mally, our model must be missing some important, and possibly unobserv- able, determinant of expenditures: An obvious candidate is the strength of business bonds and mutual aid within the Chinese community. Whichever set of estimates is used, differential returns to characteristics appear to be central. Van de Walle and Gunewardena (2001) reached broadly similar conclusions using the 1993 VLSS, albeit with greater weight on re- turns to characteristics. We should, however, add that their results are not directly comparable with ours because van de Walle and Gunewardena used a slightly different set of regressors and excluded households living in urban areas plus the Southeast and Mekong Delta regions from their sample. Overall, this analysis points toward an important, general policy impli- cation. If our concern is to close the gap between minority and majority living standards, while maintaining ethnic identities, then it will not be suf- ficient simply to improve minority education or provide minority house- holds with more land. Such endowment-increasing measures would certainly help expand income-earning opportunities for the ethnic minori- ties. However, our decomposition analysis shows that ethnic minority households appear to generate their expenditures in quite different ways from the majority. This means that antipoverty programs that are geared to minority groups will have to be different from those geared to the majority. Summary and Conclusions We conclude by drawing together the main strands of our analysis and examining their implications for ethnic minority policies in Vietnam. Using data from the 1998 VLSS, we have shown clearly that Kinh and Hoa (major- ity) households have substantially higher living standards (as measured by per capita expenditure) than ethnic minority households. This gap is also reflected in lower school enrollment rates, higher fertility, and poorer access to health services by minority households. However, ethnic minority house- holds do not appear to be more malnourished than the population at large. The sample size of the 1998 VLSS allows a crude breakdown of the 54 ethnic groups into five broad categories: the Kinh, Hoa, Khmer, and two composite categories, the Central Highlands minorities and the Northern 298 Economic Growth, Poverty, and Household Welfare in Vietnam Uplands minorities. Based on this categorization, we find that both the Kinh and Hoa experienced rapid growth in their per capita expenditures between 1993 and 1998 and are now markedly materially better off than before. The Khmer and Northern Uplands minorities also experienced reasonable growth in per capita expenditures during the 1990s and now have expendi- tures distributions that are clustered at or just below the poverty line. This indicates that as long as economic growth is distributed equitably in the fu- ture, rapid and significant reductions in poverty are likely to be experienced by these groups in the next five years or so. In contrast, the poorest people are members of the Central Highlands minorities, whose average level of expenditure per capita has remained stagnant since 1993. For a finer disaggregation of the ethnic minorities, we turned to the 3 percent enumeration sample of the 1999 Census, where we can distinguish 12 separate ethnic groups with adequate sample sizes. The census data do not include information on expenditures or incomes, but they do allow us to compute gross and net school enrollment rates and examine patterns of in- termarriage and religious observance. Although school enrollment rates are generally high in Vietnam, they are low for the Central Highlands minorities and some of the Northern Uplands minorities (especially the Hmong). These are also the ethnic groups that are least likely to intermarry and are the most likely to be religious. Because the high intermarriage­nonreligious groups (such as the Tay and Nung and, to a lesser extent, the Thai) are also the groups where school enrollments are the highest, we hypothesize that these are the ethnic groups that have assimilated the most with the Kinh and Hoa majority. Why are ethnic minority households so poor? They may lack endow- ments (physical and human capital) or they may have low returns on their endowments, perhaps because of discrimination or for cultural or informa- tional reasons. The low endowments and returns thereon are in turn partly due to the remoteness of many ethnic minority households. To tease out the relative importance of the main effects, we estimate and decompose a set of expenditure equations. The results of these decompositions suggest that ge- ographic and cultural remoteness is important. More important, our decom- position analysis shows that even if minority households had the same en- dowments as Kinh households, this would close no more than one-third of the gap in living standards. This implies that, for some reason, minority households have a lower return to their endowments than the Kinh and Hoa majority. There are thus at least two paths to prosperity for the ethnic minori- ties. One path is to assimilate, both economically and culturally, with the majority group, and in effect obtain the same return on endowments as the majority. This is the path that some ethnic groups, such as the Tay, Nung, and Muong, appear to be following quite successfully. A second path, pursued by such groups as the Khmer and Thai (and possibly the Dao), is to integrate economically with the Kinh while retaining their own group's cultural iden- tity. However, a third group of ethnic minorities, comprising almost all the Ethnic Minority Development in Vietnam: A Socioeconomic Perspective 299 minorities that are indigenous to the Central Highlands plus the Hmong, does not appear to be benefiting from the rising living standards experi- enced by the majority. If this third group of ethnic minorities is not to be left further behind by the growth process, specific interventions need to be de- signed that are appropriate to their circumstances, needs, and aspirations. The government of Vietnam and development agencies should recognize that the interventions that work to reduce poverty among the Kinh and Hoa majority will not be effective for all other minority groups. Abstractly, the diversity of the socioeconomic development experiences of the different eth- nic groups calls for greater diversity in the antipoverty and other policy in- terventions designed to assist them. Concretely, this will require far more input from ethnic minority households, and more decentralization in an- tipoverty programs, than has occurred up to now. eq. p, coeff. 0.48 0.38 0.07 0.35 0.18 0.24 0.23 0.10 0.35 0.80 0.59 0.27 a p 0.00 0.00 0.00 0.01 0.00 0.05 0.01 0.00 0.01 0.12 0.15 0.04 effects 1998 fixed Minority Co- efficient 0.075- 0.487 0.362 0.349 0.303- 0.182- 0.272- 0.341- 0.301- 0.009 0.080 0.048- sample, p Full 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.02 0.00 0.14 Households, of Kinh-Hoa Co- efficient 0.069- 0.454 0.589 0.582 0.194- 0.134- 0.186- 0.222- 0.229- 0.008 0.092 0.021- Sample eq. Full p, coeff. 0.00 0.31 0.00 0.06 0.55 0.28 0.22 0.49 0.09 0.05 0.02 0.00 p 0.00 0.00 0.00 0.00 0.15 0.38 0.14 0.09 0.03 0.86 0.99 0.02 effects Equations, fixed Minority no Co- fficiente 0.079- 0.609 0.497 0.558 )e 0.182- 0.108- 0.178- 0.225- 0.303- 0.001- 0.001 0.066- sample, Expenditure p 0.00 0.00 0.00 0.00 0.02 0.65 0.76 0.00 0.17 0.00 0.00 0.00 of Full expenditur Kinh-Hoa Co- capita efficient 0.037- 0.475 0.958 0.864 0.105- 0.025 0.031- 0.133- 0.077- 0.016 0.185 0.068 Estimates per of log 1) 16 e is en years 16 en 1,000)÷( = Regression 16­7 over uctur over household childr ed child childr str head ee (female 8A variable variables household household household size of ages of male of female one two thr squar demographics head and and and household household head, of ee-generation of of oportion members oportion members oportion members ents ents ents Appendix (dependent Independent Household Household Pr Pr Pr Thr Par Par Par Other Age Age Gender 300 1 the 0.10 0.1 0.14 0.03 0.18 0.73 0.93 0.27 0.18 0.39 0.55 0.43 0.99 0.00 oss acr 1 0.01 0.51 0.02 0.00 0.03 0.08 0.1 0.00 0.17 0.22 0.77 0.13 0.13 0.00 ficients coef of 0.025 0.000 0.093 0.254 0.339 0.168 0.068 0.194 0.029- 0.077- 0.005- 0.007- 0.022- 7.526 0.64 698 equality the 0.02 0.00 0.00 0.00 0.01 0.00 0.05 0.67 0.09 0.36 0.00 0.56 0.94 0.00 for tests "f. 0.015 0.001 0.061 0.126 0.088 0.152 0.065 0.039 0.008- 0.006- 0.007- 0.002- 0.003 7.870 0.63 5,294 coef eq.,p" 0.03 0.00 0.86 0.00 0.38 0.85 0.08 0.18 0.01 0.56 0.47 0.01 0.08 0.00 level. cent 1 per 5 0.07 0.82 0.04 0.00 0.15 0.1 0.02 0.42 0.14 0.04 0.65 0.02 0.41 0.00 the at 12 1 ent fer 0.028 0.000- 0.1 0.392 0.173- 0.177 0.075 0.103- 0.046- 0.085 0.01 0.008- 0.021 7.146 0.44 698 dif e. 1 her 0.12 0.00 0.00 0.00 0.00 0.00 0.55 0.04 0.00 0.00 0.03 0.1 0.04 0.00 statisticallye ar shown 1 face not 0.012- 0.003 0.123 0.21- 0.295- 0.156 0.030- 0.419- 0.057 0.059 0.006- 0.009 0.182 6.794 0.32 e 5,294 bold ar in fects ef ed fixed ed ha ed squar highlighted VLSS. mittanceser 1) ed education squar = ha ed 1998 land, squar of ed land, ficients commune the household (yes ha ha land, squar squar land, coef on on ha squar education, years in eceivesr observations of land education land, land, land, land, of ficients Based adults of annual land, agricultural annual land, agricultural Pairs ce: of ennial est ennial est equations. Coef Household Maximum earsY emittancesr Note: a. Sour 2 Household Household Irrigated Other Per For Other Irrigated Other Per For Other Constant Statistics R Number two 301 eq. 1998 a p, Coeff. 0.23 0.28 0.48 0.41 0.37 0.33 0.39 0.18 0.24 0.44 0.40 0.47 effects 1 p 0.00 0.00 0.00 0.05 0.00 0.06 0.01 0.00 0.01 0.06 0.1 0.07 fixed Households, 1 Minority of sample, Co- efficient 0.084- 0.526 0.405 0.299 0.314- 0.215- 0.290- 0.382- 0.318- 0.01 0.090- 0.041- Sample commune p 0.00 0.00 0.00 0.00 0.01 0.04 0.02 0.01 0.03 0.09 0.09 0.48 Mixed Kinh-Hoa Co- Commune efficient 0.065- 0.481 0.562 0.602 0.207- 0.146- 0.196- 0.238- 0.176- 0.014 0.138- 0.02- Mixed eq. 1 1 1 p, Coeff. 0.06 0.16 0.1 0.06 0.37 0.1 0.15 0.07 0.03 0.09 0.06 0.1 effects p fixed 0.00 0.00 0.00 0.00 0.02 0.12 0.03 0.00 0.03 0.53 0.77 0.01 Equations, no Minority Co- )e sample, efficient 0.083- 0.709 0.510 0.502 0.268- 0.200- 0.256- 0.355- 0.323- 0.005 0.023- 0.075- Expenditure 1 p of commune 0.1 0.00 0.00 0.00 0.22 0.57 0.50 0.22 0.75 0.01 0.01 0.68 expenditur Mixed Kinh-Hoa capita Co- efficient 0.034- 0.480 0.882 0.939 0.138- 0.045 0.053- 0.126- 0.047 0.023 0.237 0.019 Estimates per of log 1) 16 e is en years 16 en 1,000)÷( = Regression 16­7 over uctur over household childr ed child childr str head ee (female 8B variable variables household household household size of ages of male of female one two thr squar demographics head and and and household household head, of ee-generation of of oportion members oportion members oportion members ents ents ents Appendix (dependent Independent Household Household Pr Pr Pr Thr Par Par Par Other Age Age Gender 302 1 the 0.94 0.83 0.74 0.00 0.02 0.71 0.44 0.65 0.05 0.1 0.12 0.09 0.42 0.00 oss acr 0.08 0.73 0.06 0.00 0.00 0.01 0.37 0.01 0.05 0.12 0.02 0.37 0.01 0.00 ficients coef of 0.024 0.000 0.078 0.298 0.446 0.302 0.035 0.169 0.040- 0.105- 0.146- 0.004- 0.030- 7.575 0.62 510 equality the 0.01 0.92 0.02 0.04 0.51 0.06 0.98 0.47 0.28 0.72 0.08 0.10 0.97 0.00 for tests "f. coef 0.039 0.068- 0.089 0.122 0.035 0.144 0.001- 0.142 0.012- 0.003 0.009- 0.006 0.005- 6.501 0.63 993 eq.,p" level. 0.43 0.14 0.23 0.00 0.61 0.82 0.23 0.83 0.06 0.31 0.47 0.02 0.81 0.00 cent per 5 0.12 0.83 0.13 0.00 0.14 0.18 0.12 0.30 0.18 0.03 0.41 0.07 0.13 0.00 the at ent 1 fer 0.024 0.000 0.075 0.369 0.169- 0.176 0.037 0.048 0.042- 0.082 0.049- 0.005- 0.021- 7.21 0.41 510 dif e. her 0.83 0.03 0.00 0.47 0.02 0.03 0.44 0.98 0.09 0.02 0.18 0.04 0.73 0.00 statisticallye ar shown face not e 0.004 0.002 0.171 0.060- 0.236- 0.145 0.041- 0.009- 0.024 0.042 0.007- 0.012 0.066- 6.502 0.31 993 bold ar in fects ef ed fixed ed ha ed squar highlighted VLSS. mittanceser 1) ed education squar = ha ed 1998 land, squar of ed land, ficients commune the household (yes ha ha land, squar squar land, coef on on ha squar education, years in eceivesr observations of land education land, land, land, land, of ficients Based adults of annual land, agricultural annual land, agricultural Pairs ce: of ennial est ennial est equations. Coef Household Maximum earsY emittancesr Note: a. Sour 2 Household Household Irrigated Other Per For Other Irrigated Other Per For Other Constant Statistics R Number two 303 304 Economic Growth, Poverty, and Household Welfare in Vietnam Appendix 8C Multiple Adaptive Regression Spline Models The models of expenditure presented in appendixes 8A and 8B are essen- tially linear, include a large number of variables, and do not take account of possible interactions among variables. Could one build a more parsimo- nious model? To answer this, we turned to the multiple adaptive regression spline (MARS) methodology (Friedman 1991). Given a set of variables that are specified by the researcher, MARS mines the data for nonlinearities and interactions. More specifically, it creates a piecewise linear function for each continuous independent variable, starting with too many change points (knots) and then pruning the number of knots using a backward procedure. For categorical variables, MARS arranges the categories for the best fit possible. It then looks for suitable interactions be- tween independent variables. The result is a set of basis functions, which are transformations of independent variables taking into account nonlinearities and interactions. MARS then estimates a least-squares model using the base functions as independent variables. Because the models are so nonlinear, the results are typically presented with the aid of graphs. For this study, the dependent variable is the log of real per capita expen- diture; separate MARS models were estimated for the Kinh-Hoa majority and for minority households. For the Kinh-Hoa majority, the basis functions were determined to be the following: Table 8C.1. Basis Functions for MARS Model of Log of Real Per Capita Income, Kinh and Chinese Households BF1 = max(0, IRRLAND - 131.000); BF12 = max(0, 120.000 - NIRRLAND) ; BF2 = max(0, 131.000 - IRRLAND); BF13 = max(0, HEADAGE - 43.000); BF3 = max(0, WORKED98 - 6.000) × BF2; BF14 = max(0, 43.000 - HEADAGE); BF4 = max(0, 6.000 - WORKED98) × BF2; BF15 = max(0, WORKED98 + .258859E - 06) BF5 = max(0, HHSIZE - 6.000); × BF12; BF6 = max(0, 6.000 - HHSIZE); BF16 = (REMIT = 0) × BF12; BF7 = max(0, NIRRLAND - 400.000) × BF2; BF18 = max(0, PELAND + .186998E - 04) BF8 = max(0, 400.000 - NIRRLAND) × BF2; × BF1; BF9 = max(0, PAGE17M - 0.250) × BF5; BF19 = max(0, PELAND - 300.000) × BF2; BF10 = max(0, 0.250 - PAGE17M) × BF5; BF20 = max(0, 300.000 - PELAND) × BF2. BF11 = max(0, NIRRLAND - 120.000); Note: IRRLAND = area of irrigated land, in m2. WORKED98 = years of education achieved by head of household. HHSIZE = number of household members. NIRRLAND = area of nonirrigated annual land, in m2. PAGE17M = proportion of household consisting of males 17 and older. HEADAGE = age of head of household. REMIT = value of remittances received by household. PELAND = area of land planted in tree crops. The final model for Kinh-Hoa was: Y = 7.471 + .139911E - 04 × BF1 - 0.001 × BF2 + .282305E - 03 × BF3 - .198950E - 03 × BF4 - 0.045 × BF5 + 0.120 × BF6 + .389904E - 07 × BF7 + .633461E - 05 × BF8 + 0.175 × BF9 - 0.690 × BF10 + .438774E - 05 × BF11 - .435589E - 03 × BF12 - 0.004 × BF13 - 0.017 × BF14 + .219940E - 03 × BF15 - .761529E - 03 × BF16 + .212414E - 08 × BF18 + .607371E - 07 × BF19 + .388523E - 05 × BF20. Ethnic Minority Development in Vietnam: A Socioeconomic Perspective 305 This ordinary least squares model had an adjusted R2 of 0.43, much bet- ter than the R2 of 0.31 that we found for the model in appendix 8A. The MARS model achieves this with just eight variables (see table 8C.1) and so helps one to focus just on the essential elements. Three insights emerge, which can best be explained with the help of the graphs in figure 8C.1. First, as the age of the household head rises to 43, households become better off; after that, older heads are associated with poorer households (panel 1 in figure 8C.1). Second, more land planted with annual crops ("annual land"), both irrigated and unirrigated, is associated with higher per capita income. Only for households with no land does the educational level of the household head have an important effect on income, Figure 8C.1. Results of the Multiple Adaptive Regression Spline Model for Kinh-Hoa Households Curve 1: Pure Ordinal Surface 1: Pure Ordinal 0.5 IRRLAND 20 50000 0.4 WORKED9810 0 0.3 0 CONTRIBUTION 0.2 1.0 1.0 0.1 0.5 0.5 0 050000 0 10 20 30 40 50 60 70 80 90 100 20 CONTRIBUTION HEADAGE IRRLAND 10 0 0 WORKED98 Surface 2: Pure Ordinal Surface 3: Pure Ordinal NIRRLAND 1E+005 20 HHSIZE 1.0 IRRLAND 10 50000 PAGE17M0.5 0 0E+000 0 0 CONTRIBUTION 1.2 CONTRIBUTION 1.2 1.0 0.8 1.0 0.8 0.4 0.5 0.4 0.5 0 01E 0 020 1.0 10 CONTRIBUTION NIRRLAND +005 50000 CONTRIBUTION HHSIZE 0.5 0E 0 +000 0 0 IRRLAND PAGE17M Note: HHSIZE = number of household members. WORKED98 = years of education achieved by head of household. IRRLAND = area of irrigated land, in m2. HEADAGE = age of head of household. PAGE17M = proportion of household consisting of men aged 17 or older. NIRRLAND = area of nonirrigated annual land, in m2. Source: 1998 VLSS. 306 Economic Growth, Poverty, and Household Welfare in Vietnam suggesting that more education (and perhaps a move to an urban area) might be a substitute for more land (panels 2 and 3 in figure 8C.1). This raises the intriguing possibility that as population pressure leads to greater scarcity of land, there will be a stronger incentive to acquire more education, which in due course will increase the opportunities that emerge in an in- creasingly urban and nonagricultural society. Third, as household size rises, households are poorer (as measured by per capita expenditure); however, for larger households, this effect is moderated if there is a high proportion of adult males (panel 4 in figure 8C.1). The MARS model for minority households looks quite different, al- though many of the same variables come into play. The basis functions, and subsequent model, are as shown in table 8C.2. Table 8C.2. Basis Functions for MARS Model of Log of Real Per Capita Income, Minority Households BF1 = max(0, HHSIZE - 8.000); BF11 = max(0, 48.000 - HEADAGE) × BF3; BF2 = max(0, 8.000 - HHSIZE); BF13 = max(0, 11080.000 - FLAND) × BF6; BF3 = max(0, WORKED98 - 5.000); BF14 = max(0, PAGE0716 + .120596E - 07) × BF6; BF4 = max(0, 5.000 - WORKED98); BF15 = max(0, OTHELAND - 6500.000) × BF4; BF6 = max(0, 18000.000 - IRRLAND); BF16 = max(0, 6500.000 - OTHELAND) × BF4; BF7 = max(0, PELAND + .252347E - 04) BF18 = max(0, 360.000 - OTHELAND) × BF2; × BF6; BF19 = max(0, NIRRLAND - 3994.000); BF8 = max(0, HHSIZE - 3.000) × BF6; BF20 = max(0, 3994.000 - NIRRLAND); BF9 = max(0, 3.000 - HHSIZE) × BF6; Note: HHSIZE = number of household members. WORKED98 = years of education achieved by head of household. IRRLAND = area of irrigated land, in m2. PELAND = area of land planted in tree crops. HEADAGE = age of head of household. FLAND = area of forest land op- erated by household. PAGE0716 = proportion of household ages 7 to 16. OTHELAND = area of land in other uses (that is, not annual, perennial, or forest). NIRRLAND = area of nonirrigated annual land, in m2. Y = 7.349 - 0.061 × BF1 + 0.193 × BF2 + 0.046 × BF3 - 0.137 × BF4 - .486490E - 04 × BF6 + .120914E - 08 × BF7 + .344804E - 05 × BF8 + .101408E - 04 × BF9 - 0.002 × BF11 - .723555E - 09 × BF13 + .171787E- 04× BF14+ .252554E- 05× BF15+ .163910E- 04× BF16 -.987590E- 04×BF18+.194643E-04×BF19+.305935E-04×BF20. In this case, the fit of the MARS model (R2 = 0.46) is close to that of the conventional model (R2 = 0.44), but the MARS model is more parsimonious. Not surprisingly, the more land households have under irrigation or peren- nial crops, the better off they are (panel 1 in figure 8C.2); irrigated land has a particularly large effect on per capita consumption levels. However, the ability to use irrigated land profitably requires complementary labor inputs, particularly from the household, as panel 2 in figure 8C.2 shows. The third panel in figure 8C.2 shows classic age and educational effects. Reading along the age axis, one sees that income rises quickly, reaching a plateau once the head of the household is 48 years old. The effect is particu- larly pronounced for households with highly educated heads. Looking along the education axis, we see that more education is associated with higher living standards. Finally, the presence of a high proportion of Ethnic Minority Development in Vietnam: A Socioeconomic Perspective 307 Figure 8C.2. Results of the Multiple Adaptive Regression Spline Model for Minority Households Surface 1: Pure Ordinal Surface 2: Pure Ordinal 40000IRRLAN PELAND50000 20000 D HHSIZE 40000IRRLAND 10 20000 0 0 0 0 CONTRIBUTION 0.8 CONTRIBUTION 1.5 0.8 1.5 0.6 1.0 0.6 0.4 1.0 0.4 0.2 0.5 0.5 0.2040000 0 0 0 40000 IRRLAND 50000 10 20000 CONTRIBUTION IRRLAND20000 CONTRIBUTION 0 0 PELAND 0 0 HHSIZE Surface 3: Pure Ordinal Surface 5: Pure Ordinal HEADAGE IRRLAND 80 40000 15 60 WORKED98 10 PAGE0716 20000 40 0.5 5 20 0 0 0 CONTRIBUTION CONTRIBUTION 0.8 1.0 0.8 1.0 0.6 0.6 0.4 0.5 0.4 0.5 0.2 0 080 0 0.2040000 15 HEADAGE 60 IRRLAND20000 0.5 CONTRIBUTION 40 10 CONTRIBUTION 20 5 0 0 0 PAGE0716 WORKED98 Note: HHSIZE = number of household members. WORKED98 = years of education achieved by head of household. IRRLAND = area of irrigated land, in m2. PELAND = area of land planted in tree crops. HEADAGE = age of head of household. PAGE0716 = propor- tion of household ages 7 to 16. Source: 1998 VLSS. adolescents in the household appears to be associated with a slightly lower return on irrigated land (panel 4 in figure 8C.2). Notes 1. The Hoa make up approximately 2 percent of the population of Vietnam, live predominantly in urban areas, and, as will be shown below, are highly assimilated with the Kinh. 2. The 1992­93 survey spanned a full year, starting in October 1992; the 1997­98 survey began in December 1997 and also lasted a year. For brevity's sake, reference is made to the surveys as the 1993 VLSS and 1998 VLSS, respectively. 308 Economic Growth, Poverty, and Household Welfare in Vietnam 3. In conformity with usual academic usage, we use "assimilate" to mean the selective and voluntary adoption by minority groups of the economic strategies, livelihood practices, and cultural norms common among the majority group. The adoption of such strategies, practices, and norms is selective because they need to be compatible with the socioeconomic conditions of the minorities, and it is voluntary because the decision on whether to adopt them is made, usually on an individual or household basis, by the ethnic minorities themselves. As such, our usage of the word "assimilate" would best translate into Vietnamese as hoa dong or hoa nhap. 4. The full 1998 VLSS sample also included 140 communes with only Kinh or Hoa households and 6 communes where only ethnic minorities were surveyed. 5. The prices are those of January 1998. The exchange rate in January 1998 was D 12,290 to US$1. 6. The vaccinations are for tuberculosis, diphtheria-pertussis-tetanus, polio, and measles. 7. Primary school in Vietnam extends for five years, from roughly the ages of 6 through 10, although eligibility to attend primary school is determined on the basis of the calendar year of a child's birth and not on his or her age. 8. Only 10 of the 334 primary schools surveyed in the 1998 VLSS taught any lessons in ethnic minority languages. Of these 10 primary schools, 7 were in the Mekong Delta or Southeast. 9. Lower secondary school covers four years, from approximately age 11 until age 14. 10. These calculations assume monogamous marriages (de facto or de jure). Polygamy is known to have been common among affluent members of certain eth- nic groups (such as the Kinh and the Hmong) in the past, but it is now officially prohibited. None of the households enumerated in the 1999 Census recorded poly- gamous marriages. 11. Eleven percent of Tay husbands are married to Kinh wives (compared with 4.8 percent with Nung wives), and 11.2 percent of Nung husbands have Kinh wives (and a further 11.2 percent have Tay wives). Among the Dao, of the 6.2 percent of husbands who intermarry, 3.8 percent are married to Kinh women and a further 1.5 percent are married to Tay, Hoa, or Muong wives. 12. Note that the census included two questions on religion, the first asking if an individual follows a religion and a follow-up question inquiring if he or she prac- tices this religion. Table 8.6 is based on responses to the first question. 13. Examples: United Nations Development Programme (UNDP)­supported Ethnic Minority Development project (VIE/94/013­VIE/96/010), UNDP­ International Fund for Agricultural Development-supported Ha Giang Develop- ment Project for Ethnic Minorities (VIE96/027), Swedish International Development Authority (SIDA)­supported Minority Rural Development Project, SIDA-supported Vietnam-Sweden Inter-Forest (social forestry) Project, and UNDP Regional Project-- Highland People. 14. In 1992, rural Kinh and Hoa households cultivated an average of 724 m2 of "good quality" land, of which 615 m2 was irrigated; for ethnic minorities, the overall figure was 178 m2, of which just 62 m2 per household was irrigated. "Good quality" land is defined as land that yields four tonnes or more of paddy (or equivalent) per hectare per year. 15. We also estimated separate equations for urban Kinh and urban Hoa and for rural Kinh and rural Hoa, rural Khmer, rural Central Highlands minorities, and rural Northern Uplands minorities. The detailed results are not reported here, but were used in the decompositions reported in table 8.7. Ethnic Minority Development in Vietnam: A Socioeconomic Perspective 309 16. A similar effect was found by van de Walle and Gunewardena (2001) using the 1993 VLSS. 17. The Khmer have been excluded from our decomposition analysis because of the small number (95) of Khmer included in the 1998 VLSS, as well as problems of missing data for some of the Khmer households that were sampled. Bibliography The word processed describes informally reproduced works that may not be commonly available through libraries. Asian Minorities Outreach. 1998. The Peoples of Vietnam. Palestine, Texas: Asian Harvest. Blinder, Alan S. 1973. "Wage Discrimination: Reduced Form and Structural Estimates." Journal of Human Resources 8: 436­55. Cohen, Margot. 2001. "Robbing the Poor." Far Eastern Economic Review (Jan- uary 25): 28. Dang, Nghiem Van, Chu Thai Son, and Luu Hung. 2000. Ethnic Minorities in Vietnam. Hanoi: The Gioi Publishers. Deaton, Angus. 1997. The Analysis of Household Surveys: A Microeconometric Approach to Development Policy. Baltimore: Johns Hopkins University Press. Desai, Jaikishan. 2000. "Vietnam through the Lens of Gender: Five Years Later--Results from the Second Vietnam Living Standards Survey." Food and Agriculture Organization of the United Nations, Regional Office for Asia and the Pacific. Processed. The Economist. 2001 (April 7). "The Golden State Turns Brown." p. 36. Economist Intelligence Unit. 2001. Vietnam: Country Report--April 2001. London. Friedman, Jerome. 1991. "Multiple Adaptive Regression Splines." Annuals of Statistics 19 (March): 1­141. Hansen, Mette Halskov. 1999. Lessons in Being Chinese: Minority Education and Ethnic Identity in Southwest China. Seattle: University of Washington Press. Haughton, Dominique, and Jonathan Haughton. 1999. "Son Preference." In Dominique Haughton, Jonathan Haughton, Sarah Bales, Truong Thi Kim Chuyen, and Nguyen Nguyet Nga, eds., Health and Wealth in Vietnam: An Analysis of Household Living Standards. Singapore: Insti- tute for Southeast Asian Studies Press. Huy, Nguyen Van, and Le Huy Dai. 1999. Bao cao tong hop ve nghien cuu chinh sach phat trien vung mien nui va dan toc thieu so [Report on a Study of 310 Economic Growth, Poverty, and Household Welfare in Vietnam Development Policy for Mountainous and Ethnic Regions]. VIE 96/010- PS. Hanoi: United Nations Development Programme. Jamieson, Neil, Le Trong Cuc, and Terry Rambo. 1998. The Development Crisis in Viet Nam's Mountains. Honolulu: East-West Center. Minot, Nicholas, and Bob Baulch. 2004. "The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting." In P. Glewwe, N. Agrawal, and D. Dollar, eds., Economic Growth, Poverty, and Household Welfare in Vietnam. Washington, D.C.: World Bank. (Originally pub- lished for the International Food Policy Research Institute in 2001.) Oaxaca, Ronald I. 1973. "Male-Female Wage Differentials in Urban Labor Markets." International Economic Review 9: 693­709. Pham, Nguyen Quoc Pham, and Trinh Quoc Tuan. 1999. May van de ly luan va thuc tien ve dan toc va quan he dan toc o Vietnam [Some Theoretical and Practical Issues on Ethnicity and Ethnic Relations in Vietnam]. Hanoi: Nha xuat ban Chinh Tri Quoc Gia. Stata Corporation. 1999. Stata Reference Manual: Release 6. College Station, Texas: Stata Press. United Nations Economic and Social Council. 1998. "Report by Abdelfattah Amor, Special Rapporteur." Item 11, 55th Session of the United Nations Commission on Human Rights. New York. van de Walle, Dominique. 2004. "The Static and Dynamic Incidence of Viet- nam's Public Safety Net." In P. Glewwe, N. Agrawal, and D. Dollar, eds., Economic Growth, Poverty, and Household Welfare in Vietnam. Washington, D.C.: World Bank. van de Walle, Dominique, and Dileni Gunewardena. 2001. "Sources of Ethnic Inequality in Vietnam." Journal of Development Economics 65(1): 177­207. Viet Nam Living Standards Survey (VNLSS). 1992­1993. Web site: www.worldbank.org/lsms/country/vn93/vn93bid.pdf. Vietnam Living Standards Survey (VLSS). 1997­98. Web site: www. worldbank.org/lsms/country/vn98/vn98bif.pdf. Vietnam Poverty Working Group. 1999. Vietnam Development Report 2000: Attacking Poverty. Joint report of the Government-Donor-NGO Working Group. Hanoi: World Bank. Vietnam-Sweden Mountain Rural Development Project. 2000. Lao Cai: A Participatory Poverty Assessment. Hanoi: World Bank. Winrock International. 1996. "Ethnic Minorities in Vietnam: A Country Profile." Study prepared for the World Bank. Hanoi. Processed. Xinhua News Agency. 2001. "Bulletin." (March 24). Part III Progress in Health and Education in Vietnam in the 1990s 9 Poverty and Survival Prospects of Vietnamese Children under Doi Moi Adam Wagstaff and Nga Nguyet Nguyen In the mid-1980s, under a policy known as Doi Moi ("renovation"), Vietnam started dismantling its command economy, creating in its place a market- oriented domestic economy and opening its doors to international trade, foreign direct investment, and development assistance. Under Doi Moi, Vietnam has achieved impressive rates of economic growth (5­10 percent a year) and has reduced levels of absolute poverty (Glewwe, Gragnolati, and Zaman 2000). Doi Moi was also accompanied by a reduction in the scale and quality of public health services (at least in some areas), the introduction of user fees, and the encouragement of a private health sector (World Bank and others 2001). Some have expressed concerns that these developments may be having damaging effects on health outcomes, especially among poorer households (Dahlgren 2000). The broad issue this chapter addresses is how one dimension of human development--child survival--has changed under Doi Moi. In doing so, special attention is paid to the situation facing poor households. · By international standards, and especially given its relatively low per capita income, Vietnam has achieved substantial reductions in, and low levels of, the infant mortality rate (IMR) and under-five mortality rate (U5MR). By the mid-1980s, its rates were among the lowest in the developing world. The first issue this chapter will address is, has this reduction in child mortality been sustained under Doi Moi? The Viet- namese government's own goal was to reduce the IMR to 30 per 1,000 (that is, 30 per 1,000 live births) by the year 2000. There is some debate over the exact level of the IMR in Vietnam and hence over whether the government's goal for 2000 has been achieved. The evidence on recent trends in child mortality is assessed from a variety of sources, and some new estimates are derived from the 1993 and 1998 Vietnam Living Standards Surveys (VLSSs).1 313 314 Economic Growth, Poverty, and Household Welfare in Vietnam · The second question is, what are the socioeconomic differentials in child survival? Previous research by one of the authors (Wagstaff 2000) suggests that, by international standards, inequalities in IMR and U5MR between poor and better-off children were extremely low in Vietnam. These results were based, however, on VLSS data cover- ing survival and deaths among children over the period 1984­93. It is quite possible--and indeed the fear has been expressed by some-- that gaps in child survival prospects between the poor and better-off may have started to widen in Vietnam. In this chapter, data from the 1998 VLSS will be used to measure the extent of socioeconomic in- equalities in child mortality over the period 1989­98. By comparing inequalities over the periods 1984­93 and 1989­98, it can be seen if any recent reductions in child mortality have been evenly spread across the population, or whether the gains have been concentrated on better-off children. · The third question to be addressed is, what factors have caused the recent changes in child mortality and any changes in inequalities in child mortality? What role, for example, has economic growth played? Have changes in the health care sector affected child survival prospects, and if so to what extent? Have changes equally affected all groups? These questions may be answered by estimating child sur- vival models that link child survival to the underlying determinants of survival, such as the child's age and gender, the education level of the mother, the income of the household, its water and sanitation, and variables capturing the success of local health services in deliv- ering key maternal and child health interventions (immunization and prenatal care). A decomposition analysis is then undertaken to see how different factors contribute to the overall decline in mortality and the changing inequalities. · Finally, how are child survival prospects likely to evolve over the next 15 years? Vietnam is currently preparing a Poverty Reduction Strategy Paper (PRSP) to qualify for continued development assis- tance through the International Development Association. In the PRSP, countries not only set out their poverty reduction strategies but also commit themselves to achieving targets for poverty and human development indicators.2 The targets being set by other countries include the IMR and U5MR, because these are among the seven key Millennium Development Goals (MDGs).3 The latter in- volve, among other things, a reduction by all countries of the IMR and U5MR by two-thirds by 2015. This chapter addresses the issue of whether such a reduction is possible for Vietnam and whether commitment to such a path would be realistic for Vietnam's PRSP. To do this, the survival model described briefly above is used and projected forward to 2015, making assumptions about income growth, as well as the evolution of water and sanitation, and health services. Poverty and Survival Prospects of Vietnamese Children under Doi Moi 315 The section that follows this brief background reviews the evidence on recent trends in child survival at the population level. The next section, "Changing Inequalities in Child Mortality under Doi Moi," explores trends in socioeconomic inequalities in child survival. The following section out- lines the empirical model used in the remainder of the chapter. The section titled "Causes of the Recent Changes in Child Survival in Vietnam" presents and discusses the parameter estimates and then uses these parameter esti- mates to explain recent changes in the levels of and inequalities in child mortality. The next section uses the model to explore likely trends in child survival up to 2015--the year by which the MDGs are supposed to be reached. Finally, this chapter concludes with a discussion of the issues raised by the findings and discusses some policy options. Recent Trends in Child Mortality in Vietnam A number of sources are available for estimation of recent trends in the IMR and U5MR in Vietnam. Published estimates include those based on the 1989 Census, the 1994 Inter-Censual Demographic Survey (ICDS), the 1997 United States Agency for International Development­funded Demographic and Health Survey (DHS) (Government of Vietnam 1999), the 1999 Census (Central Census Steering Committee 2000), and the 2000 United Nations Children's Fund (UNICEF) Multiple Indicator Cluster Survey (MICS) (Government of Vietnam 2000).4 In addition, fertility histories were col- lected in the 1993 and 1998 VLSSs, and IMR and U5MR are reported based on these surveys. In the censuses, the ICDS, and the MICS, incomplete fer- tility histories were collected--women of fertile age were asked how many children they had ever given birth to and how many of these were still alive, but not the dates of birth and death (if applicable) of these children. Such data require that mortality estimates be estimated using the so-called indi- rect estimation method, which involves taking the data on fertility and proportions of children surviving and superimposing them on model life tables. In the DHS and VLSSs, by contrast, complete fertility histories were obtained--women reported when each child was born and if and when any of these children died. With such data, mortality estimates can be obtained using standard life table methods (the direct method of mortality estima- tion) and standard errors can be computed for the estimates. The complete fertility history is clearly more demanding in terms of recall than the incom- plete fertility history, but it potentially leads to more accurate estimates. Of course, because the data also contain the information collected in the incom- plete fertility history, the indirect method can also be used, so that the direct and indirect estimates for the DHS and VLSSs can be compared. Indirect Estimates of Infant and Under-Five Mortality in Vietnam This section compares the published estimates from the 1989 and 1999 Cen- suses, the 1994 ICDS, and the 2000 MICS with indirect estimates computed 316 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 9.1. Indirect Estimates of the Infant Mortality Rate IMR per 1,000 live births 70 60 50 40 30 45­49 age group 20 25­29 age group 10 0 1975 1980 1985 1990 1995 2000 Reference year 1993 VLSS 1993 VLSS adjusted 1989 Census from DHS 97 table 3.5 2000 MICS 1998 VLSS weighted 1994 ICDS 94 from DHS 97 table 3.5 1997 DHS 1999 Census Note: DHS = Demographic and Health Survey. ICDS = Inter-Censual Demographic Survey. IMR = infant mortality rate. MICS = Multiple Indicator Cluster Survey. Sources: VLSS: authors' calculations; other surveys: as indicated in legend. from the 1997 DHS and the 1993 and 1998 VLSSs. These indirect estimates are then compared with estimates obtained using the direct method from the DHS and the two VLSSs. Figures 9.1 and 9.2 show the IMR and U5MR obtained from the censuses and the household surveys using the indirect method (United Nations 1983). This involves superimposing on model life tables the data on live births and deaths. Estimates were obtained using the computer program QFIVE (United Nations 1983). (See appendix 9A for the full technical details and assumptions made in their derivation.) With the indirect method, separate mortality estimates are obtained for women in different age groups; the age bands typically used in these exercises are 15­19, 20­24, 25­29, 30­34, 35­39, 40­44, and 45­49. Obviously, the rates among the older age groups capture mortality rates in earlier periods than those among the younger age groups. Typically, the rates among the youngest group--and often those among the second-youngest group as well--are ignored on the grounds that they reflect the higher risks associated with pregnancies among younger women. One striking feature of both figures is the high level and odd pattern to the estimates derived from the ICDS. Hill and others (1999) noted this in Poverty and Survival Prospects of Vietnamese Children under Doi Moi 317 Figure 9.2. Indirect Estimates of the Under-Five Mortality Rate U5MR per 1,000 live births 120 100 80 60 40 45­49 age group 20 25­29 age group 0 1975 1980 1985 1990 1995 2000 Reference year 1993 VLSS 1993 VLSS adjusted 1989 Census from DHS 97 table 3.5 2000 MICS 1998 VLSS weighted 1994 ICDS 94 from DHS 97 table 3.5 1997 DHS 1999 Census Note: DHS = Demographic and Health Survey. ICDS = Inter-Censual Demographic Survey. MICS = Multiple Indicator Cluster Survey. U5MR = under-five mortality rate. Sources: VLSS: authors' calculations; other surveys: as indicated in legend. their analysis of trends in child mortality in Vietnam and decided to discard the ICDS-based estimates. The 1989 Census also gives some rather implau- sible numbers for the 25­29 and 30­34 age groups. The other estimates give a clear picture of declining IMR and U5MR, with little sign of any stagnation in recent years. There is clearly some uncertainty, however, over the actual levels of the IMR and U5MR. Direct Estimates of Infant and Under-Five Mortality Rates in Vietnam The mechanics of the direct method are illustrated in table 9.1 for data from the 1998 VLSS, where--for the purpose of the table--children born in the previous 10 years were included. This resulted in 5,316 children being se- lected. The first row of column 4 indicates that 195 children were withdrawn during the first six months (the interval used in this example), meaning that 195 children were born within the six months before the survey and there- fore had fewer than six months of exposure. The assumption is that these 195 children were, on average, exposed for only half of the six months,5 so that the total number of children exposed during the first six months was 11 rate Hazard 0.044 0.006 0.005 0.002 0.007 0.000 0.007 0.000 0.002 0.000 or 10 std. err Relative 0.0917 0.0887 0.0842 0.0819 0.0794 0.0794 0.0749 0.0749 0.0754 0.0754 or 9 err cumul. oportion 0.0020 0.0022 0.0023 0.0023 0.0025 0.0025 0.0026 0.0026 0.0027 0.0027 Std. of pr surviving 8 rate q 0x 21.8 24.8 27.3 28.1 31.5 31.5 34.7 34.7 35.8 35.8 Mortality end Cumul. oportionrp at 0.9782 0.9752 0.9727 0.9719 0.9685 0.9685 0.9653 0.9653 0.9642 0.9642 surviving Survey 6 411 15 12 4 15 0 13 0 4 0 Deaths (number) Standards 5 risk Living Exposed to 5,219 4,928 4,745 4,550 4,352 4,144 3,936 3,696 3,466 3,203 (number) ietnam V 47 195 159 176 190 198 189 226 228 232 286 ithdrawn during interval 1998 W (number) the 3 from Entering interval 5,316 5,007 4,833 4,645 4,451 4,238 4,049 3,810 3,582 3,346 (number) ableT of 2 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 Life End interval 9.1. of 1 ableT 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 Start interval 318 Poverty and Survival Prospects of Vietnamese Children under Doi Moi 319 5,316, less half of 195, or 5,219 (column 5). Of the 5,316 born during the pre- vious 10 years, 114 died during the first six months, so that the proportion surviving was (5,219 - 114)/5,219, or 0.9782. The mortality rate for the first six months, 0.5 0 q , is equal to (1 - 0.9782) × 1,000, or 21.8 per 1,000 births. The number of children starting the second six months of life is 5,316 - 195 - 114, or 5,007 (see the second row of table 9.1). Of these, 159 are exposed fewer than six months--they were born less than one year before the survey. Of the 4,928 children exposed to the risk of death in the second half of their first year of life, 15 died before their first birthday, giv- ing a cumulative proportion of children surviving from birth to their first birthday of 0.9752. Column 7 thus shows the survival function, S(t). Thus, the IMR (an infant being a child younger than one year old) is 24.8 per 1,000 live births. Column 9 gives the standard error of the cumulative proportion surviving or, equivalently, the standard error of the mortality rate from birth to the end of the interval in question. Column 10 gives this expressed as a proportion of the mortality rate. The final column shows the hazard rate, (t)--the rate at which S(t) decreases over time, -dlnS(t)/dt. Finally, the bot- tom row of column 8 gives the U5MR, q , which in this case is 35.8 per 1,000 5 0 live births, with a standard error equal to 7.5 percent of the mortality rate. Figure 9.3 compares the IMR and U5MR estimates from the 1993 and 1998 VLSSs and the 1997 DHS. In each case, for ease of comparison, the rates shown are those computed using children born in five-year intervals. The markers in the chart are placed at the year in the middle of the five-year in- terval. In the case of the DHS, all the rates reported were computed using data in the 1997 DHS,6 whereas in the case of the VLSS, the rates are shown for the corresponding VLSS. Thus, the rate for 1990 is computed on births recorded in the 1993 VLSS over the preceding five years, and the rate for 1995 is computed on births recorded in the 1998 VLSS in the five years be- fore the survey.7 Two points are worth making here. First, the DHS and VLSS, once again, show a fair amount of consistency, especially the rates computed from the 1998 VLSS, which are well in line with simple projections from the DHS es- timates. Second, the directly estimated rates are somewhat lower than the majority of the indirectly estimated rates, including those from the DHS and VLSS themselves. The direct estimates are, however, close to the unadjusted 1999 Census-based estimates shown in figures 9.1 and 9.2. Key Points The first key point to emerge from this discussion is that the IMR and U5MR appear to have continued to fall under Doi Moi--there is no sign in the data of any decrease in the rate of reduction. The second key point is that the IMR would appear to have fallen below the target figure for 2000 of 30 per 1,000. Indeed, the evidence suggests that this target was probably reached in the mid-1990s, and the figure now may well be around 25 per 1,000 or even below that number. 320 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 9.3. Direct Estimates of Infant and Under-Five Mortality Rates Mortality rate per 1,000 live births 60 50 40 30 20 10 0 1982 1984 1986 1988 1990 1992 1994 1996 1997 DHS IMR 1997 DHS U5MR 1993 VLSS MR 1993 VLSS U5MR 1998 VLSS MR 1988 VLSS U5MR Sources: VLSS: authors' calculations; other surveys: as indicated in legend. Changing Inequalities in Child Mortality under Doi Moi Previous research by one of the authors (Wagstaff 2000) suggests that by in- ternational standards, inequalities in the IMR and U5MR between poor and better-off children were extremely low in Vietnam. These results were based, however, on VLSS data covering survival and deaths among children over the period 1983­92. The question is whether under Doi Moi a gap in survival prospects between poor and better-off children has opened up. Have the ap- parent continued gains in child survival under Doi Moi been spread equally across the population? Data and Methods There are two sources of data that shed light on this issue. First, the earlier analysis of the 1993 VLSS can be replicated using the 1998 VLSS data. As in the earlier VLSS analysis, deaths over the preceding 10 years are included in the analysis of the 1998 VLSS data.8 Households are ranked by equivalent household consumption in the analysis of the 1993 and 1998 VLSS data, and then the children born during the preceding 10 years are sorted into quin- tiles. These quintiles are therefore quintiles of live births. Quintile-specific life tables are thus produced for each year, and thereby quintile-specific IMR and U5MR are obtained. Second, these two sets of VLSS results can be compared with the DHS- based results reported by Gwatkin and others (2000). These show IMR and Poverty and Survival Prospects of Vietnamese Children under Doi Moi 321 U5MR for different quintiles of wealth, where the wealth measure is obtained by means of a principal components analysis on a variety of indicators of household living standards9 along the lines proposed by Filmer and Pritchett (1999). A point to bear in mind is that the quintiles in Gwatkin and others (2000) are quintiles of households, not of live births. This makes comparison of the VLSS and DHS results somewhat awkward, because the lower-wealth groups have higher fertility rates and hence have a share of live births that is in excess of their population share. This problem can be overcome, however, using concentration curves (Kakwani, Wagstaff, and van Doorslaer 1997; Wagstaff, Paci, and van Doorslaer 1991). In this case, these are formed by ranking live births by the living standards of the child's household and then plotting the cumulative percentage of live births so ranked on the horizontal axis and the cumulative percentage of deaths (infant or under-five) on the vertical axis. If deaths are concentrated among poorer households, the resul- tant curve--the concentration curve--will lie above the diagonal, or line of equality. The farther above the line of equality it lies, the greater the degree of concentration of deaths among poorer households. Results on Trends in Socioeconomic Inequalities Figure 9.4 shows, for the 10-year period preceding each of the three surveys, the U5MR for each quintile of living standards. Over the period 1983­92, Figure 9.4. Trends in Under-Five Mortality, by Consumption Quintile U5MR per 1,000 live births 70 60 50 40 30 20 10 0 1 2 3 4 5 Quintiles 1993 VLSS 1997 DHS 1998 VLSS Note: U5MR = under-five mortality rate. Sources: VLSS: authors' calculations; DHS: Gwatkin and others (2000). 322 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 9.5. Concentration Curves for Under-Five Mortality Under-five deaths (cumulative percentage) 100 80 60 40 20 0 0 20 40 60 80 100 Cumulative percentage of births, ranked by household living standards Equality 1993 VLSS 1997 DHS 1998 VLSS Sources: VLSS: authors' calculations; DHS: Gwatkin and others (2000). there was very little difference between the survival prospects of poor and better-off children in Vietnam. By contrast, over the periods 1987­96 and 1988­97, there were marked inequalities between the poor and better-off. The VLSS data are more directly comparable, because in each year the quintiles are quintiles of live births, and the ranking variable is the same (equivalent consumption). The VLSS results suggest that the national reductions in child mortality uncovered in the previous sections have not been spread evenly across the Vietnamese population--the upper socioeconomic groups have seen appreciable reductions in child mortality, but the poorer Vietnamese children have seen little improvement in their survival prospects in recent years. The problem of the DHS data being quintiles of households can be overcome by using the concentration curve device. Figure 9.5 shows the con- centration curves for the U5MR; the nonalignment of the markers for the DHS data with the quintile cutoffs reflect the higher fertility of poorer households. It is striking how close the 1997 DHS and the 1998 VLSS concentration curves are to one another and how far they are from the curve for the 1993 VLSS. Key Points There are just two key points that can be inferred from this discussion. First, both the 1997 DHS data and the 1998 VLSS data indicate that inequalities in Poverty and Survival Prospects of Vietnamese Children under Doi Moi 323 child survival between poor and less poor children now exist in Vietnam. Second, these inequalities appear to be a recent phenomenon--they were not evident in the 1993 VLSS data. Under the recent years of Doi Moi, reductions in child mortality have not been spread evenly, being heavily concentrated among the better-off. Poorer Vietnamese children do not appear to have seen any appreciable improvement in their survival prospects in recent years. Modeling Child Survival in Vietnam The estimates discussed above suggest that the IMR and U5MR continued to fall during the past decade in Vietnam at the population level, but these improvements were heavily concentrated among better-off children. This section will outline and estimate a model intended to help explain these dif- ferential changes in survival prospects. The Basic Model Child survival could be modeled using either a production function (an equa- tion linking survival to the proximate determinants of survival, such as usage of health care, dietary and sanitary practices, and so on) or a reduced-form or demand equation for child survival (an equation linking survival to the vari- ables that influence households in the choices they make over the proximate determinants of survival) (Schultz 1984). The authors in this case opt for the demand equation and thus link survival to household resources and vari- ables affecting the "shadow price" of child survival (Grossman 1972). The coefficients in the equation reflect both technology and behavior. For exam- ple, if piped water and private inputs in child health production are substi- tutes, an increase in piped water might reduce spending on private inputs in child health production so much that health falls (Jalan and Ravallion, 2001). Variables that lower the shadow price of child survival--such as parents' education, the local availability and quality of medical services, health insur- ance coverage, and so on--should increase the household's demand for child survival. By contrast, variables that raise the shadow price of survival--such as poor local sanitary conditions and bad environment--would be expected to be negatively associated with survival in the survival demand function. In- creases in mean survival time could be due to either movements along the demand equation (rising incomes allowing households to feed their chil- dren better) or shifts in the demand equation (households being able to buy more or better-quality food from a given amount of income). The complete fertility history data over the first 10 years of life are used here to estimate the determinants of child survival. It makes sense in such a context to use a duration model. Similar to Lavy and others (1996), a Weibull model is used with covariates (Greene 1997). Let S(t) be the survival func- tion at time t, and (t) be the hazard rate at time t. The latter measures the rate at which the survival function decreases over time and is therefore 324 Economic Growth, Poverty, and Household Welfare in Vietnam equal to -dlnS(t)/dt. The basic Weibull model assumes the existence of a basic time-invariant hazard rate, , to which the hazard rate at time t is linked by equation 9.1: (9.1) (t) = p(t)p ­1 where p is a parameter, with p < 1 indicating that (t) falls continuously over time, and p > 1 indicates the opposite. In the case of child survival, it is likely that p will be less than one, because S(t) drops sharply in the first year and then starts to level out. It is linked to the basic hazard, , and the parameter p by the function in equation 9.2: (9.2) S(t) = e-(t)p. Thus, a higher basic hazard reduces the proportion of children surviving to any specific age. The econometric model then seeks to explain variations in . The model takes the form of equation 9.3: (9.3) -lni = xi where i is the basic hazard rate for child i, is a vector of parameters, and xi is a vector of determinants of child survival. Notice the dependent variable is decreasing in the hazard and hence increasing in survival duration--a posi- tive thus indicates a longer life. Although the hazard rate is unobserved at the individual level, survival times are observed. Using this information, a maximum likelihood estimator can be derived, the application of which results in estimates of the coefficient vector as well as the parameter p. This estimator takes into account any censoring--children who were alive in 1993 (or 1998) but had not yet reached their 10th birthday and who had therefore not been exposed to the risk of death for the full 10-year period over which the Weibull model is estimated. Decomposing Sources of Change The interest is in uncovering the reasons behind the uneven distribution of child survival improvements across socioeconomic groups. By using an Oaxaca-type decomposition (Oaxaca 1973) over time, a distinction can be made between two sources of mortality decline: movements along a regres- sion equation and shifts in it (see figure 9.6). The increase from -ln 93 to -ln98 in the figure represents an improvement in survival prospects. If is used to denote change, this is seen as the result of a move along the 1993 equation, giving rise to a change in -ln equal to x 93, and then a shift from the resultant point on the 1993 equation to the final point on the 1998 equation, giving rise to a further change in -ln equal to x98. Thus, as shown in equation 9.4: (9.4) [-ln ] = x · 93 + · x98. Alternatively, the final point in figure 9.6 could have been reached by first shifting from the 1993 equation to the 1998 equation, giving rise to a Poverty and Survival Prospects of Vietnamese Children under Doi Moi 325 Figure 9.6. Decomposing the Sources of Changes in Child Mortality ln (hazard) equation for 1998 ln 98 x98 equation for 1993 x 93 ln 93 x93 x98 x (determinant) change in -ln equal to x93, and then moving along the 1998 equation to the final point, giving rise to a further change in -ln equal to x 98. Thus, an alternative decomposition can be shown in equation 9.5: (9.5) [-ln ] = x · 98 + · x93. The left-hand side (LHS) in each case can be interpreted as the negative of the percentage change in the hazard between 1993 and 1998; thus, a rise in this quantity indicates improved survival prospects.10 The first term on the right-hand side (RHS) indicates the percentage change in the hazard attrib- utable to changes in the means of the covariates, and the second term on the RHS indicates the percentage change in the hazard attributable to changes in the coefficient vector. The decomposition could be undertaken for the sample as a whole to un- derstand the sources of the continued decline in child mortality at the pop- ulation level, but it can also be applied to different socioeconomic groups. Of particular interest here is understanding why mortality has declined among better-off children but not among poor children. To investigate this, the births in the sample are divided into two groups: the poorest quartile of chil- dren (the poor) and the most well-off three quartiles (the nonpoor). The smaller decline in mortality among poor children may be due to any or all of four reasons: (1) the poor experienced less advantageous changes in the x's; (2) changes in the x's mattered less for the poor, because they face a worse set of coefficients linking survival determinants to survival outcomes; (3) the 326 Economic Growth, Poverty, and Household Welfare in Vietnam poor experienced less advantageous changes in the 's; and (4) changes in the 's mattered less to the poor, because they have worse determinants to start with. Thus, in a similar vein to Makepeace and others (1999), a decom- position can be written for the differential change in mortality across poverty groups, as in equation 9.6: (9.6) [-ln NP] - [- ln P] = D1 + D2 + D3 + D4 where the LHS indicates how much more -ln has risen for the nonpoor as compared with the poor, and D1 = xNP - xP 98 NP D2 = 98 - 98 NP P xP D3 = NP - P x93 NP D4 = x93 - x93 NP P P. Terms D1-D4 correspond to points (1)­(4) above. There is no reason, of course, why D1, D2, D3, and D4 should all be positive--it may well be that some changes made a smaller reduction in mortality among the nonpoor (one or more of the terms in equation 9.6 might be negative). Components (1) and (4) are straightforward to rationalize--it is expected that the poor and nonpoor will have different mean values of the x's in any one year, and these means may change differently across the two consump- tion groups. Jalan and Ravallion (2001) develop a theoretical model that ra- tionalizes differential 's across the poor and nonpoor. In their model, child health depends on parental inputs and other inputs (such as piped drinking water). If parental inputs and, say, piped water are complements, parents will increase their own inputs when piped water increases, and it is likely that child health will increase. Jalan and Ravallion also show that if parents' inputs and water are complements, the impact of water on health is likely to rise with income, and if they are substitutes, the impact of water on health may well fall with income. All this applies equally to other nonparental in- puts, such as sanitation. If the 's differ across the poor and nonpoor at any time point, it seems likely that they may also change differently over time in the two groups-- hence D3 in equation 9.6 may be nonzero. This possibility was explored, but the estimated model in which the 's were allowed to change differently across the two consumption groups produced predicted values of the IMR and U5MR that were a long way from the actual values (see table 9.4). They therefore settled on a model that allows D1, D2, and D4 to be nonzero but D3 is constrained to be zero. This allows for poor and nonpoor children to experience different changes in the means of the health determinants, as well as for them to have different values of and different impacts of these de- terminants at the relevant time point. This does not allow, however, for dif- ferential changes in the impacts of the determinants. In the Weibull model Poverty and Survival Prospects of Vietnamese Children under Doi Moi 327 used here, the x's, a 1998 dummy, a vector of interactions between the x's, the 1998 dummy, a dummy indicating whether the child is poor, and a vector of interactions between the x's and the "poor" dummy are all in- cluded. In so doing, the parameter p is constrained to be constant over time. Tests revealed this to be a reasonable assumption. Data and Variable Definitions Three sets of variables are included in this survival demand equation-- child-level, household-level, and community-level variables (see table 9.2 for variable definitions). At the level of the child, the child's gender and age are included. The hazard is likely to decrease with age, at least up to a point. Table 9.2. Variable Definitions, Means, and Standard Deviations 1993 1998 Standard Standard Variable Mean deviation Mean deviation Regional (omitted Mekong Delta) Northern Uplands 0.18 0.38 0.21 0.41 Red River Delta 0.23 0.42 0.17 0.37 North Central Coast 0.15 0.35 0.16 0.37 Central Coast 0.10 0.30 0.11 0.31 Central Highlands 0.04 0.20 0.06 0.24 Southeast 0.10 0.31 0.10 0.30 Child's characteristics Boy 0.52 0.50 0.51 0.50 Child's age 5.12 2.84 5.69 2.77 Household characteristics Log equivalent household consumption 8.12 0.52 8.38 0.55 Years schooling of mother 6.45 3.79 6.68 3.88 Safe water 0.20 0.40 0.31 0.46 Acceptable sanitation 0.13 0.34 0.19 0.39 Community characteristics Vaccination coverage-- nonself mean (percent) 45.54 27.06 62.11 30.11 Prenatal visit coverage-- nonself mean (percent) 39.45 28.35 49.98 28.92 Medically trained delivery coverage--nonself mean (percent) 70.14 31.86 70.35 32.46 Facility delivery coverage-- nonself mean (percent) 50.93 37.45 53.93 37.94 Source: 1993 and 1998 VLSSs. 328 Economic Growth, Poverty, and Household Welfare in Vietnam From previous research, the coefficient on "boy" would be expected to be negative. Among the household-level variables, the number of years of schooling of the mother, the logarithm of equivalent household consump- tion, and dummies indicating whether the household had safe drinking water and satisfactory sanitation facilities are included.11 The equivalence scale used was simply the square root of the number of household members, which allows for economies of scale in household production but not for dif- ferences in food and other requirements across people of different ages and genders. Consumption was measured in the same way in each year, and the 1993 data were expressed in 1998 prices using the Vietnam consumer price index. For the drinking water and sanitation dummies, this study aimed to get as close as possible to the UNICEF definitions (Government of Vietnam 2000), though in the case of sanitation, the figures are probably somewhat conservative.12 At the community level, six regional dummies (the omitted region is the Mekong Delta) and variables intended to capture the local quality and avail- ability of medical services are included.13 The latter variables are not straightforward to define. Several recent studies (including Lavy and others [1996] and Thomas, Lavy, and Strauss [1996]) have included variables cap- turing whether or not local health facilities had drugs and essential medi- cines in stock, the numbers of nurses and other medical staff, and so on. This approach is not unproblematic, especially in the present context. In urban areas, it is far from clear which facilities should be included in an assessment of input availability, and indeed most facility surveys undertaken alongside a household survey do not collect such information in urban areas. The VLSS is no exception to this pattern. Furthermore, even in rural areas, there is the scope for missing important potential providers of services--especially in the private sector, which is often not included in such assessments. This is problematic in the case of the VLSS, where the focus in the facility survey (which in any case was only undertaken in 1998) was on commune health centers. Recently, these have diminished in importance as other providers-- including private physicians--have increased in importance. There is one other problem, which concerns the ability of inventory-type facility surveys to capture the ability of facilities to deliver services. Child- hood immunization--which clearly is potentially very important in the con- text of child survival--is a case in point. In Vietnam, immunization in rural areas is typically delivered through a concerted outreach campaign over a limited time, with the necessary equipment (refrigerators, cold boxes, vaccines, and so on) being borrowed or procured for the period of the campaign. An inventory approach is ill equipped to capture the ability of a facility to deliver such a service. Key Points In light of these problems, the local availability and quality of medical care were measured by variables that capture the output of the local health care Poverty and Survival Prospects of Vietnamese Children under Doi Moi 329 facilities.14 The focus is on primary care (preventive) activities; the quality of local services' coverage rates and average use rates of certain key maternal and child health services in the child's village are included as proxies. Such services include vaccinations (the full course of four key vaccinations for measles, diphtheria-pertussis-tetanus, polio, and tuberculosis; prenatal vis- its (two or more visits); and deliveries by medically trained personnel and in medical facilities. The rates and the averages were computed from house- hold data on services used in respect of last-born children. Because the number of observations can be fairly small in each village, the nonself (or dropout) mean, not the actual mean, was computed. This reduces the likeli- hood of endogeneity and ensures that the variables are the arguments of a demand function and not a production function. Causes of the Recent Changes in Child Survival in Vietnam The decomposition method is now applied to try to uncover the reasons for the faster decline in mortality among better-off children over the 1993­98 period. Estimation Results Table 9.3 shows the parameter estimates of the model. The p values are based on standard errors adjusted for clustering at the village level.15 Nonpoor children living in the Central Highlands and Southeast have significantly better survival prospects than those living in the Mekong Delta, but living in these areas significantly reduces survival prospects for the poor. This does not change over time, and none of the other regional coefficients are significant. Nonpoor children living in the Central High- lands and Southeast have significantly better survival prospects than those living in the Mekong Delta, but among the poor, living in these areas sig- nificantly reduces survival prospects. This does not change over time, and none of the other regional coefficients are significant. These results do not appear to be inconsistent with findings from Baulch and others (2004 [chapter 8 in this volume]) on ethnic minority development in Vietnam. They find that Northern Uplands minorities and the Khmer seem to be doing well out of a strategy of assimilating (both economically and cultur- ally) with the Kinh-Hoa majority, but other groups in that region are attempting to integrate economically while retaining distinct cultural iden- tities. By contrast, Central Highlands minorities and the Hmong have largely been left behind by the growth process. Such diversity in socioeco- nomic development, coupled with the fact that 67 percent of the poor in the Central Highlands were minorities in 1998, can help to explain why, other things being held constant, the poor in the Central Highlands had worse survival prospects than those living in the Mekong Delta and the poor in the Northern Uplands did not. As for other explanatory variables, the child's gender and age have no significant effect on survival prospects--this is true for 1993 and 1998 and 330 Economic Growth, Poverty, and Household Welfare in Vietnam Table 9.3. Weibull Parameter Estimates Base Year interactions Poor interactions Variable Coeff. p value Coeff. p value Coeff. p value Northern Uplands 2.054 0.13 -2.862 0.17 -0.418 0.87 Red River Delta -0.087 0.95 -1.170 0.62 -3.129 0.21 North Central Coast -0.067 0.96 0.695 0.78 0.857 0.73 Central Coast 0.609 0.66 -1.134 0.58 -0.998 0.67 Central Highlands 2.573 0.06 -2.550 0.23 -4.412 0.06 Southeast 2.869 0.04 -2.704 0.20 -5.827 0.06 Child = boy -0.657 0.31 -0.877 0.37 -1.405 0.17 Child's age -0.094 0.46 -0.061 0.74 0.084 0.62 Log equivalent house- hold consumption -1.099 0.18 1.034 0.40 0.072 0.97 Years schooling of mother 0.294 0.01 -0.118 0.51 -0.017 0.94 Safe water 1.920 0.06 0.212 0.89 -2.619 0.15 Acceptable sanitation 0.490 0.64 -1.995 0.20 2.714 0.36 Vaccination coverage 0.032 0.06 -0.012 0.52 -0.009 0.67 Prenatal visit coverage 0.000 1.00 0.005 0.82 -0.058 0.03 Medically trained delivery coverage -0.017 0.27 0.008 0.68 0.071 0.00 Facility delivery coverage 0.016 0.31 0.003 0.88 -0.010 0.72 Constant 19.600 0.01 -4.968 0.62 -0.512 0.97 Wald test of joint significance 35.53a 0.00 17.70 0.41 31.97 0.02 p value 0.236 a. Excludes constant term in Wald test of joint significance. Source: 1993 and 1998 VLSSs. for the poor and nonpoor. Consumption, of itself, has no significant effect, but the evidence suggests its impact may have risen over time--albeit not significantly. The effect of consumption appears through the interaction terms, several of which are significant. The mother's education significantly improves survival prospects for the children, and there is no significant dif- ference across consumption groups or over the two time periods. This is consistent with several other studies of child survival (Guilkey and Riphahn 1998; Hobcraft, McDonald, and Rutstein 1985; Lee, Rosenzweig, and Pitt 1997; Merrick 1985), though some studies have found little evidence of a link (Benefo and Schultz 1996, Lavy and others 1996). Having access to satisfac- tory drinking water also significantly improves a child's survival prospects. Again, this is consistent with several previous studies (Merrick 1985; Ridder and Tunali 1999), but the effect is not found everywhere (Benefo and Schultz 1996; Lee, Rosenzweig, and Pitt 1997). At the 15 percent level, the effect of drinking water is lower among the poor (indeed it is negative)--this dif- ference is consistent with an earlier study of child mortality (Esrey and Poverty and Survival Prospects of Vietnamese Children under Doi Moi 331 Habicht 1988) and with recent work on child morbidity (Jalan and Ravallion 2001). Satisfactory sanitation, by contrast, has no significant effect on sur- vival prospects, and the evidence suggests that, if anything, this effect is larger (though not significantly so) among the poor. The lack of significance of the sanitation coefficient is consistent with some recent studies (Lee, Rosenzweig, and Pitt 1997), but not others (Ridder and Tunali 1999), and the tendency toward a larger effect among the poor is consistent with Esrey and Habicht (1988), who found a larger effect of sanitation among the better- educated. In the effects of the health service variables, vaccination coverage in the child's village is associated with significantly better survival prospects--a result that is consistent with previous studies using different methods (Koenig and others 1990). There is no significant difference over time or across consumption groups, which goes against some recent work in Bangladesh that suggests that the poor benefit more from vaccination (Koenig, Bishai, and Khan 2001). The only significant coefficient on the prenatal visit coverage variable is that on the interaction with the poor dummy, which is negative. Taken at face value, this implies a negative impact of such visits on child survival among the poor and no significant effect on the nonpoor. By contrast, among the poor, the coverage of med- ically attended deliveries has a positive and significant effect on survival prospects of poor children but not on those of nonpoor children. This is consistent with recent evidence from Brazil (Furquim de Almeida and others 2000), which suggests that home births carry a higher neonatal mortality risk, but only among poorly educated mothers. All else being constant, coverage of facility deliveries has no significant effect on child survival, holding constant the proportion of deliveries by medically trained personnel. The Wald tests at the bottom of table 9.3 indicate that although the x's are jointly significant, and the interactions between the x's and the "poor" dummy are jointly significant, the interactions between the x's and the "year" dummy are not jointly significant. Dropping the year interactions from the model, however, produces predicted IMR and U5MR that are, on average, inferior to the predictions produced by the model in table 9.3 (see table 9.4). In what follows, the estimates reported in table 9.3 are main- tained, and all reported coefficients are used in the decompositions. Decomposition Results Table 9.5 shows how the various determinants of child survival have changed over time for the poor and nonpoor. Some of these changes reflect sampling variability. Some households were in the 1993 VLSS but not in the 1998 VLSS and vice versa. Children older than age five in the 1993 VLSS would have been excluded from this study's 1998 sample of births even if they had been in the 1998 VLSS sample--because they would have been too old in 1998 for the 10-year window. Furthermore, in 40 percent of n.a. n.a. 107 108 122 123 011 108 Nonpoor 1998 centage per values Poor n.a. n.a. 94 98 101 106 65 70 as essed estimated expr of n.a. n.a. 109 104 101 96 106 103 Nonpoor Rates 1993 Poor n.a. n.a. 89 89 86 86 92 91 21.80 31.40 23.36 33.97 26.58 38.62 24.07 33.94 Nonpoor 1998 Rates rates Poor 36.90 51.10 34.60 50.19 37.34 54.12 24.02 35.53 rate. Mortality Mortality mortality 32.30 49.10 35.20 51.06 32.52 47.18 34.10 50.45 -Five Nonpoor -five 1993 under = Under and Poor 38.50 55.90 34.30 49.76 32.99 47.86 35.45 51.14 U5MR rate. Infant year VLSS. mortality 9.3 each 1998 oup Predicted VLSS 9.3 interactions for gr infant and omfr table separate = table in year applicable. 1993 9.4. in as no with essions poverty IMR Not ce: ableT but egrr and n.a. Note: Sour Indicator Estimates IMR U5MR Model IMR U5MR Model IMR U5MR Model IMR U5MR 332 Poverty and Survival Prospects of Vietnamese Children under Doi Moi 333 Table 9.5. Means of Determinants of Child Survival, by Poverty Group Most well-off Poorest quartile three quartiles Mean Mean Change Mean Mean Change Variable 1993 1998 in mean 1993 1998 in mean Northern Uplands 0.23 0.34 0.11 0.17 0.17 0.00 Red River Delta 0.26 0.13 -0.14 0.22 0.18 -0.03 North Central Coast 0.21 0.18 -0.02 0.13 0.16 0.03 Central Coast 0.08 0.13 0.05 0.11 0.10 -0.01 Central Highlands 0.05 0.08 0.03 0.04 0.05 0.01 Southeast 0.05 0.02 -0.04 0.12 0.13 0.01 Child = boy 0.52 0.51 -0.01 0.51 0.51 0.00 Child's age 4.52 5.26 0.73 5.31 5.83 0.52 Log equivalent household consumption 7.52 7.76 0.24 8.31 8.59 0.27 Years schooling of mother 5.79 5.38 -0.41 6.67 7.12 0.45 Safe water 0.09 0.14 0.05 0.24 0.37 0.13 Acceptable sanitation 0.05 0.02 -0.02 0.16 0.25 0.08 Vaccination coverage 40.92 55.53 14.61 47.08 64.31 17.22 Prenatal visit coverage 30.24 39.57 9.32 42.51 53.46 10.94 Medically trained delivery coverage 62.73 57.27 -5.46 72.61 74.71 2.09 Facility delivery coverage 43.08 33.35 -9.73 53.54 60.79 7.25 Note: Calculations may not appear exact due to errors introduced by rounding. Source: 1993 and 1998 VLSSs. households in both VLSS samples, a different woman was randomly se- lected for the fertility questions, so the children (and their mothers) selected for the survival analysis in many panel households were different. Some inconsistencies in responses among the panel households are also evident, and no attempt has been made to correct them. For instance, some people reported a lower completion schooling grade in 1998 than in 1993, and some households reported worse sanitation in 1998 than in 1993. The socioeconomic profiles of the various regions have changed over time. Equivalent consumption in households with small children has risen for both the poor and nonpoor, but in the poorest quartile, the percentage in- crease has been marginally smaller than in the most well-off three quartiles (24 percent compared with 28 percent). In the poorest quartile, the data imply a decline in the average years of maternal schooling among families with small children, though the comments above need to be borne in mind. The proportions of poor and nonpoor households with satisfactory drinking water have increased equiproportionately for the poor and nonpoor. Access to satisfactory sanitation appears to have fallen among the poorest quartile in the period 1993­98. Vaccination coverage and prenatal visit coverage 334 Economic Growth, Poverty, and Household Welfare in Vietnam have increased for both consumption groups. By contrast, the proportion of babies delivered by medically trained persons and in medical facilities has fallen among the poorest quartile, while it has increased among the top three quartiles. What caused the faster reduction in mortality between 1993 and 1998 among the better-off? Tables 9.6 and 9.7 show the results of the decomposi- tions of mortality change for each poverty group, based on equations 9.4 and 9.5, respectively. For the purposes of this study, though, it is more im- portant to determine the differential decline in mortality between the poor and nonpoor. Table 9.8, which presents the results of the decomposition in equation 9.6, shows that the more rapid decline in mortality among the non- poor was roughly equally due to (a) the poor experiencing less advanta- geous changes in the determinants of child survival (D1); (b) these changes mattering less for the poor, because they faced a worse set of coefficients linking survival determinants to survival outcomes (D2); and (c) changes in coefficients mattering less for the poor, because they had worse determi- nants to start with (D4). D1 captures the fact that the poor experienced less advantageous changes in the x's. The weights applied to these changes are the nonpoor coefficients in 1998. In D1, differential changes in health service coverage stand out as the single largest contributory factor to the faster mortality decline among the better-off. This reflects the faster growth in vaccination coverage and prena- tal visit rates among the nonpoor, as well as the fact that facility deliveries apparently declined among the poor but increased among the nonpoor. Offsetting these effects is the effect of the decline in deliveries attended by medical staff, which is given a negative weight by the nonpoor coefficient in 1998. Also noteworthy in D1 are the parts played by water and mother's education. In the case of water, this stems from the faster growth of access to safe drinking water among the better-off and the positive coefficient on water among the nonpoor in 1998. In the case of mother's education, the positive contribution to D1 reflects the decline in mother's education among the poor, compared with the increase among the better-off, and the positive coefficient on mother's education among the nonpoor in 1998. Differential growth of household consumption played a negligible part, as did differen- tial changes in the regional distribution of children and differential changes in child characteristics. The offsetting effect of differential changes in satis- factory sanitation (coverage declined among the poor but increased among the better-off) is somewhat counterintuitive, reflecting the negative coeffi- cient on sanitation among the nonpoor in 1998. However, it needs to be borne in mind that none of the sanitation coefficients in table 9.3 is statisti- cally significant at conventional levels. D2 captures the fact that the changes in the x's mattered less for the poor, because they faced a worse set of coefficients. These coefficient differences are weighted by the change in the x's among the poor. Once again, health services stand out as the single largest contributory factor to the faster mor- tality decline among the nonpoor. In the cases of immunization and prenatal in x 93 1.193- 0.774- 8.594 0.784- 0.051 0.325- 0.432 4.968- 1.033 27.69 40.23 Change coefficients quartiles ee in thr x 98 0.069 0.072- 0.018- 0.080 0.284 0.127- 0.513 0.000 0.729 29.72 43.16 Change well-off determinants Most in ] cent Per hazard ln-[ 1.124- 0.846- 8.576 0.704- 0.335 0.451- 0.945 4.968- 1.763 22.76 33.10 change 9.4 in x 93 1.185- 0.731- 7.772 0.681- 0.020 0.093- 0.325 4.968- 0.460 30.83 44.76 Equation Change coefficients on Based in rate. quartile est x 98 0.417 0.026- 0.002 0.065- 0.024- 0.030- 0.771- 0.000 0.497- 38.49 55.78 Change mortality Survival, Poor determinants in -five in ] under = cent Changes ln-[ of Per hazard 0.768- 0.758- 7.774 0.746- 0.004- 0.122- 0.445- 4.968- 0.037- 34.60 50.19 change U5MR rate. VLSSs. mortality 1998 Decompositions IMR U5MR infant and consumption = 9.6. educations' 1993 services new new characteristics IMR ce: ableT ater otalT Note: Sour Indicator Region Child Household Mother W Sanitation Health Constant Implied Implied 335 in x 98 1.177- 0.801- 8.878 0.837- 0.079 0.493- 0.319 4.968- 1.000 27.90 40.54 Change coefficients quartiles ee in thr 93 x 0.053 0.045- 0.302- 0.133 0.256 0.041 0.626 0.000 0.762 29.49 42.83 Change well-off determinants Most in ] cent Per hazard ln-[ 1.124- 0.846- 8.576 0.704- 0.335 0.451- 0.945 4.968- 1.763 22.76 33.10 change 9.5 in x 98 19 1.398- 0.768- 8.018 0.633- 0.030 0.043- 0.1 4.968- 0.357 31.57 45.83 Equation Change coefficients on Based in 1 rate. quartile x 93 13 est 0.630 0.01 0.245- 0.1- 0.034- 0.079- 0.564- 0.000 0.394- 37.59 54.48 Change Survival, Poor determinants mortality in -five in ] under cent Changes Per hazard ln-[ = 0.768- 0.758- 7.774 0.746- 0.004- 0.122- 0.445- 4.968- 0.037- 34.60 50.19 of change U5MR rate. VLSSs. mortality 1998 Decompositions IMR U5MR infant and consumption = 9.7. educations' 1993 services new new characteristics IMR ce: ableT ater otalT Note: Sour Indicator Region Child Household Mother Sanitation W Health Constant Implied Implied 336 ] ence poor Nonpoor- differ ln-[ 0.356- 0.088- 0.803 0.041 0.338 0.329- 1.390 0.000 1.800 in P s x' ent 4 of P 93 D x - Differ 0.008- 0.043- 0.822 0.103- 0.030 0.232- 0.106 0.000 0.573 levels NP x 93 P 9.6 s' x ent 2 of P D - 98 Differ 0.386- 0.074- 0.017- 0.007- 0.127 0.067 0.964 0.000 0.674 Equation levels NP 98 on Based s NP x' 98 Survival, P ent in x 1 in D - 0.038 0.028 Differ 0.002- 0.151 0.181 0.164- 0.320 0.000 0.552 changes NP x Changes of VLSSs. 1998 Decompositions and 9.8. educations' 1993 services characteristics ce: ableT consumption ater otalT Sour Indicator Region Child Household Mother W Sanitation Health Constant 337 338 Economic Growth, Poverty, and Household Welfare in Vietnam coverage, this reflects the larger impact among the nonpoor and the fact that the poor experienced a rise in coverage ( xP is positive in D2). In the case of medically trained deliveries, it reflects the larger impact among the poor and the fact that the poor experienced a decline in coverage. In the case of fa- cility deliveries, the contribution is negative, reflecting the smaller impact among the poor coupled with the decline in coverage in this group. Of these terms, only two (prenatal coverage and medical deliveries) are based on sta- tistically significant differences in coefficients between the poor and non- poor. The role of water is also noteworthy in D2. This reflects the larger impact of water on survival prospects among the better-off--a difference that is significant at the 15 percent level--and the fact that water access increased among the poor. Of note, too, is the role of region, though the con- tribution reflects a variety of offsetting tendencies, and it should be borne in mind that only two of the interactions between the "poor" dummy and the regional dummies are significant. D4 captures the fact that the changes in coefficients over time mattered less for the poor than the nonpoor, because the poor had worse child sur- vival determinants to start with. These differences in determinants are weighted by the changes in the coefficients (assumed in this exercise to be the same for the poor and better-off). Given the joint insignificance of the interactions between the x's and the year dummy in table 9.3, D4 should not be interpreted too literally. Differences in levels of household consumption stand out as the largest single contributory factor--the poor, by definition, are worse off than the nonpoor--and the coefficient on consumption in- creased between 1993 and 1998, though not significantly so. A positive con- tribution is evident, too, from health services, reflecting the lower levels of coverage among the poor and the fact that, for the most part, the coeffi- cient increased between 1993 and 1998. Again, however, the change was not significant. Key Points There is no single factor explaining the faster decline in mortality among better-off children in Vietnam over the period 1993­98. Rather, it was due to three sets of factors: (1) the poor experiencing less advantageous changes in the determinants of child survival; (2) these changes mattering less for the poor, because they faced a worse set of coefficients linking survival determi- nants to survival outcomes; and (3) changes in coefficients mattering less for the poor, because they started with worse determinants. These three factors made roughly equal contributions. In (1), differential changes in health ser- vice coverage stand out as the single largest contributory factor to the faster mortality decline among the better-off, though differential changes in mother's education and access to safe water were also important. In (2), health services were also a major factor, reflecting, among other things, the lower impact of prenatal visit coverage among the poor. But other coefficient differences also left their mark, including the smaller effect of drinking Poverty and Survival Prospects of Vietnamese Children under Doi Moi 339 water among the poor. In (3), the lower levels of consumption and health service coverage among the poor meant that the increases in the impacts of these factors on survival resulted in smaller reductions in mortality among the poor. These increased impacts were not, however, significant at con- ventional levels, so that, statistically, the differential decline in mortality was due to just (1) and (2) above. Child Survival Prospects to 2015 The next question is how child survival prospects are likely to evolve over the next 15 years. Vietnam is currently preparing a PRSP--Vietnam's Com- prehensive Poverty Reduction and Growth Strategy (CPRGS)--to qualify for continued support through the International Development Association. Like other countries preparing PRSPs, Vietnam is setting out broad poverty reduction strategies as well as targets for poverty and human development indicators in its CPRGS. In other countries, these targets have included the IMR and U5MR, because these are among the seven key MDGs. The MDGs, involve among other things, a reduction by all countries of the IMR and U5MR by two-thirds by 2015. Is such a reduction possible for Vietnam? And would commitment to such a path be realistic for Vietnam's CPRGS? Methods and Assumptions Using the Oaxaca decomposition methodology, the hazard rate can be written for the year 2015 for the poor and nonpoor separately as shown in equation 9.7: (9.7) -ln15 = -ln98 + x15 - x98 98 + 15 - 98 x15, j j j j j j j j j = poor, nonpoor where the notation is self-evident. So, the (negative of the log of the) hazard in 2015 will depend on the hazard in 1998, the change in the means of the determinants of survival during the period 1998­2015, and the changes in the impacts of these determinants over the same period. The first term on the RHS is known. The other terms are unknown, and some assumptions are required to operationalize the method. Some assumptions must be made about the changes in the x's and the 's. For the nonpoor, the assumption with respect to the x's is that the annual average rate of growth (or decline) over the period 1993­98 will continue over the period 1998­2015, with two exceptions. First, the assumption is that there will be no further redistribution of young children across the seven regions in the model, and the child population is stable in the sense of show- ing no change in the age and gender composition. Second, the variables are restricted by capturing proportions to be in the interval [0,1]. Values in ex- cess of one (or 100 percent) are replaced by one (or 100 percent) in the first 340 Economic Growth, Poverty, and Household Welfare in Vietnam year it exceeds the limit and for all years thereafter. For the top three quar- tiles, full coverage is achieved for water in 2010, for sanitation in 2015, for vaccinations in 2006, and for prenatal care in 2012. With respect to the 's, a more conservative assumption has been made: simply that for the entire period 1998­2015 will be the same as that for 1993­98. For the poor, four scenarios are explored. In scenario A, the analogous as- sumptions are applied that were made above for the nonpoor: the x's grow or decline over the period 1998­2015 at the same annual rate as over the pe- riod 1993­98, subject to the qualifications as above; and for the entire period 1998­2015 will simply be the same as that for 1993­98. In scenario B, the poor group's 98 coefficient vector is used, but the poor are assumed to experience the same annual rates of growth of the x's as the nonpoor. In sce- nario C, the assumption is that the poor will experience the same growth rates as in scenarios A and B, but they will face the nonpoor group's coeffi- cient vector 98. Finally, in scenario D, the assumption is that the poor will experience the same growth rates and the same coefficient vector 98 as the nonpoor. The implied growth rates and terminal values of the x's are indicated in table 9.9 for scenario A. Table 9.10 shows the result of the analysis of likely future trends in child survival. The implied values of the U5MR corresponding to the second and third terms on the RHS of equation 9.7 are the predicted values in the event that only the x's change or only the 's change. Thus, if for the most well- off three quartiles the 's were to change between 1998 and 2015 as they did in the period 1993­98, but the x's stayed at their 1998 values, the U5MR would fall from 34 per 1,000 in 1998 to 30.1 per 1,000 in 2015. If, by contrast, Table 9.9. Assumptions Concerning Progress on Determinants of Survival Poorest quartile Most well-off three quartiles Average annual Average annual percentage Value at percentage Value at Variable change 2015 change 2015 Equivalent household consumption (log) 0.6 8.625 0.7 9.592 Years schooling of mother -1.4 4.200 1.3 8.888 Acceptable sanitation 8.6 0.584 9.3 1.000 Safe water -14.0 0.002 8.7 1.000 Vaccination coverage 6.3 100.000 6.4 100.000 Prenatal visit coverage 5.5 98.653 4.7 100.000 Medically trained delivery coverage -1.8 42.033 0.6 82.283 Facility delivery coverage -5.0 13.971 2.6 93.617 Note: Results--survival prospects to 2015. Source: 1993 and 1998 VLSSs. Poverty and Survival Prospects of Vietnamese Children under Doi Moi 341 Table 9.10. Decompositions of Changes in Survival to 2015 Implied under- Percentage five mortality change from Scenario Quartile Value rate 1998 value Baseline Top 3 -ln 98 15.86 33.97 (x15 -x98)98 1.95 21.55 -59 (15 - 98)x15 0.44 30.69 -41 -ln 15 18.25 19.46 -63 A--poorest quartile's Bottom growth and -ln 98 14.17 50.19 coefficients (x15 - x98)98 -4.20 129.50 127 (15 - 98)x15 1.11 38.87 -32 -ln 15 11.08 101.28 78 B--poorest quartile's Bottom coefficients but top -ln 98 14.17 50.19 three quartiles' (x15 - x98)98 -1.41 69.24 21 growth (15 - 98)x15 0.92 40.63 -29 -ln 15 13.68 56.16 -1 C--poorest quartile's Bottom growth but top three -ln 98 14.17 50.19 quartiles' coefficients (x15 -x98)98 1.64 34.37 -40 (15 - 98)x15 1.11 38.87 -32 -ln 15 16.92 26.57 -53 D--coefficients and Bottom growth of top three -ln 98 14.17 50.19 quartiles (x15 -x98)98 2.56 27.75 -51 (15 - 98)x98 0.92 40.63 -29 -ln 15 17.65 22.41 -61 Note: Assumes that the infant mortality rate in 1990 is 40 per 1,000 and 36 per 1,000 for the poor and nonpoor, respectively, and that the under-five mortality rate figures for 1990 are 57 per 1,000 and 52 per 1,000 for the poor and nonpoor, respectively. Source: 1993 and 1998 VLSSs. the 's were to remain unchanged at their 1998 values, but the x's were to grow (or decline) at similar rates to those observed over the period 1993­98, the U5MR for the most well-off three quartiles would fall to 21.6 per 1,000. The net effect of both sets of changes is to reduce the U5MR by just under two-thirds below its 1990 value. These two sets of changes would, in other words, ensure that among the most well-off three-quarters of children in Vietnam, the MDG for U5MR would be hit. In the case of the poorest quartile, the picture is quite different. Under the same two sets of assumptions as made above for the most well-off three quartiles, the U5MR among the poorest quartile would actually rise by 342 Economic Growth, Poverty, and Household Welfare in Vietnam 78 percent--a reduction in mortality due to beneficial changes in the 's being more than offset by deleterious changes in the x's. The prospects for the poor look somewhat better in scenario B, where they are assumed to ex- perience the same growth rates in the x's as experienced by the nonpoor. But even in this case, the U5MR among the poorest quartile would barely change from its 1990 value. In both scenarios C and D, the prospects for the poor look decidedly better. In scenario C, where the poor are assumed to experience the same growth rates as in scenario A but are assumed to face the coefficient vector of the nonpoor, a 53 percent reduction in the U5MR between 1990 and 2015 is predicted. A 61 percent reduction is predicted in scenario D, where the poor are assumed to enjoy the nonpoor's growth rates and face the nonpoor's coefficient vector. In both scenarios A and B, what holds the poor back are the negative coefficients on the water and prenatal visit variables. Caution should be exercised in interpreting these negative co- efficients too literally, but the results for contrast between the results for sce- narios Aand B, on the one hand, and scenarios C and D, on the other, do point toward the importance of ensuring that poor children benefit more from water and from prenatal visit programs than they appear to do at present. Key Points Achieving a reduction of two-thirds of the 1990 levels of child mortality in Vietnam as implied by the MDGs is very challenging. Even under quite optimistic assumptions about annual income growth, as well as the evolu- tion of water, sanitation, and health services, projected levels of child mor- tality are likely to be higher than the targets by 2015. The key problem seems to be the slow recent progress among Vietnam's poor. Reversing the poor's backward moves along the survival demand curve (especially in relation to women's education and birth-related health services), and ensuring that it is not just the better-off who benefit from improvements that increase the impact of health determinants on child survival, would seem to be central to tackling this problem. Discussion and Policy Implications The previous analysis has shown that although child mortality for the pop- ulation as a whole has continued to fall under Doi Moi, progress has been much faster at the top end of the income distribution. Indeed, the poorest quartile of children saw virtually no improvements in their survival prospects in the mid- to late 1990s. The recent slow rate of decline of child mortality among poorer Vietnamese children, which is apparently due to a variety of factors, makes it doubtful whether Vietnam will achieve the re- ductions in the IMR and U5MR envisaged by the MDGs. In this last section, we discuss some policy options for accelerating the pace of decline of child mortality among Vietnam's poor. These are somewhat tentative, given the lack of significance of some of the parameter estimates. Poverty and Survival Prospects of Vietnamese Children under Doi Moi 343 Table 9.11. Reversing Declines in Determinants of Survival among the Poorest Quartile Indicator IMR U5MR Values among poorest quartile in 1998 34.6 50.2 Mother's education back to 1993 value 34.1 49.4 Attended deliveries and deliveries at medical facility 31.3 45.4 back to 1993 values Both mother's education and deliveries back to 1993 values 30.8 44.8 Note: IMR = infant mortality rate. U5MR = under-five mortality rate. Source: 1993 and 1998 VLSSs. The previous sections of this chapter have shown that, in two key re- spects, the poorest 60 percent of Vietnamese children appear to have slipped backwards between 1993 and 1998--their mothers were less educated than their predecessors' mothers; and they were less likely to be delivered by trained birth attendants and in medical facilities. What would be the impact of reversing these declines on child survival prospects among the poorest quartile of Vietnamese children? In 1993, mothers in the bottom quartile av- eraged 5.8 years of schooling. In 1998, this figure had fallen to 5.4 years. In 1993, 62.7 percent of births in the poorest quartile were attended by a med- ically trained person, and 43.1 percent of births took place in a medical facil- ity. In 1998, these figures had fallen to 57.3 percent and 33.3 percent, respec- tively. Applying the 1998 coefficient vector to the 1993 values of these variables, keeping the other variables at their 1998 values, gives an estimate of how far mortality would fall if these declines were reversed. Table 9.11 in- dicates that the impact on the U5MR of reversing the decline in maternal schooling would be smaller than the impact of reversing the decline in de- liveries attended by medical personnel and deliveries in medical facilities. Reversing both sets of changes together would reduce the U5MR by around 5.4 per 1,000--an 11 percent reduction in the U5MR. In terms of the effects of targeted improvements in health services, drinking water, and sanitation, the poorest quartile of children lag far be- hind the most well-off three quartiles, and closing these gaps--by bringing the poor up to the levels enjoyed by the better-off--seems likely to have a sizable effect on child mortality among the poor. Table 9.12 shows the effects of raising the levels among the poorest quartile of children to the levels en- joyed by the most well-off three quartiles of children. In each case, apart from drinking water, child mortality would fall. This reflects the fact that among the poorest quartile, all determinants except water are estimated to have beneficial effects on survival, as are most health services. The largest impact would derive from extending health service coverage among the poorest quartile to the level of coverage enjoyed by the most well-off three quartiles. 344 Economic Growth, Poverty, and Household Welfare in Vietnam Table 9.12. Targeted Improvements in Health Services, Water, and Sanitation among the Poorest Quartile Indicator IMR U5MR Values among poorest quartile in 1998 34.6 50.2 Improving sanitation quality to level of top 3 quartiles 32.5 47.1 Improving drinking water quality to level of top 3 quartiles 35.5 51.5 Improving health service coverage to level of 3 quartiles 29.3 42.6 Note: IMR = infant mortality rate. U5MR = under-five mortality rate. Source: 1993 and 1998 VLSSs. The changes in table 9.12--although useful--are fairly small in absolute terms and relative to the reductions envisaged by the MDGs. For example, effecting all three sets of improvements in table 9.12 simultaneously would reduce the U5MR among the poorest quartile of children from 50.2 per 1,000 to 41.1 per 1,000--a reduction of 6 percent of the 1998 value. It is worth con- sidering what other measures might be taken to accelerate further the de- cline in mortality among Vietnam's poor. Looking back at table 9.3, it is strik- ing that when the coefficients differ significantly across the two consumption groups in all cases except one (medically trained delivery coverage) the poor fare worse. When variables contribute positively to child survival--such as living in the Central Highlands and the Southeast, good drinking water, and prenatal visit coverage--the impact is larger for the better-off. When vari- ables contribute negatively to survival--as is the case with being male--the impact is larger in absolute size for the poor. The impact of medically trained deliveries runs counter to this pattern--this contributes positively to sur- vival for the poor, but not for the better-off. This pattern suggests that some effort could usefully be directed at ex- ploring the reasons for the apparent gaps in impacts between the poor and nonpoor, and finding ways to reduce them. For example, the gap in the im- pact of drinking water may well reflect the fact that the nonpoor know bet- ter how to use water to promote the health of their children--for example, taking advantage of its plentiful supply to wash hands after defecation, before preparing meals, and before feeding their children. The gap in the gender impact may reflect a tendency for poor boys to be more likely to be involved than their better-off counterparts in accidents. The smaller impact of prenatal coverage among the poor probably reflects a failure of poorer households to take full advantage of such programs and to comply less with the instructions of the staff delivering them. The pace of mortality decline among Vietnam's poor can be accelerated by coupling targeted programs (changing the x's for the poor) with measures such as behavioral change and communication programs aimed at raising the impacts of these determi- nants (raising the 's of the poor). The potential impact of behavioral change and communication pro- grams and similar measures can be appreciated by considering what would Poverty and Survival Prospects of Vietnamese Children under Doi Moi 345 happen if, in addition to implementing the three measures in table 9.12, a program succeeded in equalizing the impacts of the x's across the poor and nonpoor, by replacing the 's of the poor with those of the nonpoor. Under this scenario the U5MR for the poorest quartile of the population would fall not to 42.6 per thousand as in table 9.12, but to 36.1 per thousand--a reduc- tion of 28 percent below the 1998 rate. This suggests strongly that a strategy for accelerating the mortality decline among Vietnam's poor should focus not just on improving their coverage of and access to drinking water to this group but also on enhancing the impacts among the poor of these key determinants of child survival. Appendix 9A Computation of Indirect Estimates of IMR and U5MR The indirect estimates pertaining to the 1989 Census, the 1994 ICDS, and the 1997 DHS were computed by the authors with the use of the computer pro- gram QFIVE (United Nations 1983) and using the data on children born and children surviving reported in table 3.5 (p. 32) of the 1997 DHS report (Government of Vietnam 1999). (The mortality estimates reported in the DHS report are computed using the direct method.) The "North" family of life tables was used throughout, as in Hill and others (1999). The IMR and U5MR reported in figure 3 (p. 46) of the MICS report (Government of Vietnam 2000) are used here. (Identical results are obtained if the data on children born and children surviving reported in table 2.2 [p. 92] of the MICS report are passed through the computer program QFIVE.) The VLSS estimates are obtained using the same methods. One issue that arises in the context of both VLSSs is that of weighting. There are two parts of this issue. First, in both surveys, fertility histories were obtained from only one woman of fertile age per household. If women in households with more than one fertile woman are alike in their fertility histories to fertile women in the population at large, this will not create a problem. Desai (1998) has shown, however, that this is not the case. There are proportionally fewer women in the lower and upper age groups responding to the fertility module of the 1993 VLSS than there are in the VLSSs as a whole. Because death rates among children tend to be higher among women ages 15­19, 20­24, and 40­45 than among women in the age groups in between, the VLSS will tend to produce lower estimates of child mortality than is war- ranted unless some adjustment is made for the underrepresentation of younger and older women. A simple way of correcting for this is to weight respondents according to their degree of underrepresentation in the fertility data--that is, by a number that is the ratio of the number of women in the age group in the VLSS generally to the number of women in the age group responding to the fertility questions. This was done for 1993 and produced estimates that differed very little from those in table 9.2. The reason for this is that the fertility rates among younger and older women in Vietnam are relatively low, with only 0.036 of births in 1993 being to women ages 15­19 346 Economic Growth, Poverty, and Household Welfare in Vietnam and only 3.79 of births to women ages 40­45. The relatively small number of births occurring to younger women is probably due in part to there being a fairly widely accepted consensus about the ideal age to start a family and in part to the widespread availability of abortion. The relatively small number of births to older women is probably in part a reflection of the official policy of discouraging more than two children per family. The second part of the weighting issue is that, unlike the 1993 VLSS sam- ple, the 1998 VLSS sample is not nationally representative. Of the 6,000 households surveyed in the 1998 VLSS, 4,305 were included in the 1993 VLSS sample. To make the households surveyed in 1998 nationally repre- sentative, the VLSS designers included household-level weights or expan- sion factors. These were applied to the fertility data throughout. Notes The authors are grateful to Nisha Agrawal, Bob Baulch, Ed Bos, Paul Glewwe, Richard Meyers, Pierella Paci, and the participants at seminars at Johns Hopkins, the World Bank in Hanoi, and the World Bank in Washington for helpful discussions on the issues covered in this chapter. 1. The 1992­93 survey spanned a full year, starting in October 1992; the 1997­98 survey began in December 1997 and also lasted a year. For brevity's sake, reference is made to the surveys as the 1993 VLSS and 1998 VLSS, respectively. 2. See http://www.worldbank.org/poverty/strategies/review/index.htm. 3. See http://www.developmentgoals.org/. 4. In this chapter, data from the 1989 Census and the 1994 ICDS were taken from the other sources listed in the text. 5. It is because of this assumption that the choice of interval width matters. The mortality rates reported are, in fact, obtained using nonfixed intervals, with smaller intervals for the first year of life than later years. This is easily accomplished within the computer program Stata. 6. The rates are those reported in table 7.4 (p. 84) of the DHS report (Govern- ment of Vietnam 1999). 7. The 1998 VLSS estimates were based on data that were weighted using the sampling weights. The mortality estimates obtained from the 1998 VLSS for the 1988­92 period were somewhat lower than those obtained from the 1993 VLSS data for the same period. 8. As elsewhere in this chapter, the 1998 data are weighted using the sample weights. 9. In the case of Vietnam the list includes the following (mostly binary) indicators: Utilities Electricity Water: Piped drinking water in residence Water: Piped drinking water in public tap Water: Inside-well drinking water Water: River, canal, or surface drinking water Water: Rain for drinking water Water: Public well Water: Water from a tanker truck Water: Bottled water Poverty and Survival Prospects of Vietnamese Children under Doi Moi 347 Sanitation Shared flush toilet Traditional pit latrine Ventilated, improved pit latrine Bush or field used as latrine Own flush toilet Appliances Radio Television Refrigerator Telephone Sewing machine Building materials Finished floor Concrete roofing Roof of galvanized iron or aluminum Earth or sand floor Rough wood or bamboo flooring Natural-material roofing Floor made of other materials Roof made of asbestos or iron sheeting Transportation Bicycle Motorcycle Car Boat Motor scooter Other Plowing machine Number of household members per sleeping room 10. The LHS is equal to ­[ln(98/93)]. 11. An ethnic minority dummy was also included, but it was dropped because its coefficient was never significant. This is probably due to the concentration of the ethnic minorities in specific regions, for which the authors control. 12. Safe drinking water was defined here as tap or standpipe, deep dug well with pump, or rain water. Satisfactory sanitation was defined as flush toilet or latrine. Both differ slightly from the definitions used by UNICEF in its analysis of the MICS data, but the categories in the VLSS data are somewhat different from those in the MICS. For both water and sanitation, use of the original VLSS categories (rather than aggregat- ing into safe dummies) was explored, but it was decided against in favor of the present approach, which has the merit of making for a straightforward interpretation. 13. An urban dummy was included, but it was dropped because its coefficient was never significant. 14. The variables used are not entirely under the control of the health system, so the term "output"--though often used in this context--is slightly misleading. 15. This adjustment of the standard errors invalidates traditional likelihood tests, hence the use of Wald tests in table 9.3. Bibliography Baulch, B., Truong Thi Kim Chuyen, D. Haughton, and J. Haughton. 2004. "Ethnic Minority Development in Vietnam: A Socioeconomic Per- spective." In P. Glewwe, N. Agrawal, and D. Dollar, eds., Economic 348 Economic Growth, Poverty, and Household Welfare in Vietnam Growth, Poverty, and Household Welfare in Vietnam. Washington, D.C.: World Bank. Benefo, K., and T. Schultz. 1996. "Fertility and Child Mortality in Côte d'Ivoire and Ghana." World Bank Economic Review 10(1): 123­58. Central Census Steering Committee. 2000. 1999 Population and Housing Census. Hanoi: The Gioi Publishers. Dahlgren, G. 2000. "Issues of Equity and Effectiveness in Equity-Oriented Health Care Policies: An Introduction." In P. Hung, I. Minas, Y. Liu, G. Dahlgren, and W. Hsiao, eds., Efficient, Equity-Oriented Strategies for Health: International Perspectives--Focus on Vietnam. Melbourne: Centre for International Mental Health. Desai, J. 1998. "Poverty and Fertility in Vietnam." In. D. Dollar, P. Glewwe, and J. Litvack, eds., Household Welfare and Vietnam's Transition. Washington, D.C.: World Bank. Esrey, S. A., and J. P. Habicht. 1988. "Maternal Literacy Modifies the Effect of Toilets and Piped Water on Infant Survival in Malaysia." American Journal of Epidemiology 127(5): 1079­87. Filmer, D., and L. Pritchett. 1999. "The Effect of Household Wealth on Educational Attainment: Evidence from 35 Countries." Population and Development Review 25(1): 85­120. Furquim deAlmeida, M., L. C. Rodrigues, G. PereiraAlenca, and H. M. Dutilh Novaes. 2000. "Neonatal Mortality by Place of Delivery in São Paulo, Brazil." Journal of Epidemiology and Community Health 54(11): 879. Glewwe, P., M. Gragnolati, and H. Zaman. 2000. "Who Gained from Vietnam's Boom in the 1990s?" Policy Research Working Paper 2275. World Bank, Washington, D.C. Government of Vietnam. 1999. "Vietnam Demographic and Health Survey 1997." National Committee for Population and Family Planning, Hanoi. _____. 2000. "Analysis of the Multiple Indicator Cluster Survey II (MICSII)." Statistical Publishing House, Hanoi. Greene, W. 1997. Econometric Analysis. Upper Saddle River, N.J.: Prentice- Hall Inc. Grossman, M. 1972. The Demand for Health: A Theoretical and Empirical Inves- tigation. New York: National Bureau of Economic Research. Guilkey, D., and R. Riphahn. 1998. "The Determinants of Child Mortality in the Philippines: Estimation of a Structural Model." Journal of Development Economics 56(2): 281­305. Gwatkin, D., S. Rutstein, K. Johnson, R. Pande, and A. Wagstaff. 2000. "Socioeconomic Differences in Health, Nutrition, and Population." Poverty and Survival Prospects of Vietnamese Children under Doi Moi 349 Health, Nutrition, and Population Discussion Paper. World Bank, Washington, D.C. Hill, K., R. Pande, M. Mahy, and G. Jones. 1999. "Trends in Child Mortality in the Developing World: 1960 to 1996." United Nations Children's Fund, New York. Hobcraft, J., J. McDonald, and S. Rutstein. 1985. "Demographic Determi- nants of Infant and Early Child Mortality: A Comparative Analysis." Population Studies 39(3): 363­85. Jalan, J., and M. Ravallion. 2001. "Does Piped Water Reduce Diarrhea for Children in Rural India?" Policy Research Working Paper 2664. World Bank, Washington, D.C. Kakwani, N. C., A. Wagstaff, and E. van Doorslaer. 1997. "Socioeconomic Inequalities in Health: Measurement, Computation and Statistical Inference." Journal of Econometrics 77(1): 87­104. Koenig, M. A., D. Bishai, and M. Ali Khan. 2001. "Health Interventions and Health Equity: The Example of Measles Vaccination in Bangladesh." Population and Development Review 27(2): 283­302. Koenig, M. A., M. A. Khan, B. Wojtyniak, J. D. Clemens, J. Chakraborty, V. Fauveau, J. F. Phillips, and J. Akbar. 1990. "Impact of Measles Vac- cination on Childhood Mortality in Rural Bangladesh." Bulletin of the World Health Organization 68(4): 441­47. Lavy, V., J. Strauss, D. Thomas, and P. de Vreyer. 1996. "Quality of Care, Survival, and Health Outcomes in Ghana." Journal of Health Econom- ics 15(3): 333­57. Lee, L.-F., M. Rosenzweig, and M. Pitt. 1997. "The Effects of Improved Nutrition, Sanitation, and Water Quality on Child Health in High- Mortality Populations." Journal of Econometrics 77(1): 209­35. Makepeace, G., P. Paci, H. Joshi, and P. Dolton. 1999. "How Unequally Has Equal Pay Progressed since the 1970s? A Study of Two British Cohorts." Journal of Human Resources 34(3): 534­56. Merrick, T. W. 1985. "The Effect of Piped Water on Early Childhood Mortal- ity in Urban Brazil, 1970 to 1976." Demography 22(1): 1­24. Oaxaca, R. 1973. "Male-Female Wage Differentials in Urban Labor Markets." International Economic Review 14(3): 693­709. Ridder, G., and I. Tunali. 1999. "Stratified Partial Likelihood Estimation." Journal of Econometrics 92(2): 193­232. Schultz, T. 1984. "Studying the Impact of Household Economic and Community Variables on Child Mortality." Population and Develop- ment Review 10: 215­35. 350 Economic Growth, Poverty, and Household Welfare in Vietnam Thomas, D., V. Lavy, and D. Strauss. 1996. "Public Policy and Anthropomet- ric Outcomes in the Côte d'Ivoire." Journal of Public Economics 61(3): 155­92. United Nations. 1983. "Indirect Techniques for Demographic Estimation." New York. Viet Nam Living Standards Survey (VNLSS). 1992­1993. Web site: www. worldbank.org/lsms/country/vn93/vn93bid.pdf. Vietnam Living Standards Survey (VLSS). 1997­98. Web site: www. worldbank.org/lsms/country/vn98/vn98bif.pdf. Wagstaff, A. 2000. "Socioeconomic Inequalities in Child Mortality: Compar- isons across Nine Developing Countries." Bulletin of the World Health Organization 78(1): 19­29. Wagstaff, A., P. Paci, and E. van Doorslaer. 1991. "On the Measurement of Inequalities in Health." Social Science and Medicine 33(5): 545­57. World Bank, Swedish International Development Agency, Australian Inter- national Development Agency, Royal Netherlands Embassy, and Ministry of Health of Vietnam. 2001. "Vietnam. Growing Healthy: A Review of Vietnam's Health Sector." World Bank, Hanoi. 10 Child Nutrition, Economic Growth, and the Provision of Health Care Services in Vietnam Paul Glewwe, Stefanie Koch, and Bui Linh Nguyen Child malnutrition is pervasive in almost every low-income country. Among all developing countries, about 30 percent of all children younger than age five are underweight (low weight for age; see United Nations Development Programme [1998]). For the least-developed countries (those with low values for the United Nations' Human Development Index), this figure rises to 39 percent. Most economists would agree that economic growth could reduce child malnutrition in these countries. However, the size of this impact is uncertain and probably varies across countries. If the impact from eco- nomic growth is small, policymakers will need to look for health policies that have a larger impact on children's nutritional status. Child nutrition is a key issue in Vietnam. It is one of the world's poorest countries, with an annual per capita gross national income (GNI) of about US$410 in 2001. It has a very high level of child malnutrition: In 1993, 50 per- cent of Vietnamese children younger than age five were stunted (low height for age), although the situation has improved since that time. The role of economic growth in improving children's nutritional status is particularly relevant for Vietnam because it had very rapid economic growth in the 1990s. Its annual rate of real economic growth since 1988 has been about 8 percent, or about 6 percent in per capita terms, yet at the same time it remains a very poor country with high rates of child malnutrition. This chapter has two objectives. The first is to estimate the impact of eco- nomic growth on child nutrition in Vietnam, using data from two household surveys recently completed in Vietnam: the 1992­93 and 1997­98 Vietnam Living Standards Surveys (VLSSs).1 A recent study of child nutrition in Vietnam, based on the 1993 data, found only a weak relationship between household income and child nutrition (Ponce, Gertler, and Glewwe 1998). 351 352 Economic Growth, Poverty, and Household Welfare in Vietnam This suggests that Vietnam's rapid economic growth in the 1990s had little impact on children's nutritional status, yet the 1998 data show that the inci- dence of stunting for children under five declined from 50 percent in 1993 to 34 percent in 1998. Given this apparent contradictory evidence, this chapter seeks to clarify the role of economic growth. One way to reconcile these findings is to investigate whether other fac- tors, such as new public health policies, improved child nutrition in the 1990s. Thus, the second objective of this chapter is to examine the impact of various health programs in Vietnam on child nutrition. A very rich analysis is possible because the 1998 VLSS contains data on health infrastructure that were not collected in the 1993 VLSS. The section following this brief background presents some basic infor- mation about child nutrition and economic growth in Vietnam in the 1990s. The data used and the analytical framework are then discussed. Estimates of the impact of household income on child nutrition are given in the next sec- tion, and estimates of the impact of health programs and health care prices are presented. The chapter closes with a brief summary of the results and several concluding comments. Child Nutrition and Economic Growth in Vietnam This section presents data on the nutritional status of Vietnamese children in the 1990s and Vietnam's economic performance during the same period. Before examining the data, it is useful to discuss methods for measuring children's nutritional status. Measurement of Children's Nutritional Status The nutritional status of children can be assessed using data on their age, sex, height, and weight. In particular, such data can be used to calculate three indicators of children's nutritional status: stunting (low height for age), wasting (low weight for height), and underweight (low weight for age). Each indicator describes different aspects of malnutrition. Stunting is defined as growth in a child's height that is low compared with the growth in height of a reference healthy population. Slow growth in height over long periods of time causes children to fall further and further behind the height of the reference population. Thus, stunting is a cumulative indicator of slow physical growth. In developing countries, stunting is caused primarily by repeated episodes of diarrhea and other childhood dis- eases, as well as by insufficient dietary intake. In contrast, wasting is an indicator of very recent malnutrition, which leads to weight loss. It indicates current nutritional problems, such as diar- rhea, other childhood diseases, and insufficient dietary intake. Although stunting is usually not reversed--children who become stunted typically re- main so throughout their lives--the weight loss associated with wasting can be restored quickly under favorable conditions. Child Nutrition, Economic Growth, and the Provision of Health Care Services 353 The third indicator, underweight, can reflect stunting, wasting, or both. It does not distinguish between long-term and short-term malnutrition. All three measures are commonly expressed in the form of z scores, which compare a child's weight and height with the weight and height of a similar child from a reference healthy population. More precisely, the z score for stunting (low height for age) of a child i is the difference between the height of that child, Hi, and the median height of a group of healthy children of the same age and sex from the reference population, Hr, divided by the standard deviation of the height of the same group of children (same age and sex) from the reference population, SDr: z score = Hi - Hr . SDr Relatively short children have negative height for age z scores; thus, stunted children are commonly defined as those who have z scores of -2 or lower. The z scores for underweight (low weight for age) children are calculated in the same way, using the weight of the child (instead of height) and the median weight (and standard deviation) of children of the same age and sex from a healthy reference population. The z scores for wasting (low weight for height) are obtained by comparing the weight of the child with the me- dian weight (and standard deviation) of children from the reference popula- tion who have the same height as that child. The reference population was selected by the National Center for Health Statistics, in accordance with World Health Organization (WHO) recommendations (WHO 1983). The two preferred anthropometric indexes for the measurement of nutri- tional status of children are stunting and wasting, because they distinguish between long-run and short-run physiological processes (WHO 1986). The wasting index has an additional advantage: It can be calculated without knowing the child's age. It is particularly useful in describing the current health status of a population and evaluating the benefits of intervention pro- grams, because it responds more quickly to changes in nutritional status than does stunting. Adisadvantage of this index, however, is that it classifies children with poor growth in both weight and height as normal (Gibson 1990). Stunting measures long-run social conditions because it reflects past nutritional status. Thus, the WHO (1986) recommends it as a reliable mea- sure of overall social deprivation. Children's Nutritional Status in Vietnam Young children who receive sufficient breast milk, infant foods, and adult foods grow quickly and attain their potential weight and height, unless di- arrhea or other illnesses intervene. In developing countries, children who fail to attain their potential growth typically suffer from inadequate dietary intake, illness, or both. During the first few years of life, the single most important factor is the incidence of diarrhea. Children who are exclusively breastfed are much less likely to be exposed to pathogens that lead to 354 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 10.1. Diarrhea, Stunting, and Wasting, 1993 Percent 70 60 50 40 30 20 10 0 0­5 months monthsmonths months2 years 3 years 4 years 5 years 6 years 7 years 8 years 9 years 6­11 ­17 ­23 12 18 Age With Diarrhea Wasted Stunted Source: Authors' calculations using the 1993 VLSS. diarrhea and other gastrointestinal diseases; they also receive immunogens from breast milk. Yet when weaning foods are introduced, typically in the first three to six months of life, infants are exposed to many pathogens that often lead to diarrhea and other diseases. This typical pattern is found in the 1993 VLSS data. Figure 10.1 shows that the incidence of diarrhea (in the four weeks preceding the survey interview) in the first 18 months of life steadily rises to about 6 percent before declining to less than 1 percent among chil- dren ages 3 years and older. Repeated bouts of diarrhea interfere with human growth, leading to low weight gain. This is seen in the data on wasting in figure 10.1. The incidence of wasting (defined as a weight-for-height z score below -2) is only about 2 percent among children ages 0­5 months who are still primarily breastfed. Yet wasting increases steadily during the first 2 years of life, peaking at about 11 percent for children ages 18­23 months. For children 2 years and older, wasting fluctuates around 5 percent; by this age, wasting is more likely caused by inadequate food intake because diarrhea is relatively rare. The long-run consequence of diarrhea, other illnesses, and inadequate food intake is stunting. As seen in figure 10.1, stunting (defined as a Child Nutrition, Economic Growth, and the Provision of Health Care Services 355 Figure 10.2. Diarrhea, Stunting, and Wasting, 1998 Percent 70 60 50 40 30 20 10 0 0­5 months monthsmonths months2 years 3 years 4 years 5 years 6 years 7 years 8 years 9 years 6­11 ­17 ­23 12 18 Age With Diarrhea last 4 weeks Wasted Stunted Source: Authors' calculations using the 1998 VLSS. height-for-age z score below -2) rises dramatically during the first 2 years of life, from about 14 percent for children ages 0­5 months to 65 percent for children ages 18­23 months, and then settles down to about 55 percent for children ages 2­9 years. Thus, Vietnamese children follow the typical pat- tern, in that their worst bouts of malnutrition occur during the first 2 years of life, and as a consequence slightly more than one-half of them become stunted by age 2 and remain so for the rest of their lives. This was the situation in the early 1990s. The situation in the late 1990s showed substantial improvements, at least in terms of stunting. Figure 10.2 shows the same general pattern, with wasting and diarrhea peaking in the second year of life, so that stunting again typically develops during the first two years of life, after which it remains relatively high. However, stunting in 1998 was less common than in 1993 for all age categories. Figures 10.1 and 10.2 demonstrate typical patterns of malnutrition and show that stunting declined during the 1990s, but two anomalies stand out. First, diarrhea appears to have increased dramatically for almost all age groups. This apparent increase is spurious, because the question was asked differently in the two surveys. In the 1993 VLSS, each person was asked (or, 356 Economic Growth, Poverty, and Household Welfare in Vietnam for small children, parents were asked) whether he or she had been sick in the last four weeks and, if so, what illness. This underestimates the inci- dence of diarrhea for two reasons. First, some people may think of diarrhea as normal and so would answer that they had not been sick during the past four weeks. Second, persons suffering from more than one illness in the past four weeks were allowed to report only one illness. Thus, if a child had di- arrhea and another illness, the other illness may have been reported instead of diarrhea. In the 1998 VLSS, all individuals were asked directly whether they had had diarrhea in the past four weeks, which resulted in a much higher reported incidence of diarrhea. The second anomaly is that wasting appears to have increased, which is inconsistent with the dramatic decline in stunting. More specifically, the data show that, for each age category, average weight and height increased substantially between 1993 and 1998, indicating that the nutritional status of Vietnamese children greatly improved in that period. However, height increases were larger than weight increases, so that the change in weight for height indicates increased wasting. This suggests that examining changes in weight for height over time may provide a misleading picture of changes in children's nutritional status when overall nutritional status increases rapidly. Similar contradictory results over time have been found in Sub- Saharan Africa (Sahn, Stifel, and Younger 1999). More information about the nature of malnutrition (as measured by stunting and wasting) is provided in table 10.1. The first three columns pre- sent data from the 1993 VLSS. At that time, the overall incidence of stunting for children ages 0­60 months was about 50 percent, and the incidence of wasting was about 6 percent. Stunting was most prevalent in rural areas, af- fecting about 53 percent of the population, while only 33 percent of urban children were stunted. This is not surprising, because real per capita expen- ditures in urban areas were almost double those of rural areas (dong [D] 1,899,000 compared with D 990,000), as seen in the third column. The figures on wasting are somewhat surprising in that they are almost the same for urban and rural children (5.7 percent and 5.9 percent, respectively). Regional rates of stunting and wasting are also instructive. Vietnam has seven regions, as shown in table 10.1. The two regions with the highest rate of stunting in 1992­93 were the Northern Uplands and the North Central Coast. As one would expect, these two regions also had the lowest average per capita expenditures. Stunting was least common in the Southeast region, which includes the largest city in Vietnam (Ho Chi Minh City, formerly known as Saigon) and has the highest per capita expenditures. This strong correlation of stunting and per capita expenditures is not found in the data on wasting. Wasting is most common in the Mekong Delta (7.7 percent), even though it had the second highest per capita expenditures. The North Central Coast had the lowest incidence of wasting (4.1 percent), despite having the second lowest per capita expenditures. Overall, there is no clear correlation between wasting and per capita expenditures, which casts doubt on its use as an indicator of nutritional status. es dong) capita 1,593 2,462 1,716 1,959 1,855 4,340 2,154 4,099 1,816 2,227 Per expenditur (thousand 1998 cent) asting 9.8 1.01 15.8 8.1 6.8 7.3 2.3 8.4 1.41 10.8 W (per cent) 42.3 26.6 40.7 37.1 43.3 17.8 33.5 18.4 38.2 34.6 Stunting (per es dong) Months capita 804 877 983 990 60 1,049 1,232 1,754 1,333 1,899 1,128 to Per 0 expenditur (thousand Children 1993 cent) asting 6.2 5.4 4.1 5.5 5.5 5.5 7.7 5.7 5.9 5.8 VLSSs. W (per 1998 Region, and by 1993 asting cent) the W 58.7 54.2 57.8 46.5 52.7 29.8 45.0 33.2 53.2 50.2 Stunting (per using and calculations' Stunting Coast Uplands Delta 10.1. Coast Highlands Delta Authors Central River bleaT urban uralr ietnamV ce: Sour Location Northern Red North Central Central Southeast Mekong All All All 357 358 Economic Growth, Poverty, and Household Welfare in Vietnam Columns 4­6 of table 10.1 present information from the 1998 VLSS. Per capita expenditures increased in real terms in all regions (although deflated numbers are not presented), and the incidence of stunting declined by al- most one-third, from 50 percent to 35 percent. This decline is seen in both urban and rural areas and in all seven regions. The region with the largest percentage increase in per capita expenditures--the Red River Delta, which moved from fourth highest to second highest per capita expenditures--had the largest decline in stunting, from 54 percent to 27 percent. In contrast, the wasting data are rather puzzling. Wasting increased in urban and rural areas of Vietnam and in all seven regions. It shows no clear relationship with income or with changes in income. Given that other indi- cators of child health show improvement over this period--for example, the infant mortality rate dropped from 44 to 39 (see Wagstaff and Nguyen [2004; chapter 9 in this volume] and World Bank [2001a])--the rest of this chapter will focus on the stunting data. Further explanation of the counterintuitive findings in the wasting data would probably require a detailed nutritional study, which is beyond the scope of this chapter. Vietnam's Economic Performance in the 1990s In the 1980s, Vietnam was one of the poorest countries in the world. A rough estimate of its GNP per capita in 1984 is US$117. This would place it as the second poorest country in the world, barely ahead of Ethiopia and just be- hind Bangladesh (as reported in World Bank [1986]). By 1999, Vietnam's GNP per capita had increased to US$370, so that it ranked 167 out of 206 countries (World Bank 2001b). This rapid improvement in Vietnam's economic performance began in 1986, when the first Doi Moi ("renovation") economic policies started to transform Vietnam from a planned to a market-oriented economy. In partic- ular, the government disbanded state farms and divided agricultural land equally among rural households, removed price controls, legalized buying and selling of almost all products by private individuals, stabilized the rate of inflation, and opened up the economy to foreign trade and investment (see Chapter 1 for a more detailed discussion of these policy changes). In the 1990s, Vietnam was one of the 10 fastest-growing economies in the world, with an average real gross domestic product growth of 8.4 percent per year from 1992 to 1998. This rapid economic growth has led to a dramatic decline in the rate of poverty, from 58 percent in 1993 to 37 percent in 1998 (World Bank 1999). As seen in table 10.1, it also appears to have led to large decreases in the rate of stunting among Vietnamese children. Are these dramatic increases in the incomes of Vietnamese households the main cause of the large decreases in stunting among young children? Table 10.2 provides evidence. For each VLSS, households were divided into five groups of equal size on the basis of their per capita expenditures. The first group, quintile 1, is the poorest Child Nutrition, Economic Growth, and the Provision of Health Care Services 359 Table 10.2. Malnutrition by Expenditure Quintiles, Children 0 to 60 months 1998 with Quintile 1993 1998 1993 quintile Stunting (percent) 1 58.6 41.3 45.0 2 59.1 42.1 38.6 3 45.3 32.6 41.6 4 44.4 27.5 34.0 5 29.2 14.2 18.5 Wasting (percent) 1 5.4 13.2 15.0 2 5.5 12.4 12.1 3 6.9 10.0 12.5 4 5.9 5.3 9.2 5 5.8 9.0 7.2 Source: Authors' calculations using the 1993 and 1998 VLSSs. 20 percent of the population. In 1993, about 59 percent of the children in that group were stunted. The second poorest group, quintile 2, had about the same rate. Quintiles 3, 4, and 5 had steadily lower rates of 45 percent, 44 per- cent, and 29 percent, respectively. The same pattern is seen in the 1998 VLSS; the incidence of stunting among the poorest quintile is 41 percent and steadily drops to 14 percent for the wealthiest quintile. This pattern, based on cross-sectional data, suggests that higher incomes reduce child malnutri- tion. In contrast, the data on wasting show no such pattern, adding to doubts about the informational content of this nutritional indicator, at least in the context of Vietnam. The stunting data in table 10.2 show that stunting rates decline over time within each quintile. This suggests that something in addition to income growth was leading to reduced malnutrition in Vietnam in the 1990s. Yet these quintiles are not strictly comparable, because the poorest 20 percent of the population in 1998 had a higher income than the poorest 20 percent in 1993. The last column in table 10.2 adjusts for this difference, classifying households in the 1998 VLSS according to the quintile categories used in the 1993 VLSS. Even after this adjustment is made, there are still dramatic de- clines in stunting for households in the same expenditure group. This sug- gests that increased household income is not the only factor that improved the nutritional status of Vietnamese children. The rest of this chapter will examine this phenomenon more formally. 360 Economic Growth, Poverty, and Household Welfare in Vietnam Data and Analytical Framework The 1993 VLSS covered 4,800 households, and the 1998 VLSS covered 6,002 households. Both surveys are nationally representative. About 4,300 house- holds were interviewed in both surveys and thus constitute a large, nation- ally representative panel dataset. In both surveys, the household question- naire covered many different topics, including education, health (use of health care facilities and anthropometric measurements of all household members), employment, migration, housing, fertility, agricultural activities, small household businesses, income and expenditures, and credit and sav- ings. In each year, community questionnaires were completed in rural areas (where about 80 percent of Vietnamese households live) and detailed price questionnaires were completed in both urban and rural areas. The 1998 VLSS also included health facility and school questionnaires. For further information on the VLSS, see the appendix to chapter 1. These two surveys are well suited for examining the determinants of children's nutritional status. All household members, children and adults, were measured for height, weight, and arm circumference. The vast amount of household information, including detailed income and expenditure data, reduces problems of omitted variable bias. The panel data allow for estima- tion that controls for unobserved household fixed effects. Finally, the 1998 data include a large amount of information on the prices of medicines and the types of medical services (and their costs) provided by local health care facilities. The data presented in the section above show changes over time but can- not explain what caused those changes, or more generally what determines children's nutritional status. Such causal analysis is much more difficult, and it requires a clear analytical framework to avoid drawing false infer- ences from the data. The starting point for thinking about the determinants of a child's nutri- tional status is a health production function, because nutritional status is a major component of child health. In general, a child's health status (H) is de- termined by three kinds of variables: health inputs (HI), the local health en- vironment (E), and the child's genetic health endowment (). Equation 10.1 shows child health status as a function ( f ) of these three types of variables: (10.1) H = f(HI, E, ). The child's genetic health endowment () is defined as all (genetically) inher- ited traits that affect his or her health. It is exogenous (cannot be altered by the child or anyone else) but rarely observed in any data. The local health en- vironment (E) consists of the characteristics of the community in which the child lives that have a direct effect on his or her health, such as the prevalence of certain diseases and the extent of environmental pollution. It is also ex- ogenous, although one could argue that it is endogenous to the extent that households migrate to areas with healthier environments or take measures to improve the local health environment. (This issue will be discussed further Child Nutrition, Economic Growth, and the Provision of Health Care Services 361 below.) Finally, there are a wide variety of health inputs (HI) that the house- hold provides to the child, including prenatal care, breast milk, infant for- mula, all other foods, medicines, and medical care. In addition, the quality of the household's drinking water, toilet facilities, and other hygienic condi- tions around the home can be treated as health inputs. Researchers would often like to estimate a health production function, but it is almost impossible to do so: Complete data on health inputs and the local health environment are rarely available, and data on the child's genetic endowment are rarer still. This incompleteness may well lead to serious problems of omitted variable bias. Analysis is further complicated by the need to have this information not only for the current time period but for all past time periods of the child's life. A more practical alternative is to con- sider what determines health inputs and substitute out that variable from equation 10.1. In general, the health inputs that households choose for their children are determined by the household's income level (Y), the education levels of both parents (MS and FS, for mother's schooling and father's schooling), their preferences for child health (), the local health environ- ment, and the child's genetic health endowment. Equation 10.2 shows this relationship, depicted in terms of a function g( ): (10.2) HI = g(Y, MS, FS, , E, ). Note that family size and the presence of other siblings are not included as determinants of health inputs. They are excluded because they are clearly endogenous, and including endogenous variables can lead to biased esti- mates unless suitable estimation methods, such as instrumental variables, are used. Thus, it is best to include in equation 10.2 only variables that are clearly exogenous. Of course, one could rightly claim that household in- come is endogenous; for example, parents may change their hours of work in response to the health status of their children. However, removing this variable from equation 10.2 would preclude estimation of the key relation- ship of interest in this chapter, so it is retained. The approach used to deal with possible estimation biases from retaining this variable is discussed below. Substituting (10.2) into (10.1) gives the basic equation that this chapter attempts to estimate: (10.3) H = g(Y, MS, FS, , E, ). The child's height-for-age z score will be used as the indicator of a child's health status, H. As mentioned above, both surveys have data on household income and expenditures. Household per capita expenditures will be used instead of household per capita income to measure Y, for two reasons. First, expenditure data are likely to be more accurate than income data (Deaton 1997). Second, expenditure data are more likely to reflect a household's per- manent income, which is more appropriate in this case because Y represents the household's income stream since the child was born, not just current income. 362 Economic Growth, Poverty, and Household Welfare in Vietnam The remaining variables in equation 10.3 merit further comment. The schooling of each parent is provided in both surveys, even for children no longer living with one or both parents (9 percent of the children in the sam- ple are not living with their father, and 2 percent are not living with their mother). However, parental preferences for child health, , are difficult to ascertain and no attempt was made to do so here. Dropping this variable from the estimation altogether is risky; doing so would relegate it to the error term, and it could be correlated with household income (which would lead to biased estimates of the impact of household income on child health). For example, some parents may be irresponsible, which often implies low preferences for child health and low income. This would lead to overesti- mation of the impact of income on child health. This chapter uses three approaches to deal with this problem. First, dummy variables representing different ethnic and religious groups are included to approximate, albeit only partially, preferences for child health. Second, in some estimates, in- strumental variables are used for household per capita expenditures, which should eliminate some or perhaps even all of the bias due to correlation between income and unobserved preferences for child health. Third, some estimates presented below are based on panel data; if parental preferences can be specified as an additive fixed effect, that variable will be differenced out of the panel estimates. The last two variables in equation 10.3 are the local health environment, E, and the child's innate healthiness (genetic health endowment), . The estimates presented in the section titled "Income Growth and Child Nutri- tion" use community fixed effects to control for all differences across com- munities, including differences in the local health environment. In "Health Programs and Child Nutrition," a different approach is used; data from the 1998 VLSS on local health conditions, medicine prices, and the availability of medical services are used to measure explicitly the impact of the local health environment on child health. Finally, consider the child's genetic health endowment, . In the cross-sectional estimates, this is partially represented by the height of each parent (which reflects both normal variation in height that is not associated with health status and the innate healthiness of each parent) and by the sex of the child (because girls are typically healthier than boys--note that this masks any sex discrimination that may be taking place). In estimates using panel data, the average healthiness of each household's children is treated as a fixed effect and thus is differenced out. The last issue to address is the problem that household income is endogenous, which raises the possibility of simultaneity bias. In general, households make decisions about their children's health at the same time that they make decisions about income-earning activities, and these two decisions could be related. For example, parents whose children are chroni- cally ill may decide to purchase costly medicines or medical services. To do this, some household members may work more hours to pay for those medicines. If so, ordinary least squares (OLS) estimates would tend to un- derestimate the impact of household income (expenditures) on child health Child Nutrition, Economic Growth, and the Provision of Health Care Services 363 because unobserved negative shocks to child health would increase house- hold income. Alternatively, households may reduce hours worked because of a child's illness--for example, the mother may work fewer hours so that she can spend more time caring for the child. In this case, OLS would over- estimate the impact of household income (expenditures) on child health. Another problem with both household income and expenditure data is that they are often measured with random error, simply because it is diffi- cult for households to report accurately their incomes and expenditures. This chapter uses household expenditures instead of household income be- cause it is likely to be more accurate. However, even household expendi- tures may have a significant amount of measurement error, much of which will be random. This will lead to underestimation (attenuation bias) of the true impact of household expenditures on child health. Instrumental variable methods can, in principle, remove the bias caused by either endogeneity or measurement error in the household expenditures variable. The difficulty is to find plausible instrumental variables--variables that are correlated with household income but uncorrelated with unob- served determinants of child health and uncorrelated with the measurement error in the household expenditure variable. Two plausible categories of instrumental variables are types of agricultural land allocated to the house- hold and certain sources of nonlabor income. In Vietnam, agricultural land is tightly controlled by the government, and markets for land simply do not exist in most rural communities. (Fewer than 3 percent of households in the 1993 VLSS reported that they had bought or sold land in the previous year.) Thus, households' land assets are unlikely to be influenced by children's health status. Similarly, some types of nonlabor income are received regard- less of children's health status. Thus, the following instrumental variables are used to predict households' per capita expenditures: irrigated land for annual crops, unirrigated land for annual crops, perennial cropland, water surface (fish ponds), and income from social funds, social subsidies, dowries, inheritances, and lottery winnings. Finally, relatives (more specifi- cally, children of household members) living overseas may also indicate an additional source of income; although the amount of remittances sent by such relatives may respond to child illnesses, the existence of such relatives is unlikely to be affected by those illnesses. Two such variables are used: overseas relatives in other Asian countries and overseas relatives in Western countries. As will be seen below, although these instrumental variables have statis- tically significant predictive power, they are rather weak in terms of the R2 coefficient in the first-stage regressions. If the main problem is measurement error, as opposed to household expenditures being correlated with unob- served determinants of child health, then one could use household income as an instrumental variable for household per capita expenditures. In the re- gression results presented below, two sets of instruments are used: one with- out household income, which should be robust to both endogeneity and measurement error in the income variable, and one that adds income, which 364 Economic Growth, Poverty, and Household Welfare in Vietnam is robust to measurement error but is invalid if household income is en- dogenous with respect to child health. Income Growth and Child Nutrition This section presents estimates of equation 10.3 above. In all estimates, the dependent variable is the child's height-for-age z score, and the sample in- cludes only children of ages 0­60 months. Separate estimates are presented for urban and rural areas. For cross-sectional estimates, results are given for both 1993 and 1998. The cross-sectional estimates are presented first, followed by panel data estimates. Cross-Sectional Estimates Table 10.3 presents estimates of equation 10.3, the determinants of child malnutrition (as measured by height-for-age z scores), for urban areas of Vietnam in 1993. The first column presents OLS estimates, which are likely to suffer from omitted variable bias because of unobserved characteristics of local communities (such as the local health environment). OLS estimates may also be biased because they do not account for endogeneity or measurement error in the household expenditure variable. The second column of estimates includes community fixed effects, which avoid bias due to unobserved com- munity characteristics as long as those variables enter equation 10.3 in a sim- ple additive form without interaction terms with household- or child-level variables. Yet fixed effects estimates are not robust to endogeneity or mea- surement error in the expenditure variable. The third and fourth columns use both community fixed effects and instrumental variables for household expenditures. The third column does not use household income as an instru- ment, so it should be robust to both measurement error and endogeneity, and the fourth adds household income as an instrument and thus controls only for measurement error. Although the OLS estimates in the first column are likely to be biased, it is useful to begin with them because the results for many variables change only slightly when other estimation methods are used. As one would expect given the data in figures 10.1 and 10.2, the age of the child has a strong rela- tionship to malnutrition as measured by stunting. In addition to a linear term (age in months), quadratic and cubic terms were added to allow for flexibility in this relationship. Mothers' height and fathers' height are both strongly and positively related to child health, which partially controls for unobserved children's health endowment but also reflects natural variation in height across a healthy population. For some mothers and fathers (4 per- cent of mothers and 16 percent of fathers), the height variable was missing. In this case, the parent is assigned the average height and a dummy variable is added to indicate this type of observation. In all the regressions in table 10.3, girls in urban areas appear to be slightly healthier than boys, but this apparent advantage is never statistically 1.00 page) Mean 31.5 n.a. n.a. 0.04 0.16 0.48 1.95 2.09 7.34 153.0 162.4 following on b n.a. 2SLSFE 0.219***- 0.007*** 0.046*** 0.721** 0.052*** (0.037) (0.001) 0.00006***- (0.012) (0.362) (0.013) 0.251*- 0.101 0.162 (0.152) (0.109) (0.128) 0.224*- 0.502 continues (0.122) (0.308) (0.00002) (table a 18) n.a. 2SLSFE 0.220***- 0.007*** 0.046*** 0.666 0.052*** (0.039) (0.002) 0.00007***- (0.012) (0.407) (0.013) 0.257*- 0.100 0.180 (0.148) (0.109) (0.132) 0.207*- 0.441 (0.1 (0.443) (0.00002) 1993 effects n.a. 0.221***- 0.007*** 0.046*** 0.652** 0.052*** 0.100 0.185* (0.038) (0.001) (0.012) (0.285) (0.013) (0.153) (0.109) (0.105) (0.92) 0.388*** Areas, Community fixed 0.00007***- 0.257*- 0.203**- (0.140) (0.00002) Urban in least es 12) 14 squar 19.245***- (2.62) 0.223***- 0.007*** 0.051*** 0.905*** 0.051*** (0.034) (0.001) 0.00007***- (0.010) (0.266) (0.012) 0.252*- 0.073 0.126 (0.144) (0.1 (0.098) 0.1- 0.493*** (0.100) (0.108) (0.00001) Malnutrition Ordinary Child of se schooling schooling (centimeters) missing Determinants (centimeters) missing expenditur (months) ed (months) yearss' 10.3. heights' heights' yearss' heights' heights' child capita (months) squar cubed mother father per bleaT ariableV Constant Age Age Age Mother Mother Father Father Female Log Log Log 365 1 Mean 0.30 0.14 0.002 0.02 0.1 0.02 n.a. n.a. n.a. funds, social omfr b 415 countries. come 2SLSFE 0.204- (0.162) 0.121- (0.337) 2.306***- (0.241) 1.202- (0.874) 0.631- 0.325 0.350 0.364 36.29 in (0.405) (0.681) fects. ef surface; on-Asiann fixed in water and a and 15 415 community 2SLSFE 0.207- (0.164) 0.1- (0.331) 2.346***- (0.256) 1.221- (0.891) 0.616- 0.267 0.351 0.284 41.39 countries (0.405) (0.736) with opland, es cr Asian squar ennial other least per in a. note 14 effects 415 opland, elativesr in 0.207- (0.158) 0.1- (0.327) 2.356***- (0.174) 1.226- (0.898) 0.612- 0.251 0.351 n.a. n.a. (0.401) (0.657) two-stage cr of = Community fixed annual variables 2SLSFE existence the umental and least entheses. unirrigated instr es 415 par the squar 0.018- 0.129 (0.129) (0.215) 1.992***- (0.223) 0.966***- (0.347) 0.161- 0.386 0.283 n.a. n.a. in to (0.218) (0.642) data. opland, winnings; cr Ordinary VLSS design) income lottery annual level. level. level. 1993 and sample the cent cent cent for household per per per irrigated using ) value) 10 5 1 e p( uments at at at ar capita inheritances, (adjusted per test instr ors add continued( significant err variables calculations' significant significant d dowries, excluded applicable. umental eligionr estimates Authors 10.3. minority Standar on Not subsidies, ce: Instr The bleaT ariableV otestant n.a. *Statistically **Statistically ***Statistically Note: a. b. Sour 2 test Buddhist Catholic Pr Other Chinese Ethnic R Overidentification F Observations social 366 Child Nutrition, Economic Growth, and the Provision of Health Care Services 367 significant. The impact of mothers' and fathers' schooling is usually not sta- tistically significant, which is somewhat surprising, especially for mothers. One would think that better-educated mothers are more able to care for their children's illnesses, other things being equal. Perhaps better-educated women are also more likely to work outside the home, which could have negative consequences for their children's health, so that the net effect of the mother's education is zero. Finally, there are few differences across ethnic and religious groups in urban areas (the omitted groups are Vietnamese and "no religion"), the two exceptions are that Protestant households and house- holds practicing religions other than Buddhism and Christianity had chil- dren who were significantly less healthy. Both groups are relatively rare in urban areas, and it is not clear what to make of this result; indeed, the result for Protestants is based on a single child and so should be treated with cau- tion. Because the focus of this chapter is on the impact of household income and health care services, these apparent impacts of religion on child health will not be discussed further. The discussion turns now to the impact of per capita household expen- ditures (expressed in natural logarithm) on child health. The OLS estimate is 0.493, which is fairly precisely estimated (the standard error of 0.108 yields a t statistic of 4.56). This is higher than the estimate of 0.22 found by Ponce, Gertler, and Glewwe (1998), but that study included older children (up to nine years old); thus, the estimates are not strictly comparable. Even if household expenditures were exogenous and measured without error, the OLS estimate of the corresponding coefficient is likely to be biased by the correlation between household income and unobserved community differences. The basic problem is that wealthier communities may have a better health environment--for example, better sanitation and health care facilities. If these community characteristics have effects that are primarily additive, community fixed effects estimates will remove this bias. Such esti- mates are shown in the second column of table 10.3. As expected, the impact of household per capita expenditures is smaller, falling from 0.493 to 0.388. Yet the impact of household expenditures on child health is still statistically significant (the standard error is 0.140, yielding a t statistic of 2.77). The last two specifications in table 10.3 attempt to correct for endogene- ity and measurement error in household expenditures. The third column presents estimates that instrument household expenditures using the land asset and nonlabor income variables. Although these instrumental variables have strong explanatory power in the sense that they have a high F statistic (41.39), they do not by themselves explain a large percent of the variation of per capita household expenditures (the R2 coefficient of a regression of the expenditure variable on the excluded instruments is only 0.08). Thus, although the coefficient on per capita expenditures does not change appre- ciably (it is 0.441), it is not statistically different from zero because the stan- dard error has increased to 0.443. This imprecision implies that one can say almost nothing about the impact of household expenditures on child health in urban areas of Vietnam in 1993. 368 Economic Growth, Poverty, and Household Welfare in Vietnam Somewhat higher precision can be obtained if one assumes that house- hold expenditures are exogenous, so that the only estimation problem is measurement error. This allows per capita income to be used as an instru- mental variable, which increases the precision of the estimates. When this is done, the coefficient rises slightly to 0.502. Although the standard error falls from 0.443 to 0.308, the coefficient is still quite imprecisely estimated and thus not significantly different from zero (t statistic of 1.63). Note finally that the standard overidentification test (see Davidson and MacKinnon [1993]) suggests that the instrumental variables are uncorrelated with the residual, although the power of the test to detect this problem may not be very high. Overall, it is difficult to estimate with any precision the impact of household expenditures on children's nutritional status in urban areas of Vietnam in 1993 once one accounts for the possibility that the expenditure variable may be endogenous and may be measured with error. Cross-sectional results for rural areas of Vietnam in 1993 are reported in table 10.4. The age and parental height variables show the same patterns as in urban areas. As in urban areas, girls are somewhat healthier than boys, but the difference is not statistically significant. Mothers' schooling has a marginally significant negative effect in the OLS results, but this counterin- tuitive finding disappears in the fixed effects and two-stage least squares with fixed effects (2SLSFE) estimates. Fathers' schooling has a significantly positive effect in the fixed effects results, but this is not seen in the other specifications. Most estimates regarding religious and ethnic groups are statistically insignificant, except that, again, Protestant children are more malnourished and, in some specifications, ethnic minorities are more likely to be malnourished. Focusing on the (log) household expenditure variable, the OLS results show a precisely estimated impact of 0.336 (the standard error is 0.072). As in urban areas, this figure declines, when community fixed effects are intro- duced, to 0.185 (with a standard error of 0.92). The third column in table 10.4 specifies the expenditure variable as en- dogenous, using the land and nonlabor income variables as instruments. The point estimate is quite large, at 0.724, but the precision of the estimate is quite low because the standard error increases to 0.437. This imprecision is not surprising because a regression of household expenditures on the ex- cluded instruments alone yields an R2 coefficient of only 0.060. When (log) per capita income is added as an instrument, the coefficient drops to 0.500; although the standard error is smaller (0.305), this estimate is not quite sig- nificant at the 10 percent level (t statistic of 1.64). Finally, note that both 2SLSFE specifications easily pass the overidentification test. The 1998 VLSS had a larger sample size, which may provide more precise estimates. The results for urban and rural areas are presented in tables 10.5 and 10.6, respectively. Many results for urban areas in 1998 are similar to those for 1993. The age effects and parental height impacts are similar, although somewhat weaker; the sex of the child and parental schooling show no consistently significant effects; and the impacts of the religion variables are 1 1.00 Mean 30.9 n.a. n.a. 0.03 0.1 0.49 1.63 1.82 6.77 page) 151.7 161.9 following on b 19) n.a. 2SLSFE 0.208***- 0.006*** 0.039*** 0.009 0.021*** (0.022) (0.001) 0.00005***- (0.006) (0.215) (0.008) 0.000- (0.1 0.081- 0.008 0.069 0.500 continues (0.060) (0.073) (0.076) (0.305) (0.00001) (table a n.a. 2SLSFE 0.209***- 0.006*** 0.038*** 0.002 0.019** 0.004 (0.023) (0.001) 0.00005***- (0.006) (0.215) (0.008) (0.121) 0.079- (0.061) 0.029- 0.034 0.724* (0.086) (0.090) (0.437) (0.00001) 1993 19** effects n.a. 0.205***- 0.006*** 0.040*** 0.018 0.023*** 0.060 0.1 0.185** (0.022) (0.001) Areas, Community fixed 0.00005***- (0.006) (0.219) (0.007) 0.007- (0.120) 0.083- (0.059) (0.055) (0.060) (0.092) (0.00001) Rural in es 14) 14) squar Ordinary 12.909***- (1.283) 0.207***- 0.006*** 0.042*** 0.028*** 0.067 0.336** (0.021) (0.001) least 0.00005***- (0.006) 0.013- (0.1 (0.007) 0.026- (0.1 0.084- (0.061) 0.086*- (0.050) (0.058) (0.072) (0.00001) Malnutrition Child of se schooling schooling (centimeters) missing (centimeters) Determinants missing (months) (months) yearss' expenditur ed 10.4. heights' heights' yearss' heights' heights' child capita (months) squar cubed mother father per bleaT ariableV Constant Age Age Age Mother Mother Father Father Female Log Log Log 369 Mean 0.23 0.09 0.01 0.02 0.01 0.19 n.a. n.a. n.a. funds, social omfr b 19) countries. come 2SLSFE 0.017- (0.126) 0.212- (0.140) 0.529***- 0.094 0.208 (0.078) (0.375) (0.370) 0.198*- 0.252 0.591 16.59 2,372 in (0.1 fects. ef surface; on-Asiann fixed in water and a and community 0.006 (0.137) 2SLSFE 0.206- (0.139) 0.510***- 0.068 0.144 (0.091) (0.389) (0.384) 0.170- 0.242 0.559 10.84 2,372 countries (0.129) with opland, es cr Asian squar ennial other least per in a. note effects 17) 14) in 0.050- (0.1 0.220- (0.136) 0.556***- 0.130 0.297 (0.058) (0.355) (0.343) 0.238**- 0.257 2,372 opland, lativeser n.a. (0.1 n.a. two-stage cr of = Community fixed annual variables 2SLSFE existence the umental es and 11 entheses. unirrigated instr par the squar 0.042 (0.080) 0.1- (0.125) 0.248- 0.203 0.076 (0.157) (0.336) (0.250) 0.322***- 0.196 n.a. n.a. 2,372 in (0.093) to Ordinary opland, winnings; least cr VLSS. design) income lottery annual level. level. level. 1993 and sample the cent cent cent for household per per per irrigated using ) value) 10 5 1 e p( uments at at at ar capita inheritances, (adjusted per test instr ors add continued( significant err variables calculations' significant significant d dowries, excluded applicable. umental eligionr estimates Authors 10.4. minority Standar on Not subsidies, ce: Instr The bleaT ariableV otestant n.a. *Statistically **Statistically ***Statistically Note: a. b. Sour 2 test Buddhist Catholic Pr Other Chinese Ethnic R Overidentification F Observations social 370 1.00 page) Mean 31.6 n.a. n.a. 0.02 0.12 0.49 1.99 2.08 8.13 152.5 162.5 following on b )1 n.a. 2SLSFE 0.127***- 0.003** 0.033** 0.018 0.032*** (0.041) (0.001) 0.00003**- (0.013) (0.373) (0.01 0.185- 0.070 (0.200) (0.144) 0.026- (0.140) 0.256*- 0.674*** continues (0.150) (0.226) (0.00001) (table a n.a. 0.121**- 0.003* 0.034** 0.034*** 0.082 0.059 0.401 (0.047) (0.002) 2SLSFE 0.00003- (0.015) 0.033- (0.406) (0.012) 0.206- (0.203) (0.146) (0.305) 0.196- (0.214) (0.865) (0.00002) 1998 15*** )1 effects n.a. 0.1- 0.003** 0.036*** 0.036*** 0.092 0.138 0.146 (0.039) (0.001) Areas, Community fixed 0.00002*- (0.013) 0.080- (0.361) (0.01 0.225- (0.186) (0.137) (0.102) 0.139- (0.148) (0.120) (0.00001) Urban in es 10) squar Ordinary 15.028***- (2.306) 0.140***- 0.004*** 0.038*** 0.041*** 0.073 0.181** 0.341*** (0.034) (0.001) least 0.00003***- (0.013) 0.064- (0.417) (0.010) 0.140- (0.181) (0.126) (0.090) 0.063- (0.134) (0.1 (0.00001) Malnutrition Child of se schooling schooling (centimeters) missing Determinants (centimeters) missing expenditur (months) ed (months) yearss' 10.5. heights' heights' yearss' heights' heights' child capita (months) squar cubed mother father per bleaT ariableV Constant Age Age Age Mother Mother Father Father Female Log Log Log 371 Mean 0.25 0.07 0.01 0.01 0.06 0.01 n.a. n.a. n.a. funds, social omfr b 469 countries. come 2SLSFE 0.122- 0.087 (0.143) (0.169) 0.881***- (0.132) 0.597- 1.016*** 1.129 0.357 0.608 26.05 in (0.492) (0.328) (0.787) fects. ef fixed surface; on-Asiann in water and a and 469 community 0.121- 0.078 (0.146) (0.170) 0.842***- (0.174) 0.648- 0.990*** 1.132 0.379 0.499 14.25 countries (0.452) (0.325) (0.777) 2SLSFE with opland, es cr Asian squar ennial other least per in a. note effects 469 opland, elativesr in 0.120- 0.070 (0.150) (0.172) 0.806***- (0.202) 0.696**- 0.966*** 1.134 0.386 n.a. n.a. (0.338) (0.313) (0.773) two-stage cr of = Community fixed annual variables 2SLSFE existence the umental es and entheses. unirrigated instr 16) 1) 469 par the squar 0.145- (0.1 0.061- (0.185) 0.641***- (0.21 0.574*- 0.857*** 0.862 0.260 n.a. n.a. in (0.301) (0.306) (0.721) to opland, Ordinary winnings; cr least VLSS. design) income lottery annual level. level. level. 1993 and sample the cent cent cent for household per per per irrigated using ) value) 10 5 1 e p( uments at at at ar capita inheritances, (adjusted per test instr ors add continued( significant err variables calculations' significant significant d dowries, excluded applicable. umental eligionr estimates Authors 10.5. minority on Standar Not subsidies, ce: Instr The bleaT ariableV otestant n.a. *Statistically **Statistically ***Statistically Note: a. b. Sour 2 tests Buddhist Catholic Pr Other Chinese Ethnic R Overidentification F Observations social 372 1.00 page) Means 32.3 n.a. n.a. 0.02 0.10 0.49 1.61 1.78 7.42 151.9 161.8 following on b continues n.a. 2SLSFE 0.203***- 0.006*** 0.083*** 0.135 0.027*** 0.128 (0.022) (0.001) 0.00005***- (0.007) (0.174) (0.010) (0.107) 0.001- (0.070) 0.122- (0.071) 0.016- 0.203 (0.105) (0.335) (0.00001) (table a 11 15) n.s. 0.202***- 0.006*** 0.041*** 0.104 0.029*** 0.1 0.042 (0.023) (0.001) 2SLSFE 0.00005***- (0.008) (0.165) (0.009) (0.106) 0.014- (0.071) 0.063- (0.093) (0.1 0.220- (0.539) (0.00001) 1998 effects n.a. 0.202***- 0.006*** 0.039*** 0.130 0.027*** 0.126 0.134 (0.022) (0.001) Areas, Community fixed 0.00005***- (0.007) (0.173) (0.009) (0.103) 0.003- (0.067) 0.104- (0.073) 0.006- (0.082) (0.095) (0.00001) Rural in es 16 squar Ordinary 12.715***- (1.862) 0.205***- 0.006*** 0.037*** 0.242 0.035*** 0.123 0.018 (0.022) (0.001) (0.008) (0.189) (0.009) (0.103) (0.068) 0.1- 0.035 0.298*** (0.073) (0.082) (0.068) Malnutrition least 0.00005***- (0.00001) Child of se schooling schooling (centimeters) missing (centimeters) Determinants missing expenditur (months) ed (months) yearss' 10.6. heights' heights' yearss' heights' heights' child capita (months) squar cubed mother father per bleaT ariableV Constant Age Age Age Mother Mother Father Father Female Log Log Log 373 0.15 0.12 0.03 0.03 0.002 0.22 n.a. n.a. n.a. funds, Means social omfr b countries. come 0.173 (0.108) 2SLSFE 0.228*- (0.124) 0.210- 0.043 (0.279) (0.234) 0.126- (0.595) 0.064- 0.326 0.506 14.54 1,672 in (0.179) fects. ef fixed surface; on-Asiann in water and a )11 and community 0.186* (0.1 0.196- (0.131) 0.184- 0.033 0.055 (0.280) (0.242) (0.663) 0.149- 0.320 0.561 13.54 1,672 countries (0.193) with 2SLSFE opland, es cr Asian squar ennial other least per in a. note effects opland, lativeser in 0.175 (0.108) 0.223*- (0.129) 0.206- 0.041 (0.277) (0.235) 0.096- (0.598) 0.078- 0.326 n.a. n.a. 1,672 two-stage cr of (0.158) = Community fixed annual variables 2SLSFE existence the umental and es entheses. unirrigated instr par the squar 0.080 (0.093) 0.170*- (0.098) 0.269- (0.257) 0.148- (0.242) 0.170- (0.339) 0.086- 0.215 n.a. n.a. 1,672 in to (0.127) data. opland, Ordinary winnings; cr least VLSS design) income lottery annual level. level. 1993 and sample the cent cent for household per per irrigated ) 1 using e value) 10 p( uments at ar at capita inheritances, (adjusted per test instr ors add continued( significant err variables calculations' significant d dowries, excluded applicable. umental eligionr estimates Authors 10.6. minority on Standar Not subsidies, ce: Instr The bleaT ariableV otestant n.a. *Statistically ***Statistically Note: a. b. Sour 2 tests Buddhist Catholic Pr Other Chinese Ethnic R Overidentification F Observations social 374 Child Nutrition, Economic Growth, and the Provision of Health Care Services 375 similar. The one change is that the Chinese variable now has a positive effect that is statistically significant at the 5 percent level, but this is of little interest for the purposes of this chapter. The OLS estimate of the impact of household expenditures in urban areas is lower in 1998 than in 1993 (0.341 and 0.493, respectively), and the same holds when fixed effects are introduced (0.146 in 1998 compared with 0.388 in 1993). The first set of 2SLSFE estimates shows an effect very similar to that of 1993 (0.401 compared with 0.441), but neither is statistically signif- icant from zero. If the assumption is made that household expenditures are exogenous, although measured with error, then household income is a valid instrumental variable, and a much more precise and statistically significant estimate of 0.674 (standard error of 0.226) is obtained. Both 2SLSFE esti- mates for 1998 easily pass the overidentification test. In rural areas in 1998 (table 10.6), again the results for most variables are similar to those for rural areas in 1993. Turning to the variable of primary in- terest, household expenditures, the OLS and fixed effects estimates are quite similar across the two years. However, the first set of 2SLSFE estimates is very different: In 1993 the estimated impact was 0.724, and in 1998 it was -0.220. Yet both of these estimates have very large standard errors (0.473 in 1993 and 0.593 in 1998), and it is clear that the difference between them is not statistically significant (t statistic of 1.36). The second set of 2SLSFE esti- mates, which adds household income as an instrumental variable, is closer to the estimates for 1993, with point estimates of 0.500 in 1993 and 0.203 in 1998. Again, neither of these is very precisely estimated (with standard errors of 0.305 and 0.335, respectively), so they are not significantly different from each other. Panel Data Estimates In principle, there are two ways to use panel data to estimate the impact of household expenditures on children's nutritional status in Vietnam. First, one could examine data on the same children over time and estimate the impact of changes in income on changes in their height-for-age z scores. However, this is rather problematic because, as seen in figures 10.1 and 10.2, stunting develops in the first two years of life, after which there is little change. Thus, any child who was covered in the 1993 VLSS was already at least five years old in the 1998 VLSS, and the impact of the household's expenditure levels in the latter survey should have almost no effect on the stunting of those children because their stunting developed three or more years before that survey. The other possibility is pursued in this chapter, which is to compare chil- dren five years or younger in the first survey to children from the same house- hold who were five years or younger in the second survey. This can be done using data for those households in the panel that had children of that age in both 1993 and 1998, a situation that occurs for 1,663 of the 4,300 households in the panel dataset. For households that had two or more children in this 376 Economic Growth, Poverty, and Household Welfare in Vietnam age range in either year (or both years), all variables used are averages over those children. Before examining the estimates, a discussion of their usefulness is in order. Recall that parental preferences for child health () and the child's genetic health endowment () are unobserved variables that could be corre- lated with household expenditure (which represents household income). One way to try to get around this problem is to use instrumental variables for expenditure that are not correlated with these variables. The approach with panel data is somewhat different. Instead, one assumes that the impact of these variables on child health takes an additive form and that these additive components do not change over time. If so, changes in household expenditure will be uncorrelated with these household fixed effects, so re- gressing changes in height-for-age z scores on changes in household expen- ditures (and other variables that may change over time) should eliminate bias due to these two types of unobserved household characteristics. Although this seems to be a promising approach, there are at least two problems with it. First, children's health endowments vary at the child level, not at the household level, so that although a household's average child health endowment differences out, the variation across different children within the household does not, and this could (at least in principle) lead to biased estimates for the same reason that it would do so in OLS estimation of cross-sectional data. Second, regressing differences in variables on each other greatly exacerbates bias due to measurement error, as stressed in Deaton (1997). Thus, it is preferable to find instrumental variables that can predict changes in household expenditure over time. This excludes many of the instrumental variables used above. Here, one instrumental variable is used: changes in household income over time. Table 10.7 presents panel data estimates for urban and rural areas. The only variables that change over time are the (average) age and gender of the children and (log) per capita expenditures. The female child (gender) dummy variable has no significant impact in any of the regressions. The age variable is again specified in a flexible way, with linear, quadratic, and cubic terms. The coefficients on these age terms are quite similar to those seen in tables 10.3­10.6. The three urban regressions (OLS, fixed effects, and 2SLSFE) yield a rather odd finding: negative point estimates for the impact of household ex- penditures on child health. However, note that the standard errors on these coefficients are very large (0.168, 0.185, and 0.920, respectively), which re- flects the small sample size. Thus, the positive estimates in tables 10.3 and 10.5 are not necessarily inconsistent with these results. Unfortunately, the standard errors of the panel data estimates are so large that little can be inferred from them. In rural areas, the sample sizes are much larger. In the OLS and fixed effects estimates, the estimated impacts of household expenditures are very close to zero, and the standard errors are small enough (0.077 and 0.084) to exclude the point estimates in tables 10.4 and 10.6 from the associated 1*** that 0.004 (0.147) 0.087***- (0.009) 0.001 0.376 0.226 1,426 2SLSFE (0.0002) (0.512) 0.000005***- (0.000001) households panel 1*** all effects Rural 0.003- 0.240 1,426 (0.147) 0.086***- (0.009) 0.001 (0.0002) 0.046- (0.084) Community fixed 0.000005***- (0.000001) includes Sample es fects. 1*** ef squar 0.017 (0.157) 0.083***- (0.009) 0.001 0.004 0.138 1,426 fixed Ordinary (0.0002) (0.077) least 0.000005***- estimation. 0.000001)( for enced community fer 237 with 0.037- (0.395) 0.054*- (0.029) 0.0008 (0.0005) 0.939- 0.178 dif (0.920) e es 2SLSFE Estimates 0.000004- (0.000003) wer squar least Data variables effects 237 All Panel 0.032 (0.345) 0.067***- (0.022) 0.0010** (0.0004) 0.106- 0.239 (0.185) two-stage Urban Community fixed 0.000004*- = (0.000002) entheses. par VLSSs. in 2SLSFE es 1998 Malnutrition: 237 and squar 0.033 (0.329) 0.059**- design) (0.022) 0.0009** (0.0004) 0.016- 0.068 surveys. (0.168) Ordinary Child least 0.000004*- 1993 (0.000002) level. level. level. both sample in the of cent cent cent for per se per per using 10 5 1 months at at at 60 (adjusted to 0 ors Determinants expenditur significant significant err calculations' (months) significant d child ed (months) one 10.7. child capita size Authors Standar least ce: (months) squar cubed per bleaT at ariableV *Statistically **Statistically ***Statistically Note: Sour 2 Female Age Age Age Log R Sample had 377 378 Economic Growth, Poverty, and Household Welfare in Vietnam confidence intervals. However, recall that such differenced estimates may suffer from considerable attenuation bias due to increased influence of mea- surement errors. The final column of estimates in table 10.7 uses household income to correct for measurement error in the household expenditures variable (but recall that this assumes that expenditures can be considered exogenous). The point estimate of 0.376 is much larger and comparable with estimates in tables 10.4 and 10.6. Unfortunately, this point estimate also has a very large standard error (0.512), so even in rural areas the panel data esti- mates are probably too imprecise to add anything to what has been learned from the cross-sectional estimates. Impact of Income Growth on Child Nutrition Given the estimates in tables 10.3­10.7, what can be said about the impact of Vietnam's economic growth on child nutrition? More precisely, is the rapid increase in household incomes and expenditures the main cause of Vietnam's substantial decrease in child stunting? This question is examined in table 10.8, which shows changes in mean height-for-age z scores and in the percentage of children who are stunted from 1993 to 1998. The first three lines of the table show the actual changes for rural and urban areas separately, and the rest of the table uses the esti- mated impacts of household expenditures from tables 10.3­10.7 to examine how much of the change was brought about by directly raising households' expenditure levels. Table 10.8 shows that the mean height-for-age z score in urban areas of Vietnam increased by 0.56 standard deviations, while the mean in rural areas increased by 0.49 standard deviations. These increases are quite dra- matic over a period of only five years; they correspond to a drop of about 15 percentage points in the incidence of stunting in both urban and rural areas. Given the high rate of income growth over this time, it is tempting to conclude that this large improvement in children's nutritional status is due to higher household income. The remaining lines of table 10.8 assess whether this conclusion is valid. For each estimator, the predicted change in the mean height-for-age z score is given, which is simply the estimated coefficient of the impact of household expenditures multiplied by the change in (the log of) average household expenditures. For estimates based on cross-sectional data, the estimated im- pact used is a simple average of the 1993 and 1998 estimates. In addition, these estimated impacts are added to each child's z score in 1993 to see how they change the incidence of stunting. Those calculations are reported in the third and fourth columns of table 10.8. The obvious conclusion to draw from the results in table 10.8 is that growth in household expenditures accounts for only a small proportion of the improvement of children's nutritional status in Vietnam from 1993 to 1998. In urban areas, the mean height-for-age z score increased by 0.56, but Child Nutrition, Economic Growth, and the Provision of Health Care Services 379 Table 10.8. Role of Economic Growth in Reducing Child Malnutrition Mean height for age z score Stunted (percent) Indicator Urban Rural Urban Rural Actual figures 1993 -1.455 -2.009 33.2 53.2 1998 -0.895 -1.524 18.4 38.2 Change (over 5 years) 0.560 0.485 -14.8 -15.0 Estimates of change due to economic growth, by different estimation methods Ordinary least squares 0.199 0.083 -5.7 -3.0 [0.344] [0.133] Fixed effects 0.128 0.042 -3.3 -1.5 [0.300] [0.109] 2SLSFEa 0.201 0.066 -6.6 -2.4 [1.112] [0.421] 2SLSFEb 0.281 0.092 -9.0 -3.2 [0.639] [0.324] Ordinary least squares (panel) -0.008 0.001 0.2 -0.0 [0.156] [0.041] Fixed effects (panel) -0.051 -0.013 2.3 0.6 [0.130] [0.031] 2SLSFE (panel) -0.449 0.098 15.0 -3.2 [0.451] [0.363] Note: 2SLSFE = two-stage least squares with community fixed effects. Cross-sectional esti- mates are based on the mean of the 1993 and 1998 estimates. Increase in real expenditures per capita was 29.8 percent in rural areas and 61.3 percent in urban areas (GSO 1999). This implies that the changes in log per capita expenditures were +0.261 in rural areas and +0.478 in urban areas. Numbers in brackets are upper bounds of 95 percent confidence intervals. a. Instrumental variables are irrigated annual cropland, unirrigated annual cropland, peren- nial cropland, and water surface; income from social funds, social subsidies, dowries, inheri- tances, and lottery winnings; and the existence of relatives in other Asian countries and in non-Asian countries. b. The estimates add per capita household income to the instrumental variables in note a. Source: Authors' calculations using the 1993 and 1998 VLSSs. the highest predicted change among seven different specifications is only 0.28 (from the 2SLS specification with income as an instrumental variable), one-half of the total amount. Similarly, the incidence of stunting dropped by 14.8 percentage points, but the predictions from the econometric estimates are much smaller, the highest one showing a drop of only 9 percentage points. The same conclusion holds even more forcefully for rural areas: The mean height-for-age z score dropped by 0.49 standard deviations, but the largest predicted drop is only 0.09 standard deviations; the incidence of stunting dropped by 15 percentage points, but the largest predicted drop is only 3.2 percentage points. 380 Economic Growth, Poverty, and Household Welfare in Vietnam Given the imprecision of the estimated impacts, it is useful to check the upper bound of the 95 percent confidence interval of the estimated impacts, because it is possible that even though the point estimates are low, the actual change may still lie within that confidence interval. The mean changes in height-for-age z scores using the upper bounds of the 95 percent confidence intervals are shown in brackets in the first two columns of table 10.8. In only 2 of 14 cases does the actual change lie within that confidence interval. Thus, one must conclude that growth in household incomes accounts for only a proportion, and probably a rather small proportion, of the improvement in children's nutritional status in Vietnam during the 1990s. Health Programs and Child Nutrition The results of the preceding discussion strongly suggest that something else happened in Vietnam in the 1990s that improved children's nutritional sta- tus. One possibility is that health services in Vietnam dramatically increased in their quantity, quality, or both. This section reviews changes in the quan- tity and quality of health services in Vietnam and uses the 1998 VLSS data to examine the impact of health services on child nutrition in rural areas. Growth in Health Programs The community and price questionnaires in the 1993 and 1998 VLSSs have some information that can be used to examine expansion in health services in Vietnam. In general, publicly provided services did not expand their cov- erage in the 1990s (World Bank 2001a). However, the economic reforms al- lowed private individuals to sell medicines and provide health services. Both surveys collected price data on a variety of commonly used medicines in Vietnam. The medicines that were covered differed in the two surveys: Only ampicillin and penicillin prices were collected in both surveys. No price data were recorded if communities did not have a given medicine available, which in principle allows use of the price data to see whether allowing private individuals to sell medicines increased their availability. In urban areas in 1993, all 30 communes in the sample had price data for all medicines, but in rural areas, only 3 of 120 communes reported no data for ampicillin prices and only 10 of 120 reported no data for penicillin prices. This completeness of the 1993 price data implies that it is not possible to use the price data to check for increased availability of these medicines from pri- vate providers. Although one may think that this demonstrates that medi- cine availability did not improve, because these medicines were already available almost everywhere, this is not necessarily the case because the dis- tance traveled to obtain the medicine may have decreased. Thus, the only conclusion is that the price data from the two surveys are not very informa- tive about changes in the availability of medicines. The other way to check the survey data to see if the availability of medi- cines and medical services from private providers increased in the 1990s is to examine the community questionnaires in both surveys, which asked Child Nutrition, Economic Growth, and the Provision of Health Care Services 381 about distances from the communities to various medical facilities, includ- ing the distance to the nearest private pharmacy. Distance information is available for both years from 111 of the 120 rural communes covered in the 1993 VLSS. In 67 communes (60 percent), the distance to the nearest private pharmacy did not change during the two surveys. For 18 communes (16 per- cent), the distance increased; for 27 communes (24 percent), the distance de- creased. There may be some noise in these data, but overall they suggest that the distance to the nearest private pharmacy was more likely to decrease than to increase. In fact, among the 18 communes where the distance in- creased, the median distance of the increase was 3 kilometers; in the 27 that experienced a decrease, the median distance dropped by 7 kilometers. Econometric Estimates The 1998 VLSS price questionnaire collected detailed data on the prices of nine medicines in both urban and rural areas. In rural areas, the community questionnaire also collected data on the distance from the sampled com- munes to 14 different kinds of health facilities or health service providers. That questionnaire also collected information from community respondents on specific illnesses that were common to the community and on reported problems with the commune health center. Commune health centers are the first line of defense in the Vietnamese health care system. Almost every rural commune has a commune health cen- ter; of the 156 communes in the 1998 VLSS for which community data were collected, only 2 did not have their own commune health center (and in both cases, there was a commune health center within 5 kilometers). The 1998 VLSS administered a commune health center questionnaire in 155 of the 156 communes in the rural areas covered by that survey. That questionnaire col- lected information on (a) the number of medical staff (doctors, physician assistants, nurses, and nurse's aides); (b) hours of operation; (c) the number of beds; (d) the availability of 11 kinds of medical services; (e) the availabil- ity of electricity, clean water, and a sanitary toilet; (f) 13 different types of medical equipment, ranging from thermometers to laboratories; (g) the availability and prices of 9 kinds of medicines (the same ones surveyed in the price questionnaire); and (h) fees for 5 kinds of services, and some informa- tion on whether those fees are waived for certain types of clients. Here, community-level variables that are relevant for child health have been added to the regression analysis to see whether they have any explana- tory power as determinants of child health. Because there are so many vari- ables, and they vary only at the level of the community (and there are only 156 communities in the rural sample), they are not added all at once; rather, the most basic are added first, and then other sets of variables (sometimes in the form of an index) are added to see what explanatory power they have. The first data to be used are medicine prices. Of the nine types of medi- cines available, the one most relevant for child nutrition is oral rehydration salts. Other potentially relevant medicines are the antibiotics ampicillin and 382 Economic Growth, Poverty, and Household Welfare in Vietnam Table 10.9. Descriptive Statistics of Selected Community Variables in Rural Areas, 1998 Communities with Standard Variable observations Mean deviation Minimum Maximum Price of oral rehydration salts (thousand dong) 146 1.3 0.4 0.4 2.5 Price of ampicillin (thousand dong) 152 4.1 1.1 0.7 8.0 Price of penicillin (thousand dong) 151 2.6 0.9 1.0 8.0 Price of iron tablets (thousand dong) 119 3.4 5.5 0 37.3 Price of vitamin A (thousand dong) 126 0.7 0.9 0 4.8 Price of acetaminophen (thousand dong) 151 1.1 1.1 0 9.0 Distance to provincial hospital (kilometers) 154 38.6 33.4 0 180 Distance to state pharmacy (kilometers) 144 6.1 12.1 0 100 Distance to private pharmacy (kilometers) 146 3.1 7.8 0 50 Diarrhea is local health problem? 156 0.46 n.a. 0 1 Characteristics of community health center Lacks electricity 155 0.13 n.a. 0 1 Lacks clean water 155 0.26 n.a. 0 1 Lacks sanitary toilet 155 0.34 n.a. 0 1 Total equipment index 152 9.29 1.11 7 14 Price of ampicillin (thousand dong) 135 4.07 5.40 0 50 Injection price (thousand dong) 153 0.28 0.56 0 5 Price of oral rehydration salts (thousand dong) 150 1.41 1.12 1 5 n.a. Not applicable. Source: Authors' calculations using the 1998 VLSS. penicillin, acetaminophen (to reduce fevers--also known as paracetemol), iron tablets, and vitamin A tablets. Unfortunately, these price data are very noisy, as seen in the first six rows of table 10.9. Although prices were sup- posed to be collected for a given number of doses for a particular brand, it is likely that some observations are for a different number of doses or perhaps Child Nutrition, Economic Growth, and the Provision of Health Care Services 383 for a different brand. The variation in the price of iron tablets is particularly egregious: The standard deviation is nearly twice as large as the mean, and the maximum value is 20 times larger than the median value. To see whether these data had any explanatory power, the OLS regression in table 10.6 (for rural areas in 1998) was reestimated five times, each time adding one of the price variables for each of these five types of medicines (iron tablet prices were deemed too noisy to be used). In all cases, the point estimates were very close to zero and far from any statistical significance. As an example, consider the medicine most likely to have an effect, namely, oral rehydration salts (which also displayed the least amount of noise in the data). The price of oral rehydration salts had the expected negative sign, but it had a t statistic of only 0.83 and thus was not statistically different from zero. Another "price" of medical care is the distance to nearby health facilities. Although the distance to commune health centers is trivial, those centers are not equipped to handle the most serious medical problems; seriously ill in- dividuals must go to a hospital in a district or provincial capital or perhaps to some other kind of health facility (including private health facilities). Of the 13 other types of health facilities or health service providers considered in the commune questionnaire, 4 had missing data for nearly one-third or one-fourth of the observations (private nurse, medicine peddler, midwife, and practitioner of Eastern medicine) and thus were not considered. For the remaining 9, the same procedure used with the medicine price data was used for the distance data. For 6 types of facilities (family planning center, polyclinic, district hospital, other hospital, private doctor, and private physi- cian assistant), no significant relationship was found. However, for 3 types of health facilities (provincial hospital, state pharmacy, and private phar- macy), a negative effect--significant at the 5 percent level--was found. The first column of table 10.10 presents the results when all three dis- tance variables are added simultaneously. The distance to the nearest provincial hospital is statistically significant at the 10 percent level, with a coefficient of -0.0023. The distance to the nearest state pharmacy is not sta- tistically significant; it has a coefficient of -0.0022 and a standard error of 0.0019. Finally, the distance to the nearest private pharmacy is statistically significant at the 1 percent level, with a coefficient of -0.0158. Despite the statistical significance of two of the distance variables, the policy signifi- cance of the estimated impacts is not particularly large. Reducing by one- half the current mean value of the distance to the nearest provincial hospital implies an improvement in children's height-for-age z scores of about 0.044, and halving the distance to the nearest private pharmacy implies an im- provement of only 0.025. The next variables to consider are those from the community questionnaire concerning local health problems and problems with the commune health center. Five illnesses cited seem relevant for small children: malaria, respi- ratory illnesses (other than tuberculosis), childhood illnesses (diphtheria, whooping cough, measles, polio, tetanus, and encephalitis), diarrhea, and 384 Economic Growth, Poverty, and Household Welfare in Vietnam Table 10.10. Impact of Community Health Services on Child Malnutrition in Rural Areas, 1998 Adding Adding Adding oral electricity equipment rehydration Variable Base model and toilet index salts Age (months) -0.211*** -0.209*** -0.208*** -0.207*** (0.024) (0.024) (0.025) (0.025) Age squared (months) 0.006*** 0.006*** 0.006*** 0.006*** (0.001) (0.001) (0.001) (0.001) Age cubed (months) -0.00005*** -0.00005*** -0.00005*** -0.00005 (0.00001) (0.00001) (0.00001) (0.00001) Mother's height (cm) 0.035*** 0.035*** 0.035*** 0.0036*** (0.008) (0.008) (0.008) (0.008) Mother's height missing 0.150 0.127 0.110 0.160 (0.219) (0.219) (0.108) (0.208) Father's height (cm) 0.032*** 0.032*** 0.031*** 0.031*** (0.010) (0.010) (0.010) (0.010) Father's height missing 0.094 0.099 0.110 0.092 (0.107) (0.107) (0.108) (0.109) Female child -0.030 -0.031 -0.023 -0.033 (0.068) (0.068) (0.069) (0.071) Log mother's years schooling -0.202*** -0.185*** -0.178** -0.181** (0.069) (0.070) (0.0070) (0.075) Log father's years schooling -0.002 0.000 -0.001 -0.002 (0.085) (0.084) (0.083) (0.088) Log per capita expenditures 0.0259*** 0.240*** 0.232*** 0.219*** (0.074) (0.078) (0.080) (0.083) Buddhist 0.047 0.029 0.028 0.053 (0.098) (0.097) (0.097) (0.103) Catholic -0.233* -0.213* -0.269** -0.288*** (0.124) (0.116) (0.108) (0.109) Protestant - 0.357* -0.278 -0.343 -0.346 (0.195) (0.204) (0.221) (0.217) Other religion -0.330 -0.364 -0.387 -0.178 (0.227) (0.231) (0.247) (0.184) Chinese -0.946*** -0.971*** -0.894*** -0.959*** (0.112) (0.110) (0.106) (0.113) Ethnic minority -0.066 -0.036 -0.007 -0.016 (0.134) (0.135) (0.132) (0.136) Distance to provincial hospital (in kilometers) -0.0023* -0.0016 -0.0022 -0.0018 (0.0012) (0.0015) (0.0016) (0.0017) Distance to state pharmacy (in kilometers) -0.0022 -0.0020 -0.0018 -0.0024 (0.0019) (0.0020) (0.0021) (0.0021) Child Nutrition, Economic Growth, and the Provision of Health Care Services 385 Table 10.10. (continued) Adding Adding Adding oral electricity equipment rehydration Variable Base model and toilet index salts Distance to private pharmacy (in kilometers) - 0.0158*** - 0.0136** - 0.0138** - 0.0133** (0.0053) (0.0062) (0.0062) (0.0063) Commune health center variables Lacks electricity n.a. - 0.104 - 0.028 - 0.052 (0.165) (0.173) (0.166) Unsanitary toilets n.a. - 0.151* - 0.142 - 0.192** (0.087) (0.089) (0.091) Equipment index n.a. n.a. 0.051 0.052 (0.041) (0.038) Availability of oral rehydration salts n.a. n.a. n.a. 0.110** (0.048) Sample size 1,446 1,443 1,411 1,342 R2 0.247 0.250 0.252 0.261 n.a. Not applicable. *Statistically significant at 10 percent level. **Statistically significant at 5 percent level. ***Statistically significant at 1 percent level. Note: Standard errors (adjusted for sample design) in parentheses. Source: Authors' calculations using the 1998 VLSS. child malnutrition. Using the same procedure described above, only respira- tory illnesses approached statistical significance, with a point estimate of -0.130 and a t statistic of -1.34. Even less statistical significance was seen in the variables citing problems with local health facilities (lack of equipment and supplies, lack of medicines, inadequate staff, inability of staff to provide services, inadequate training opportunities, and lack of sanitation). None of these variables had a t statistic greater than 1.3 when added to the regression. Finally, turn to the data from the commune health center questionnaire. The numbers of different kinds of staff in the center, divided by the popu- lation of the commune, never had any explanatory power, either separately or as a group. The same is true of weekly hours of operation and number of beds (divided by the commune population). Of the 11 kinds of services offered, 1 was offered by all centers in the sample (immunizations), 1 was offered by all but two (prenatal care), and 2 appear to be irrelevant for child nutrition (eye exams and dental exams). Of the remaining 7, 3 are closely tied to child health (obstetrics, child health exams, and education on nutrition), 2 concern birth control (intrauterine device [IUD] insertion and abortion), and 2 are very general (Eastern medicine and simple opera- tions). The last 4 had no explanatory power when entered individually. Neither did the 3 that are most closely tied to child health. In addition, 386 Economic Growth, Poverty, and Household Welfare in Vietnam combining all these variables into a general health services index also had no explanatory power. Next consider the three general variables concerning amenities at the fa- cility: electricity, clean water, and sanitary toilet. Of these, lack of a sanitary toilet and lack of electricity both had significantly negative effects on child health when added separately, but the clean water variable had no effect. A regression adding the two significant variables is shown in the second col- umn of table 10.10. The electricity variable loses statistical significance, and the toilet variable is still significantly negative. The policy significance of lacking a sanitary toilet is much larger than that of the distance variables; taking the coefficient at face value implies that remedying this deficiency will increase the typical child's height-for-age z score by 0.15 point. The next variables examined are the 13 equipment variables. There are so many types of equipment that the best thing to do is to develop an index. First, 4 variables with almost no variation were dropped: blood pressure monitor and stethoscope (only one commune did not have these) and ther- mometer and laboratory (only three communes did not have thermometers and only three had laboratories). Two variables were also dropped that have no clear relevance for child malnutrition: eye charts and family planning equipment (for abortions). This leaves 7 variables for the general index: re- frigerator, sterilizing equipment, delivery bed, microscope, examining bed, child growth chart, and child scale.2 The results, after adding the equipment index, are shown in the third column of table 10.10. The index has the ex- pected positive sign but is not statistically significant (t statistic of 1.24). The last type of information in the commune health center questionnaire is data on the availability of drugs, the prices of those drugs (if available), and the prices of common health services. The drug availability index ranges from 1 (not available) to 5 (always available). This variable was statistically significant for only one drug, oral rehydration salts, which had the expected positive impact with a t statistic of 1.74. Turning to drug prices, missing data were a serious problem because prices were not recorded for commune health centers that rarely or never had the medicine. Of the four for which price data were available at almost all commune health centers, three had statistically significant effects: ampicillin, penicillin, and oral rehydration salts (the fourth was acetaminophen, which is used to treat fevers). The prices of ampicillin and penicillin were highly correlated, and only ampi- cillin was retained because it had fewer missing values. An odd finding for both ampicillin and penicillin is that the sign was positive: Higher prices im- proved child health. Finally, prices of three kinds of medical services were examined: general examination, birth, and injection. Only the injection price had a statistically significant impact, which was unexpectedly positive. When all four of the statistically significant variables (price of ampicillin, price of injection, and availability and price of oral rehydration salts) are added to the regression (not shown in table 10.10), only one retains statisti- cal significance: The availability of oral rehydration salts has the expected negative sign. Because the sample size drops considerably as a result of Child Nutrition, Economic Growth, and the Provision of Health Care Services 387 missing data in all of these variables, only the oral rehydration salts variable is added in the regression shown in the last column of table 10.10 (even then, the sample size drops from 1,411 to 1,342). The impact of the index is statis- tically significant at the 5 percent level, with a parameter estimate of 0.110. Because the index ranged from 1 to 5, this estimate implies that a change from never being available to always being available raises child height-for- age z scores by 0.44 point, which is almost as large as the increase in rural areas from 1993 to 1998 reported in table 10.8. Note, however, that 84 percent of the commune health centers report that they already have this medicine available all the time, so the benefits of oral rehydration salts are already reaching most Vietnamese children. Summary and Conclusion This chapter has investigated the impact of household income growth, as measured by household expenditures, on child nutritional status in Vietnam. Vietnam was doubly fortunate in the 1990s: Household incomes rose dramat- ically and children's nutritional status improved rapidly. Although one might conclude that the former caused the latter, the estimates presented here do not support such a conclusion. Using many different estimation methods, this chapter has shown that the impact of household expenditures on children's nutritional status (as measured by height-for-age z scores) is not necessarily significantly different from zero. More specifically, the impact may well be positive, but it is not very large. In particular, none of the estimates is large enough to account for even one-half of the measured improvement in chil- dren's nutritional status from 1993 to 1998. Some observers may argue that this finding casts doubt on the benefits of economic growth for children's health status, but such a conclusion would be misleading. Economic growth may lead to other changes in society, such as improved health care services. That is, economic growth typically in- creases government budgets through higher tax revenues, some of which can then be used to provide better health care services. The question then be- comes: What kinds of health services are most effective at raising child (and adult) health? A first attempt at answering this question was made in the "Health Programs and Child Nutrition" section. The community-level data on health services suggest that the distance to private pharmacies has a sta- tistically significant, though not very large, negative effect on child nutri- tion. It also suggests that providing commune health centers with sanitary toilets and ample supplies of oral rehydration salts could have substantial positive impacts on child health in Vietnam. These findings regarding community-level health services are tentative; much more research is needed in Vietnam on the impact of different kinds of health care services and programs on children's nutritional status. A better understanding is important for policy, given current trends in the provision and use of government health care services; in particular, government sup- port for commune health centers is stagnant (World Bank 2001a), which may 388 Economic Growth, Poverty, and Household Welfare in Vietnam explain the decline in their use documented in chapter 11 of this volume. A particularly crucial factor may be parents' health knowledge, especially mothers' health knowledge. A recent study of Morocco found that mothers' health knowledge was the main pathway by which their education affects child health (Glewwe 1999). In Vietnam, several new community programs supported by donor agencies focus on raising parents' health knowledge (see World Bank [2001a]). Such programs could lead to substantial improve- ments in children's nutritional status, but rigorous analysis is needed to test this hypothesis. Notes 1. The 1992­93 survey spanned a full year, starting in October 1992; the 1997­98 survey began in December 1997 and also lasted one year. For brevity's sake, refer- ence is made to the surveys as the 1993 VLSS and 1998 VLSS, respectively. 2. As an exploratory exercise, each piece of equipment was added separately. The three that approached statistical significance were delivery beds, examining beds, and child scales, with the expected positive sign and t statistics ranging from 1.9 to 2.6. Because there are many other kinds of equipment that are not observed but may be correlated with specific observed equipment, it seemed best to combine all types of equipment into a single index. Bibliography Davidson, Russell, and James MacKinnon. 1993. Estimation and Inference in Econometrics. New York: Oxford University Press. Deaton, Angus. 1997. The Analysis of Household Surveys. Baltimore: Johns Hopkins University Press. Gibson, Rosalind. 1990. Principles of Nutritional Assessment. Oxford: Oxford University Press. Glewwe, Paul. 1999. "Why Does Mother's Schooling Raise Child Health in Developing Countries? Evidence from Morocco." Journal of Human Resources 34(1): 124­59. General Statistical Office. 1999. "Viet Nam Living Standards Survey 1997­98." Hanoi. Ponce, Ninez, Paul Gertler, and Paul Glewwe. 1998. "Will Vietnam Grow Out of Malnutrition?" In D. Dollar, P. Glewwe, and J. Litvack, eds., Household Welfare and Vietnam's Transition. Washington, D.C.: World Bank. Sahn, David, David Stifel, and Stephen Younger. 1999. "Intertemporal Changes in Welfare: Preliminary Results from Nine African Coun- tries." Cornell Food and Nutrition Policy Program Working Paper 94. Cornell University, Ithaca, N.Y. Child Nutrition, Economic Growth, and the Provision of Health Care Services 389 United Nations Development Programme. 1998. Human Development Report. New York: Oxford University Press. Viet Nam Living Standards Survey (VNLSS). 1992­1993. Web site: www. worldbank.org/lsms/country/vn93/vn93bid.pdf. Vietnam Living Standards Survey (VLSS). 1997­98. Web site: www. worldbank.org/lsms/country/vn98/vn98bif.pdf. Wagstaff, Adam, and Nga Nguyet Nguyen. 2004. "Poverty and Survival Prospects of Vietnamese Children under Doi Moi." In P. Glewwe, N. Agrawal, and D. Dollar, eds., Economic Growth, Poverty, and House- hold Welfare in Vietnam. Washington, D.C.: World Bank. WHO (World Health Organization). 1983. Measuring Change in Nutritional Status. Geneva. . 1986. "Use and Interpretation of Anthropometric Indicators of Nutri- tional Status." Bulletin of the World Health Organization 64(6): 929­41. World Bank. 1986. World Development Report. New York: Oxford University Press. . 1999. Vietnam: Attacking Poverty. Joint Report of the Government of Vietnam and the Donor­NGO Poverty Working Group. Hanoi. . 2001a. Vietnam Health Sector Review. Hanoi. . 2001b. World Development Indicators. Washington, D.C. 11 Patterns of Health Care Use in Vietnam: Analysis of 1998 Vietnam Living Standards Survey Data Pravin K. Trivedi The changes to the Vietnamese health sector initiated by Doi Moi ("renova- tion") have been described and analyzed in previous World Bank empirical studies based on the 1992­93 Vietnam Living Standards Survey (VLSS). These studies noted, in particular, the decline of the traditional public sector provider of health care to the poor, the commune health center, and the in- cipient rise in the early 1990s of private sector health providers. The deregu- lation of the pharmaceutical industry was followed by a dramatic growth of private pharmacies as the single most important source of drugs for self- medication. Contacts with pharmacies, both public and private, became the most important type of contact between the provider and patient while the role of the commune health center declined. Deregulation permitted the emergence of private care facilities provided by doctors and nurses, some of whom are simultaneously employed in government hospitals and other public facilities. Another major new feature of the health care sector, largely absent in 1993, is the Vietnam Health Insurance (VHI) program, with mandatory coverage for some sections of the community and voluntary coverage for others. This is an important new development that has signifi- cant implications about the rate at which the relative importance of different providers in the health sector has changed and is likely to change in the future. These emergent trends, also highlighted and explored in the Vietnam Health Sector Review (VHSR; World Bank 1999), have raised important ana- lytical and policy questions about the access to and use of health services by different socioeconomic groups. This chapter provides a further analysis of the direction and scope of changes in the health sector. Here, the emphasis is on econometric modeling of demand for different types of health care in Vietnam rather than supply 391 392 Economic Growth, Poverty, and Household Welfare in Vietnam aspects. In contrast, the VHSR (World Bank 1999) surveys all aspects of the health sector. In modeling the demand for health care, both individual- and household-level data from the 1997­98 VLSS are used.1 The focus is on four major features of health care use: · Determinants of the largely self-prescribed use of pharmaceutical drugs · Use of government hospitals · Low level of use of commune health centers · Increasing use of private health care facilities. In each case, the major interest is the role of price and income variables. Given the high rate of economic growth, accompanied by major structural changes in health care delivery, it is useful to consider how these changes have affected the pattern of health care use and what further changes we can expect if these trends persist. The major analytical tool is multivariate re- gression analysis using models that respect the discrete nature of many of the outcome variables that are reported in the VLSSs. The analysis in this chapter complements that provided in the compre- hensive VHSR (World Bank 1999), which provides much descriptive statisti- cal information on the structure and organization of health care delivery and on the broad pattern of health care use in Vietnam. The VHSR also provides many points of comparisons between the 1993 and 1998 VLSSs. Although the tabular summary information provided in the VHSR is both rich and in- formative, and also highly suggestive of factors that influence health care use, no formal modeling of the data is undertaken. Relationships between variables are studied or interpreted on a bivariate basis--for example, the relation between health care usage and household income. Because this method cannot control for the presence of other relevant factors, and can at most only establish informal associations, there is a danger of misleading in- terpretations as a result of the neglected confounding variables. This chap- ter addresses the task of measuring and testing various hypotheses about health care use within a multivariate econometric framework. In the second section of the chapter, there is a preliminary look at data and an outline of the substantive issues, as well as a summary of the changes and trends affecting the health sector since the 1993 VLSS. The next section surveys the main empirical issues and is followed by a section that consid- ers the modeling framework and related statistical issues. This is followed by a section that analyzes the determinants of enrollment in the VHI pro- gram. Next, the use of the major components of health care is modeled, ig- noring the remaining providers who collectively account for only a small share of the health care budget. Econometric methods used in this section are those appropriate for binary-valued and count variables. These are used to model the probability of or number of visits, or both, to commune health centers, pharmacies, government hospitals, private providers, and public hospitals. The next section discusses results for individual and household health care expenditures, which is followed by a section that discusses Patterns of Health Care Use in Vietnam 393 the policy implications of the results. The last section summarizes and concludes. The 1998 VLSS The health component of the 1998 VLSS is the main source of the data used here. Usage data are collected for seven types of providers: government hos- pital (GOVHOSP), commune health center (CHC), regional polyclinic, other government health facility, private health facility (PHF), traditional Eastern medical practitioner, and pharmacy visits or self-medication (PHARVIS). For each type of provider, the questionnaire seeks information on the num- ber of contacts, total expenditure, the amount spent on medicines, and trans- portation and other costs associated with the visits. For the self-medication part, the questionnaire also sought information on whether the visit to the drug vendor was self-initiated or requested by another provider. The survey also provides information on government hospital admis- sions (HOSPADM) in the 12-month period preceding the survey and the number of nights spent in hospitals (HOSPNITE). The responses to questions about expenditure on health care refer to a period of 12 months. The data make it possible to calculate this figure for the household both inclusively and exclusively of health insurance expenditure. Finally, the questionnaire collected information on whether the respon- dent had health insurance (HLTHINS) and the amount spent on it in the pre- vious 12 months. The survey includes information on the current health status, such as occurrence of illness (ILL) or injury (INJ) in the preceding four weeks, the number of days of illness (ILLDAYS), and the number of days of restricted activity (ACTDAYS).2 However, there are some gaps in the information available that affect econometric modeling of health care use. For example, there is still no infor- mation on long-term health status, such as the presence of limiting and non- limiting chronic conditions. The general health status of an individual is an important conditioning variable in most analyses of demand for health care. Further, the direct information on HLTHINS lacks some necessary detail as to which of the several levels of coverage of insurance applied to a sur- vey respondent (this issue is elaborated further in the section on health insurance).3 The survey also provides information on various sociodemographic variables such as gender (SEX), years of schooling (EDUC), age (AGE), and marital status (MARRIED), as well as total household expenditure (INC). Table 11.1 provides the data definitions and descriptive summary statistics. These data can support empirical investigations at both the individual and household levels. The available frequency-of-use data can be used in re- gression modeling of the probability of contact between provider and pa- tient and also in modeling the frequency of such contact. Models of proba- bility of contact attempt to explain the factors that distinguish those who 394 Economic Growth, Poverty, and Household Welfare in Vietnam Table 11.1. Definitions and Descriptive Statistics Standard Variable Definition Mean deviation PHARVIS Number of pharmacy visits 0.51 1.31 PHARDUM 0/1 dummy for pharmacy visit 0.26 0.44 GOVHOSP Number of hospital outpatient visits .049 0.41 PHF Number of visits to private health facility 0.11 0.80 CHC Number of visits to commune health centers 0.04 0.34 HOSPADM 0/1 dummy for hospital admissions .051 0.22 HOSPNITE Number of hospital nights 13.53 21.82 MEDEXP Total medical expenditure, 1998 Dong 1,520 5,139 (4W+12m) (4 weeks plus 12 months) HHEXP Total nominal household expenditure, 1998 Dong 15,273 13,020 log(INC) log(HHEXP) 2.60 0.62 lnmedexp (>0) lnmedexp (>) for positive expenditures 2.14 1.08 lnmedexp (insured) lnmedexp for the insured 2.32 1.11 lnmedexp (uninsured) lnmedexp for the uninsured 2.12 1.08 HLTHINS 0/1 dummy for health insurance status 0.16 0.37 AGE Age in years 29.7 9.67 SEX 0/1 dummy for gender 0.51 0.49 MARRIED 0/1 dummy for marital status 0.40 0.49 EDUC Completed years of education 3.38 1.94 ILL 0/1 dummy for illness in 4 weeks before survey 0.41 0.49 INJ 0/1 dummy for injury in 4 weeks before survey 0.009 0.098 ILLDAYS Number of days of illness/injury in 4 weeks before survey 2.80 5.45 ACTDAYS Number of days of limited activity in 4 weeks before survey 0.06 1.11 URBAN 0/1 dummy for urban household 0.29 0.45 HHSIZE Household size 4.73 1.96 Source: Author's calculations. received some care from those who did not receive any care. Such models distinguish only between zero and positive levels of use. Count data models distinguish between different levels of usage, but do not distinguish be- tween high-quality contacts that may have cost more and those of lower quality. Models of expenditure, however, allow us to take into account expenditure variations that may be due to variation in the quality of care, but they do so indirectly. Count models are analogous to models that Patterns of Health Care Use in Vietnam 395 explain quantities, but expenditure models attempt to explain the product of quantity and price of service. Aggregated health care expenditure data can be used to develop total household expenditure models. The two modeling approaches are largely complementary and mutually reinforcing. Survey of Main Issues The VHSR (World Bank 1999, pp. 49­50) considers five factors in health care use: income, price, quality of care, access (especially by income levels), and the role of education. In this section, the main issues concerning the role of these factors are qualitatively outlined. Regression models are considered in another section. First, the main features of health care use data provided in the 1998 VLSS are summarized. Table 11.2 summarizes the contact rates for the main types of providers in 1993 and 1998. The data show that other government health facilities and traditional providers are a small part of the total number of visits. Pharma- cies and drug vendors, GOVHOSP, PHF, and CHC account for the bulk of the total usage. All types of use have grown since 1993, but the use of drug vendors has shown the fastest rate of growth: It has more than tripled. This attests to the overwhelming importance of self-medication. Pharmacies are both private and public. Private pharmacies are more prevalent in urban areas and public ones are more prevalent in rural areas. The VHSR (World Bank 1999, p. 56) points out that the increase in the average number of pharmacy visits is accompanied by a decline in out-of-pocket expenditure on drugs. Also notable is the growth of use of private health facilities, which has increased more than 2.5 times. Table 11.2. Annualized Health Service Contact Rates, by Provider Public Year GOVHOSP CHC OTHER providers 1993 0.32 0.19 0.03 0.54 1998 0.60 0.57 0.25 1.43 All Year PHARVIS PHF TRAD providers 1993 2.14 0.66 0.03 3.37 1998 6.78 1.76 0.36 10.33 Note: Definition of variables: CHC = commune health center. GOVHOSP = government hospital. OTHER = other government facilities. PHARVIS = visits to pharmacy or drug ven- dor. PHF = private health facility. TRAD = traditional (Eastern) practitioner. Source: World Bank 1999, pp. 37­38. 396 Economic Growth, Poverty, and Household Welfare in Vietnam Previous analyses of the 1993 VLSS data have attributed the impressive rise in pharmacy visits to a combination of factors. The first is the improve- ment in the supply and availability of pharmaceutical drugs between 1993 and 1998 after the deregulation of the retail markets and liberalization of the pharmaceutical industry in 1989. Evidence suggests that the real price of drugs may have declined over the 1993­98 period by as much as 30 percent or more. The second factor is the persistence of self-medication, which is in- duced in part by the ease of access to medicines relative to the alternatives. In rural areas, especially, distance from government health facilities and poor quality of health services at commune health centers have been cited as possible reasons for the continued growth of self-medication. The growth of private health providers is another important facet of health care use. There are two main types of private health providers: full- time providers who own private facilities and part-time providers employed by the public health facilities who also engage in private practice during off- hours (World Bank 1999, p. 101). Both licensed and unlicensed practices are included in this category; hence, the quality of care in this sector may be vari- able. Nearly 70 percent of PHFs are estimated to be in urban areas. Access and Costs The annualized health service contact rate by provider type is a rough mea- sure of access (table 11.3). This measure can be misleading if it is not supple- mented with other information. In any sample survey, one is likely to ob- serve zero contact frequency for some respondents, in part because the respondent was healthy in the survey period and did not need health care. Table 11.4 compares the overall contact rates with the subset consisting of those classified as sick or injured and extends the same comparison across different income levels. The figures show that commune health care is sought at a three to four times higher rate by the sick in the lowest income quartile, compared with the sick in the highest income quartile. The situation is reversed in the case of government hospitals. The contact rate in that case is three to four times higher for the better-off sick than for the sick poor. This differential usage is further in evidence for private health care providers, but the difference mul- tiple is closer to two than three or four. The differential usage is the smallest for pharmaceutical providers. This appears to indicate that access to private drug vendors is roughly equal for the sick, whether they have low or high income. However, this needs a caveat, because government hospitals that are accessible to the VHI enrollees and favored by the high-income groups also dispense pharmaceutical drugs and act as a substitute for pharmacies. Table 11.4 shows that at all income levels, the pharmacy is the most fre- quently contacted care provider for the sick. Private health care providers and government hospitals, respectively, are the next most frequently con- tacted by those in the top income quartile. For the sick in the lowest income quartile, private providers are relatively more frequently contacted than are Sick -- 10%, 0.0586 0.0537 0.0086 0.038 0.091 0.037 0.086 0.41 hospital. pharmacy HOMEVIS to lowest OTHER visits = All 0.0279 0.0256 0.0221 0.0096 sample: government VIS sick = TRAD 0.022 0.054 0.014 0.033 0.33 for PHAR Sick size 1.0688 1.1368 1.2323 1.1407 visits. VIS GOVHOSP. Sample . VIS home center = PHAR 0.512 1.214 0.406 0.971 All 2,775. facility 92.17 0.5660 0.5595 0.4526 0.3985 health PHAR 10%, health HOMEVIS Sick highest commune 13 0.1967 0.2089 0.3878 0.3989 = private = PHF hospital. 0.1 0.274 0.084 0.198 2.61 6,939; . PHF CHC PHF. 25%, Providers Status All 0.1002 0.0995 0.1406 0.1398 vendor government = practitioner variables: highest ug Care of 13 dr 0.049 0.1 0.105 0.248 3.22 Health or 6,922; GOVHOSP (Eastern) by Sick Health GOVHOSP. 0.0532 0.0672 0.2005 0.2920 25%, Definition pharmacy 945. to ferent center traditional lowest = Providers, GOVHOSP All 10%, Dif visits CHC 0.040 0.096 0.041 0.100 0.96 health 0.0299 0.0336 0.0762 0.1081 2,773; = to TRAD. with 10%, highest VIS isits 1 V commune 2,483; PHAR of N = facility Rates Sick 0.1271 0.1219 0.0326 0.021 lowest 27,331 1,3221 4,496 1,818 8,081 25%, CHC health CHC Contact sample: facilities. 19 Number private All full highest es variables: = 0.0659 0.0596 0.01 0.0075 for of calculations. 3,273; calculations. Mean PHF. 's verage sample A size government 's ed 25%, 1.3.1 sample Author other expenditur Definition vendor 1.4.1 class 10% 25% 25% 10% available. Sample = Author cent) sample sample ed ce: ce: ug Not lowest ableT Note: dr Sour (per Category Full Sick Insur Sick/insur Health or ableT -- Note: Sour Income Lowest Lowest Highest Highest 1,408; OTHER 397 398 Economic Growth, Poverty, and Household Welfare in Vietnam commune health centers. The contact rates for the latter are very low for high-income individuals. Thus, commune health centers appear to primarily serve the low-income groups, and government hospitals primarily serve the high-income group. However, those who are covered by the VHI program are served by govern- ment hospitals. Also, insurance coverage under VHI is more extensive for the relatively better-off groups; hence, the observed higher usage for these groups may be due to a combined income and price effect, as will become clear in this chapter's econometric analysis. Finally, as others have also pre- viously noted (Gertler and Litvack 1998), self-medication through drugs purchased at pharmacies appears to be the first line of defense against sick- ness, irrespective of income class. For each type of service contact, the questionnaire collected responses on the total cost of transportation, room and board, and other related costs. Of course, these data were collected only for those who actually had nonzero usage. The data are censored for those who had zero usage. For those who had a positive usage level, the data can be used to estimate average extrane- ous cost of health service. Although this is useful information, it is insuffi- cient for modeling individual choice of the type of service. Standard eco- nomic theory suggests that in choosing between two types of providers (for example, commune health center and government hospital), the relative ex- traneous cost of the services of the two providers is relevant. The survey data pertain only to the average extraneous cost of the service actually cho- sen by the patient. By itself, average extraneous cost cannot be used to con- struct the relative price for each user that is needed for modeling purposes. Even a simple measure, such as distance from different types of providers, may be used to construct a more appropriate measure of the extraneous costs under the assumption that such costs are closely related to the dis- tance. However, the average extraneous cost of access, for those who used a provider, may still be useful as a rough benchmark comparison. Average household medical expenditures in 1998 dong for different types of households are shown in table 11.5. The medical expenditure for the average Table 11.5. Medical Expenditure by Household Type (1998 dong) Household type N Expenditure 2 Expenditure 1 Average 5,999 768 1,520 Average |y > 0 5,006 788 1,822 Urban 1,730 997 2,000 Rural 4,269 674 1,325 Farm 2,561 942 1,772 Nonfarm 3,438 637 1,331 Note: Expenditure 1 = medical expenditure. Expenditure 2 = medical expenditure excluding insurance. Source: Author's calculations. Patterns of Health Care Use in Vietnam 399 urban household is nearly 50 percent higher than the average rural house- hold. Greater insight is provided by budget share of health expenditure at different income levels. The VHSR (World Bank 1999) indicates that, overall, 11 percent of total consumption expenditure and 24 percent of nonfood ex- penditure were devoted to medical care in 1998. Medical care constitutes a much larger share of nonfood expenditure for the poor than the rich; the richest quintile devoted 15 percent of the nonfood expenditure to health care, compared with 30 percent for the poorest quintile. Health Insurance A major new development in the health sector since the 1993 VLSS is the emergence of a national health insurance program, VHI; initiated in late 1992, the program began effective operation in 1993. Three health insurance programs in Vietnam are provided under gov- ernment sponsorship, a compulsory national health insurance program and two voluntary programs. During its first phase, insurance coverage was provided to current and retired civil servants and salaried employees of state-owned enterprises and large, private enterprises. Benefits include the full cost of pharmaceuticals and ambulatory and inpatient care at govern- mental facilities to which enrollees are referred. A district or provincial hos- pital acts as the primary care provider. The mandatory VHI coverage does not extend to dependents of employees. A voluntary VHI plan provides for group coverage to dependents of those enrolled in VHI and some other groups, such as communes. That is, groups rather than individuals must enroll in the program. The benefit level varies. A third tier of national health insurance is the voluntary plan called Comprehensive Student Insurance (CSI). The benefits of the compulsory VHI are less variable than those of the voluntary component. The CSI plans and premiums are locally designed and administered, and they show sub- stantial variation in premiums and benefits among localities. One estimate of the number of total (compulsory and voluntary) en- rollees in the VHI program comes from the VHSR (World Bank 1999), which estimates this at 9.8 million in 1998, including about 38 percent voluntary enrollees. This covers roughly 12 percent of the population. The coverage of the target population for the compulsory component is around 77 percent, but it is much lower for the voluntary component and largely consists of students. Having health insurance is positively related to income class. In the low- est income quartile, the insurance coverage rate is 9.2 percent. A high pro- portion of this group may be those enrolled in the voluntary scheme. In the top income quartile, 24.5 percent have health insurance (see table 11.6). Tabular analysis of the impact of insurance is provided in table 65 of the VHSR (World Bank 1999). Controlling for income (by quintiles), the insured have significantly higher rates of service use of public providers, especially for inpatient services in government hospitals. 400 Economic Growth, Poverty, and Household Welfare in Vietnam Table 11.6. Health Insurance and Income Status Income class Lowest 10% Lowest 25% Highest 25% Highest 10% Percentage insured 8.7 9.2 24.5 27.0 Source: Author's calculations. Because the VHI premium for the compulsorily insured is a fixed per- centage of the employee's base salary, the cost of insurance varies and in- come serves as a partial proxy for the cost of insurance. One of the objectives of the empirical investigations in this chapter is to estimate the impact of health insurance on health care use, an issue that was not relevant in analyzing 1993 VLSS data. The foregoing account raises an important econometric issue concerning the treatment of health insur- ance. For those who are compulsorily enrolled in the program--that is the majority--insurance status can be treated as exogenous, but for those who are voluntarily enrolled, there may be an element of individual choice, which is an argument for treating the variable as endogenous. However, as noted, enrollment is on a group, and not individual, basis. This factor di- minishes the role of individual preferences in the choice of health insurance. There is also a related data problem. The health component of the question- naire asked only two questions about insurance: whether the respondent had health insurance and the cost of health insurance in the previous 12 months. Without additional data, one cannot distinguish between those who were enrolled in the compulsory insurance program and those who were not. Therefore, insurance status is treated as exogenous in the health care use equation--that is, it is postulated that causality runs from health in- surance to health care use. In theoretical models with unrestricted choice of insurance, the choice of insurance and health care use will be interdepen- dent or jointly (rather than recursively) determined. It is of some interest to compare usage patterns among insured and unin- sured individuals, conditional on positive expenditure over the previous 12 months. Total expenditure of the insured sample is about 20 percent higher, and this difference is statistically significant. The average difference in government hospital use between the insured and the uninsured is also statistically significant: that for the insured population is higher by a factor of about 2.5. The average difference in the use of a private health facility and drug vendors is significantly higher for the uninsured sample than for the in- sured. This general pattern is also consistent with the results of regression analysis in which many sociodemographic variables are controlled for. Statistical Issues in Analysis of Individual Data In this section, econometric models are developed for health insurance and for four categories of health care services that jointly account for about Patterns of Health Care Use in Vietnam 401 99 percent of the total health care expenditure. The largest component (92.7 percent) is due to visits to drug vendors and pharmacy; government hospitals and private health facilities account for another 5.5­6.0 percent, with the former being slightly larger. The commune health centers account for close to 1 percent. Traditional (Eastern) medicine providers and other smaller components are not analyzed in this chapter. The Problem of Zeros Individual use data are available for 27,731 cases. However, this includes a high proportion of cases of zero use, in part because the survey period is truncated at four weeks (see table 11.7). For PHARVIS, the zero proportion is about 76 percent, but for the other three major provider categories it is be- tween 96 and 98 percent each. Typically, the observed frequency distribution shows positive probability mass at only a few other integer values--such as 1, 2, and 3--and very small mass at higher integer values. For example, for PHARVIS, the frequency of one visit is about 10 percent and of two visits, less than 5 percent. In the case of GOVHOSP, PHF, and CHC, the corre- sponding percentages are even smaller. The handling of the zero problem depends on whether one's objective is to model health care expenditures or frequency of contact with the provider. In modeling expenditures, the zeros pose a problem because they introduce a discontinuity in the distribution of expenditures. But in modeling a dis- crete random variable, such as contact frequency, this is not an issue. Table 11.7. Frequency Distribution of Health Service Contacts Number of contacts PHARVIS GOVHOSP PHF CHC OTHER HOSPNITE 0 20,639 26,796 26,481 27,041 27,158 42 1 3,827 736 540 486 254 71 2 1,716 133 316 111 124 100 3 776 30 180 52 66 122 4 359 17 99 23 36 74 5 174 8 36 6 24 92 6 64 4 12 3 12 54 7 43 1 18 4 16 274 8 16 3 4 1 9 32 9 4 2 3 3 2 20 10+ Note: Sample size, 27,731. Sample size for HOSPNITE, 1,463. Outpatient contacts only for GOVHOSP. Definition of variables: CHC = commune health center. GOVHOSP = government hospital. HOSPNITE = nights in government hospitals. OTHER = other government facilities. PHARVIS = pharmacy visits. PHF = private health facility. Source: Author's calculations. 402 Economic Growth, Poverty, and Household Welfare in Vietnam For the 1993 data, Gertler and Litvack (1998) chose the two-part model (TPM), in which the first part models the split between zero and nonzero expenditures through a binary outcome model. That is, the focus is on mod- eling the probability of contact with the provider, using the logit, the probit, or the linear probability model. The second part of the model is a linear regression in which the outcome variable is health care expenditure, or the logarithm of it, for those who had at least one contact with a provider. If y denotes the measure of health care use--for example, expenditure--and X denotes the explanatory variables, then according to the two-part model: E[y|X] = probability [y > 0|X] × E[y|y > 0, X]. The TPM framework is attractive because it provides a solution to the awkward statistical issues arising from the presence of significant probabil- ity mass at y = 0.4 Note also that the zeros have two possible interpretations. The first is that they correspond to "corner solutions" in the consumer choice problem. That is, they indicate nonconsumption, given current in- come, price, and health status. A second interpretation is that zeros indicate that the good under consideration is not in the consumer's choice set for a variety of possible reasons (see Cameron and Trivedi [1998, chapter 6]). The first part of the TPM may be interpreted as a model of the probability of an interior solution to the choice problem, and the second part models the level of consumption, conditional on an interior solution being realized. There- fore, both parts yield estimates of economically interesting parameters. A disadvantage of this framework is that the number of observations available for the second part of the model can be proportionately quite small, leading to a loss of precision in estimation. In these cases, it is attrac- tive to use a count data model, which can naturally accommodate a signifi- cant probability mass at zero, making it unnecessary to separate the zero and nonzero observations as in the case of TPM.5 Further, count data mod- els work well in those cases where the outcomes are concentrated on a rela- tively few values of the outcome variable, which is the case for three of the four usage measures that need to be modeled. An important exception is the number of nights spent in a GOVHOSP. In all count models, the following specification is used to model the number of visits: log(E[#visits to provider |X|]) = Xjj. This chapter uses Gertler and Litvack's 1998 binary outcome framework in those instances where the frequency distribution of contacts is strongly concentrated on just two (0 and 1) or three values (0, 1, and 2). For other cases, most notably PHARVIS, a count data model is used in which the mod- eling focus is on the average number of contacts as a function of observed characteristics of individuals. Aggregate health care expenditures are modeled at both the individual and household levels, but expenditure is not modeled on individual com- ponents of use. Patterns of Health Care Use in Vietnam 403 The Problem of Clustering Another significant factor is the clustering of responses by the primary sam- pling unit, the commune, used in the complex (stratified) sample survey methodology. In the case of VLSS data, clustering is by commune, which is the sampling unit. The sample covers fewer than 200 communes, with a variable number of observations per commune. Clustering implies lack of independence of observations. It affects both the discrete and continuous outcome variables studied in this chapter. If there is significant within-commune homogeneity, perhaps due to common unobserved fixed or random components that affect all individual behavior within a commune, then assuming independence of observations will pro- duce estimates with a spuriously higher degree of precision than is war- ranted. The correct sampling variances are larger than estimated under the independence assumption. Two statistical approaches to deal with the effects of clustering are used. The first is based on adjusting all standard errors for clustering by using the "cluster-robust" variance estimator. This adjustment typically inflates the es- timated variance of the coefficients. This approach is analogous to the use of the Eicker-White robust variance estimator. All estimates of standard errors reported in this chapter are "cluster robust" unless stated otherwise.6 A second approach to clustered observations uses a different statistical model. In this case, it is assumed that each commune has its own intercept, denoted j, where j is the commune subscript. Correlation between re- sponses for a given commune reflect the presence of a common intercept, which is treated as a cluster (fixed) effect. In this formulation of the cluster effect, both the point estimates of parameters and their standard errors are affected, whereas the cluster-robust approach adjusts only standard errors. An example of a linear fixed effects model is shown in Deaton (1997, pp. 288­92). In this chapter, it is necessary to allow for commune fixed effects, but the main interest is not in estimating the fixed effects but in elim- inating their impact on economically interesting parameters such as income elasticity. Neglect of the fixed (cluster) effects can bias the coefficients of in- terest if the relevant regressors are correlated with fixed effects. In fixed effect models, parameters of interest may be obtained after "sweeping out" commune effects. However, unlike linear regression models (where the sweep step is always possible), for the nonlinear regression models used in this chapter, this approach is feasible only for a class of models, such as the logit (but not the probit) and Poisson regression models (Cameron and Trivedi 1998, chapter 9.3; Hsiao 1986). An interesting feature of the use of the fixed effects count model is that it indirectly reduces the impact of the excess zeros problem. This comes about because the fixed effects model will drop all observations from a commune if the responses within it are identical. In general, therefore, fixed effect models are based on fewer observations because (as is the case in the sample here) all observations from a commune are thrown out if every survey 404 Economic Growth, Poverty, and Household Welfare in Vietnam respondent records the same response (such as zero). For example, if in any particular commune the sample shows no usage of the CHC, then all obser- vations for that commune are dropped. This reduces the sample size and also reduces the impact of the excess zeros problem. The number of "lost" observations will vary with the type of provider being considered. The two approaches for handling clustered data involve nonoverlapping statistical assumptions; hence, there is no guarantee that the results from the fixed effects model will coincide with those from the standard models. Nor is there a simple way of choosing between the two models if they yield dif- ferent results. Econometric Models The preferred model for counted data in this chapter is some variant of the count regression model that can handle the statistical problems of cluster- ing, excess zeros, or both. In those cases where the excess zeros problem is very severe, the binary outcome model is used to model the probability of nonzero outcome. The justification here is that the ability to distinguish be- tween factors that lead to, say, n + 1 units of use rather than $n$ is seriously reduced when the cell counts for those outcomes are small. Thus, to gain precision, an attempt to distinguish between zero usage and positive usage is made, ignoring the extent of positivity. In all cases, the fixed effects version of an estimator is used to check the robustness of the estimates. Fixed effects estimators are usually adopted in the analysis of longitudinal data to deal with individual-specific hetero- geneity under the assumption that this component is fixed for any individ- ual and can be "swept out" by an appropriate data transformation that is feasible given repeated observations on the same individual. In the case of VLSS data, it is assumed instead that the fixed effect is commune specific and can be handled in an analogous fashion, given more than one observa- tion per commune. That is, the longitudinal (panel) models can be adapted to handle the present case. The underlying general method for handling clustered data is the conditional maximum likelihood approach, which can be applied to binary, counted, or continuous data. Fixed effects variants of the Poisson, the logit, and the linear regression available in the literature for panel data can be adapted to handle clustered observations. Counted data are sometimes modeled using the ordered probit or the Tobit regression. These rely on the normality assumption. Violation of nor- mality due to the excess zeros and clustering is so pervasive in the VLSS sample that these estimators are inappropriate. To summarize, the following alternative models specifications were esti- mated. For data on individuals: · Probability of having health insurance · Probability of visit to CHC, PHF, GOVHOSP, or PHARVIS · Number of visits to CHC, GOVHOSP, PHF, or PHARVIS · Probability of HOSPADM Patterns of Health Care Use in Vietnam 405 · Number of nights spent in government hospital (HOSPNITE) · Aggregate expenditure on health care provided by all providers in the four weeks preceding the survey and in government hospitals in the previous 12 months (MEDEXP). For household data, model specifications were estimated on aggregate expenditure on health care provided by all providers in the four weeks pre- ceding the survey and in government hospitals in the previous 12 months. Results The main focus of the discussion of results is on the role of income and health insurance, after conditioning on a set of relevant covariates. This con- ditioning applies to all models unless stated otherwise. As is customary, the income variable is proxied by logarithm of total household expenditure, denoted log(INC). The only price variable is HLTHINS. The conditioning covariates are AGE, SEX, MARRIED, EDUC, ILL, INJ, ILLDAYS, and ACTDAYS. These variables are defined in table 11.1. Determinants of Health Insurance Status Initially, there is no differentiation between voluntary and mandatory en- rollees into the VHI program. A logit model for HLTHINS status, using the full sample, shows that AGE, EDUC, and income (INC) are all strongly pos- itively related to having insurance. (See the first two columns of table 11.8 for detailed regression results.) The coefficient on log of total household ex- penditure, the proxy for income (INC), is precisely determined with a t ratio exceeding 5 in most specifications. High income and high education both increase the probability of compulsory coverage; hence, the observed result is quite plausible. Surprisingly, being married is negatively related to having health insurance. One possible interpretation of this result is that it reflects the higher rate of health insurance among students through the CSI pro- gram, but it could also reflect relatively higher enrollment into the VHI by unmarried males. Controlling for these factors, there is a negative association between hav- ing insurance and being female. In the third and fourth columns, the results are for a specification with a more flexible functional form for the income variable to allow for different response coefficients in the four income quar- tiles, denoted INC1, INC2, INC3, and INC4. Essentially, log(INC) is split into four ranges to consider whether the income coefficient varies and whether this spline functional form improves the fit of the model. This spec- ification fits the data better. However, the fixed effects logit version of the same specification, which allows for commune fixed effects, fits even better. Overall, this regression gives a similar picture to that from the simple logit, but it shows that the insurance decision is insensitive to income in the low- est quartile and most sensitive in the two middle quartiles. . err 10 ors n.a. n.a. n.a. n.a. n.a. err logit Std. 0.0452 0.0018 0.0388 0.0616 0.01 d dummy 60< oss-section education. cr of 0/1 effects 8,220 1 age<22 of standar = case years logit, Fixed n.a. n.a. n.a. n.a. n.a. het Coeff. 0.5868 0.021 0.2456- 0.8005- 0.2144 to MARRIED obustr e). data completed For = . err 1 sample. EDUC expenditur n.a. n.a. logit Std. 0.001 0.0365 0.0464 0.0105 0.1509 0.2234 0.1684 0.0762 longitudinal the for years. omfr in effects 9,422 Full household model age opped = Fixed n.a. n.a. 171 dr Coeff. 0.0105 0.3509- 0.4266- 0.2045 0.1 1.0391 0.6203 0.3004 e AGE nominal ar esponding corr log(total . variables: = the of Enrollment err individuals n.a. n.a. 177 oss Std. 0.6039 0.0366 0.0733 0.0227 0.2330 0.2871 0.2343 0.1 adapts acr log(INC) logit e. logit Definition likelihood. Insurance 1,2421 Full of fects ef variation log Robust expenditur clustering. Coeff. Health 3.9217- n.a. n.a. 0.2966- 0.5008- 0.2963 0.0169- 1.3532 0.3967 0.0837 Fixed esponser for denotes of no household specific. . adapted -log-lik. 1 show err n.a. n.a. n.a. n.a. that nominal gender Probability Std. 0.2985 0.061 0.0412 0.0368 0.0726 0.0226 commune formula logit be total for of 1,2681 Full to by Communes Robust oxied dummy -White-type n.a. n.a. n.a. n.a. pr Coeff. 4.53- 0.3645 0.3786 0.2940- 0.5099- 0.2952 assumede 0/1 = ar Eicker communes. income SEX calculations. Determinants 's fects ef by using applicable. ed status. 1.8.1 Author Fixed Not household ce: cluster = ableT ariableV n.a. Note: calculated marital e Sour Constant log(INC) AGE SEX MARRIED EDUC INC1 INC2 INC3 INC4 -log-lik Sample data ar INC for 406 Patterns of Health Care Use in Vietnam 407 Several variants of the fixed effects logit model were estimated by three age categories: younger than 22 years, between ages 22 and 60, and older than 60. The motivation for this disaggregation comes from the differences in types of health insurance. It is expected that the student enrollees are pre- dominantly concentrated in the youngest group and the retired individuals in the oldest group, leaving the middle group containing the largest number of those mandatorily enrolled. It is interesting that the average rate of en- rollment in the three age groups varies only between 16 percent and 19 per- cent. The regressions for the middle group are shown in the last two columns of table 11.8, and the main features of this regression conform to those mentioned before in this section. The results for the youngest group show the highest income coefficient (0.72), and those of the oldest group show statistically zero income sensitivity. (To save space, the separate results for the young and old groups are not reported in table 11.8.) Although this disaggregation is rough, and a more careful modeling of the insurance deci- sion is desirable, the results do not suggest that there is serious distortion due to the absence of distinction between types of health insurance. It is, however, possible that the main impact of different types of insurance will be on use. This issue will be explored in later sections. An important issue is whether there is adverse selection into the insur- ance program. To infer that there is, one would need to show that controlling for other factors, the insurance program enrolls a disproportionately larger number of bad risks. Identifying enrollees at high risk is not easy because of the lack of information about the long-term health status of individuals in the health survey. The inference has to be more indirect. The survey pro- vides information on injury or illness in the four weeks preceding the sur- vey, as well as information on the number of illness days (ILLDAYS) and days of limited activity (ACTDAYS). Information is also available on the smoking or nonsmoking status of the respondent. This latter variable can serve as a proxy for future health problems. The logit regression results sug- gest that there is no significant association between insurance status and the number of ILLDAYS, ACTDAYS, or both, or on smoking habit. There is, however, a statistically significant, but weak, positive association between occurrence of illness (ILL) and injury (INJ) and having health insurance. This is consistent with those who are insured having a greater proclivity to report illness or injury. Employment plays an important role in insurance status because insur- ance is mandatory in some government and private sectors. If, as seems rea- sonable, membership in this sector is independent of the level of health care use, then exogeneity of the insurance variable is justified. In such cases, it seems valid to argue that causality runs from insurance to health care use. To put this argument on a sound footing, it is desirable to enter sector of occu- pation as an additional factor in the insurance equation and confirm its role as an important factor after conditioning on income and educational attain- ment. Implementing this step requires additional data that were not avail- able when this study was done. Regression analysis of health care use, treat- ing the insurance variable as exogenous, is needed. 408 Economic Growth, Poverty, and Household Welfare in Vietnam Table 11.9. Models for Number and Probability of Commune Health Center Visits Fixed effects Poisson Fixed effects Poisson Fixed effects logit Variable Coeff. Std. err. Coeff. Std. err. Coeff. Std. err. Constant n.a. n.a. n.a. n.a. n.a. n.a. log(INC) n.a. n.a. -0.2214 0.0775 -0.3574 0.1116 INC1 0.0117 0.1807 n.a. n.a. n.a. n.a. INC2 -0.0704 0.3240 n.a. n.a. n.a. n.a. INC3 -0.3847 0.3592 n.a. n.a. n.a. n.a. INC4 -0.9222 0.3285 n.a. n.a. n.a. n.a. HLTHINS 0.1855 0.0941 0.1885 0.0996 0.3820 0.1310 SEX 0.0699 0.0616 0.0608 0.0616 0.0637 0.0877 AGE -0.0078 0.0018 -0.0082 0.0018 -0.0082 0.0026 MARRIED 0.3015 0.0764 0.3124 0.0760 0.2958 0.1108 ILLDAYS 0.0392 0.0041 0.0394 0.0041 0.0180 0.0065 ACTDAYS -0.0221 0.0122 -0.0213 0.0121 -0.0194 0.0301 INJ 1.1758 0.1788 1.1447 0.1788 1.6969 0.3422 ILL 3.7315 0.1992 3.7248 0.1991 3.8622 4.1359 EDUC -0.0496 0.0235 -0.0510 0.0234 0.0065 0.0316 -log-lik 2,695 2,700 1,640 N 15,132 15,132 15,132 n.a. Not applicable. Notes: Fixed effects are assumed to be commune specific. Fixed effects logit and Poisson adapt the corresponding models for longitudinal data to the case of cross-section data clustered by communes. Communes that show no response variation across individuals are dropped from the sample. Definition of variables: ACTDAYS = number of days of limited activity in 4 weeks before survey. AGE = age in years. EDUC = completed years of education. HLTHINS = 0/1 dummy for health insurance status. ILL = 0/1 dummy for illness in 4 weeks before survey. ILLDAYS = number of days of illness/injury in 4 weeks before survey. INC = household in- come proxied by total nominal household expenditure. INJ = 0/1 dummy for injury in 4 weeks before survey. log(INC) = log(total nominal household expenditure). MARRIED = 0/1 dummy for marital status. SEX = 0/1 dummy for gender. -log-lik denotes log of likelihood. Source: Author's calculations. Commune Health Centers Detailed regression results are given in table 11.9. In the case of the CHC, the relative infrequency of use--the excess zeros problem--has already been noted. This makes the results from direct application of the Poisson regres- sion unreliable. Thus, the reported results use either the fixed effects (condi- tional maximum likelihood) variant of the Poisson regression or simply the fixed effects logit model for the probability of CHC use. The three regressions in table 11.9 suggest that the CHC is treated by users as an inferior good. The marginal impact of rising household income on both the probability and level of usage is negative and significant (see the last four columns of table 11.9). The impact of rising educational level is also negative but only marginally significant. That is, CHCs are typically not Patterns of Health Care Use in Vietnam 409 used by the higher-income and better-educated groups. The VHSR (World Bank 1999, p. 51) notes that "the quality of commune health centers is regarded by most users to be very low." It is to be expected that with rising incomes, users may substitute toward other, relatively higher-quality providers. However, to the extent that CHCs act as suppliers of pharmaceu- ticals, and as the supply of these has improved both in the CHCs and gener- ally, the absolute level of their use (especially by the low-income groups) could rise. In their analysis of the 1993 VLSS, Gertler and Litvack (1998) have also noted the low quality of CHC services (also see World Bank [1999]). The evidence here is consistent with their observation. However, it would have been more satisfactory to have introduced additional variables in the regression to reflect various observed CHC features to pin down precisely why CHC usage continues to be low. For example, it would be useful to know whether there are differences in their ability to supply drugs or provide higher levels of service. However, this requires more data than are currently available. To throw more light on the relation between income [log(INC)] and the level of use of CHCs, an additional regression is reported in the first two columns of table 11.9, which uses the spline functional form to allow for dif- ferent response coefficients in the four income quartiles, denoted INC1, INC2, INC3, and INC4. The fit of the model for probability of using a CHC improves only slightly. These results indicate that the income coefficient is not significantly different from zero for the three lowest income quartiles, and it is significantly negative for the highest income quartile. Qualitatively, the pattern of coefficients is that which would be expected for an inferior good. The coefficient of HLTHINS is positive. This is an unexpected result. When additional regressions were estimated for the three age groups, the re- sults indicated a positive coefficient in each case but with a low degree of precision. One possible explanation is that the result reflects the role of the CHCs as drug providers to eligible insured individuals. The results also indicate that AGE and CHC use are negatively related-- that is, older individuals avoid using CHCs. The strongest positive relation of CHC use is with short-term health sta- tus. Those who are ill or injured and have suffered limitation in physical activity do use CHCs. For young and sick or injured individuals from low- income households, CHCs may serve as a first step in seeking health care. Government Hospitals This subsection first discusses the results from models of probability of out- patient visits. This is followed by discussion of results for inpatient hospital admissions and number of nights spent in the hospital. PROBABILITY OF OUTPATIENT USE. A variety of estimation methods have been used to model the probability of use of public hospital outpatient ser- vices. Detailed regression results are given in table 11.10. The results from the regular logit and fixed effects logit are qualitatively similar and precise. 410 Economic Growth, Poverty, and Household Welfare in Vietnam Table 11.10. Government Hospital Inpatient and Outpatient Use Robust Fixed effects Fixed effects Fixed effects logit logit logit Poisson GOVHOSP GOVHOSP HOSPADM HOSPNITE Variable Coeff. Std. err. Coeff. Std. err. Coeff. Std. err. Coeff. Std. err. Constant -5.6949 0.1915 n.a. n.a. n.a. n.a. n.a. n.a. log(INC) 0.4474 0.0571 0.4183 0.0839 0.1045 0.0643 0.4283 0.0175 HLTHINS 1.1155 0.0834 0.9449 0.0928 0.2070 0.0768 0.1583 0.0214 SEX 0.1723 0.0742 0.1709 0.0746 0.0115 0.0553 -0.0452 0.0160 AGE -0.1783 0.0518 -0.1933 0.0485 -0.0349 0.0370 0.0068 0.0004 MARRIED 0.4947 0.1006 0.5085 0.0966 0.1035 0.0718 -0.1812 0.0194 ILLDAYS 0.0906 0.0041 0.0964 0.0046 0.0169 0.0054 0.0111 0.0013 ACTDAYS 0.0392 0.0300 0.0451 0.0199 0.0555 0.0259 -0.0046 0.0070 INJ 1.7518 0.2653 1.9980 0.2440 -0.7760 0.4478 0.4671 0.0900 ILL 0.5322 0.0318 0.6057 0.0384 0.1851 0.0712 0.0388 0.0207 EDUC 0.0003 0.0192 -0.0289 0.0222 -0.0441 0.0173 -0.0063 0.0047 -log-lik 3,297 2,568 4,981 10,479 N Full 25,227 27,380 1,412 n.a. Not applicable. Note: Fixed effects are assumed to be commune specific. Fixed effects logit and Poisson adapt the corresponding models for longitudinal data to the case of cross-section data clustered by communes. Communes that show no response variation across individuals are dropped from the sample. For robust logit, standard errors are calculated using Eicker-White-type formula adapted for clustering. Definition of variables: ACTDAYS = number of days of limited activity in 4 weeks before survey. AGE = age in years. EDUC = completed years of education. HLTHINS = 0/1 dummy for health insurance status. ILL = 0/1 dummy for illness in 4 weeks before survey. ILLDAYS = number of days of illness/injury in 4 weeks before survey. INC = household income proxied by total nominal household expenditure. INJ = 0/1 dummy for in- jury in 4 weeks before survey. log(INC) = log(total nominal household expenditure). MAR- RIED = 0/1 dummy for marital status. SEX = 0/1 dummy for gender. -log-lik denotes log of likelihood. Source: Author's calculations. The probability of hospital use is strongly and positively related to household income. The elasticity of probable use with respect to income is estimated around 0.4. There is also a strong positive relationship with having health insurance (HLTHINS), which confirms the strong connection between the use of public hospital outpatient services and having health insurance. These two results confirm that income and price effects in the use of public hospital outpatient services are strong. Additional robustness checks do not show that the fit of the model can be improved by using a spline specification for the income variable. The use of public hospitals is also strongly related to being ill or injured and to the length of illness. Usage is also higher for married persons and for females. Somewhat unexpectedly, AGE is negatively related to use. Patterns of Health Care Use in Vietnam 411 However, inpatient use is strongly and positively related to age, as dis- cussed in the next section. HOSPITAL ADMISSIONS AND NIGHTS IN THE HOSPITAL. About 5 percent of the sample reported admission into hospital in the 12 months before the survey. About 26 percent of those admitted are insured. Of these, the overwhelming majority, more than 97 percent, spent at least one night in the hospital. The average number of nights spent in the hospital, conditional on admission, is between 13 and 14, but the median figure is about 7, indicating a skewed, fat-tailed distribution. The probability of hospital admission is higher for the insured than for the uninsured, and the average length of hospital stay is also longer for the insured, about 18 days versus 12 for the uninsured. The distribution of hospital expenditure is also correspondingly skewed and fat-tailed, with very high nonnormal kurtosis. That is, a relatively small number of individuals account for a high proportion of the total hospital ex- penditure. Detailed regression results for hospital admission and hospital nights are given in table 11.11. The results indicate a strong positive relation between being insured and the probability of admission into the hospital. There is also a statistically significant positive relation between income and hospital admission, but this link is weaker than that with insurance. Other factors that indicate bad health status also increase the probability of hospital admission. When HOSPADM regressions are run separately for the insured and uninsured subsamples, hardly a single explanatory variable has a sta- tistically significant coefficient for the insured subsample, but for the unin- sured, both health status and income levels are important factors. When the HOSPNITE regressions are run separately for the insured and uninsured subsamples, most of the coefficients in the uninsured equation are ab- solutely larger, indicating their greater sensitivity to other factors. That is, having health insurance reduces the role of other factors but does not elimi- nate it. Developing a robust regression model for the number of nights in the hospital is difficult because of the awkward frequency distribution of the data. Several linear and nonlinear (count data) models were tried. None fit the data really well because the data are intrinsically rather noisy. Overall, these results indicate that AGE, INC, and HLTHINS are the most important explanators of hospital usage. When different specifications of equations are considered, INC is not a robust explanatory variable, but HLTHINS and AGE remain consistently significant. Private Health Facilities The contact rate for private health facilities is significantly higher among the uninsured than the insured population. It is also slightly higher among the younger age group. Because the excess zero problem is very evident (95.5 percent of the sample report no use), the estimated regressions model 412 Economic Growth, Poverty, and Household Welfare in Vietnam Table 11.11. Models for Probability of Use of Private Health Care Fixed effects logit Robust logit Variable Coeff. Std. err. Coeff. Std. err. Constant 0.0 n.a. -6.93 0.6177 log(INC) 0.0414 0.0723 n.a. n.a. INC1 n.a. n.a. 0.5434 0.3033 INC2 n.a. n.a. -0.3038 0.5098 INC3 n.a. n.a. 1.5588 0.3783 INC4 n.a. n.a. 0.1425 0.1572 HLTHINS -0.2388 0.1064 -0.4003 0.1135 SEX 0.1512 0.0653 0.1396 0.0544 AGE -0.3908 0.0364 -0.3628 0.0373 MARRIED 0.3187 0.0849 0.3034 0.0815 ILLDAYS 0.0390 0.0046 0.0336 0.0045 ACTDAYS -0.0376 0.0242 -0.0325 0.0237 INJ 1.3521 0.2512 1.3208 0.2760 ILL 4.3357 0.2160 4.2822 0.2211 EDUC 0.0242 0.0207 -0.0451 0.0244 -log-lik 3,097 3,086 N 27,733 27,783 n.a. Not applicable. Note: Fixed effects are assumed to be commune specific. Fixed effects logit adapts the same model for longitudinal data to the case of data clustered by communes. Communes that show no response variation across individuals are dropped from the sample. For robust logit, stan- dard errors are calculated using Eicker-White-type formula adapted for clustering. Definition of variables: ACTDAYS = number of days of limited activity in 4 weeks before survey. AGE = age in years. EDUC = completed years of education. HLTHINS = 0/1 dummy for health insurance status. ILL = 0/1 dummy for illness in 4 weeks before survey. ILLDAYS = number of days of illness/injury in 4 weeks before survey. INC = household income proxied by total nominal household expenditure. INJ = 0/1 dummy for injury in 4 weeks before survey. log(INC) = log(total nominal household expenditure). MARRIED = 0/1 dummy for marital status. SEX = 0/1 dummy for gender. -log-lik denotes log of likelihood. Source: Author's calculations. the probability of contact. Detailed regression results are given in table 11.11. In the full-sample regressions, HLTHINS has a negative coefficient, which reflects the higher contact rate among the uninsured. Because insured indi- viduals are eligible for outpatient care in GOVHOSP, the private facilities are relatively more frequented by the uninsured. Health insurance, therefore, diverts usage away from the private facilities. Once insurance status is controlled for, there is no clear evidence that in- come and private health care are positively related. The size of the income coefficient is found to be sensitive to changes in specification. In the fixed effects logit model, the income coefficient is small and relatively imprecise. The use of the spline specification also does not completely resolve the am- biguity, but there is slight evidence that in the higher-income quartiles, there Patterns of Health Care Use in Vietnam 413 may be a significant positive relation between income and the use of private health facilities (PHF). In an attempt to examine the role of aggregation bias in estimating the in- come effect, the fixed effects Poisson model was reestimated by age groups, using both the regular and the spline specification of the income variable. For the middle-income group (ages 22­60 years), the insurance impact is sig- nificantly negative, and the income coefficient is small (.064; t ratio, 1.06). In the spline version, the insurance impact is again negative and significant, and two of the four income coefficients are also negative, with t ratios greater than 2, and one positive with a t ratio also greater than 2. That is, PHFs appear to be an inferior good at the higher range of income but possi- bly a normal good at a lower level. However, the link between income and use is not robust. The ambiguity in the results persists also for the younger (ages younger than 22 years) and older (ages older than 60 years) groups. These results suggest that as yet there is only weak evidence of income- induced demand shift toward PHFs and quite clear evidence of the negative impact of the VHI plan on PHFs. In other respects, the pattern of use is qualitatively similar to that of pub- lic hospitals. As in that case, usage is positively related to being ill or injured, being female (SEX) or married, and with the length of illness, and negatively related to age. The last result may simply indicate a greater willingness on the part of the young to try out the emergent private health facilities. Pharmacy Visits Detailed regression results for the number of visits and the probability of visit are given in table 11.12. Because of the overwhelming importance of expenditure on purchased medicines, the results for the frequency of phar- macy visits are of special interest. Although many variants of the Poisson regression and the logit model were used, the reported results are based on the commune fixed effects formulation.7 The most interesting results shown in the first two columns of table 11.12 are that the overall income effect is significantly negative and the health in- surance effect is also negative.As in the case of GOVHOSP, using the fixed ef- fects model lowers the absolute size of the income coefficient. This result is plausible and consistent with lower-income households relying overwhelm- ingly on self-medication in the event of illness, injury, and activity limitation. Evidence has already been cited that indicates an increasing reliance on self- medication as the supply of drugs has improved and the retailing of drugs has been deregulated. Drugs can also be dispensed at public hospitals, but the evidence presented above suggests that this particular channel is available to, and more likely to be used by, the high-income insured individuals. To more closely investigate the connection between income and phar- macy visits, separate regression models were fitted for probability and num- ber of pharmacy visits, using a flexible spline specification for the income 414 Economic Growth, Poverty, and Household Welfare in Vietnam Table 11.12. Models for Pharmacy Visits Fixed effects Poisson Fixed effects Poisson Fixed effects logit PHARVIS PHARVIS PHARDUM Variable Coeff. Std. err. Coeff. Std. err. Coeff. Std. err. Constant n.a. n.a. n.a. n.a. n.a. n.a. log(INC) -0.1025 0.0190 n.a. n.a. n.a. n.a. INC1 n.a. n.a. 0.0623 0.0544 -0.0336 0.1257 INC2 n.a. n.a. -0.0939 0.0939 0.2239 0.2234 INC3 n.a. n.a. -0.0346 0.0823 -0.6112 0.1972 INC4 n.a. n.a. -0.3042 0.0461 -0.4301 0.1028 HLTHINS -0.1614 0.0273 -0.1589 0.0273 -0.2613 0.0592 SEX 0.0255 0.0171 0.0024 0.0171 0.1267 0.0399 AGE 0.0564 0.0004 0.0283 0.0004 0.0059 0.0011 MARRIED 0.1071 0.0201 0.1010 0.0201 0.1158 0.0491 ILLDAYS 0.0229 0.0012 0.0230 0.0012 -0.0401 0.0032 ACTDAYS 0.0203 0.0056 0.0207 0.0056 0.0438 0.0190 INJ 0.3019 0.0788 0.2965 0.0789 0.2533 0.2195 ILL 3.5072 0.0486 3.5062 0.0486 4.9302 0.0706 EDUC -0.0167 0.0055 -0.0169 0.0056 0.0162 0.0126 -log-lik 18,564 18,547 7,617 N 27,671 27,671 27,671 n.a. Not applicable. Note: Fixed effects are assumed to be commune specific. Fixed effects Poisson adapts the same model for longitudinal data to the case of data clustered by communes. Communes that show no response variation are dropped from the sample. Definition of variables: ACTDAYS = number of days of limited activity in 4 weeks before survey. AGE = age in years. EDUC = com- pleted years of education. HLTHINS = 0/1 dummy for health insurance status. ILL = 0/1 dummy for illness in 4 weeks before survey. ILLDAYS = number of days of illness/injury in 4 weeks before survey. INC = household income proxied by total nominal household expendi- ture. INJ = 0/1 dummy for injury in 4 weeks before survey. log(INC) = log(total nominal household expenditure). MARRIED = 0/1 dummy for marital status. PHARDUM = 0/1 dummy for pharmacy visit. SEX = 0/1 dummy for gender. -log-lik denotes log of likelihood. Source: Author's calculations. variable. These results suggest that pharmacy use is a normal good, with a positive (but imprecisely determined) income elasticity, in the lower-income quartile. But with high probability, it is an inferior good, with a negative income elasticity, in the two highest quartiles. Unfortunately, the relatively large standard errors on the coefficients preclude a stronger statement. That is, pharmacy visits appear to be an inferior good for the rich, but a normal good for the poor. This result is different from that in some previous analy- ses based on the 1993 VLSS data (Gertler, Litvack, and Prescott 1996), which suggests that pharmacy visits are a normal good at all income levels. The impact of HLTHINS on self-medication is found to be negative, sta- tistically significant, and sign-wise robust across a range of alternative spec- ifications. The size of the impact is larger in models that do not control for Patterns of Health Care Use in Vietnam 415 clustering, and they are not reported in table 11.12. The fixed effects Poisson model yields the lowest estimate, but even this is unambiguously negative and significant. The interpretation that self-medication is a risky form of health care is consistent with the results, and it is avoided as income rises and as alternative, higher-quality health care becomes available through health insurance. In Vietnam, the higher-quality care is provided in public hospitals. Note that the more highly educated individuals, and hence pre- sumably those better aware of the risks of self-medication, also avoid phar- macy visits. The role of other factors--such as being female, being married, having illness or injury, and the length of illness--is similar to that found for other types of health care; that is, they all increase the frequency of pharmacy vis- its. One difference is that AGE does have a positive effect. This picture is broadly consistent with the predictions of a theoretical model that treats self-medication as a risky alternative to professional care (Chang and Trivedi 2003). The reported results are robust and do not quali- tatively change if the econometric analysis is carried out by insurance status or by using disaggregated age categories, or using other econometric esti- mators based on more flexible assumptions. Analysis of Health Care Expenditure Individual Data This section is devoted to a regression analysis of medical expenditure. The dependent variable is log(MEDEXP)) for each member of the household, conditional on a positive level of expenditure for that individual. All types of health care expenditure in the four-week period preceding the survey are included. The sample size is 8,081. The main focus in this analysis is again on the role of household income and HLTHINS. As before, AGE, SEX, MARRIED, EDUC, and health status (ILL, INJ, ILLDAYS, ACTDAYS) are controlled for. The detailed regression results are shown in table 11.13. The results indicate that whereas household income continues to show a strong explanatory power, the insurance variable is much less significant. The point estimate of the elasticity of individual health expenditure with re- spect to income is of the order of 0.3­0.4. The estimates of income elasticity from the fixed effects model are slightly smaller than those without fixed effects. The response to HLTHINS, however, is positive but with a relatively large standard error. This can be interpreted as follows: HLTHINS acts to di- vert demand from lower-quality care, such as that provided by commune health centers, to higher-quality care, such as that provided in public hospi- tals. The aggregate response of expenditure to insurance would then be small or zero if most of the impact takes the form of such substitution. How- ever, substitution in terms of number of visits need not have zero impact on 416 Economic Growth, Poverty, and Household Welfare in Vietnam Table 11.13. Models for Positive Medical Expenditure for Individuals Fixed effects Ordinary least squares, robust Ordinary least squares Variable Coeff. Std. err. Coeff. Std. err. Constant 0.6096 0.0797 n.a. n.a. log(INC) 0.3623 0.0195 0.2963 0.0248 HLTHINS 0.0922 0.0327 0.0254 0.0332 SEX 0.0021 0.0228 0.0033 0.0212 AGE 0.1071 0.0136 0.0961 0.0126 MARRIED 0.0321 0.0284 0.0468 0.0267 ILLDAYS 0.0427 0.0022 0.0422 0.0022 ACTDAYS 0.0455 0.0127 0.0380 0.0119 INJ 0.1488 0.1275 0.2395 0.1170 ILL -0.0659 0.0587 -0.0363 0.0578 EDUC 0.0085 0.0063 0.0047 0.0071 R2 0.1208 N 8,081 8,081 n.a. Not applicable. Note: Robust standard errors are calculated using Eicker-White-type formula adapted for clustering. Definition of variables: ACTDAYS = number of days of limited activity in 4 weeks before survey. AGE = age in years. EDUC = completed years of education. HLTHINS = 0/1 dummy for health insurance status. ILL = 0/1 dummy for illness in 4 weeks before survey. ILLDAYS = number of days of illness/injury in 4 weeks before survey. INJ = 0/1 dummy for injury in 4 weeks before survey. INC = household income proxied by total nominal household expenditure. log(INC) = log(total nominal household expenditure). MARRIED = 0/1 dummy for marital status. SEX = 0/1 dummy for gender. Source: Author's calculations. total expenditure. If the substitution is toward higher-quality care, then total medical expenditure may increase. Such a situation seems consistent with the estimates obtained. The insurance effect on health care expenditure re- flects substitution toward higher-quality care. The reported estimates are robust. Point estimates of income elasticity similar to those from the fixed effects model are obtained from several other variants (including separate models for insured and uninsured samples, a random effects version of the estimated model, and a model in which in- come coefficient is allowed to differ by income quartile). To save space, the details of these results are not included in table 11.13. Household Data Analysis of medical expenses aggregated across all household members serves as a useful check on the results from individual data. It also yields es- timates of Engel curves for medical expenditure. The main limitation of this approach is that one cannot control for health status of individual members Patterns of Health Care Use in Vietnam 417 of the household, and to do so is important. However, one can control for some of the other relevant variables, such as location (urban or rural) and size of the household, SEX, AGE, and educational attainment of the head of the household. Descriptive sample statistics show that, on average, households with an insured head spend about 20 percent more on health care than those without. Average household health care expenditure is higher for urban than rural households, although rural households are typically slightly larger. Detailed regression results are shown in table 11.14. The most interesting result is that point estimates of the income elasticity are around 0.6­0.7, varying somewhat with the exact definition of total expenditure used. Health care in total is thus found to be a normal good, but not a luxury good. If total health care expenditure is a small part of the household budget, en- dogeneity of the total expenditure may be ignored. However, this assump- tion is easy to relax. Instrumental variable estimates are given in the last two columns of table 11.14. The income elasticity still remains around 0.6. Additional robustness checks do not indicate that this estimate varies significantly by urban or rural location or by level of income. This point estimate of income elasticity is larger than the corresponding estimate for individual expenditures. One possible explanation is that the estimate from aggregate household data may be upward biased because of the failure to take into account the effect of HLTHINS, which, as was seen earlier, is positively correlated with the household income level. That is, the role of Table 11.14. Models for Positive Medical Expenditure for Households Random effects Fixed effects Robust Ordinary least Ordinary least Ordinary least Instrumental squares, robust squares squares variable Variable Coeff. Std. err. Coeff. Std. err. Coeff. Std. err. Coeff. Std. err. Constant -0.8852 0.3806 -0.3008 0.4163 0.1103 0.4568 0.0281 0.5573 log(INC) 0.7340 0.0436 0.6542 0.0479 0.6000 0.0519 0.6210 0.0692 SEX 0.1189 0.0531 0.1172 0.0513 0.1091 0.0519 0.0760 0.0517 AGE 0.0100 0.0017 0.0115 0.0016 0.0116 0.0016 0.0105 0.0016 EDUC -0.0296 0.0127 -0.0628 0.0129 -0.0519 0.0135 -0.0922 0.0140 URBAN -0.3024 0.0595 -0.2648 0.0917 n.a. n.a. n.a. n.a. HHSIZE 0.0163 0.0135 0.0111 0.0136 0.0112 0.0141 0.0360 0.0160 R2 0.094 N 5,006 5,006 5,006 5,006 n.a. Not applicable. Note: The variable log(INC) has been "instrumented" using the urban, farm, age group, edu- cational attainment, and province instruments. Definition of variables: AGE = age in years. EDUC = completed years of education. HHSIZE = household size. INC = household income proxied by total nominal household expenditure. log(INC) = log(total nominal household ex- penditure). SEX = 0/1 dummy for gender. URBAN = 0/1 dummy for urban household. Source: Author's calculations. 418 Economic Growth, Poverty, and Household Welfare in Vietnam insurance has been absorbed into the income elasticity, causing it to become somewhat inflated. A second possibility is simply that demand for health care at the level of the household is indeed more elastic than it is for an indi- vidual. Age and sex of the head of the household are also significant factors. On average, households with female heads spend more on health care, and households with older heads also spend more. Educational level of the head of the household is not found to be a significant explanator. Discussion of Policy Issues Three health care policy issues are discussed in this section: the implications of the pervasive phenomenon of self-medication, the future of the CHCs, and the expected future changes in the pattern of health care use. This chapter has documented the pervasiveness of self-medication and the factors that promote it. The phenomenon is a common one in most de- veloping countries. The World Health Organization explicitly recognizes that self-medication has an important role to play in most health care sys- tems and has observed that with the continued improvement in people's education, general knowledge, and socioeconomic status, self-medication has been successfully integrated into many health care systems around the world. In a developing economy like Vietnam, where both public and private health care infrastructures are relatively basic, self-medication is an important form of health care. Freer availability of drugs contributes to household welfare. Worldwide, the purchase of prescription-only drugs without a prescription is far more common than the sale of over-the-counter drugs.8 Opportunities for self-medication are enhanced and aided by the In- ternet and deregulation of over-the-counter sales of pharmaceutical prod- ucts. These tendencies have harmful consequences, of which the growing in- effectiveness of antibiotic drugs is the most alarming and most visible. When small doses are used to treat bacterial infections, presumably because the user cannot afford the cost of the prescribed full course, the practice pro- motes the growth of antibiotic-resistant strains of bacteria instead of clearing the infection. This reduces the future potency of antibiotics for all users. Pre- vious World Bank reports reflect the concern expressed by the medical pro- fession and public health organizations. Inadequately supervised and ad- ministered drugs, often in incorrect dosages, are a major contributing cause of the growth of antibiotic-resistant bacteria. The problem is a serious concern because it involves a negative intertemporal externality--current actions of an individual have a negative future impact on the society as a whole. The public health issue is how to combat this problem. This chapter has found econometric evidence that the practice of self- medication is negatively associated with educational attainment, income lev- els, and the relative price and accessibility of alternative providers. There is also strong a priori reason to believe that supply-side deregulation has lowered the price of, and improved access to, drugs and thereby encouraged Patterns of Health Care Use in Vietnam 419 self-medication. On the demand side, we can expect that economic growth will reduce the seriousness of the problem. But this view needs to be quali- fied: Although it appears that in the highest income quartile, self-medication is an inferior good, any reduction from this source due to income growth will make only a small contribution to net reduction. The larger positive contri- bution from lower-income groups may outweigh the negative tendencies. Moreover, economic growth has been more rapid in urban than in rural areas, so the reduction in harmful types of self-medication in rural areas is likely to be even slower. Moreover, if affordable alternatives to self-medication exist, then its use is more likely to decline as awareness of its dangers increases. Again, such alternatives are less accessible in rural areas. These arguments suggest that in the absence of regulatory constraints, the practice of self- medication is likely to persist. A discussion of the appropriate form of regu- lation is outside the scope of this chapter. Previous World Bank analyses, in expressing concern about the extent and continued growth of (harmful) self-medication in Vietnam (Gertler and Litvack 1998; World Bank 1999), have suggested that this growth is in part a consequence of the low quality of care available to the lower-income groups. Various policy prescriptions have been put forward. Gertler and Litvack (1998, pp. 246­47) suggest that improving the quality of service at the CHCs--for example, by improving the supply of low-priced generic drugs--would reduce self-medication. The VHSR (World Bank 1999, pp. 120­21) also mentions the proliferation of counterfeit and substandard drugs available on the market. It goes on to describe the recently initiated Vietnam national drug policy also aimed at rational and safe use of drugs. The results in this chapter suggest that an expansion of the voluntary health insurance program would also have a qualitatively similar effect. It appears that enrollment in the voluntary health insurance program has stalled, al- though the reasons for such stagnation are unclear. However, results here indicate that enrollment in the insurance program is responsive to income growth. Again, this suggests that continued growth will reduce the problem. However, the poorer rural sections of the population will benefit from this development more slowly and to a lesser extent. This chapter finds evidence that strongly suggests that CHCs provide an inferior service whose consumption declines with income and education. The analysis here is not sufficiently detailed to pinpoint the CHC character- istics that are responsible for their decline, but it is plausible that the readily available alternative of self-medication is partly responsible. Note that it has been suggested that CHCs fail to carry adequate stocks of cheap generic drugs and are generally poorly staffed and equipped. Given the continua- tion of the present trends, the CHCs may become even less important in the future. A factor that may counter this trend concerns the eligibility of CHCs to provide treatment to those with health insurance. The data in this chapter pertain to the period in which the insured were required to obtain treatment at government hospitals. If, however, this were to change, then the use of the CHCs, especially in rural areas, may increase. 420 Economic Growth, Poverty, and Household Welfare in Vietnam The growth of PHFs is a relatively recent phenomenon. Results in this chapter do not suggest that income growth has a strong impact on the growth of this sector. It seems likely that this sector provides services to those who desire a higher quality of care than that available outside public hospitals, but who are also either ineligible or unable to get treated at public hospitals. Evidence indicates that higher-income individuals with insurance get treatment and drugs at public hospitals. Currently, public hospitals seem to be the main and perhaps the only source of quality care. Increased de- mand for health care of higher quality will come with economic growth, es- pecially in urban areas. As long as the catchment area of the private health facilities is restricted directly or indirectly, future economic growth will put greater pressure on public hospitals unless other alternatives superior to those currently available can be found. If, however, the insured individuals can be treated at private health facilities, there will be an alternative to pub- lic hospital care. This will reduce pressure on public hospitals that is bound to arise if income growth continues to be robust. Whether this will raise the average quality of care, or reduce rampant self-medication, would seem to depend upon the regulatory constraints that apply to the private health facilities. Summary and Conclusions A stylized pattern of health care use in Vietnam has emerged from previous analyses of 1993 VLSS data. Broadly, according to this stylized view, the better- off sections of the population get their health care at public hospitals, the poorer at the commune health centers, and all groups use self-medication heavily, causing self-medication to dominate as the principal source of health care. In this picture, other providers such as private health facilities, tradi- tional medicine practitioners, home providers, and so forth play a relatively minor role. The stylized description has little to say about the impact of health insurance on the observed health care use pattern. This chapter provides some confirmations, as well as modifications, of that stylized description: · Evidence supports the view that those who are either ill or injured have ready access to some form of health care. At all levels of income, the most common response to the need for health care is some form of self-medication. · The private providers collectively are more important as a fraction of total health care spending than the commune health centers. · The results suggest that both self-medication and commune health centers are inferior goods by the usual definition, because their de- mand declines with rising household incomes. Self-medication ap- pears to be an inferior good, especially for high-income households, but a normal good at low-income levels. In the aggregate, however, the net effect of income on pharmacy visits is estimated to be close to Patterns of Health Care Use in Vietnam 421 zero. These results are consistent with the view that both self- medication and commune health centers are low-quality and risky forms of health care in Vietnam. · Within the existing income distribution, estimates show a negative relation between the use probability of commune health centers and income. This negative relation is less robust in the lowest income quartile. · Health care provided by private health facilities is weakly related to income overall but may be positively related for the lower-income groups. · There is a strong positive relation between income and the use of both inpatient and outpatient care provided at public hospitals. · The net impact of health insurance on self-medication and the use of private health facilities is negative. That is, under the current organi- zation of health care delivery, having health insurance diverts pa- tients from private health care and self-medication mainly toward public hospitals and, to a lesser extent, toward commune health cen- ters. The growth of services in the private health facilities is therefore affected in opposite directions by rising income and the rising pro- portion of the insured population. · Income, insurance status, and age are the three major determinants of inpatient care (hospital nights) in public hospitals. · There appears to be no evidence that health insurance has had a sig- nificant impact on the total household out-of-pocket health care ex- penditures (excluding inpatient care at public hospitals) in either direction. That is, much of the impact seems to be in the form of redistribution of care between types of providers. · Previous analyses have expressed serious concern about the dominant role of unsupervised and unregulated self-medication in Vietnam. It has been suggested that this is made possible by easy availability of a very wide range of pharmaceutical drugs. Both rising incomes and growth of health insurance reduce the extent of self- medication, but it is not clear whether these deterrents are strong enough. · Aggregate household income elasticity for health care is higher than indicated by previous studies. There are some qualifications. This chapter has treated, with some justi- fication, both health insurance and household incomes as weakly exogenous variables. No allowance has been made for measurement error in income. Standard statistical arguments suggest that this may cause underestimation of the impact of income on the demand for health care. Second, the chapter distinguished between different types of health insurance only indirectly, and this may have caused some aggregation bias of an indeterminate nature. Although a statistical model of the probability of enrollment in the health insurance program was provided, because of lack of data, no 422 Economic Growth, Poverty, and Household Welfare in Vietnam estimates of price sensitivity of insurance demand could be provided. This is an important qualification to the finding that enrollment in the insurance program is strongly associated with income. Finally, it is useful to review the main policy implications of the findings. First, the results suggest several avenues for reducing the heavy (and often inappropriate) reliance of the Vietnamese on over-the-counter antibiotics. Increases in household income and education levels should reduce this re- liance over the long term. In addition, expansion of the new voluntary health insurance program would reduce reliance on self-medication with antibiotics and other drugs. Second, there is clear evidence of dissatisfaction with health services provided by commune health centers. People switch to other sources of medical care as their income rises. Further research is needed on the source of this dissatisfaction. Among the possible sources are inadequate stocks of medicine and inadequate staff. If no changes are made, commune health centers will become an even less important source of health care for the Vietnamese population. Finally, the role of private health care providers in Vietnam needs further development. In the long run, they can provide high-quality hospital care unless restrictions prevent them from doing so. This would serve as an additional source of such care for individ- uals in Vietnam's voluntary health insurance program, and as that program expands, the capacity to meet the demand for such health care must be ex- panded as well. Notes 1. The 1992­93 survey spanned a full year, starting in October 1992; the 1997­98 survey began in December 1997 and also lasted a year. For brevity's sake, reference is made to the surveys as the 1993 VLSS and 1998 VLSS, respectively. 2. There is a suspicion, based on information collected in the field, that the re- sponses to the question about recent illness or injury may be biased. The bias is thought to arise from the greater propensity to report illness or injury of those in higher socioeconomic categories. 3. By combining datasets, it may be feasible in the future to impute the type of insurance of each individual and thereby make it possible to study the differential impact of each on use measures. 4. Continuous probability distributions cannot allow for nontrivial probability mass at zero frequency. 5. A qualification is that the proportion of zeros may be excessive relative to the count model specified, thereby requiring use of as flexible a count data model as fea- sible (see Cameron and Trivedi [1998]). 6. For example, this is possible in the computer program Stata 6.0 for a variety of estimators. 7. See Cameron and Trivedi (1998, chapters 3 and 9) for a detailed discussion of the count data models used here. 8. WHO Drug Information (World Health Organization 2000). This article cites a consumer interview study carried out in six Latin American countries that found that only 34 percent of the dispensed medicines were classified as over-the-counter drugs. Patterns of Health Care Use in Vietnam 423 Bibliography The word processed describes informally reproduced works that may not be commonly available through libraries. Behrman, J. R., and J. C. Knowles. 2000a. "The Demand for Health Insurance in Vietnam: An Application of Contingent Valuation." Working paper. University of Pennsylvania, Philadelphia. ----. 2000b. "The Demand for Student Health Insurance in Vietnam." Work- ing paper. University of Pennsylvania, Philadelphia. Cameron, A. C., and P. K. Trivedi. 1998. Regression Analysis of Count Data. Econometric Society Monograph 30. New York: Cambridge Univer- sity Press. Cameron, A .C., P. K. Trivedi, F. Milne, and J. Piggot. 1988. "A Microecono- metric Model of the Demand for Health Care and Health Insurance in Australia." Review of Economic Studies 55(1): 85­106. Chang, F.-R., and P. K. Trivedi. 2003. "Economics of Self-Medication: Theory and Evidence." Health Economics 12(9): 721­39. Deaton, A. 1997. The Analysis of Household Surveys. Baltimore: Johns Hopkins University Press. Dollar, D., P. Glewwe, and J. Litvack, eds. 1998. Household Welfare and Vietnam's Transition. World Bank Regional and Sectoral Study. Washington, D.C.: The World Bank. Gertler, P., and J. Litvack. 1998. "Access to Health Care during Transition: The Role of the Private Sector in Vietnam." In D. Dollar P. Glewwe, and J. Litvack , eds., Household Welfare and Vietnam's Transition. World Bank Regional and Sectoral Study. Washington, D.C.: The World Bank. Gertler, P., J. Litvack, and N. Prescott. 1996. "Health Care during Transition: The Role of the Private Sector in Viet Nam." World Bank, Washington, D.C. Processed. Hsiao, C. 1986. Analysis of Panel Data. Cambridge: Cambridge University Press. Viet Nam Living Standards Survey (VNLSS). 1992­1993. Web site: www. worldbank.org/lsms/country/vn93/vn93bid.pdf. Vietnam Living Standards Survey (VLSS). 1997­98. Web site: www. worldbank.org/lsms/country/vn98/vn98bif.pdf. World Bank. 1995. Vietnam: Poverty Assessment and Strategy. Report of the World Bank Country Operations Division, Country Department 1. East Asia and Pacific Region. Washington, D.C. ----. 1999. Vietnam Health Sector Review. Washington, D.C. World Health Organization. 2000. WHO Drug Information. 14(1): 1­2. 12 Trends in the Education Sector Nga Nguyet Nguyen In the 1990s, Vietnam achieved remarkably high economic growth. At the same time, it experienced steady improvement in many social indicators, which were already high despite the country's low per capita income. As ex- plained in chapter 1 of this volume, during the past decade the country has seen a huge reduction in poverty incidence, which occurred for all income groups and in all regions. Over the period from 1993 to 1998, school enroll- ment rates increased substantially, but the increases were not spread equally across different income groups, regions, and ethnic groups. The government of Vietnam is embarking on a very challenging agenda to provide higher- quality basic education for all. Much more remains to be done to ensure that this ambitious, but not impossible, goal will be achieved. The purpose of this chapter is to document and analyze changes in en- rollment and financing for education in Vietnam over the period from 1993 to 1998, using the Vietnam Living Standards Survey (VLSS) data collected in 1992­93 and 1997­98.1 The first issue discussed in the chapter is recent change in school enrollment, focusing on primary and lower secondary ed- ucation because universal enrollment at these levels is an explicit goal of the Vietnamese government. Enrollment rates have increased significantly in the past five years at all levels of education. To see whether these enrollment rate increases have been evenly spread across the population, as opposed to being concentrated in certain groups or regions, the chapter compares changes in enrollments for all expenditure groups, all regions, all ethnic groups, and boys and girls. Second, the chapter examines education finance, including the relative role of the public and private sectors in financing education in Vietnam. Pri- vate spending on education is compared by expenditure groups, regions, ethnic groups, and gender to assess any changes in the financial burden of sending children to school, especially among the poor. This chapter investi- gates possible links between household spending on education, enrollment, and the quality of education, and it checks whether poor households benefit 425 426 Economic Growth, Poverty, and Household Welfare in Vietnam as much as better-off households from public spending on education. Does government spending reduce inequalities? Can any unequal patterns be cor- rected by reallocating the government's education budget? The final issue considered is whether there have been changes in rates of returns to education, including whether any changes have varied across different groups. This is done by estimating Mincerian earning functions that link wage income to the underlying determinants of earnings, such as schooling, work experience, age, economic sector, region, and gender. The structure of the chapter follows the three issues just mentioned. The section that follows this brief background documents and analyzes changes in enrollment. This is followed by a discussion of the pattern of financing for education and the relative role of the public and private sectors in financing education in Vietnam. The section titled "Returns to Education" looks at the recent changes in rate of returns to education with limited focus on wage earners in the private sector. Finally, the chapter concludes with some policy suggestions. Enrollment This section documents and analyzes changes in enrollment in Vietnam using the 1993 and 1998 VLSS data. It discusses recent changes in school enrollment, focusing on primary and lower secondary education because universal enrollment at these levels is an explicit goal of the Vietnamese government. Enrollment rates have increased significantly in the past five years at all levels of education. To see whether these enrollment rate increases have been evenly spread across the population, as opposed to being concentrated in certain groups or regions, the section compares changes in enrollments for all expenditure groups, all regions, all ethnic groups, and boys and girls. General Trends Vietnam made impressive progress in expanding enrollment in basic educa- tion in the 1990s. Figure 12.1 shows total student enrollment at the primary and secondary levels for every year from 1987 to 2000, based on official gov- ernment education statistics. Vietnam has achieved high rates of literacy and school enrollment despite its low per capita income, while maintaining good social indicators (infant and under-five mortality rates, life expectancy, fertility rate, child nutrition, and access to basic services) compared with other countries with similarly low per capita income. The enrollment statistics shown in figure 12.1 were collected by Vietnam's Ministry of Education and Training through its statistical reporting system. Absolute numbers of students attending primary school increased dramati- cally between the early 1990s and 1998, rising by 7 percent from 9.7 million in 1993 to 10.4 million in 1998. From 1998, a downward trend due to Trends in the Education Sector 427 Figure 12.1. Trends in School Enrollment, 1987­2000 Enrollment (millions) 12 10 8 6 4 2 0 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 Year Primary Lower secondary Upper secondary Source: Ministry of Education and Training (2001). demographic change was observed. The downward trend in enrollments in the years before 1990 was reversed by the mid-1990s at the secondary levels. At the lower secondary level, enrollment increased even faster, by almost 43 percent, from 3.0 million to 5.3 million in 1998. Upper secondary school enrollment has risen even more dramatically compared with 1993, from 0.6 million to 1.4 million, and has far surpassed the peak in 1987. Trends by Expenditure Group Education progress for different expenditure groups can be assessed by looking at the number of children actually in school and whether children are in school at the appropriate age.2 From 1993 to 1998, net enrollments increased significantly at all levels, especially at higher general education levels. Net enrollments rates (NERs) for primary schools increased from 87 percent in 1993 to 91 percent in 1998. At the lower secondary level, NERs more than doubled in five years, from 30 percent in 1993 to 62 percent in 1998. Increases in NERs at higher levels were even more impressive; they more than quadrupled at the upper secondary level and tripled at the post- secondary level (table 12.1). 428 Economic Growth, Poverty, and Household Welfare in Vietnam Table 12.1. Net and Gross School Enrollment Rates, by Quintile and Education Level (percent) Lower Upper Enrollment Primary secondary secondary Postsecondary rate/indicator 1993 1998 1993 1998 1993 1998 1993 1998 Net Vietnam 87 91 30 62 7 29 3 9 Poorest quintile 72 82 12 34 1 5 0 0.4 Most well-off quintile 96 96 55 91 21 64 9 29 Gross Vietnam 120 115 42 78 9 36 4 12 Poorest quintile 100 112 15 47 1 8 0 0.4 Most well-off quintile 130 104 77 105 24 75 13 37 Source: Author's estimates based on the 1993 and 1998 VLSSs. Table 12.1 shows that, for the poorest quintile, enrollments expanded at all education levels during the period from 1993 to 1998. Net enrollment of poor children increased in primary schools from 72 percent in 1993 to 82 percent in 1998, and it increased in lower secondary schools from 12 percent in 1993 to 34 percent in 1998. At higher levels, although enrollment increased, very few children from the poorest households went to upper secondary and postsecondary schools (figure 12.2). Enrollment in primary education became more equitable in the 1990s, but inequities still exist. In 1998, 82 percent of poor children of ages 6 to 10 years were in primary school, compared with 96 percent for the most well- off children. Moreover, gaps in enrollment in secondary and postsecondary education between the better-off and the poor are still large in Vietnam, with the NER for the poorest quintile only about one-third of that for the most well-off quintile at lower secondary schools. In 1998, 91 percent of children of ages 11 to 14 years from the most well-off quintile attended lower sec- ondary schools, compared with only 34 percent of children from the poorest quintile. Although more children from the poorest quintile went to upper secondary and postsecondary schools in 1998, enrollment rates for this group at these higher education levels were extremely low compared with those from the highest expenditure groups. Sixty-four percent of the most well-off children of ages 15 to 17 years attended upper secondary schools, in contrast with only 5 percent of children from the poorest quintile. The gap in enroll- ment in postsecondary schools is even larger: The NER was 29 percent for the most well-off children, compared with 0.4 percent for the poorest chil- dren (figure 12.2 and table 12.1). 1998 1998 secondary Quintile Quintile 1993 1993 1998 Postsecondary Lower and 1993 12345 12345 Level, and Percent 100 80 60 40 20 0 Percent 100 80 60 40 20 Quintile by VLSSs. 1998 Rates, 45 and 1998 1998 1993 the Enrollment Quintile Quintile omfr 1993 secondary 1993 Net Primary Upper School calculationss' 123 12345 12.2. Author ce: Figure Percent 100 80 60 40 20 0 Percent 100 80 60 40 20 00 Sour 429 430 Economic Growth, Poverty, and Household Welfare in Vietnam Table 12.2. Repetition Rate by Level of Education, 1998 (number of students who repeated classes as percent of total students, for each level) Poorest Most well-off Level quintile quintile Vietnam Primary schools 24.6 8.0 19.2 Lower secondary 28.6 14.0 21.8 Upper secondary 26.0 12.4 16.0 Source: Author's estimates based on the 1998 VLSS. In Vietnam and other developing countries, many children start the first grade of primary school at a later age than normal, and many children re- peat grades. This can cause the number of children enrolled at a certain level to be higher than the number of children in the age range associated with that level. At the primary school level, over the 1993­98 period, NERs in- creased while gross enrollment rates (GERs) declined for all expenditure groups except the poorest group (table 12.1). This shows that Vietnam is moving from the first to the second phase of education development: Chil- dren are increasingly starting school at the right age, and fewer children are repeating grades. However, the high GERs for the poorest quintile in 1998 imply that either their repetition rates were still significant and much higher than those for other expenditure groups, or that many children were starting primary school at a late age (or perhaps were leaving school for a few years and then returning). In 1998, about 25 percent of the poorest primary school children repeated at least one grade, compared with only 8 percent of the most well- off children (table 12.2). Furthermore, table 12.1 also implies that the poorest quintile had the lowest rate of continuation over the 1993­98 period. Fewer than 50 percent of children from the poorest quintile continued into lower secondary school in 1998, but about 94 percent of primary school children from the most well-off quintile in 1993 moved to lower secondary schools in 1998. As can be seen in table 12.3, despite high rates of repeating and delayed enrollment in the first grade, over 1993­98 there was a significant improve- ment in age-grade matching. In 1993, about 23 percent of students enrolled in primary schools were older than 10 years, and almost 28 percent of students enrolled in lower secondary schools were older than 14 years. The proportions for 1998 were 19 and 20 percent, respectively. All these trends have important implications for Vietnam. Although enrollment in basic education (defined as primary and lower secondary education) has been improved for poor children, inequalities persist at the primary level, and the gaps are widening at higher levels. This suggests that poor children face disadvantaged conditions more often than do better-off children, such as poorer quality of education, lack of sufficient time spent Trends in the Education Sector 431 Table 12.3. Age and Grade Matching, 1993 and 1998 (percentage of total enrolled in level) Age group Date/level < 6 6­10 11­14 15­17 18­24 >24 Total 1993 Primary 4.8 72.0 22.1 1.0 0.1 0 100 Lower secondary 0 1.0 71.3 25.9 1.8 0 100 Upper secondary 0 0 1.4 83.7 13.5 1.4 100 Postsecondary 0 0 0 17.1 67.5 15.4 100 1998 Primary 1.2 79.7 17.9 1.0 0.2 0 100 Lower secondary 0 1.2 79.2 18.7 0.8 0 100 Upper secondary 0 0 0.7 79.1 19.9 0.3 100 Postsecondary 0 0 0.2 4.2 79.7 16.0 100 Note: Rows may not add to 100 due to errors introduced by rounding. Source: Author's estimates based on the 1993 and 1998 VLSSs. studying because of housework or income-earning activities, and poor health. Gender Differences Vietnam has almost achieved gender equality of enrollment in general edu- cation (including primary and lower and upper secondary schools). Testing the hypothesis that girls and boys had the same NERs showed that this equality is statistically significant. In 1998, 91 percent of school-age girls were enrolled in primary school, compared with 92 percent of boys. Gender inequality is not found at the lower secondary level (NERs in 1998 were 33.5 percent for boys and 33.7 percent for girls), although a gap between girls' and boys' net enrollment at the upper secondary level is larger and in- creased slightly from 2 percentage points in 1993 to 3 percentage points in 1998 (figure 12.3). Table 12.4 shows both GERs and NERs, for 1993 and 1998. By 1998, using NERs showed that Vietnam had achieved a quite equal enrollment in basic education for both girls and boys, although the gender gap was slightly larger at the upper secondary level. It is interesting that if the GER is used instead of the NER, the whole picture changes significantly. GERs for girls were lower than those for boys at all levels--and much lower at the upper secondary level. For example, at the primary level, the girls' NER was 99 per- cent of that of boys, but the girls' GER was only 93 percent of the boys' GER. The difference is clearer at the upper secondary level. This implies that school-age boys and girls had equal opportunity to attend basic education. Boys repeated classes more often than girls, which is one of the reasons for a lower percentage of GER for girls compared with that for boys. However, it 432 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 12.3. School Net Enrollment Rate, by Gender and Education Level, 1993 and 1998 Percent 100 90 80 70 60 50 40 30 20 10 0 Primary Lower Upper Primary Lower Upper secondary secondary secondary secondary Education level Boys Girls Source: Author's calculations from the 1993 and 1998 VLSSs. Table 12.4. Girls' Enrollment Rates as a Percentage of Boys' Enrollment Rates, by Education Level, 1993 and 1998 Enrollment rate/level 1993 1998 Net Primary education 101 99 Lower secondary education 93 101 Upper secondary education 73 91 Gross Primary education 96 93 Lower secondary education 80 96 Upper secondary education 62 82 Source: Author's estimates based on the 1993 and 1998 VLSSs. may also imply that overaged girls were less likely to stay or go back to school than overaged boys. This is somewhat consistent with findings from participatory poverty assessments done in Vietnam--that girls contend with a complex mix of barriers to their right to formal education. For example, Trends in the Education Sector 433 when a household's decision to send children to school is weighted against labor contributions, girls are often the last to be sent to school and the first to be withdrawn. Similarly, some parents still consider that an investment in a daughter's education, especially at higher levels, will be lost when she marries into another family. Gender equality can be achieved and main- tained only by the conscious integration of gender issues within all policies, strategies, and practices focused on addressing educational disadvantage, especially at higher levels of education. Ethnic Groups' Differences Despite recent improvements, ethnic minorities experience a much lower level of net enrollment at all levels of education compared with the majority ethnic Vietnamese (Kinh) and Chinese (figure 12.4). In 1998, only 82 percent of primary school-age children from ethnic minorities were enrolled in pri- mary school, compared with 93 percent of majority children. At the lower secondary level, the disparity is much more pronounced: In the same year, the lower secondary NER was only 37 percent for minority children, com- pared with 66 percent for the majority. Low participation rates mean that Figure 12.4. School Net Enrollment Rate, by Ethnicity and Level of Education, 1993 and 1998 Percent 100 90 80 1998 70 60 1993 50 40 30 20 10 0 Primary Lower Upper Primary Lower Upper secondary secondary secondary secondary Education level Kinh and Chinese Ethnic Source: Author's calculations from the 1993 and 1998 VLSSs. 434 Economic Growth, Poverty, and Household Welfare in Vietnam approximately 0.4 million of the 2.8 million minority children ages 6 to 14 years were not enrolled in school. Of these, 0.3 million were ethnic girls. Many ethnic minority girls are disadvantaged by a combination of late en- rollment in first grade (if they enroll at all) and a tradition of early marriage. Minority people's concentration in the Northern Uplands, the Central Highlands, and the Mekong Delta means that they also experience the disadvantages found more generally in these locations. Fortunately, compared with the Kinh and Chinese children, ethnic mi- nority children experienced a much larger improvement in enrollments in primary and lower secondary schools from 1993 to 1998, given their low starting points in 1993. The higher the level of education, the larger the gap in school enrollment among different socioeconomic groups. Over this pe- riod, the enrollment rates of ethnic minorities increased by 18 percentage points at the primary level and 29 percentage points at the lower secondary level, compared with increases of 2 percentage points at the primary level and 22 percentage points at the lower secondary level for the Kinh and Chinese. The Kinh and the Chinese enjoyed larger improvement in enroll- ment at the upper secondary and postsecondary levels, however. Over the same period, the increase in enrollment was 24 percentage points at the upper secondary level and 7 percentage points at the postsecondary level for the majority, compared with 6 percentage points at the upper secondary level and 0.6 percentage point at the postsecondary level for the ethnic mi- norities (figure 12.4). Regional Differences Despite overall improvement, enrollment gaps between urban and rural areas in Vietnam are still considerable, and the higher the level of education, the larger the enrollment gap between urban and rural areas. Enrollment gaps between urban and rural areas in 1998 were smaller than those in 1993 at the primary and lower secondary levels. For rural children, enrollment in primary education was improved significantly, from 85 percent in 1993 to 91 percent in 1998 (table 12.5). Improvement in NERs at higher levels was even more impressive: NERs more than doubled at the lower secondary level and improved fourfold at the upper secondary level, compared with slightly less than double at the lower secondary level and threefold at the upper secondary level for urban children. However, the increase in enroll- ment rates at the upper secondary level over the same period was higher in urban areas than in rural areas, increasing the gap at this level. Specifically, the urban-rural enrollment gap at the upper secondary level increased from 12 percentage points in 1993 to 32 percentage points in 1998 (table 12.5). Table 12.5 shows clear regional disparities at all levels of education, with the Mekong Delta and the Central Highlands below the average national NER in 1998. Surprisingly, the Northern Uplands had a very high enroll- ment level at primary school, behind only the Red River Delta, despite the relatively low living standards in the Northern Uplands. This achievement Trends in the Education Sector 435 Table 12.5. Net Enrollment Rates by Location, 1993 and 1998 (percent) Primary Lower secondary Upper secondary Location 1993 1998 1993 1998 1993 1998 Vietnam 86.7 91.4 30.1 61.7 7.2 28.6 Urban-rural Urban 96.6 95.5 48.5 80.3 17.3 54.5 Rural 84.8 90.6 26.3 57.9 4.7 22.6 Region Northern Uplands 85.7 94.1 22.2 56.4 5.6 22.2 Red River Delta 95.1 95.6 46.5 83.2 10.3 45.2 North Central Coast 88.1 92.5 30.6 62.3 5.8 29.6 Central Coast 84.4 88.1 38.1 64.2 11.6 31.8 Central Highlands 67.8 80.2 15.0 43.9 2.0 10.5 Southeast 93.5 93.4 35.2 71.7 9.7 36.3 Mekong Delta 79.0 86.9 19.4 45.0 3.6 17.4 Source: Author's estimates based on the 1993 and 1998 VLSSs. may be a result of the sizable educational assistance that this region has received in recent years from both the government and external aid to pro- mote universal primary education. However, at the secondary education level, this region was still among the worst-performing regions. The Red River Delta and the Southeast had the highest NERs at all three levels of education. Despite the enrollment improvement for the whole country at all levels, enrollment improvement was not evenly spread across all regions (table 12.5). The enrollment gap between the best- and the worst-performing regions over 1993­98 was smaller at the primary level, with the gap reduced from 27 percentage points in 1993 to 15 percentage points in 1998. However, the gaps were wider at secondary levels, with enrollment disparity increasing from 32 percentage points in 1993 to 39 percentage points in 1998 at the lower secondary level. It is worrisome that there was a huge increase in the enroll- ment gap at the upper secondary level, from only 8 percentage points in 1993 to 35 percentage points in 1998. This widening gap at the upper secondary level was due to considerable increased enrollment for the Red River Delta and relatively slight increased enrollment for the Central Highlands; it was also due to a low starting point for the Central Highlands (only 2 percent). Financing for Education This section examines education finance, including the relative role of the public and private sectors in financing education in Vietnam. Private spending on education is compared by expenditure groups, regions, ethnic groups, and gender to assess any changes in the financial burden of sending 436 Economic Growth, Poverty, and Household Welfare in Vietnam children to school, especially among the poor. This section investigates pos- sible links between household spending on education, enrollment, and the quality of education, and checks whether poor households benefit as much as better-off households from public spending on education. Does govern- ment spending reduce inequalities? Can any unequal patterns be corrected by reallocating the government's education budget? Public and Private Role in Providing and Financing Education The public sector in Vietnam still dominates the provision of education. Even though the share of the private sector in providing school places in 1998 increased at the lower and upper secondary levels compared with 1993, provision of education is still dominated by the public sector. Table 12.6 shows that the private sector provided more school places, and its share at the upper secondary level increased from 2 percent in 1993 to 5 percent in 1998. Surprisingly, at the primary level, the private share in provision of service fell from about 1 percent to 0.3 percent, and the semipublic school (joint between public and private) share fell from 2 percent to 0.4 percent. One possible explanation is that public schools expanded quickly as a result of government efforts to achieve universal primary education. Moreover, there seems to be no space limit in public primary school because no admis- sion examination is required, so the demand for private primary school is not as high as that for secondary levels. Table 12.6. Public and Private Shares of School Enrollment, by Education Level, 1993 and 1998 (percent) Share of total enrollments Type of school 1993 1998 Public Primary 97.3 99.3 Lower secondary 97.6 96.6 Upper secondary 95.7 83.0 Private Primary 0.9 0.3 Lower secondary 1.7 0.5 Upper secondary 2.1 4.9 Semipublic Primary 1.9 0.4 Lower secondary 0.7 2.9 Upper secondary 2.1 12.6 Source: Author's estimates based on the 1993 and 1998 VLSSs and public expenditure data from the Ministry of Finance (1998). Trends in the Education Sector 437 In secondary schools, the most obvious explanation for the increased role of the private sector is the limited capacity of the public sector to pro- vide education at that level. Demand for private and semipublic education at the secondary level has increased, because many more children who want to attend secondary schools cannot gain admission to public schools be- cause the space available is limited. Admission to lower and upper sec- ondary school used to be governed by admission examination, though this practice was recently abolished by the government. Instead of an admission examination, performance in the previous level of education is used. But criteria vary considerably between geographic locations and schools. How- ever, semipublic lower secondary schools developed faster than private lower secondary schools, perhaps because they are a better investment than their purely private counterparts. Because the public sector cannot meet the increasing demand for recur- rent spending on schools in Vietnam, many schools were made semipublic; they continue using public infrastructure (classrooms and facilities), but self-finance for most of their increased recurrent expenditure (such as salaries for new-hire teachers and increasing salaries for experienced teach- ers). As a result, at the lower secondary level, the private provision of edu- cation fell from 1.7 percent to 0.5 percent, and the semipublic provision increased from 0.7 percent to 2.9 percent. Private schools at the upper sec- ondary level experienced a considerable increase, from 2 percent in 1993 to 5 percent in 1998, compared with a huge increase for semipublic schools, from 2 percent in 1993 to 13 percent in 1998 (table 12.6). The overall state budget expenditure for education and training (includ- ing funding from both central and local government levels) increased from about 10 percent of current expenditures in 1993 to 17 percent in 1998 (see table 12.7). This implies that spending on education by the government increased from 1.8 percent of gross domestic product (GDP) in 1992 to 3.5 percent in 1998. Compared with 1993, budgetary spending on primary and secondary education and training has increased from 61 percent of total public spending in 1993 to 76 percent in 1998, at the expense of spending on higher education and training. In 1998, the largest share of the state budget for the education and training sector, 36 percent, was allocated to primary education, an increase from 33 percent in 1993. The share of the budget expenditure on education for lower secondary schools in 1998 was the same as that in 1993--18 percent, and the budget share for upper secondary edu- cation increased slightly from 7 percent in 1993 to a little more than 8 percent in 1998. Government spending on higher education and training dropped from 39 percent in 1993 to 24 percent in 1998. Most of the reduction in the budget for higher education and training came from reduction in the state budget for technical and vocational training, not from reductions in spend- ing on universities. Despite an overall increase in public spending on education, there is a large variation in the amount of financial resources available at the local level. Provincial governments have a considerable responsibility in generating 438 Economic Growth, Poverty, and Household Welfare in Vietnam Table 12.7. Trends in Government Expenditure on Education, 1992­98 Indicator 1992 1993 1994 1995 1996 1997 1998 Per capita expenditure (thousand dong at 1994 prices) 39.2 58.1 71.3 82.8 80.4 100.0 115.0 Percentage of discretionary budget 9.3 10.0 11.7 13.4 12.3 14.1 17.4 Percentage of gross domestic product 1.8 2.6 2.9 3.0 2.7 3.2 3.5 Share of total public education expenditure (percent) On education 63.9 60.7 60.4 72.3 71.9 72.6 75.7 Primary school 40.0 32.6 29.6 29.1 33.2 34.3 36.4 Lower secondary school 14.8 17.6 17.7 19.9 19.2 19.2 17.6 Upper secondary school 6.0 6.9 8.6 7.4 9.0 8.6 8.3 On higher education and training 36.1 39.3 39.6 27.7 28.1 27.4 24.3 Source: Author's estimates based on public expenditure data from the Ministry of Finance (1998). resources for preuniversity education and enjoy discretion in spending alloca- tion both between sectors and within sectors.As a result, richer provinces have been able to collect more revenues and also to spend substantially more on education. However, many provinces still find that their resources are insuffi- cient to provide a national standard of education. Provinces normally adjudge national norms according to their financial capacity to fund their education systems.At the district level, further adjustment in norms has been carried out on spending allocation between salary and nonsalary items. Schools also have certain autonomy in collecting additional contributions from parents. As a re- sult, schools in wealthier areas are able to collect and spend more than those in poorer areas, and the gaps in school spending become even wider. The allocation by level of public spending on education determines the availability of resources for primary education. Public funding for primary education is critical for providing basic education skills and opportunities for poor children to be able to continue on to higher education. The previous bias against primary education in the allocation of resources within the sec- tor as a whole in Vietnam is demonstrated by estimating how many primary students could be financed by the cost of one student in either secondary or tertiary education. In 1993, compared with a primary-level student, a lower secondary school student was 1.7 times as expensive, an upper secondary student was 4.3 times as expensive, and a postsecondary student was 26.5 times as expensive. By 1998, the situation had been significantly improved, but bias against lower levels of education still existed. A lower secondary student in Vietnam was only 0.9 times as expensive as a primary-level stu- dent, an upper secondary student in Vietnam was 1.5 times as expensive, and a postsecondary student was 6 times as expensive (table 12.8). 100 20 100 levels 1998 9,549 four of 100 otalT 1993 100 2,761 14.9 1998 and 52 5 1.0 5.9 1998 2,381 2,287 1993 3 Level, Postsecondary 1993 953 53 0.4 2,779 26.5 968 01 2.1 11 1.5 Education 1998 568 by secondary 6 3 Upper 1993 179 0.4 448 4.3 Enrollments, 22 6.3 32 346 0.9 of 1998 2,096 and secondary Lower 1993 477 17 2.8 19 173 1.7 VLSSs. Education 1998 on 43 53 1998 4,104 10.6 388 1.0 and 1993 Primary Spending 42 the 1.31 76 105 1.0 1993 1,152 on based Public of dong) student (times) spending per Share total oss) estimatess' of of (gr public (thousand student spending Author 12.8. education billion ce: bleaT centage ollments cent student spending equivalent public Sour Indicator Public Dong, Per Enr Million Per Per Primary 439 440 Economic Growth, Poverty, and Household Welfare in Vietnam Table 12.9. Public and Private Shares of Financing, by Education Level, 1993 and 1998 (percentage of total expenditure at each education level) Public financing Private financing Level 1993 1998 1993 1998 Primary education 45 61 55 39 Lower secondary education 34 41 66 59 Upper secondary education 40 33 60 67 Postsecondary education 71 46 29 54 Source: Estimates based on government budget data (Ministry of Finance 1998) and the 1993 and 1998 VLSSs. Despite its dominant role in providing school places, the public sector accounts for only slightly more than 50 percent of total expenditure on edu- cation. Expenditures by the private sector (that is, by households) have emerged as an important complement to budget outlays at all levels of edu- cation. Estimates based on the 1993 and 1998 VLSSs suggest that total private spending on education has increased dramatically--from about 1.7 percent of GDP in 1993 to 3.4 percent in 1998. Of this aggregate total increase, 14 per- cent is due to the increase in total number of children enrolled in schools and 84 percent is due to increased private spending per student. Almost all of this--nearly 97 percent--was spent for students enrolled in public schools. Putting these public and private expenditures together (table 12.9) sug- gests that the state budget financed only 52 percent of overall education ex- penses in 1993 and 50 percent in 1998. More important, as a result of the major reallocation of public spending within the education sector, public spending plays an increasing role in financing public primary education; its share in total education expenditure for this level increased from 45 percent in 1993 to 61 percent in 1998. Private spending in public primary schools fell from 55 percent in 1993 to 39 percent in 1998, though the absolute level of private spending almost doubled in real terms. Similarly, the share of public spending in total spending on lower secondary education increased from 34 percent in 1993 to 41 percent in 1998. In contrast, the private sector played an increasingly important role in financing upper secondary and postsec- ondary education. The share of private spending in these schools in total ed- ucation expenditure increased from 60 percent and 29 percent, respectively, in 1993 to 67 percent and 54 percent, respectively, in 1998 (table 12.9). This trend means that more money from public funds for education will be avail- able for basic education levels. Private Financing of Education Table 12.9 highlights the diminished role that the public sector now plays in financing upper secondary and postsecondary education, as distinct from the provision of education. On the one hand, this means that the education Trends in the Education Sector 441 Table 12.10. Per Student Household Expenditure on Schooling, 1993 and 1998 (percentage of total household nonfood expenditure) Most well-off Date/level Poorest quintile quintile Vietnam 1993 Primary 4.4 3.0 3.1 Lower secondary 11.4 5.1 7.9 Upper secondary 20.9 7.6 15.8 1998 Primary 4.9 3.8 3.4 Lower secondary 8.9 5.2 6.4 Upper secondary 21.1 7.9 13.8 Source: Author's estimates based on the 1993 and 1998 VLSSs. sector has been successful in mobilizing a considerable volume of private sources to finance schooling. On the other hand, it means that private costs already play an important role in rationing enrollment in public schooling-- the costs of official fees, private fees, unofficial contributions, books, uni- forms, transport, and so forth. This factor is especially likely to influence ac- cess by students from poor families and may limit the scope for further cost recovery to finance expanded access and a better quality of education. In Vietnam, households still consider that the private sector offers lower-quality education than that offered by the public sector. Although fees are no longer compulsory at the primary level, house- holds must pay many other school-related costs, such as parent contribu- tions for the parent-teacher association (PTA), books, uniforms, private tu- toring fees, transportation, and lunches at school. Though none of these are compulsory by any regulation, they are at best quasi-voluntary; and many children have been punished and humiliated when their families do not pay for these extras. These costs are a considerable financial burden on the poor. Total private expenses per primary school child amounted to 4.4 percent of a typical poor household's nonfood expenditure (table 12.10). Private costs at public lower and upper secondary levels accounted for 9 percent and 21 percent, respectively, of total household nonfood expenditure in 1998, compared with 11.4 percent and 21 percent, respectively, in 1993. As a result, if a typical poor household (from the poorest quintile) had two children in school--one at a primary and the other at a lower secondary school--it would cost 14 percent of this household's total nonfood expendi- ture in 1998 compared with 16 percent in 1993. It should be noted that 75 per- cent of the poorest households were "food poor" (table 12.11). This means these poor households do not have enough money to spend on food to get the minimum standard of necessary calories, so 3 percent of household spending on education is considerable and much more difficult for them, compared with better-off households, which are not food poor. 442 Economic Growth, Poverty, and Household Welfare in Vietnam Table 12.11. Composition of Consumption, 1993 and 1998 (percent) Most well-off Poorest quintile quintile Vietnam Indicator 1993 1998 1993 1998 1993 1998 Food poor 100 75 0 0 25 15 Total consumption expenditure On food 70 68 44 38 55 48 On nonfood 30 32 56 62 45 52 On education 2 4 4 8 3 6 Source: Author's estimates based on the 1993 and 1998 VLSSs. Table 12.12. Composition of Private Spending on Primary Education, by the Poorest and Most Well-Off Quintiles, 1993 and 1998 (percent) Most well-off Date/expense Poorest quintile quintile Vietnam 1993 Fees 5.1 6.7 5.6 PTA and school contribution 19.4 8.0 11.5 Uniforms 9.9 12.9 12.7 Textbooks and school supplies 44.9 14.5 25.0 Transportation, lodging, and food 14.5 45.4 35.1 Other 6.3 12.5 10.1 1998 Fees 0.9 11.9 5.6 Private tutoring 4.9 24.3 14.9 PTA and school contribution 22.1 10.5 15.4 Uniforms 12.0 10.7 13.6 Textbooks and school supplies 45.8 16.3 23.8 Transportation, lodging, and food 11.3 27.8 30.1 Other 7.2 5.3 6.3 Note: PTA = Parent-teacher association. Source: Author's estimates based on the 1993 and 1998 VLSSs. Composition of Private Spending on Primary Education Table 12.12 shows the composition of private spending on primary educa- tion by the poorest and most well-off quintiles in 1993 and 1998. In 1993, PTA and other school contributions (19.4 percent) and textbooks and school supplies (50 percent) were the largest items in total private spending by the poorest quintile. In contrast, among better-off households, most private Trends in the Education Sector 443 spending in primary education was on food and lodging (45 percent) and books and school supplies (15 percent); PTA and other school contributions accounted for only 8 percent. In 1998, PTA and school contributions accounted for an even larger share (22 percent) of total education spending per primary school child in poor households. Another important feature in education in the 1990s was greater self-financing of textbooks and school supplies, which emerged as the largest expenditure item (46 percent) in- curred by poor households. In contrast, better-off households devoted most of their private spending to private fees for tutorials (24.3 percent); trans- portation, food, and lodging (28 percent) (which is very important spending for high-quality education); and health and nutrition for their children. The finding that the most well-off households spend as much as 24 per- cent of their total education spending per primary student on private tutors, and the poorest households spend less than 5 percent on the same line item, implies that access to higher-quality education is biased against the poor. Moreover, because primary education is free, the 12 percent of education ex- penditure that the most well-off families spend for fees may reflect more payments for private tutors. This can help to explain why the better-off children repeated grades less frequently. Because the public budget for edu- cation is limited, having teachers supplement their low salaries by teaching children after school is quite a common phenomenon in urban and wealthy areas in Vietnam. Children who cannot afford these tutorials are at a disad- vantage, because they may not get needed guidance on how to do home- work or more detailed explanations of challenging course content. The cur- riculum is overloaded, and the coursework may even be incomprehensible without additional teacher assistance. It is important to note that much of the material covered in tutorial sessions should, in fact, be taught within the normal school curriculum. Teachers often hold back on teaching these mate- rials so that they can increase their low salaries by teaching extra sessions after school hours. It should also be noted that tutorials are typically used only in urban and wealthy areas. In poorer and rural areas, where house- holds cannot afford these extra classes, teachers must supplement their low salaries by doing other jobs, which also affects the quality of education they can offer. The composition of private spending on education also varied widely by region (table 12.13). For the northern regions, the greatest burden for house- holds of sending children to primary school was not the fees but textbooks and teaching aids, as well as PTA and school contributions, in both 1993 and 1998. It is interesting that in 1993, household expenses for textbooks and teaching aids were much higher in the north compared with the central and southern regions. Households in the northern regions spent about one- half of their total spending on primary education on textbooks and teaching aids, but these accounted for about one-quarter in the central regions and about 15 percent in the southern regions. In 1998, households in the north spent relatively less on these items, while households in the central and southern regions increased the share of their spending on these items. 2.7 6.7 Delta 12.5 16.0 57.0 5.1 0.8 5.3 6.5 17.5 19.0 46.9 4 Mekong 12.6 6.8 15.8 14.0 43.5 7.3 13.1 17.6 9.7 12.5 18.3 26.8 2 Southeast 1.3 16.7 12.2 26.3 39.9 3.6 1.2 2.8 29.3 21.8 31.7 12 1.2 Central Highlands Region 4.0 Coast 1.71 23.1 24.1 16.8 20.3 0.9 22.7 16.0 16.6 29.5 16.3 0.5 by Central Education, Central 5.1 Coast 22.8 16.0 46.0 3.0 7.1 1.3 10.8 23.3 10.2 41.0 4.5 8.9 North Primary on VLSSs. River 3.8 Delta 18.7 1.0 42.7 1.61 22.2 1.91 25.6 14.3 6.5 33.2 7.2 1.3 1998 Spending Red and 1993 Private the of 3.1 23.5 0.1 55.4 13.0 4.9 1.1 9.0 23.9 1.7 51.5 5.8 7 on Northern Uplands association. based Composition food food ent-teacher estimatess' par and and and and = school supplies school supplies A 12.13. tutoring Author PT ce: cent) and and ableT A contribution lodging, A contribution school lodging, Note: Sour (per Date/expense 1993 Fees PT Uniforms extbooksT school ransportation,T Others 1998 Fees Private PT Uniforms extbooksT ransportation,T Others 444 Trends in the Education Sector 445 PTA and school contributions account for a large proportion (24 percent) of total household education spending in the north and central regions in both 1993 and 1998. In principle, there are no tuition fees for primary educa- tion, but in practice there are several other fees and voluntary contributions. An example of these fees is parent contributions for school construction ex- penses. These contributions are quasi-voluntary, because they are not offi- cially regulated by the government but are considered to be necessary by schools and teachers. The importance of these contributions to total private spending for education varies widely among regions. The PTA and school contributions were much higher in the northern regions than in the southern and central regions, with the exception of the Central Highlands in 1998. In the Northern Uplands and Red River Delta in 1998, items such as books, PTA contributions, and school supplies accounted for 77 percent of total household spending per primary school child; these same items accounted for 87 percent of total household spending per primary school child in the North Central Coast. However, these items amounted to only 26 percent of household spending in the Mekong Delta in 1998. In contrast, in the southern regions, especially in the Mekong Delta, uni- forms, textbooks, and transportation, food, and lodging were the largest spending items. In the Mekong Delta, transportation, food, and lodging ac- counted for about one-half (57 percent in 1993 and 47 percent in 1998) of household education expenditure per primary school child. It is interesting to note that in southern regions, PTA and school contributions were rela- tively small (less than 10 percent of total expenditures). Because data did not allow for a further breakdown of expenditure on transportation, food, and lodging, it is impossible to draw conclusions about why they were so expensive in these regions compared with other regions and which item-- transportation, food, or lodging--was the driving factor for such a high cost for these items in primary education. Because these items were such a large primary education expense in the Mekong Delta, further study is needed to better understand the situation in this income-poor and educationally poor region. There is considerable variation in public spending per student across re- gions, however, which, when coupled with variation in enrollment rates across regions, resulted in a public spending pattern that was not pro-poor, although public spending on primary education was neutral in 1998. The low level of public budget expenditure on education makes the additional private resources essential, but they clearly have serious implications for eq- uity, both in exacerbating existing inequalities in education finance and in the potential barriers they represent to participation by poor children. Figure 12.5 shows all public and private education spending per student and by region for both 1993 and 1998. A great variation in both public and private education spending can be seen across regions, though the variation in public education spending per student was less in 1998 than in 1993. In both years, the Southeast had the highest level of private spending--much higher than the rest of the country. Surprisingly, public spending per student 446 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 12.5. Per Student Public and Private Spending on Primary Education, by Region, 1993 and 1998 Spending (thousand dong) 1,200 1,000 1998 800 600 1993 400 200 0 I II III IV V VI VII I II III IV V VI VII Region Private Public Note: I = Northern Uplands, II = Red River Delta, III = North Central Coast, IV = Central Coast, V = Central Highlands, VI = Southeast, and VII = Mekong Delta. Source: Author's calculations from the 1993 and 1998 VLSSs. was highest in the Southeast, the most well-off region in the country. This is partly because Ho Chi Minh City is in this region, which has more high-level education institutions than the other regions. Another explanation is the de- centralization effect: The better-off cities and provinces have more resources to spend than others and thus can spend much more on education. In 1998, the Mekong Delta, the North Central Coast, and the Central Coast had the lowest levels of public spending on education. There may be some link be- tween high performance in terms of enrollment and the level of public edu- cation spending per student. This is explored more in the next section by looking at public spending per student by level and region (see table 12.14). Who Benefits from Public Education Spending? The overall picture in education has been seen to have improved from 1993 to 1998 in all quintiles and all regions of Vietnam. In response to increased enrollment, there was a sizable increase in public spending for education in the 1990s and an intrasectoral reallocation toward the primary and lower secondary levels. Since policy changes have increased the availability of public funds at these levels, in absolute as well as in relative terms, public Trends in the Education Sector 447 Table 12.14. Share of Public Education Spending and School-Age Children, for Poorest and Most Well-Off Quintiles, 1993 and 1998 (percent) Poorest quintile Most well-off quintile Share of Share of public Share of public Share of education school-age education school-age Date/level spending children spending children 1993 All education 16.5 19.0 22.9 19.0 Primary 20.0 24.0 17.4 16.0 Lower secondary 7.2 20.0 33.8 19.0 Upper secondary 2.1 16.0 60.3 21.0 1998 All education 18.1 20.0 21.4 18.0 Primary 26.0 26.0 12.5 14.0 Lower secondary 12.4 20.0 21.7 16.0 Upper secondary 4.0 18.0 41.0 20.0 Source: Author's estimates based on the 1993 and 1998 VLSSs. primary and lower secondary spending per student has increased. Public spending on education has varied considerably across regions, however, be- cause a large part of public education spending, especially at the primary and lower secondary levels, is locally funded as a result of the decentraliza- tion process. Moreover, poor students still have significantly lower enrollments than better-off students, even after the recent surge in enrollment. In general, the public sector's share of total enrollment is fairly constant across quintiles; thus, public school enrollment rates are higher among the better-off than among the poor. This enrollment gap is particularly wide at higher levels of education: The small part of the population that reaches a higher level of ed- ucation, among whom the better-off are overrepresented, receives a dispro- portionate share of the education budget. To examine the changes in benefits from public spending on education accrued to different groups of the popu- lation, and to see whether these changes are more or less pro-poor, the remainder of this section analyzes the distribution of per capita public spending across socioeconomic groups, regions, and genders for all educa- tion levels. This section uses the benefit incidence analysis technique to measure how well public services are targeted to certain groups in the population, such as the poor, particular regions, and females. The observed patterns of public spending across socioeconomic groups are determined by two factors. The first is government spending allocations 448 Economic Growth, Poverty, and Household Welfare in Vietnam within the sector on each level of public education and across regions, and the second is household behavior regarding use rates of public education. Thus, the incidence analysis integrates two sources of information: unit ex- penditure by level of education and by region (for example, the per student annual public spending for primary education by region) and information on individual use rates of education at different levels disaggregated by so- cioeconomic group, region, and gender. In this way, public expenditure on education is distributed across socioeconomic groups subject to their enroll- ment rates and the annual public expenditure per student by education level. The incidence of public expenditure is the result of both a public policy decision, the allocation of public expenditure to and within the edu- cation sector across regions, and a private decision, the behavior of house- holds (that is, to send their children to school or not). Assessing how well public spending on education is targeted to the poor requires a profile of who uses publicly provided education, together with measures of subsidy received by these users. This analysis uses the 1993 and 1998 VLSS data to generate the distribution of public school enrollment by per capita consumption, region, and gender, together with per student sub- sidies, estimating from the public finance data. If public spending were equally distributed across the population, then every expenditure quintile, region, and gender group would receive exactly the same percentage share of total public resources in the sector compared with their share in the total population. A Lorenz curve will be used to illustrate benefit incidence of public spending for education. In this chapter, an adjusted Lorenz curve is used instead of a normal one, because the number of school-age children was not equally distributed across quintiles. The two poorest quintiles in Vietnam have a higher propor- tion of primary school-age children than their proportion in the population. In 1998, 26 percent of the total number of school-age children were from the poorest quintile, compared with only 14 percent of school-age children from the most well-off quintile. In contrast, the poorest quintile contained only 18 percent of all upper secondary students, compared with 23 percent from the fourth quintile and 20 percent from the most well-off quintile (table 12.14). Thus, on the Lorenz diagram, cumulated share of public spending for edu- cation accrued is drawn against the cumulated share of school-age children instead of the cumulated share of population. The diagonal line (or 45- degree line) is also known as the line of absolute equality because it goes through those points where the cumulative share of school-age children equals the cumulative share of total public spending on education. Targeting of public education spending to the poor improved signifi- cantly between 1993 and 1998, partly as a result of the surge in enrollments by poor children at the primary and lower secondary education levels and partly because of the recent policy shift favoring primary education. The majority of school enrollment in Vietnam has traditionally consisted of primary-level students. Students at the primary level made up 78 percent of the total general education enrollments in 1993 and 58 percent in 1998. Trends in the Education Sector 449 As a result, the share of public education spending received by the poor increased between 1993 and 1998. The share of all public education expen- diture received by the poorest household expenditure quintile increased from 17 percent in 1993 to 18 percent in 1998, while the share of the most well-off household expenditure quintile fell from 23 percent in 1993 to 21 percent in 1998. Primary education demonstrated the most dramatic shift. In 1998, the poorest expenditure quintile received 26 percent of public primary education expenditure, up from 20 percent in 1993. The share going to the most well-off expenditure quintile fell to 13 percent in 1998 from 17 percent in 1993. When these cumulative shares of public spending in primary edu- cation are put against cumulative shares of school-age children in Lorenz diagrams for 1993 and 1998, it is shown that public spending for primary education in 1998 was quite neutral, though in 1993 it was still slightly biased against the poor. In 1993, public expenditure at the primary level was slightly biased against the poor because its distribution was below but close to the diagonal line, which means that the poor receive a slightly smaller share of total education spending than their share of primary school-age children. In contrast, in 1998 public expenditure at the primary level was pro-poor because its distribution was just above the diagonal line (fig- ure 12.6). Despite this recent progress in education spending and enrollments, by 1998 inequities in public education spending were still considerable at the secondary level. Public expenditure at secondary education levels was still not favoring the poor because their distributions were under the diagonal line, though the Lorenz curves moved upward between 1993 and 1998, re- flecting improvement supporting the poor. The poorest quintile received a 12 percent share of lower secondary education expenditures while it con- sisted of 20 percent of lower secondary school-age children (see table 12.14). Similarly, in 1998 the bottom quintile received a much smaller share of pub- lic spending on upper secondary education (4 percent) than its share of school-age children (18 percent in 1998). In contrast, the most well-off quin- tile received 22 percent of lower secondary and 41 percent of upper sec- ondary education expenditures, although it accounted for only 16 percent and 20 percent, respectively, of school-age children at these levels. The Lorenz distribution of public expenditure on upper secondary education was most biased against the poor, compared with that at the primary and lower secondary levels. Regional Benefit Incidence of Public Spending on Education Table 12.15 shows public spending per capita by level of education and across regions in Vietnam for 1993 and 1998. Per capita public spending for education increased more than threefold between 1993 and 1998, from D (dong) 40,000 to D 126,000. Improvement in public spending for education was highest in the Northern Uplands, the Central Highlands, and the North Central Coast over the five-year period. These three regions received the Figure 12.6 Lorenz Distributions of Public Education Expenditure, by Level of Education, 1993 and 1998 Cumulative share of public spending on education (%) 100 1993 80 60 40 20 0 0 20 40 60 80 100 Cumulative share of school-age children (%) 45-degree line Lower secondary Primary Upper secondary Cumulative share of public spending on education (%) 100 1998 80 60 40 20 0 0 20 40 60 80 100 Cumulative share of school-age children (%) 45-degree line Lower secondary Primary Upper secondary Source: Author's calculations from the 1993 and 1998 VLSSs. 450 16.6 6.9 2.6 ietnamV 13.7 40.0 50.4 27.9 12.2 31.9 122.0 Delta 16.3 6.0 1.6 6.5 30.0 48.3 21.7 7.9 21.2 99.0 Mekong 16.7 7.8 3.5 22.2 50.0 37.7 27.2 15.1 66.4 146.0 1998 Southeast and 1993 18.5 3.8 0.6 0.0 23.0 63.4 24.1 5.5 7.3 Central 100.0 Highlands Region, and 15.0 9.4 4.5 15.0 44.0 45.5 24.2 13.5 35.9 Central Coast 19.01 Level by Central 17.6 6.1 1.9 12.2 38.0 58.8 32.0 13.1 21.0 Coast 125.0 Education, North VLSSs. for 1998 River 16.1 8.2 3.2 and 22.8 50.0 44.3 33.0 16.7 45.0 Delta 139.0 Spending dong) Red 1993 the 1998 on Public 17.6 4.9 2.0 7.9 32.0 62.7 30.2 10.0 16.7 based 120.0 January Northern Uplands Capita Per constant estimatess' of Author 12.15. secondary secondary secondary secondary ce: ableT otalT otalT Sour (thousands Date/level 1993 Primary Lower Upper Postsecondary 1998 Primary Lower Upper Postsecondary 451 452 Economic Growth, Poverty, and Household Welfare in Vietnam highest per capita public spending in primary education compared with other regions. This may be an explanation of their good performance in terms of primary school enrollment in 1998. The Red River Delta and the Mekong Delta had the highest per capita public education spending, be- cause they spent much more on postsecondary education. Table 12.16 shows benefit incidence by region in 1993 and 1998. Overall, public spending for education in Vietnam was more equal in 1998 compared with 1993. In terms of benefiting from public education spending, the regions that were disadvantaged in 1993 gained more in 1998, and vice versa for the regions that received more in 1993. Although inequalities still exist, the Northern Uplands, North Central Coast, Central Highlands, and Mekong Delta benefited more from public spending on education relative to their share of population. In contrast, the Red River Delta, Central Coast, and Southeast benefited less from public spending on education compared with 1993, although the two most well-off regions--the Red River Delta and Southeast--still received much more compared with their share in total popu- lation (table 12.16). The major explanation is that, on the one hand, the change in the allocation of public education spending moved relatively toward primary and lower secondary education (as opposed to upper secondary and postsecondary education). On the other hand, the two most well-off regions had fewer children attending primary and lower secondary education. The above standard benefit incidence analysis assumes that all regions have the same share of school-age children in the population. However, it is school-aged children who need to be in school, not all population. There- fore, the proportion of school-age children matters more in benefit inci- dence analysis than the proportion of the population. As can be seen from table 12.17, the Northern Uplands and the Southeast received a far greater proportion of total public spending on primary education than their pro- portion of school-age children in 1993, as occurred in 1998 in the Northern Uplands and the Mekong Delta. It is interesting that the Mekong Delta moved from a relatively less benefited status in public spending on primary education in 1993 to relatively more in 1998. The opposite trend happened in the Southeast. Data from table 12.17 partially reflect a result of the recent decentraliza- tion process in Vietnam, which allows more authority at the local level to make decisions on local budget allocations and has resulted in an increased inequality in education opportunities. In fact, the provinces that may have more public funding resources or external support can thus spend more on education in terms of absolute per student costs, in terms of share of total public expenditure spent on education, or both. In contrast, other provinces have fewer funding resources or external assistance, which in turn limits their ability to spend more on education. Two central regions, the Central Coast and the Central Highlands, and the Southeast spent relatively less from public funding on primary education in 1998 than in 1993. There is a need to review and improve the system of budget planning and manage- ment, in general, and the mechanism for budget allocation, in particular, to ietnamV 100 100 100 100 100 100 100 100 100 100 Delta 21.9 19.5 13.5 17.0 22.4 20.6 16.7 14.0 18.0 21.5 Mekong 12.7 14.2 17.0 16.0 12.6 9.5 12.4 15.9 14.0 12.7 Southeast 1998 3.6 1.8 0.7 2.0 3.2 4.6 3.2 1.7 3.0 3.7 and Central Highlands 1993 9.7 9.3 Central Coast 10.8 16.3 20.6 13.0 1.91 1.91 10.0 10.7 Region, by Central Coast 13.5 1.31 9.2 12.0 12.8 16.1 15.8 14.9 14.0 13.8 Education, ounding.r North by for level) VLSSs. oduced 1998 River intr and Spending education Delta 20.9 25.7 27.0 27.0 21.6 17.2 23.2 26.9 22.0 19.6 ors Red 1993 err each to the Public for on due of 16.6 1.21 100 12.1 13.0 15.6 22.2 19.4 14.8 18.0 17.9 based to Northern Uplands spending add Incidence not estimatess' public may of cent cent 12.16. cent secondary secondary per secondary secondary per Author Rows ce: population population bleaT per schools schools of of Note: Sour (as Date/indicator 1993 Primary Lower Upper All Region 1998 Primary Lower Upper All Region 453 ietnamV 100 100 100 100 100 100 100 100 100 100 1998 and 71 41 22 02 42 19 02 12 71 81 Delta 1993 Mekong Region, 71 21 31 21 31 12 11 31 51 71 by Southeast Education, 3 1 3 3 3 5 4 4 8 8 Central Highlands Secondary 11 13 12 11 13 9 11 11 12 12 cost) Central Coast Lower unit and egionalr Central 11 19 13 14 12 14 13 14 14 13 Coast Primary using North ounding.r by for VLSSs. adjusted oduced 1998 River 19 30 22 23 20 16 17 20 17 17 Delta intr and Subsidies Red ors 1993 err education, to the Public for on due of 22 10 16 17 14 24 24 18 17 16 100 based Northern Uplands to spending add spending spending Incidence not age age estimatess' public public population public Population may of of of of of school age school age Author 12.17. cent secondary secondary secondary secondary Rows ce: bleaT per centage centage otalT school centage centage otalT school Note: Sour (as Date/indicator 1993 Per Primary Lower Per Primary Lower 1998 Per Primary Lower Per Primary Lower 454 Trends in the Education Sector 455 allow opportunities to redirect more public resources toward poorer provinces and toward those that need more resources for social services, such as education and health for the poor. Returns to Education This section looks at the recent changes in rate of returns to education, focusing on private sector wage earners. The Vietnamese labor force is rela- tively well educated, given its low income level, and there was a clear improvement in education level attained from 1993 to 1998 (table 12.18). The average number of years of schooling of wage earners increased from eight years in 1993 to nine years in 1998. The proportion of wage earners with upper secondary-level educations or higher increased from 23 percent in 1993 to 29 percent in 1998. Real average earnings (converted to January 1998 prices) have increased 11 percent annually over the 1993­98 period. Fig- ure 12.7 shows that in 1993, there was no clear difference in real earnings be- tween completed education levels. Instead, a worker who had completed upper secondary school did not earn more than those with less education. Similarly, holding a university degree did not make a significant difference in earnings. The whole picture changed in 1998: the higher the education level completed, the higher the earnings. This is especially true at the uni- versity level: University graduates earn 50 percent more than workers with no more than an upper secondary school diploma. Among those working in the private sector, university graduates can earn almost three times more than workers with only an upper secondary school diploma. Rates of returns to investment in education are estimated here following up a work that was itself based on the 1993 VLSS (Moock, Patrinos, and Venkataraman 1998). The estimates in this study are based on wage data from the 1993 and 1998 VLSSs in an effort to capture the impact of recent economic reforms in Vietnam. Basic and extended Mincerian earning func- tions are used to estimate returns to schooling and levels of education. In the extended model, four levels of general education systems in Vietnam were considered. The levels were primary, lower secondary, upper secondary, and university or college. The impact of vocational training (with three cate- gories: vocational training after primary, lower secondary, and upper sec- ondary schools) has been estimated separately from the benefit of general education (from primary to university and college). Because the earning dif- ference between public and private sectors was statistically significant, models were estimated separately for the private and public sectors. The basic Mincerian earnings function links logarithm of average earn- ings to the approximate determinants of earnings such as years of schooling (S), years of experience (EXP), squared experience, sex, northern or southern region, urban or rural location, and seniority. The Mincerian earning func- tion has the following semilog form: ln Yi = + 1Si + 2EXPi + 3EXP2i + 4SEXi + 5NORTHi + 6URBANi +7SENIORi + i 9 1998 32.0 9.0 17.0 493 47 32 24 23 22 0.45 1,257 Females 6 1993 31.0 8.0 17.0 279 45 0.49 893 42 24 28 19 8 33.0 9.0 18.0 620 48 12 27 26 18 0.37 1998 1,942 1998 Males 7 32.0 8.0 17.0 383 47 12 29 29 14 0.39 and 1993 1,358 1993 0 7.0 542 49 33 34 24 9 1 Sex, 30.0 17.0 1998 1,906 by Private 0 7 0 and 29.0 6.0 16.9 365 48 43 35 23 1993 1,285 5 37.0 12.2 18.8 612 1.0 45 14 27 35 19 1998 1,293 Employment of Public 1 6 15 36 28 15 1993 34.0 1.11 31 1.0 44 17.4 966 Sector by 9.0 570 47 12 26 25 20 9 32.9 17.9 1998 0.40 3,199 ariables,V otalT 8.2 342 46 22 27 28 16 7 1993 31.3 17.1 0.43 2,251 Selected of VLSSs. month week cent) Mean dong) 1998 per per (per of and observations level schooling experience of secondary secondary 1993 12.18. of of sector worked (years) earnings education ce: ableT ariableV Age earsY earsY (thousands No Primary Lower Upper University/college Sour Real Public Hours Number Education 456 Trends in the Education Sector 457 Figure 12.7. Monthly Earnings of Private and Public Sector Workers, 1993 and 1998 Earnings (thousand dong) 2,000 1,800 1,600 1,400 1,200 1,000 800 600 400 200 0 None Primary Lower Upper University secondary secondary Highest education level completed All 1993 Public 1993 Private 1993 All 1998 Public 1998 Private 1998 Source: Author's calculations from the 1993 and 1998 VLSSs. where Si stands for number of years of schooling that individual i had com- pleted; EXPi and EXP2i are number of years of experience and its square; SEXi represents sex of worker with value one for male and zero for female workers; NORTHi is the residence location of workers that takes value one if the workers were from the north (including the Northern Uplands, the Red River Delta, the North Central Coast, and the Central Coast) and value zero otherwise; URBANi distinguishes workers from urban areas (value one) from workers from rural areas (value zero); SENIORi identifies whether the worker has participated in the work force for a period of fewer than five years, which has value one if fewer than five years and zero otherwise. In this semilog earnings function specification, the coefficient on years of schooling can be interpreted as the average private rate of return to one ad- ditional year of education, regardless of the education level to which this year of schooling refers. As can be seen from table 12.19, returns to one additional year of school- ing, regardless of the level of education, increased from 3 percent in 1993 to 4 percent in 1998. Despite this change, returns to schooling in Vietnam are 8.64 2.95 statistict 3.20- 10.51 9.04- 3.10 0.39 17.74 1998 and 1998 0.039 0.019 0.0004- 0.292 0.389- 0.122 0.019 3.358 0.330 1,897 1993 Coefficient orkers, W 6.8 17.4 0.7 0.3 0.3 0.3 Mean 440.2 Sector Private for 5.96 6.30 statistict 6.37- 1.381 6.99- 1.62- 0.07- 22.27 Schooling of 1 earsY 1993 0.031 0.032 0.0006- 0.380 0.268- 0.057- 0.004- 4.31 0.200 1,285 on Coefficient Based ean 6.01 M 16.92 0.65 0.28 0.32 0.16 444.18 Functions Earning VLSSs. 1) ed Basic 1) 1) = 1998 = squar 1) = and = observations school experience (junior of 1993 12.19. of of (north (urban ce: (male ableT ariableV arseY earsY Sour 2 Experience Sex North Urban Seniority Constant R Number 458 Trends in the Education Sector 459 still low by international standards and are below those of some other devel- oping countries, such as the Philippines (6.2 percent in 1998) (Schady 2001). Worldwide, one more year of schooling can bring about a 10 percent increase in earnings. However, low returns to schooling were observed in some de- veloping countries in the 1980s--the estimates of returns to schooling in China ranged from 1 to 5 percent in the mid- and late 1980s, 2.9 percent in Poland in 1986, and 4.3 percent in Hungary in 1987 (Psacharopoulos 1994). Among factors other than schooling, the experience, gender, region, and urban or rural location of the worker were statistically significant in deter- mining earnings in 1993 and 1998. An additional year of experience brings 3 percent higher earnings in 1993, an amount equivalent to the impact of one more year of schooling. However, experience was relatively less important in 1998, so that an additional year results in only a 2 percent increase in earn- ings (table 12.19). Female workers are better educated than male workers. Although males and females have nearly the same level of education attainment in terms of number of years of schooling, the female work force has a slightly higher proportion with a university degree or who have completed upper sec- ondary school than the male work force. However, males' earnings were still 37 percent and 26 percent higher than females' earning in 1993 and 1998, re- spectively. In both years, a male worker earned much more than an equally skilled female worker even after controlling for education, experience, and the other variables in table 12.19. In 1993, a male worker earned 46 percent more than a female worker with the same (observable) characteristics. This differential decreased slightly by 1998, but it was still 40 percent. These re- sults mean that there is still some kind of discrimination between males and females, and something that keeps females' wages low. Further study is needed to fill the gap in this area. A structural test was used to see if there is any difference in returns to ed- ucation between male and female workers for the 1993 and 1998 surveys. It is interesting that returns to schooling of male workers in 1993 were higher than those of female workers, but the situation was reversed in 1998, when females had higher returns to schooling than males. However, the differ- ences in 1998 were not statistically significant. The wages of workers also varied by their locations. Urban workers in 1998 earned 12 percent more than identical workers in rural areas. Workers in the south earned more, on average, than those in the north, even though the average education level in the north is higher. The work force in the north had, on average, one more year of schooling than the work force in the south, and it had higher proportions with upper secondary- and university- level education than the south. After taking into account other factors, in 1993 a worker in the south still earned 24 percent more than an identical worker in the north. In 1998, the gap widened to 32 percent. A variable on seniority was included in the model to see if there was any impact from re- cent changes in the labor market. It is assumed that younger workers are more affected by the recent changes because they are entering directly into a 460 Economic Growth, Poverty, and Household Welfare in Vietnam free market wage economy, and higher levels of education can bring them a significant difference in earnings. The variable takes the value one if a worker had been in the labor market for fewer than five years and zero otherwise. It turns out that, other things being equal, there is no statistical difference in terms of earnings between young and senior workers. An extended earning model was estimated to look at the rates of return for different levels of education. It estimates average private returns to dif- ferent levels of schooling by converting continuous years of schooling (S) into a series of dummy variables. The dummy variables represent the com- pletion of the respective level of general education or vocational training after general education. The following is the extended model: ln Yi = + 1PRIMi + 2LSECi + 3USECi + 4UNIVi + 5VOCPi +6VOCLSi + 7VOCUSi + 8EXPi + 9EXP2i + 10SEXi +11NORTHi + 12URBANi + 13SENIORi + i where PRIMi, LSECi, USECi, and UNIVi are dummy variables indicating pri- mary, lower secondary, upper secondary, or university education completion by individual i; VOCPi , VOCLSi, and VOCUSi reflect the kind of vocational training that individual i took after completing primary, lower secondary, or upper secondary school. The private rates of return to one year at different levels of schooling then can be derived by dividing the estimated rates of return to each level of education by the correspondent number of additional years required at each level. For primary education, this is assumed to be equal to one, because there will be no opportunity cost involved for a child at this level. A student in Vietnam is required to spend at least four, three, and four years at lower secondary, upper secondary, and university education, respectively. Table 12.20 presents regressions that allow for a different impact of dif- ferent levels of education. Unlike table 12.19, the coefficients in table 12.20 represent returns to different completed education levels (and to completed vocational training), not the returns per year of education at different levels. Returns per year of education at the different levels can be easily calculated and are given below. Other economic variables were kept unchanged. Investment in education at the primary and lower secondary levels was profitable in both surveys. Returns to primary education declined slightly, from 15 percent in 1993 to 14 percent in 1998, however. Similarly, returns to lower secondary education declined from 3 percent in 1993 to 2 percent in 1998. In contrast, upper secondary and university education turned out to be statistically significant only in 1998. One more year of attending an upper secondary school did not make any statistical difference in a worker's earn- ings in 1993, but it could bring as much as 4 percent more to the worker's earnings for each additional year spent in an upper secondary school in 1998. Similarly, returns to one more year in a university or college could bring a 24 percent increase in earnings in 1998. It is interesting that, after taking into account differences in education attainment level, attending vocational 3.86 2.14 1.90 4.17 statistict 0.13- 0.65 0.68 3.24 3.57- 10.81 8.57- 0.54 2.99 18.01 1998 and 1993 1998 0.143 0.072 0.122 0.940 0.025- 0.068 0.057 0.021 0.000 0.300 0.369- 0.026 0.121 3.448 0.340 1,897 Coefficient orkers, W Sector 0.67 0.34 0.09 0.01 0.01 0.03 0.05 Mean 17.41 0.65 0.32 0.33 0.34 440.22 Private for 3.75 2.64 statistict 0.18- 1.01- 0.83- 0.44 0.19- 6.18 6.29- 1.651 6.82- 0.01 1.36- 22.51 Education of Level 18 on 1993 0.153 0.1 0.013- 0.265- 0.334- 0.063 0.027- 0.032 0.001- 0.389 0.269- 0.001 0.048- 4.361 0.200 1,285 Coefficient Based Functions ean 0.66 0.31 0.08 0.00 0.00 0.01 0.01 M 16.92 0.65 0.28 0.16 0.32 444.18 Earning VLSSs. 1) ed Extended 1) = 1) 1998 = squar 1) = and = observations experience (junior of 1993 12.20. secondary secondary vocational vocational of (north (urban ce: (male ableT ariableV Primary Lower Upper University Postvocational Lower Upper earsY Sour 2 Experience Sex North Seniority Urban Constant R Number 461 462 Economic Growth, Poverty, and Household Welfare in Vietnam training did not make any statistical difference in earnings in the private sector in both years (table 12.20). As in the basic earning model, experience, gender, number of hours worked, and region seemed to have similar impacts on earnings in this extended model. Conclusions Over the period from 1993 to 1998, school enrollment in Vietnam increased dramatically at all levels, but especially at higher education levels. These im- provements occurred for all expenditure groups and regions. Changes in the allocation of public spending on education in the 1990s increasingly favored lower levels of education. As a result, the share of public spending on edu- cation going to the poor increased from 16.5 percent in 1993 to 18.1 percent in 1998. This chapter documents major changes and conditions in the edu- cation sector in Vietnam, using data on public expenditure on education and the VLSSs. There are four main conclusions: Enrollment The improvement in enrollment rates was not equal across the different expenditure groups. At the primary level, there was a huge increase in enroll- ment for the bottom quintile, but still 18 percent of the poorest children of primary school age were not enrolled in school. The enrollment gaps between the better-off and the poor are even wider for higher levels of education. Improvements in enrollment rates also varied considerably across regions, with wider disparities at higher levels of education. It is important to look at the reasons underlying these disparities and understand why some regions were doing much better than others. Regional variation in public spending, private contributions, or both, is an important factor that calls for government attention, especially in the context of strong decentralization in Vietnam. On a more positive note, improvements in enrollment were considerable and equally spread for boys and girls. Gaps in improvement in enrollment between the majority and minority groups widened, however, given low enrollment rates for minority children--especially at the higher education levels. This may be an area that merits special attention from policymakers, researchers, and others. Considerable improvement in primary enrollment for ethnic minority groups shows that recent programs of the government and the international community have generated positive results. Addi- tional efforts will be needed to provide access to higher levels of education and encourage ethnic minority children to enroll (and stay) in school. Financing for Education Public spending on education more than tripled in real terms from 1993 to 1998, and a larger share of public spending was allocated to primary and lower secondary education. Despite this overall increase, several issues in spending allocation and revenue collection accounted for a large variation in public spending on schools. Trends in the Education Sector 463 Although fees are no longer compulsory at the primary level, households purchased many school-related items, such as books, uniforms, private tu- toring, transportation, and lunch. There were also other contributions; thus, even at the primary level, paying for education can be a considerable finan- cial burden on the poor. This burden is even greater for poor children con- tinuing on to higher levels of education. Findings from many participatory poverty assessments in Vietnam show that financial contribution was one of the main reasons why many poor children dropped out of school at a rela- tively young age (World Bank 1999). Moreover, better-off households spent more on such quality-related items as private tutoring after school hours, transportation, food, and lodging. As a result, the poor children not only re- ceive less education compared with better-off children, but the education they receive is of lower quality. It is thus important that although reducing the financial burden on poor households is necessary, it is equally important that everyone have access to quality education. It also should be noted that getting these marginal children to school may be much harder than for the average child. Efforts in this regard must come from many sides: the gov- ernment, civil society, the local community, and households. The level and composition of private spending on education also vary greatly across regions. In the Mekong Delta and the Central Highlands, which are poor regions, private spending by households on secondary edu- cation was much higher than in the Northern Uplands and the Red River Delta. In the northern regions, the highest burdens faced by households with children in primary schools were PTA contributions, textbooks, and paper. In contrast, in the southern and central regions, especially in the Southeast and the Mekong Delta, uniforms, transportation, and lunch were the largest spending items. Given the poor performance in enrollment in the Mekong Delta and the Central Highlands, it is important that the role of, and regional variations in, private payments for the PTA, textbooks, and other school supplies are taken into consideration if these are major imped- iments to enrollment of poor children. Pro-Poor Budget Planning and Management The proportion of school-age children in the population varies greatly among regions, with poorer regions likely to have more children who are of school age. When the proportions of school-age children were taken into consideration in the benefit incidence analysis, it turned out that the Northern Uplands received a far greater proportion of total public spending on primary education than its proportion of school-age children in both 1993 and 1998. It is interesting that the Mekong Delta moved from benefiting rel- atively less from public spending on primary education in 1993 to benefiting relatively more in 1998. This, together with low performance profiles in some of the less-benefited regions, may have implications for policy in real- location of public spending on education, especially between regions, to im- prove enrollment performance in these regions. There is a need to review and improve the system of budget planning and management in general and 464 Economic Growth, Poverty, and Household Welfare in Vietnam the mechanism for budget allocation in particular to allow opportunities to redirect more public resources toward poorer provinces and toward those that need the most. Returns to Education Finally, this chapter shows that the labor market in Vietnam is changing rapidly. Returns to schooling at higher levels of education increased sub- stantially between 1993 and 1998, especially at the upper secondary and university levels. Investment in education at all levels raised wages in the private sector, but returns to primary and lower secondary education dropped slightly between 1993 and 1998. Linking these findings with those from previous sections in this chapter shows that helping the poor to com- plete primary education can provide them with an opportunity to improve their earnings and living standards (by entering secondary and postsec- ondary education). In contrast, vocational training had no statistically signif- icant influence on earnings in the private sector. This result raises questions about the value of such education, which is thought to play an important role in generating technical skills to meet the demand of a fast-changing market economy. Further study is needed on the quality of vocational training and its capacity to provide the right skills that new businesses need most. Notes The author is grateful to Nisha Agrawal, Dominique van de Walle, Carolyn Turk, Bob Baulch, and especially Paul Glewwe for helpful discussions on the issues cov- ered in this chapter. 1. The 1992­93 survey spanned a full year, starting in October 1992; the 1997­98 survey began in December 1997 and also lasted a year. For brevity's sake, reference is made to the surveys as the 1993 VLSS and 1998 VLSS, respectively. 2. The gross enrollment rate is defined as the ratio of the number of children currently in school at a given education level to the total number of children of the right age for that level. For example, the right age for primary school is 6 to 10 years. The net enrollment rate is defined as the ratio of the number of children currently in school at the right age for a given education level to the total number of children of the right age for that level. Bibliography The word processed describes informally reproduced works that may not be commonly available through libraries. Behrman, J. R., and N. Birdsall. 1987. "Comment on Returns to Education: A Further International Update." Journal of Human Resources 22(4): 603­07. Behrman, J., and J. C. Knowels. 1999. "Household Income and Child School- ing in Vietnam." World Bank Economic Review 13(1): 211­57. Glewwe, P., and H. G. Jacoby. 1998. "School Enrollment and Completion in Vietnam: An Investigation of Recent Trends." In D. Dollar, P. Glewwe, Trends in the Education Sector 465 and J. Litvack, eds. Household Welfare and Vietnam's Transition. World Bank Regional and Sectoral Studies. Washington, D.C.: World Bank. _____. 2000. "Economic Growth and the Demand for Education: Is There Any Wealth Effect?" Journal of Development Economics. Glewwe, P., and H. A. Patrinos. 1996. "The Role of the Private Sector in Education in Vietnam: Evidence from the Vietnam Living Standard Survey, 1992­93." Background Paper for the Vietnam Education Finance Sector study (VEFSS). Policy Research Department, and Human Development Department, World Bank, Washington, D.C. Ministry of Training and Education. 2001. Education Ten Year Strategy 2001­2010. Hanoi. Moock, P., A. Patrinos, and M. Venkataraman. 1998. "Education and Earn- ings in a Transition Economy (Vietnam)." World Bank Policy Re- search Working Paper 1920. World Bank, Policy Research Depart- ment, Washington, D.C. Psacharopoulos, G. 1994. "Returns to Investment in Education: A Global Update." World Development 22(9): 1325­43. Schady, Norbert R. 2001. "Convexity and Sheepskin Effects in the Human Capital Earnings Function: Recent Evidence for Filipino Men." World Bank Policy Research Working Paper 2566. World Bank, Policy Research Department, Washington, D.C. Schultz, T. P. 1998. "Education Investments and Returns." In H. Chenery and T. N. Srinivasan, eds., Handbook of Development Economics. The Netherlands: Elsevier Science Publishers. Truong, T. K. C., T. N. D. Thai, and H. V. Bach. 1999. "Educational Enroll- ments in Lower Secondary School." In D. Haughton and J. Haughton, eds., Health and Wealth in Vietnam: An Analysis of Household Living Standards. Singapore: Institute of Southeast Asian Studies. Viet Nam Living Standards Survey (VNLSS). 1992­1993. Web site: www. worldbank.org/lsms/country/vn93/vn93bid.pdf. Vietnam Living Standards Survey (VLSS). 1997­98. Web site: www. worldbank.org/lsms/country/vn98/vn98bif.pdf. World Bank. 1995. "Vietnam Poverty Assessment and Strategy." Country Report. Washington, D.C. _____. 1996. "Vietnam: Education Financing Sector Study." Human Resources Operations Division, Country Department I, East Asia and Pacific Region, Washington, D.C. _____. 1999. "Voices of the Poor in Vietnam: A Synthesis Report of the Four Participatory Poverty Assessments." Hanoi. 13 An Investigation of the Determinants of School Progress and Academic Achievement in Vietnam Paul Glewwe Many economists have claimed that a well-educated population leads to higher economic growth (Barro 1991; Lucas 1988; Mankiw, Romer, and Weil 1992). The World Bank concurs with this view; for example, it has claimed that education played an important role in the economic success of many East Asian countries (World Bank 1993). In almost all developing countries, the majority of schools are run by the government, so government education policy has a very direct effect on education outcomes. Vietnam is no excep- tion in this regard: 99 percent of Vietnamese primary school students and more than 90 percent of secondary school students attend public schools. Despite its recent economic success, Vietnam is still a low-income coun- try; its gross national income per capita in 2000 was US$390, which implies a rank of 164 among 206 countries (World Bank 2002a). Yet Vietnam's performance in education is much higher than that of other low-income countries. According to World Bank data, the average (net) primary school enrollment rate for low-income countries (those with per capita incomes below US$755) was 76 percent in 1997, yet Vietnam's rate was 100 percent. Analogous figures for (net) secondary school enrollment rates are 51 percent for low-income countries and 55 percent for Vietnam (World Bank 2001a). Vietnam's success in education is commendable, but it is still below the educational achievements of the top-performing countries in East Asia. For example, China's (net) secondary school enrollment rate was 70 percent and the Republic of Korea's rate was 100 percent. In view of Vietnam's desire to close this gap, this chapter examines the determinants of school completion and school performance (as measured by test scores) in Vietnam, focusing on primary and lower secondary schools. 467 468 Economic Growth, Poverty, and Household Welfare in Vietnam The next section of this chapter presents trends in education outcomes in Vietnam in the 1990s, followed by a description of the data and a presenta- tion of the empirical framework for estimation. The results are presented in the subsequent section, and the last section summarizes and presents concluding comments. Primary and Secondary School Outcomes This section presents some basic facts about education outcomes in Vietnam in the 1990s, focusing on trends over time and variation across different socioeconomic groups. Trends over Time Vietnam enjoyed high rates of economic growth in the 1990s: The average annual rate of economic growth from 1990 to 2000 was 7.9 percent (World Bank 2002b). Wealthier countries typically have relatively high rates of school enrollment, so one would expect increased enrollment in primary and secondary school in Vietnam in the 1990s. This is confirmed by house- hold survey data from Vietnam, which show that school entrance rates increased at all levels from 1992­93 to 1997­98. As seen in table 13.1, the percentage of children who eventually enter primary school (grades 1­5) increased from 95 percent to 97 percent, so that virtually all Vietnamese chil- dren begin primary school. The percentage of children who eventually enter lower secondary school (grades 6­9) increased from 66 percent to 72 percent, and the percentage who eventually enter upper secondary school (grades 10­12) increased from 23 percent to 29 percent.1 These recent increases in enrollment stand in sharp contrast to the changes that took place from the late 1980s to the early 1990s. Although in- come growth was also robust during this earlier period, school enrollment at the secondary level declined. Table 13.2 illustrates this fact, using school enrollment rates calculated from two different data sources: the official rates reported in the United Nations Educational, Scientific, and Cultural Table 13.1. School Entrance Rates in Vietnam, 1993 and 1998 (percent) Education level 1993 1998 Primary 94.7 97.4 Lower secondary 66.1 72.3 Upper secondary 22.7 29.4 Note: Because late enrollment and grade repetition are common in Vietnam, primary entrance is based on children ages 11­12 years. The ages for lower and upper secondary levels are 16­17 and 19­20, respectively. Source: Author's calculations based on the 1993 and 1998 VLSSs. An Investigation of the Determinants of School Progress 469 Table 13.2. School Enrollment Rates in Vietnam, 1980­98 School enrollment 1998 VLSS UNESCO data (gross rates) Primary Secondary Year Primary Secondary Postsecondary Net Gross Net Gross 1980­81 109 42 2 -- -- -- -- 1981­82 -- -- -- -- -- -- -- 1982­83 -- -- -- -- -- -- -- 1983­84 -- -- -- -- -- -- -- 1984­85 104 40 -- -- -- -- -- 1985­86 103 43 2 62 75 39 44 1986­87 104 -- -- 66 82 37 41 1987­88 106 44 -- 68 85 35 39 1988­89 104 40 2 68 85 32 35 1989­90 102 35 2 72 89 32 35 1990­91 103 32 2 74 91 32 35 1991­92 104 31 2 76 93 35 38 1992­93 109 32 2 79 96 37 41 1993­94 111 35 2 81 102 41 44 1994­95 113 41 3 85 107 44 48 1995­96 114 47 4 87 110 48 53 1996­97 115 52 7 89 113 53 58 1997­98 113 57 -- 92 113 57 63 -- Not available. Note: Gross enrollment rates measure the number of children enrolled in the particular level of schooling, regardless of their ages, divided by the number of children in the age range as- sociated with that level of schooling (ages 6­10 for primary, ages 11­14 for lower secondary, and ages 15­17 for upper secondary). Net enrollment rates are similarly defined, except that the number of children enrolled excludes children enrolled who are not in the associated age range. Sources: UNESCO, Statistical Yearbook, various years; author's calculations based on the 1998 VLSS. Organization (UNESCO) statistical yearbook and rates calculated using retrospective data from the 1997­98 Vietnam Living Standards Survey (VLSS).2 The UNESCO figures show that from 1987­88 to 1992­93, the sec- ondary gross enrollment rate fell from 44 percent to 32 percent, but rose steadily thereafter. The VLSS data show a similar trend, although the timing is somewhat different. Both sources of data show increases in enrollment rates starting in the early 1990s, reaching a gross secondary school enroll- ment rate of about 60 percent in 1997­98; they also show a modest increase in the gross primary school enrollment rate during the 1990s. These increases are consistent with substantial increases in government spending on educa- tion, especially secondary education, in the 1990s. Recent trends in education finance are described in detail in chapter 12. 470 Economic Growth, Poverty, and Household Welfare in Vietnam Variation across Different Socioeconomic Groups Table 13.3 shows entrance rates into lower and upper secondary schools in 1993 and 1998 for the two VLSSs (described in the section on data and methodological framework). For Vietnam as a whole, the proportion of chil- dren entering lower secondary school (grades 6­9) increased from 66 per- cent to 72 percent, a respectable but not spectacular increase over five years. For upper secondary school, the increase was sharper, from 23 percent to 29 percent. Yet these school entrance rates vary widely across different pop- ulation groups in Vietnam. As in most countries, urban children in Vietnam are more likely to enter lower secondary school than rural children. The rates among urban children were 83 percent in 1993 and 87 percent in 1998. Thus, in urban areas lower secondary school is approaching universal coverage. In rural areas, the rates are lower, but the increase is higher--from 62 percent in 1993 to Table 13.3. Entrance Rates for Lower and Upper Secondary Schools, 1993 and 1998 (percent) Lower secondary Upper secondary Indicator 1993 1998 1993 1998 All Vietnam 66.1 72.3 22.7 29.4 Urban 82.6 87.2 46.9 53.1 Rural 62.2 68.5 15.2 22.5 Region Northern Uplands 62.4 72.3 20.7 22.5 Red River Delta 83.3 91.7 35.8 45.7 North Central Coast 72.1 84.4 31.1 38.2 Central Coast 68.6 73.8 23.2 31.0 Central Highlands 45.7 57.2 0.0 8.2 Southeast 71.1 71.1 24.1 39.2 Mekong Delta 49.2 54.4 9.2 16.4 Quintile 1 (poorest) 40.1 47.4 9.1 7.6 2 53.7 64.7 7.3 10.9 3 67.8 70.6 21.1 22.6 4 75.8 81.2 25.0 35.6 5 (most well-off) 84.2 92.8 43.3 58.8 Ethnic group Vietnamese (Kinh) 71.2 72.8 25.4 29.1 Chinese 80.0 82.9 31.0 53.8 Other ethnic minority 28.6 64.0 3.4 24.5 Note: Because late enrollment and grade repetition are common in Vietnam, lower and upper secondary school entrance rates are based on children ages 16­17 and 19­20, respectively. Source: Author's calculations based on the 1993 and 1998 VLSSs. An Investigation of the Determinants of School Progress 471 69 percent in 1998. Although rural areas are slowly catching up to urban areas in terms of enrollment in lower secondary school, there is still a large gap at the upper secondary school level. In urban areas, 47 percent of the population entered upper secondary school in 1993, a figure that increased to 53 percent in 1998. Yet in rural areas the rate was only 15 percent in 1993, although it did increase to about 23 percent by 1998. Thus, the largest gap in schooling between urban and rural areas in Vietnam is at the upper secondary level. There are also large regional disparities in entry into lower and upper secondary school. The best performer among Vietnam's seven regions3is the Red River Delta. In 1993, about 83 percent of the children in that region had entered lower secondary school, and this increased to 92 percent by 1998. Entry into upper secondary school was also the highest in the Red River Delta among the seven regions, increasing from 36 percent in 1993 to 46 per- cent in 1998. The next-best performer among the seven regions was the North Central Coast: By 1998, it had an enrollment in lower secondary school of 84 percent and an upper secondary enrollment of 38 percent. At the other end of the spectrum are the Central Highlands and the Mekong Delta. The percentage of children who entered lower secondary school was in the mid- to upper 40s in 1993 and the mid-50s in 1998. Entry into upper sec- ondary school was lowest in the Central Highlands, at about 8 percent in 1998, while the Mekong Delta had the second lowest rate among the seven regions, at 16 percent. In almost any country, including Vietnam, school progress is positively correlated with income levels. Among the poorest 20 percent of the popula- tion (as measured by consumption expenditures per capita), only about 40 percent of children had entered lower secondary school in 1993, although the rate increased to 47 percent by 1998. Among the most well-off 20 percent of the population, the respective rates are doubled: 84 percent in 1993 and 93 percent in 1998. In upper secondary education, the disparities are even wider and appear to be increasing. Only 9 percent of children in the poorest 20 percent of households had entered upper secondary school in 1993; this rate was even lower in 1998, at 8 percent. In stark contrast, among the most well-off 20 percent of the population, entry into upper secondary school increased from 43 percent in 1993 to 59 percent by 1998. Finally, it is worthwhile to examine differences by ethnic groups. About 85 percent of the population in Vietnam is ethnically Vietnamese (Kinh). Another 2 percent is Chinese, and the remaining 13 percent is spread across a wide variety of groups that are found predominantly in remote rural areas. Entrance into lower secondary school increased slightly for both Kinh and Chinese children; in contrast, entry into lower secondary education in- creased dramatically for ethnic minority children, from 29 percent in 1993 to 64 percent in 1998. This suggests that policies to promote schooling among minority populations were quite successful. However, minority entrance into lower secondary schools (64 percent) still lagged behind the rates for Kinh (73 percent) and Chinese (83 percent). At the upper secondary level, 472 Economic Growth, Poverty, and Household Welfare in Vietnam enrollment among Kinh increased modestly (from 25 percent to 29 percent), while enrollment for Chinese and ethnic minorities increased dramatically (from 31 percent to 54 percent and from 3 percent to 25 percent, respec- tively). It is particularly surprising that by 1998, ethnic minorities appear to have almost caught up with Kinh in upper secondary enrollments. Data and Methodological Framework The analysis in the remainder of this chapter uses data from the 1998 VLSS. This survey collected extremely detailed information from 6,002 households in urban and rural areas in all regions of Vietnam. In rural areas, a commu- nity questionnaire was used to collect information on community character- istics, including the distance to the nearest schools. The survey is one of the World Bank's Living Standards Measurement Study household surveys, which are described in Grosh and Glewwe (1998). See the appendix to chap- ter 1 for more information on the VLSS. For further details on the 1998 VLSS, see World Bank (2001b). Data The 1998 VLSS collected a large amount of data on education. The education of all household members is recorded in detail, including years repeated and diplomas or certificates obtained. For individuals who had less than an upper secondary education, simple tests of reading comprehension and mathematics computation were administered. In all rural areas, a school questionnaire was used to collect information from primary schools and from lower and upper secondary schools. This allows for an analysis of the relationship between household and school characteristics and various educational outcomes, including test scores. For two reasons, the analysis in the remainder of this chapter examines only rural areas. First, as seen in table 13.3, rural areas lag far behind urban areas, so concern about children leaving school early or, more generally, about children not doing well in school leads to a focus on rural areas, where 80 percent of the Vietnamese population lives. Second, data on communities and local schools in the VLSS are available only for rural areas. To provide some context for the data analysis presented in this chapter, table 13.4 presents basic descriptive statistics on selected variables of interest, with separate figures for the primary school sample and the secondary school sample. The typical Vietnamese child has a mother with about five years of education and a father with about seven years. The nearest primary school is less than one kilometer away, but the nearest lower and upper sec- ondary schools are about two and six kilometers away, respectively. Poor material conditions are the most common complaint about both primary and lower secondary schools, and concerns about crowding and teacher quality are far less common (data on these school problems are from the community questionnaire). An Investigation of the Determinants of School Progress 473 Table 13.4. Descriptive Statistics for Child, Household, and School Variables Primary Lower secondary Standard Standard Variable Mean deviation Mean deviation Mother's years in school 4.69 3.33 4.94 3.28 Father's years in school 6.56 3.55 6.91 3.41 Distance to primary school (km) 0.61 1.00 n.a. n.a. Distance to lower sec. school (km) 2.26 2.65 1.79 1.83 Distance to upper sec. school (km) n.a. n.a. 5.77 4.57 Poor material conditions 0.82 0.38 0.73 0.45 Lack school supplies 0.59 0.49 0.62 0.49 Crowded 0.09 0.29 0.06 0.25 Low teacher quality 0.19 0.39 0.13 0.33 Teachers with credentials (percent) 0.76 0.25 0.86 0.18 Teachers with 5­10 years experience (percent) 0.28 0.18 0.21 0.16 Teachers with 10+ years experience (percent) 0.44 0.25 0.57 0.24 Percent students without desks 0.02 0.06 0.02 0.08 Poor-quality classroom 0.40 0.27 0.30 0.32 Blackboard 0.76 0.24 0.84 0.25 Books per pupil 0.05 0.16 0.17 0.51 School day (minutes) 189.5 17.2 210.5 20.4 Electricity 0.62 0.40 0.86 0.32 Clean water 0.47 0.42 0.66 0.46 Sanitary toilet 0.44 0.41 0.54 0.48 Library 0.26 0.35 0.59 0.48 Multigrade classrooms 0.14 0.30 0.02 0.14 Number of shifts 1.94 0.23 1.75 0.41 n.a. Not applicable. Source: Author's calculations based on the 1998 VLSS. Data from the school questionnaires show that three-fourths of primary school teachers and almost 9 of 10 lower secondary school teachers have the credentials appropriate for their positions. About half of the teachers in both types of schools have more than 10 years of teaching experience. Almost all of the students have desks, and about 80 percent of classrooms have black- boards. The school days are quite short--a little more than three hours for primary school and just three and one-half hours for lower secondary school. This is not surprising, given that both types of schools typically have two shifts. About 62 percent of primary schools and 86 percent of lower secondary schools have electricity. Approximately one-half of both types of schools have clean drinking water and sanitary toilets. About one-fourth of primary schools have libraries, compared with 59 percent of lower 474 Economic Growth, Poverty, and Household Welfare in Vietnam secondary schools with libraries. Finally, multigrade classrooms (more than one grade in the same class) are relatively rare. Methodological Framework Although the education data in the VLSS provide one of the best opportuni- ties to analyze education outcomes in Vietnam, there are formidable methodological problems that limit the inferences that can be drawn from any analysis of them. This chapter focuses on two basic educational out- comes: school completion and academic performance as measured by test scores. The remainder of this section will explain the regression analysis that was done and the limitations of that analysis. First consider school progress, which can be measured in terms of years of schooling completed or level of schooling completed. The equation to be estimated, the determinants of school progress, takes the form: SP = f (Xc, Xh, Xs). In this equation, SP indicates school progress. It is a function (denoted by f ) of Xc, a vector of child characteristics; Xh, a vector of household characteris- tics; and Xs, a vector of school characteristics. If all child, household, and school characteristics were observed and measured without error, standard linear or nonlinear regressions methods, such as ordinary least squares, could be used to estimate this relationship. In reality, it is never possible to observe all components of the X vectors-- although the VLSS measures more than is done in most other surveys--and those components that are observed may be measured with error. For exam- ple, it is difficult, if not impossible, to collect information on children's innate abilities or parents' wishes for their children's schooling. Even more daunt- ing is the collection of data on school characteristics. Schools can vary in hundreds of ways, some of which are difficult to measure. Examples of difficult-to-measure school variables are the quality and motivation of the teachers, the quality (as opposed to quantity) of school supplies and equip- ment, and the pedagogical methods used by teachers. The following simple example illustrates the difficulty of estimating the determinants of school progress. Suppose that the above equation is esti- mated and shows that children living in areas where the schools have ade- quate numbers of desks and other basic equipment are much more likely to be enrolled in school. Although it is probably true that better-equipped schools attract students, other factors that are not observed could also be at work. For example, communities in which the parents place an unusually high value on education are likely to take actions that improve the quality of those schools, so better school equipment is positively correlated with the values held in the community. Because these values are not in the VLSS data, and they almost certainly also have a direct impact on whether children at- tend school, there will be a tendency to overestimate the impact of school equipment; part of the estimated effect of school equipment is in fact due to An Investigation of the Determinants of School Progress 475 the higher value placed on education in communities with ample equip- ment. To make matters worse, it is also possible that the impact of better- equipped schools on school progress is underestimated. This would be the case if there were government programs that provided equipment and sup- plies to areas where school progress is particularly low. The problems with estimating the determinants of school progress also confound attempts to estimate the determinants of children's academic achievement. The equation to be estimated in this case can be depicted as: TS = g(Xc, Xh, Xs) where TS is the student's test score, a different function g( ) is to be esti- mated, and the three sets of X variables may be different from those in the school progress equation. An example of the difficulty here is measuring the impact of textbooks on academic achievement. In Vietnam, as in many other developing countries, schools do not provide textbooks; instead, parents are expected to purchase them for their children. Of course, parents who decide to purchase those textbooks probably place a higher value on education than parents who do not buy textbooks. Moreover, parents who place a high value on education probably do other things to help their children, many of which may not be observed in the VLSS data (such as helping children with schoolwork). When estimating the above equation for test score perfor- mance, standard estimation methods will tend to overestimate the impact of textbooks on learning because textbooks are positively correlated with parental actions that are not observed in the data. A full treatment of the difficulties involved in estimating the determi- nants of school progress and academic achievement in Vietnam is beyond the scope of this chapter. See Glewwe (2002) for a thorough review of the problems encountered when estimating the determinants of academic achievement in developing countries. The point to bear in mind is that the estimates in the next section should be treated as suggestive only, not as definitive. Results This section presents estimates of the determinants of school progress in terms of level of schooling completed, and of academic achievement, as measured by test scores and examination results. As explained above, the re- sults may suffer from biases and thus should be treated with caution; further analysis is needed before using these results to provide policy advice. School Progress As seen in table 13.1, almost all Vietnamese children (97 percent) enter pri- mary school, so there is little point in estimating the determinants of entry into primary school. However, only about 80 percent finish primary school. Table 13.5 presents logit regressions for finishing primary school. Because n.a. 1.15 0.50 3.33 3.55 0.47 0.35 0.39 0.31 6.7 0.41 1.00 2.65 0.38 0.49 0.29 Standard deviation 1 1.00 Mean 17.45 0.51 4.69 6.56 7.72 0.15 0.18 0.1 0.21 0.61 2.26 0.82 0.59 0.09 155.4 statistict 2.07- 3.37- 0.47- 5.06 5.75 5.62 0.05- 0.56- 0.30 0.10 0.75- 1.15 1.21- 1.70- 0.94 1.72- variables Adding 1 15 school 6.772 0.21- 0.091- 0.149 0.171 1.302 0.015- 0.1- 0.070 0.001 0.127- 0.083 0.058- 0.372- 0.203 0.533- Coefficient 1 n.a. statistict 3.69- 0.12- 3.57 5.45 7.65 0.90- 1.38 0.05- 1.1 0.36- n.a. n.a. n.a. n.a. n.a. effects Community fixed 19 n.a. 0.239- 0.021- 0.1 0.159 1.703 0.304- 0.292 0.017- 0.015 0.067- n.a. n.a. n.a. n.a. n.a. Coefficient 5.88 6.54 5.35 0.68 n.a. n.a. n.a. n.a. n.a. Completion statistict 2.88- 3.06- 0.01 1.50- 0.66- 0.99 0.73- model School Base 11 7.254- 0.185- 0.001 0.183 0.182 1.163 0.355- 0.121- 0.245 0.009 0.1- n.a. n.a. n.a. n.a. n.a. Coefficient Primary of capita) school school per secondary Determinants in in e conditions primary lower supplies years years to (km) to (km) 13.5. 's 's minority (cm) missing material school bleaT (Logit) ariableV school school owded Constant Age Female Mother Father Log(expenditur Ethnic Buddhist Christian Height Height Distance Distance Poor Lack Cr 476 0.39 0.25 0.18 0.25 0.06 0.27 0.24 0.16 17.2 0.40 0.42 0.41 0.35 0.30 0.23 0.18 0.10 n.a. n.a. variation. 0.19 0.76 0.28 0.44 0.02 0.40 0.76 0.05 0.62 0.47 0.44 0.26 0.14 1.94 0.04 0.01 n.a. n.a. 189.5 uster 0.38- 2.72 0.85 1.65 2.46- 1.51 1.01- 0.06- 0.26 0.68 0.03- 1.25- 1.14- 0.72- 0.54- 0.35- 0.01 within-cl by o 1,939 N solely 0.106- 0.819 0.649 0.794 0.025- 0.569 0.446- 0.041- 0.002 0.193 0.007- 0.338- 0.334- 0.203- 0.201- 0.141- 0.007 identified is n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. -- which 1,639 seY ession, egrr n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. second the except n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. design, 1,983 No sample VLSS. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. ed 1998 cluster the on for based desks adjusted ors years fects? years cent) oom language ooms ef err edentials + cent) without teaching d calculations cr 5­10 's quality (per 10 (per classr fixed pupil (minutes) classr shifts ethnic minority with with with d standar students toilet of in applicable. Author cent) per day water language size All teacher Not ce: -quality ethnic cent Low eachersT (per eachersT experience eachersT experience Per Poor Blackboar Books School Electricity Clean Sanitary Library Multigrade Number achingeT × n.a. Note: Sour Ethnic Sample Community 477 478 Economic Growth, Poverty, and Household Welfare in Vietnam some children start primary school one or two years late, and grade repeti- tion is common, it is often not clear whether a particular child will complete primary school until he or she is 16 years old. Thus, the regressions in table 13.4 are based on outcomes for children ages 16 to 19. The dependent vari- able is one if the child has finished primary school and zero otherwise. As explained above, these regressions are for rural areas only, because there are data on school characteristics only in those areas. The results shown in table 13.5 are based on a sample of 2,075 children ages 16 to 19. Some of these children have already moved away from home, but any problems with sample selection are avoided because the VLSS data include information on children of household members who no longer live with their parents (including questions on their schooling).4 Of the 2,075 children, 92 were dropped because of missing data on one or more child and household variables (the most common missing variable was father's edu- cation), so the estimates are based on a sample of 1,983 children. For about 21 percent of the children, data on their height were missing; to retain them, their heights were set at the average height, and a dummy variable was created to indicate children whose height data were missing. The first two regressions in table 13.5 focus on child and household char- acteristics. The variables included are the age, sex, and height of the child; the years of schooling of each parent; household expenditures per capita; and dummy variables indicating whether the household is an ethnic minority, Buddhist, or Christian. The means and standard deviations of all variables are shown in the last two columns of the table. In the first regression, the impacts of these variables are generally as ex- pected. Age has a negative impact on primary school completion. This indi- cates that older children did not go as far in school as younger children, which is consistent with the trends in school enrollment shown in table 13.3. The "female" dummy variable has no impact at all, indicating no gender dif- ferences at the primary school level. The years of schooling of both parents are strongly and positively associated with primary school completion. This probably reflects two distinct effects: Better-educated parents are more able to help their children with their schoolwork, and they are also likely to place a higher value on education. Another strong effect is that of household expenditure per capita; as ex- pected, it has a strong and significantly positive impact. The magnitude of this impact can be measured in terms of its effect on the probability of attend- ing primary school. Consider a child who is average in all of his or her char- acteristics. Such a child would have an average (per capita) expenditure level that would put him or her in the middle of the distribution of per capita expenditures--that is, in the middle of quintile 3 (see table 13.3). If the per capita expenditure of the child's household dropped to one standard devi- ation below the mean, he or she would be in quintile 1, albeit near the top of that quintile. The estimated parameter on per capita expenditures implies that the probability of finishing primary school would drop by 7.7 percentage points (recall that 80 percent of Vietnamese children finish primary school). An Investigation of the Determinants of School Progress 479 Similarly, for an average child whose (log) per capita expenditure increases from average to one standard deviation above average (which would put him or her near the top end of quintile 4), the probability of finishing primary school is predicted to increase by 5.3 percentage points. Another way of depicting these impacts is to consider that a child who is average in every re- spect, except that his or her household per capita expenditure rises from quin- tile 1 to quintile 4, is 13 percentage points more likely to finish primary school. Children from ethnic minority households are somewhat less likely to complete primary school, even after controlling for parents' schooling and income (expenditures), but the statistical significance of this effect is weak (t value of 1.50). This suggests that minority children face barriers even after income and parental education are controlled for (this will be dis- cussed further later in this section). The "Buddhist" dummy variable is neg- ative but insignificant, and the "Christian" dummy variable is positive but also insignificant. A final variable of interest is child height. In general, height is partially determined by nutritional status in early childhood. Many Vietnamese chil- dren are stunted (see chapter 10 in this volume), and child malnutrition has been linked to poor school performance in other developing countries (see Glewwe, Jacoby, and King [2001] for the Philippines). The coefficient on the height variable is positive, as expected, but not statistically significant, and the coefficient is very small; an increase of one standard deviation in height raises the probability of finishing primary school by less than 1 percentage point. The coefficient on the dummy variable indicating no data on height is also insignificant, which indicates no systematic differences between chil- dren with and without height data. (Recall that the height variable is set to the mean for children without data on height.) The estimates in the first column of table 13.5 could be biased because they include no variables pertaining to local schools. For example, part of the impacts of expenditures per capita and parental education could be due to better-off, better-educated communities having higher-quality schools. To check for such bias, the second set of results in table 13.5 includes community fixed effects, which essentially control for variation in local schools and in all other community characteristics. In general, there is little change in the im- pacts seen in the first regression.5 The impacts of parental education decline somewhat, and the impact of household expenditure increases by about 50 percent. This suggests that part of the impact of parental education is in fact due to differences within communities, and the impact of household ex- penditures is partially masked by variation in community characteristics. The analysis so far has excluded school characteristics. Yet the impact of school characteristics is of particular interest because governments have much more control over school characteristics than they do over child and household characteristics. However, regressions that include school charac- teristics could suffer from serious biases, as explained in this chapter's sec- tion on data and methodological framework, so the results must be treated with caution. These results are shown in the third regression in table 13.5. 480 Economic Growth, Poverty, and Household Welfare in Vietnam Note that, compared with the first regression in table 13.5, the sample size is slightly lower (1,939 instead of 1,983); 44 observations were dropped because of missing data on local schools. Before turning to the impact of the school variables, it is worthwhile to check whether the effects of the child and household variables in the first re- gression were altered by the addition of school variables. In almost all cases, the answer is no. The impacts of the religion variables change somewhat but are still completely insignificant. The only change of interest is that the mar- ginally significant negative impact of the ethnic minority variable is much smaller and completely insignificant. This suggests that much of the nega- tive impact of the ethnic minority variable in the first two regressions is due to differences in the characteristics of schools available to ethnic minorities and Kinh children. The school variables used in table 13.5 represent almost all of the infor- mation on local primary schools contained in the school questionnaire that was a part of the 1998 VLSS. The two exceptions to this are that variables relating to school cost and the student-teacher ratio have not been used. Regarding the first exception, primary schools in Vietnam do not charge tuition, and this is reflected in the data. However, there are reports that some schools charge fees that families must pay. Unfortunately, because such pay- ments are contrary to government policy, it is doubtful whether they are reported accurately in the VLSS data. A variable of particular interest was a dummy variable indicating whether the school reduced or eliminated fees for particular households, such as poor households, ethnic minority house- holds, and so on. Similar data were collected in the household question- naire, but these data were not closely correlated with the data in the school questionnaire. Thus, this variable was also dropped from the analysis. Finally, the student-teacher ratio was dropped because it is clearly endoge- nous; any unobserved school characteristic that makes schools more attrac- tive is likely to increase this ratio, which implies that including it as an explanatory variable will almost certainly underestimate any negative im- pact that it may have on school quality.6 Returning to the third regression in table 13.5, begin with the two dis- tance variables. The distance to the nearest primary school has an unex- pected positive impact, but it is not statistically significant. This statistical insignificance is not surprising because every community (commune) in the VLSS data has a primary school, so the average distance from the house- hold's village to the nearest school was only 0.6 kilometer (the longest distance was only 5 kilometers). Although the distance to the nearest lower secondary school is longer (an average of 2.3 kilometers), and this does have the expected negative impact, it is not statistically significant (t statistic of ­1.21). This suggests that the distances to primary and lower secondary schools have little effect on whether children finish primary school. The next four school variables are based on data collected in the commu- nity questionnaire, which asked community leaders about problems with local primary schools. Four dummy variables were created that indicate An Investigation of the Determinants of School Progress 481 whether these problems were cited: (a) poor material conditions, (b) lack of school supplies, (c) overcrowding, and (d) low teacher quality. None of these four variables was statistically significant at the 5 percent level, although two were significant at the 10 percent level; in rural communities where schools have poor material conditions or are overcrowded, children were less likely to go to school. Unfortunately, the meaning of the "poor material conditions" variable is rather imprecise. The data in the school question- naire provide more detail; the estimated impacts of those variables are shown in the remaining rows of the table. The rest of the variables in table 13.5 are from the school questionnaire of the 1998 VLSS. The first three pertain to observable characteristics of teach- ers: the percentage of teachers who have the proper credentials to teach pri- mary school (that is, they have completed lower secondary school and have 2 additional years of teacher training), the percentage who have 5 to 9 years of teaching experience, and the percentage who have 10 or more years of teaching experience. The percentage of teachers with proper credentials has a positive and highly statistically significant impact. Approximately 75 per- cent of primary school teachers have these qualifications. The magnitude of this impact can be seen by predicting the change in the probability of finish- ing primary school for an otherwise average child when a school in which only one-half of the teachers have these credentials is modified to become one in which all teachers have them. The predicted probability increases by about 5 percent. In contrast, teacher experience has little effect: There is no significant impact of the proportion of teachers with 5 to 9 years of experi- ence, and the impact of teachers with 10 or more years of experience is barely significant at the 10 percent level (t statistic of 1.65). Although the impact of teachers' credentials suggests that better-quality teachers encourage parents to send their children to school, the policy im- plications may be somewhat more nuanced. In particular, it may be that teachers who obtain credentials are more motivated and thus their motiva- tion is the essential characteristic. If this is the case, providing training to teachers who are currently without credentials may have little effect if the training does not affect their motivation. In fact, it may not be teacher char- acteristics at all but something else about the school (such as a good princi- pal) that leads to more teachers with these qualifications, in which case the key policy lever is choosing the principal, not the teachers. Such alternative scenarios are possible, but it could also be true that teacher credentials do in fact encourage parents to keep their children in school; further analysis is needed to verify or reject this hypothesis. The next four variables refer to basic pedagogical materials in the school. Lack of desks has a negative and statistically significant impact on primary school completion. The size of this impact can again be shown in terms of how changes in it affect the probability that children will finish primary school. About 6­7 percent of children in Vietnam attend schools where 10 percent or more of children do not have desks. Providing enough desks to a school that does not have desks for 10 percent of the children (but is 482 Economic Growth, Poverty, and Household Welfare in Vietnam average in all other respects) is predicted to raise the probability of finishing primary school by 3 percentage points. The other variables--books per pupil, the presence of blackboards, and poor-quality classrooms--have no significant effects; indeed, all of them have unexpected signs. The insignifi- cant effect of the books per pupil variable may reflect the fact that Viet- namese parents are expected to purchase textbooks for their children; the variable used here refers to any books owned by the school that are lent to students, which occurs rarely (the mean of this variable is 0.05). It is less clear why the blackboard and poor-quality classroom variables are insignificant. One possibility is that they are measured with error, which in general leads to underestimation of their true effects. Alternatively, there may be unob- served school characteristics that are correlated with all four of these vari- ables that could be causing biased estimates for all of them. A variable of particular interest in table 13.5 is hours per day that the school is in session. Primary schools in Vietnam have unusually short days. The mean number of minutes that a primary school is in session was 189: 3 hours and 9 minutes (see the mean value reported in table 13.4). It would be very useful to have a reliable estimate of the impact of expanding the school day on school completion and student performance. Somewhat sur- prisingly, the regression in table 13.5 shows no impact at all of hours per day on school completion. Taken at face value, this suggests that parents do not find longer school days more attractive when deciding whether their chil- dren should complete primary school. This may reflect the opportunity cost of children's time, but once again unobserved factors correlated with the length of the school day could be producing biased estimates. The next three school variables in table 13.5 show the availability of some basic amenities--electricity, clean water, and sanitary toilet. The effects of all three on primary school completion are statistically insignificant, providing no evidence that these variables have significant impacts on primary school completion. The last five variables in table 13.5 cover several different topics. For the first three, there is no significant impact of the presence of a library, the pro- portion of multigrade classrooms, and the number of shifts operated by the school. It would seem intuitive that a library should attract students, but in this case the estimate was not only insignificant but also had an unexpected negative sign. Multigrade classrooms (teaching students of different grades in the same classroom, often a sign of a small school) and multiple shifts are often thought to be negative factors. Both did have negative coefficients, but neither coefficient was significant. The last two variables focused on whether teaching in ethnic (non-Kinh) languages attracted students; one variable indicates whether some lessons are taught in ethnic languages and the other interacts this variable with the "ethnic minority" dummy variable. Both variables were completely insignificant, suggesting that this policy did not help ethnic minority students complete primary schooling. As pointed out in table 13.3, rural areas have much lower entrance rates into lower and upper secondary education than do urban areas. Vietnamese An Investigation of the Determinants of School Progress 483 policymakers would like to understand why this occurs. This is examined in table 13.6, which presents estimates of the determinants of completion of lower secondary education in rural areas of Vietnam. The sample consists of all individuals ages 20 to 23 who have completed primary school (and thus have had the opportunity to enter lower secondary school). This age range was used because many children in their late teens are still enrolled in lower secondary school (because of late entry into primary school, grade repeti- tion, or both); thus, it is not yet known whether they will complete lower secondary school. The first regression in table 13.6 includes only child and household vari- ables. As in the primary school completion regressions, the signs of most of these variables are consistent with what one would expect. One surprise is that age has no significant impact. This may reflect the sample consisting of persons ages 20 to 23; many of these individuals finished lower secondary school several years before the 1998 survey was implemented, so the recent trends seen in table 13.2 would not apply. Most of the remaining child and household variables in the first regres- sion in table 13.6 yield unsurprising results. Female children are somewhat less likely to complete lower secondary school, but this impact is statistically significant only at the 10 percent level. The mother's schooling has a positive effect, but it is not quite significant at the 10 percent level (t statistic of 1.60). The father's schooling has a strong and statistically significant positive effect, as does per capita expenditure. Unlike finishing primary school, there is no significantly negative impact of being a member of an ethnic minority group--indeed, the point estimate is positive, although not at all significant. There is no significant effect of being Christian, but being Buddhist has a strongly significant negative impact on the probability of completing lower secondary school. Finally, child height has a positive sign, but again it is far from being statistically significant. The second set of results in table 13.6 uses community fixed effects to reduce bias caused by correlation between household characteristics and community characteristics. The impacts of three of the variables change. First, the negative effect for girls is halved and thus becomes completely insignifi- cant. Second, the impact of father's years of schooling is reduced by more than one-half and consequently is not even significant at the 10 percent level (t statistic of 1.60). Third, the strongly negative impact of being Buddhist is reduced by more than one-half and is no longer statistically significant. These results suggest that some of the effects seen in the first set of results reflect correlation between those variables and some community characteristics. The third set of results in table 13.6 adds the same school variables used in the analysis of primary school completion.7 The distances to the nearest lower secondary school and to the nearest upper secondary school both have negative impacts, as would be expected, but neither impact is statisti- cally significant. Thus, access to secondary schools, at least as determined by distance, has no discernible effect on finishing lower secondary school. Only one of the four types of school problems identified in the community 1 n.a. 1.1 0.50 3.28 3.41 0.44 0.27 0.38 0.28 6.34 0.50 1.83 4.57 0.45 Standard deviation 1.00 Mean 21.48 0.49 4.94 6.91 7.86 0.08 0.17 0.09 0.53 1.79 5.77 0.73 157.1 statistict 2.71- 0.50 1.04- 1.57 3.20 5.37 1.36 3.68- 0.65 1.08 0.62 1.29- 0.22- 1.43- Coefficient 1.3161- 0.030 0.262- 0.047 0.092 1.075 0.470 0.668- 0.195 0.021 0.096 0.064- 0.004- 0.271- n.a. statistict 0.17- 0.73- 1.56 1.60 4.37 1.13 1.45- 1.50 0.87 0.55 n.a. n.a. n.a. 1 n.a. 0.01- 0.193- 0.051 0.049 1.009 0.625 0.392- 0.570 0.018 0.088 n.a. n.a. n.a. Completion Coefficient School statistict 2.41- 0.07- 1.70- 1.60 4.76 4.78 0.51 4.61- 0.33- 0.53 1.55 n.a. n.a. n.a. Secondary 8.426- 0.004- 0.400- 0.042 0.125 0.876 0.131 0.836- 0.092- 0.009 0.242 n.a. n.a. n.a. Coefficient Lower of capita) school school per secondary secondary Determinants in in e conditions lower upper years years to (km) to (km) 13.6. 's 's minority (cm) missing material bleaT (Logit) ariableV school school Constant Age Female Mother Father Log(expenditur Ethnic Buddhist Christian Height Height Distance Distance Poor 484 0.49 0.25 0.33 0.18 0.16 0.24 0.08 0.32 0.25 0.51 20.4 0.32 0.46 0.48 0.48 0.14 0.41 n.a. n.a. variation. 0.62 0.06 0.13 0.86 0.21 0.57 0.02 0.30 0.84 0.17 0.86 0.66 0.54 0.59 0.02 1.75 n.a. n.a. 210.5 uster 2.32 0.12 0.77- 1.04 0.83- 2.36 0.07 2.09- 0.29 0.22 1.69- 1.29 1.24- 0.10 0.32 2.56- 0.81- within-cl by o 961 N solely 0.402 0.055 0.166- 0.371 0.584- 1.141 0.001 0.696- 0.124 0.032 0.008- 0.287 0.259- 0.018 0.064 0.939- 0.175- identified is which n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. ession, 940 seY egrr n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. second the except n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. design, 1,072 No sample VLSS. ed n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. 1998 cluster the on for based desks adjusted ors years fects? years cent) oom ooms ef err edentials + cent) without d calculations cr 5­10 (per 10 's supplies quality (per classr fixed pupil (minutes) classr shifts with with with d standar students toilet of applicable. size Author school cent) per day water All teacher Not ce: -quality owded cent Lack Cr Low eachersT (per eachersT experience eachersT experience n.a. Note: Sour Per Poor Blackboar Books School Electricity Clean Sanitary Library Multigrade Number Sample Community 485 486 Economic Growth, Poverty, and Household Welfare in Vietnam questionnaire--poor material conditions, lack of school supplies, over- crowding, and low teacher quality--has a statistically significant impact on completion of lower secondary school; lack of school supplies has an unex- pected positive effect that is statistically significant at the 5 percent level. One possible interpretation of this result is that parents who care more about their children's education, and thus whose children are more likely to be in school, are more vocal about problems at their local schools. Of the variables in the school questionnaire, one of the three teacher vari- ables is strongly significant: The presence of teachers with 10 or more years of experience significantly increases the probability of students finishing lower secondary school. The magnitude of this impact implies that improv- ing a school where currently only one-half of the teachers have such experi- ence to the point that three-fourths of the teachers have that experience will increase the probability of finishing lower secondary school by 6.9 percent- age points. Yet even though this suggests that teacher quality is important, it is possible that unobserved aspects of teacher quality are the real key factors, as explained above. Only two of the remaining school variables are statistically significant at the 5 percent level. The percentage of classrooms that are of poor quality has a strongly negative impact on completion of lower secondary school, and the same is true of multigrade classrooms. The impact of poor-quality class- rooms implies that a lower secondary school with 30 percent of its class- rooms of poor quality (which is true of about one-third of lower secondary schools) would increase the probability of a student finishing that level of schooling by 5 percentage points if it repaired all of those rooms. In compar- ison, the negative impact of multigrade classrooms is huge: Although only about 2 percent of students are in such schools, they reduce the probability of completing lower secondary school by 23 percentage points. A final, somewhat unexpected result in table 13.6 is that longer school days reduce the probability of completing lower secondary school, although this impact is significant only at the 10 percent level. At a minimum, it suggests that increasing the length of the school day will not attract more students. A final issue to consider regarding school progress in Vietnam is the role of child labor. In general, children who leave school usually begin to work for the family, either for the household farm or business or, less often, for wages. Although it is natural to ask whether child labor "causes" children to drop out of school, one could also ask whether dropping out of school "causes" child labor. In fact, leaving school and starting work are really two comple- mentary parts of a single decision, and it is misleading to think that one part causes the other. Whatever its ultimate causes, child labor is an important issue in Vietnam. For a detailed discussion of child labor, see chapter 14. Academic Achievement The remaining regressions in this chapter investigate the determinants of academic achievement in Vietnamese schools, measured in two ways: scores on mathematics and reading tests administered to children in primary and An Investigation of the Determinants of School Progress 487 lower secondary schools and self-reported academic achievement. The mathematics and reading test scores come from brief tests administered as part of the 1998 VLSS. The math test consisted of four problems: adding two two-digit numbers, subtracting a one- or two-digit number from a two-digit number, multiplying a two-digit number by a one-digit number, and divid- ing a two-digit number by a one-digit number. The interviewer then recorded the outcome as one of three codes: can do without difficulty, can do but with difficulty, or cannot do. The reading test consisted of reading two sentences in Vietnamese, with the interviewer recording the result using the same three codes.8 These tests were not administered to people with an upper secondary or higher education, because it was thought that anyone with such an education would be able to do them without difficulty (this is confirmed below in the discussion of table 13.10). Finally, the self-reported academic achievement question is the response given to the following ques- tion (asked only of household members currently in school): "What score did you achieve on the final exams taken in your previous grade?" Four re- sponses were available, ranging from "poor" to "excellent." Table 13.7 begins by presenting ordinary least squares estimates of the determinants of reading test scores in primary schools. The sample of 3,260 consists of all children ages 6 to 13 who are currently enrolled in primary school. The first set of estimates uses only household and child characteris- tics. Most of the estimated impacts have the expected signs. Children in higher grades have higher scores, as do children with better-educated fa- thers and mothers. There are no significant impacts of age (after controlling for grade) and sex. Children from better-off households do somewhat better, but the impact is not statistically significant (t statistic of 1.40). Ethnic minority children perform significantly worse, which is not particularly sur- prising because the reading test was in Vietnamese, but there are no signifi- cant differences by religion. Finally, taller children scored significantly higher, which indicates that poor health and nutrition in early childhood have a negative impact on current school performance. The second set of estimates in table 13.7 imposes community fixed effects to control for unobserved differences in the communities in which the chil- dren live. The estimated impacts of most variables change very little, but a few show more sizable changes. First, the impact of parents' education de- clines and loses much of its statistical significance (both mother's and fa- ther's education are now significant only at the 10 percent level). Second, the impact of expenditures per capita doubles and becomes significant at the 1 percent level, indicating that community differences tend to mask its true impact. Finally, the impact of height declines and loses statistical signifi- cance (t statistic of 1.46), and the Buddhist variable has a positive effect that is significant at the 10 percent level. The last set of estimates in table 13.7 adds detailed characteristics of the schools the children attend. These are the same variables used in tables 13.5 and 13.6, except that the four problems cited at the community level are ex- cluded because they are not school specific (and, in fact, when added they are almost never significant). The impacts of the household and child level n.a. 1.64 2.01 0.50 3.46 3.31 0.46 0.39 0.37 0.32 10.7 0.08 0.28 0.20 0.26 Standard deviation 1 1.00 3.32 9.32 0.47 6.04 7.10 7.58 0.19 0.16 0.1 0.01 0.75 0.28 0.43 Mean 124.3 0.54 statistict 13.34 0.57- 0.94 1.76 2.43 2.42 2.71- 1.79 0.24- 2.21 2.17 0.36- 1.08 2.10 0.209 0.218 0.006- 0.018 0.008 0.009 0.069 0.142- 0.060 0.012- 0.004 0.152 0.024- 0.108 0.164 Coefficient n.a. statistict 13.18 0.84- 1.05 1.84 1.91 2.64 2.06- 1.82 0.55 1.45 2.96 n.a. n.a. n.a. School n.a. 0.182 0.009- 0.018 0.007 0.006 0.069 0.134- 0.060 0.024 0.002 0.156 n.a. n.a. n.a. Coefficient Primary in 1.25 2.56 3.28 1.40 2.19 Scores statistict 1.701 1.26- 0.93 3.00- 0.05 0.15- 2.30 n.a. n.a. n.a. estT 0.339 0.185 0.014- 0.017 0.009 0.012 0.036 0.143- 0.002 0.007- 0.004 0.096- n.a. n.a. n.a. Coefficient Reading of capita) years school years school per cent) cent) Determinants in in e edentials + cr 5­10 (per 10 (per years years with with with 13.7. 's 's minority (cm) missing cent) bleaT ariableV Constant Grade Age Female Mother Father Log(expenditur Ethnic Buddhist Christian Height Height eachersT (per eachersT experience eachersT experience 488 The 7.92 0.26 0.31 0.25 19.8 0.47 0.50 0.50 0.47 0.36 0.25 0.20 0.12 n.a. n.a. variation. 2.28 0.43 0.73 0.05 0.67 0.48 0.45 0.32 0.15 1.95 0.04 0.01 n.a. n.a. uster 187.5 within-cl 0.66- 0.23 1.87- 1.33 0.77- 0.96 0.94- 0.29 1.29- 1.59- 1.25- 0.57- 3.67- by o solely 2,675 N 15 1 0.002- 0.012 0.1- 0.089 0.001- 0.034 0.029- 0.01 0.041- 0.082- 0.068- 0.059- 0.442- identified is which n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. ession, 3,260 seY egrr second n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. the except 0.57. of n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. design, deviation 3,260 No d sample VLSS. ed n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. standar 1998 a cluster the for and on 1.70 ethnic of based desks × adjusted ors mean fects? a oom language err without ooms ef teaching d has calculations 's classr (minutes) classr ethnic fixed d pupil shifts standar students toilet of in applicable. variable Author per day water language size All Not ce: -quality cent Per Poor Blackboar Books School Electricity Clean Sanitary Library Multigrade Number achingeT minority n.a. Note: Sour Ethnic Sample Community dependent 489 490 Economic Growth, Poverty, and Household Welfare in Vietnam variables are similar to what they were in the first two regressions, so the focus is on the school variables, which are the most relevant variables for education policy. One of the three teacher variables, the percentage of teachers with more than 10 years of experience, has a significantly positive impact. Yet the size of the impact is not particularly large--increasing the proportion of teachers with this experience from one-fourth to one-half increases the predicted test score (which ranges from one to three) by 0.04 point. Of the classroom equipment variables, none is significant at the 5 percent level. One variable, the proportion of rooms with blackboards, is significant at the 10 percent level, but it has an unexpected negative sign. Indeed, with one exception, none of the remaining school characteristics in table 13.7 is statistically sig- nificant, not even at the 10 percent level. The exception is the interaction term between teaching in ethnic languages and being an ethnic minority student; this combination has a strong and significantly negative impact on students' reading scores. This is not surprising, given that the reading test was in Vietnamese. Table 13.8 examines mathematics scores in rural primary schools in Vietnam. As before, the first set of estimates includes only child and house- hold variables. The results are very similar to those for the reading scores, the only exception being that there is no significantly negative impact of being a member of an ethnic minority. This may simply reflect the fact that ethnic minorities are less familiar with the Vietnamese language but have no particular problem learning basic mathematics. The next set of estimates controls for differences across communities. Most of the results are un- changed, the exceptions being that father's education becomes much less important (and loses statistical significance), the impact of per capita expen- ditures becomes more important (and attains statistical significance at the 1 percent level), and being Buddhist has a positive impact that is statisti- cally significant at the 5 percent level. Note that, in contrast with the results for reading scores, child height retains some statistical significance (at the 10 percent level) when community fixed effects are included. The final set of estimates in table 13.8 adds the same school variables used in table 13.7. Unfortunately from the viewpoint of policy formulation, only one of the variables has a statistically significant impact; all variables are insignificant at the 10 percent level except that the interaction between teaching in ethnic languages and being an ethnic minority has a signifi- cantly negative impact on math test performance. Although there are poten- tial problems concerning biased estimates, these results raise questions about the benefits of teaching in ethnic languages in Vietnamese primary schools, at least the way it is done currently. The reading and mathematics results just examined are based on very simple tests that may not capture overall student learning very well. An alternative is to use students' self-reported end-of-year examination results. This is done in table 13.9 for the same sample of primary school students examined in tables 13.7 and 13.8 (except that 162 observations were dropped page) statistict 1.67- 17.59 0.38- 0.36 1.66 1.81 2.49 0.60- 0.28 0.67- 2.31 0.54 0.40- 0.36 1.13 following on 0.779- 0.349 0.005- 0.008 0.009 0.009 0.090 0.042- 0.012 0.051- 0.005 0.072 0.041- 0.056 0.139 continues Coefficient (table n.a statistict 18.74 0.27- 0.51 1.85 0.78 2.89 1.32- 1.98 0.12- 1.73 0.98 n.a. n.a. n.a. n.a 0.295 0.003- 0.010 0.007 0.003 0.096 0.080- 0.076 0.005- 0.003 0.089 n.a. n.a. n.a. Coefficient School 1 statistict 1.16- 15.01 0.64- 0.12 2.27 2.26 1.03 1.07- 0.61- 0.48- 2.04 0.1 n.a. n.a. n.a. Primary in 1 1 Scores 0.01 0.01 0.039 0.012 n.a. n.a. n.a. estT oefficientC 0.413- 0.301 0.010- 0.002 0.069- 0.028- 0.034- 0.004 Math of cent) (per capita) years school years school per cent) cent) Determinants in in e edentials + cr 5­10 (per 10 (per years years with with with 13.8. 's 's minority (cm) missing bleaT ariableV Constant Grade Age Female Mother Father Log(expenditur Ethnic Buddhist Christian Height Height eachersT eachersT experience eachersT experience 491 For statistict 0.93- 0.93 1.52- 0.30 0.33- 0.26- 1.27 0.68- 0.82- 1.51- 1.04- 0.48- 2.50- o variation. 0.77. 2,675 N of uster 0.003- 0.070 0.146- 0.021 0.001- 0.014- 0.068 0.038- 0.050- 0.121- 0.072- 0.067- 0.457- deviation within-cl d Coefficient by standar a solely and n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. 1.43 statistict identified of is 3,260 seY mean a which has n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. ession, Coefficient egrr variable second dependent the n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. The statistict except 13.6. 3,260 No table design, see n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. sample VLSS. Coefficient ed variables, 1998 cluster the for on explanatory ethnic of based ) desks × adjusted ors fects? oom language err without ooms ef deviations teaching d d calculations continued( 's classr (minutes) classr ethnic fixed d pupil shifts toilet of in applicable. standar standar 13.8. students Author per day water language size All Not and ce: -quality bleaT cent ariableV Per Poor Blackboar Books School Electricity Clean Sanitary Library Multigrade Number eachingT minority n.a. Note: Sour Ethnic Sample Community means 492 page) 0.13 3.43 statistict 4.88- 4.24 2.54 5.89 4.67 2.07- 1.48 0.43- 1.58 0.60 1.32 0.88- 0.45- 0.60 0.89 1.35 2.00 1.07 following on 0.059 0.069 0.079- 0.124 0.012 0.031 0.186 0.088- 0.059 0.020- 0.004 0.088 0.096 0.103- 0.040- 0.002 0.047 0.084 0.138 0.001 continues Coefficient (table n.a. 4.61 statistict 5.31- 4.51 3.44 5.74 3.85 0.57- 1.66 0.21- 0.62 1.22 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. 13 n.a. 0.081 0.079- 0.1 0.020 0.028 0.162 0.025- 0.074 0.013- 0.001 0.122 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. Coefficient School 1.92 4.16 statistict 5.34- 4.32 2.74 5.48 5.05 3.19- 0.73 1.08- 1.59 0.24 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. Primary in Rank 11 0.654 0.081 oefficientC 0.086- 0.1 0.014 0.027 0.191 0.129- 0.027 0.049- 0.004 0.027 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. Examination of cent) (per desks capita) years school years school per cent) cent) oom without Determinants in in e edentials + cr 5­10 (per 10 (per classr (minutes) years years with with with d pupil 13.9. 's 's minority (cm) missing students per day -quality bleaT cent ariableV Constant Grade Age Female Mother Father Log(expenditur Ethnic Buddhist Christian Height Height eachersT eachersT experience eachersT experience Per Poor Blackboar Books School 493 For 1.40 0.37 1.89 1.82 0.77 0.89 statistict 2.07- 1.15- o variation. 0.69. 2,512 N of uster 0.055 0.014 0.066 0.075 0.041 0.058 0.108- 0.122- deviation within-cl d Coefficient by standar a solely and n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. 2.58 statistict identified of is 3,098 seY mean a which has n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. ession, Coefficient egrr variable second dependent the n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. The statistict except 13.6. 3,098 No table design, see n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. sample VLSS. Coefficient ed variables, 1998 cluster the for on explanatory of based ) adjusted ors fects? language ooms ef err deviations teaching d d calculations continued( 's classr shifts ethnic fixed minority toilet of in applicable. standar standar Author 13.9. water language size All Not and ce: ethnic bleaT ariableV Electricity Clean Sanitary Library Multigrade Number achingeT × n.a. Note: Sour Ethnic Sample Community means 494 An Investigation of the Determinants of School Progress 495 because the examination results were missing). As before, the first set of results examines only child and household variables. As one would expect, most of the results are very similar to those in tables 13.7 and 13.8. Grade has a strong positive effect,9 as does parents' schooling. Being a member of an ethnic minority has a negative effect. The two religion variables have no significant effects. Yet some results in table 13.9 differ from those in tables 13.7 and 13.8. Some variables that were insignificant in the previous tables display signifi- cant effects in table 13.9 (with the same sign as in the earlier table). In partic- ular, age has a significantly negative effect, and the "female" dummy variable and per capita expenditures both have significantly positive effects. The significantly negative effect of age has an obvious interpretation; after grade is controlled for, older students are more likely to be repeaters and thus have shown themselves to be weak students. Finally, the height vari- able is not quite statistically significant (t statistic of 1.59), although it was statistically significant in tables 13.7 and 13.8. Overall, the results are not very different, the main difference being that the results in table 13.9 are often more precisely estimated. The second set of results in table 13.9 checks whether controlling for community differences changes the results. There is little difference for most variables, which suggests that (for these variables) the relationships of these household and child variables are not artificially induced by correlation of those factors with community characteristics. The two exceptions are that the ethnic minority variable no longer has a statistically significant negative impact and the height variable loses the little significance that it had. These changes suggest that the differences observed in academic achievement of ethnic minority students are due to unobserved differences in communities with and without large numbers of ethnic minorities, and they cast doubt on the impact of child height (an indicator of past nutritional status) on acade- mic performance. The final regression in table 13.9 adds the same school variables that were used in tables 13.7 and 13.8. On the one hand, if the examination grade variable is more informative than the reading and math tests, one would expect to find more statistically significant impacts. On the other hand, one should keep in mind that the dependent variable is an average over all aca- demic subjects. Of the 16 variables used, 2 were significant at the 5 percent level and 2 more were significant at the 10 percent level. This indicates about the same precision found in the reading test results of table 13.7 (2 variables were significant at the 5 percent level and 1 at the 10 percent level) and more precision than the math results of table 13.8 (1 variable was significant at the 5 percent level and none at the 10 percent level), but it should be remem- bered that the dependent variables are somewhat different. Turning to the significant results in table 13.9, more books per pupil, libraries, and having sanitary toilets appear to improve students' academic performance, but teaching in ethnic languages reduces their performance. These results are all reasonable, but they are very different from those in 496 Economic Growth, Poverty, and Household Welfare in Vietnam tables 13.7 and 13.8. The only consistency is that all three tables indicate that teaching in ethnic languages has negative effects, although even here the re- sults differ (in tables 13.7 and 13.8, the impact is felt only by ethnic minority students, but in table 13.9, the impact is felt by all students). The fragility of these results suggests that caution is needed when trying to infer policy re- sults from this analysis. Perhaps the main conclusion is that teaching in eth- nic languages may cause more harm than good, but further study is needed before changing policies. The last set of results, in table 13.10, examines (self-reported) student ex- amination performance in lower secondary school. It is based on a sample of 1,601 students between the ages of 11 and 16 who were enrolled in lower sec- ondary school at the time of the survey. It is not possible to compare these results with similar results based on the math and reading tests, because in both tests 99 percent of the students received the top score (that is, they could read the two sentences and do the math problems without difficulty), leaving no variation to analyze. The first estimates in table 13.10 examine only household and child vari- ables. The results are very similar to those for primary schools: Grade, par- ents' education (father's but not mother's), being female, and per capita expenditures all have positive impacts, but age and being an ethnic minor- ity member have negative impacts. Height also has a positive impact, although it is significant only at the 10 percent level, and being Christian has a negative impact. Most of these results are similar after controlling for com- munity differences (the second set of results in table 13.10), except that mother's education regains some statistical significance, and per capita expenditures, being an ethnic minority, being Christian, and height lose their significance. These losses of significance suggest that these variables do not have causal effects, but instead are correlated with community vari- ables that have causal effects. In the third set of results in table 13.10, five variables have positive and strongly significant effects on (self-reported) academic achievement in lower secondary schools: presence of a blackboard, electricity, presence of a sanitary toilet, libraries, and multigrade classrooms.10 Of these, the first four make intuitive sense, but it is hard to see why multigrade classrooms have a positive effect. To summarize, most household- and child-level variables have statisti- cally significant effects in the expected direction. There is some evidence, although not very consistent, that children who were malnourished early in life do not go as far or perform as well in school. Several school variables showed statistically significant effects, but these results need to be treated with caution, because they may reflect in part other school characteristics that are more difficult to observe. One particularly interesting result is that there was no evidence that teaching in ethnic languages helps ethnic minor- ity students--indeed, there was some evidence that this had a negative effect--but further study is needed before drawing any policy conclusions. page) n.a. 1.08 1.42 0.50 3.72 3.15 0.45 0.32 0.35 0.30 9.0 0.10 0.21 0.16 0.25 10.72 0.33 0.28 0.40 22.0 Standard deviation following 1 on 1.00 7.31 Mean 13.43 0.46 6.49 7.88 7.76 0.1 0.14 0.10 0.01 0.83 0.21 0.56 3.17 0.32 0.82 0.14 144.7 208.7 continues 2.99 3.36 statistict 5.36- 4.72 1.15 3.84 1.27 2.35- 0.38 2.55- 1.44 0.36- 1.00- 1.04- 1.48- 0.16- 1.60 2.70 0.89- 0.57 (table 1.580 0.088 0.109- 0.151 0.008 0.029 0.059 0.126- 0.020 0.164- 0.004 0.046- 0.090- 0.214- 0.171- 0.000- 0.120 0.189 0.055- 0.001 Coefficient School n.a. 3.85 statistict 6.64- 4.42 1.69 4.31 0.89 1.27- 1.70 0.72- 1.34 0.01 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. 11 17 Secondary n.a. 0.087 0.1- 0.135 0.010 0.028 0.040 0.1- 0.096 0.070- 0.003 0.001 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. Coefficient Lower in 3.45 3.69 statistict 6.50- 4.29 0.78 3.58 3.08 2.99- 0.04- 2.94- 1.86 0.34- n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. Rank 16 1.370 0.089 0.1- 0.127 0.005 0.023 0.127 0.160- 0.002- 0.154- 0.005 0.037- n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. Examination Coefficient of desks capita) years school years school per cent) cent) oom without Determinants in in e edentials + cr 5­10 (per 10 (per classr (minutes) years years with with with d pupil 13.10. 's 's minority (cm) missing students cent) per day -quality cent ableT ariableV Constant Grade Age Female Mother Father Log(expenditur Ethnic Buddhist Christian Height Height eachersT (per eachersT experience eachersT experience Per Poor Blackboar Books School 497 The 0.36 0.48 0.50 0.49 0.19 0.43 n.a. n.a. Standard deviation variation. 0.84 0.63 0.51 0.57 0.04 1.75 Mean n.a. n.a. uster 1 1 within-cl 3.79 statistict 0.1- 2.62 2.1 2.22 1.63- by o solely 1,323 N 19 0.197 0.005- 0.1 0.098 0.174 0.097- identified is Coefficient which n.a. n.a. n.a. n.a. n.a. n.a. statistict ession, 1,601 seY egrr second n.a. n.a. n.a. n.a. n.a. n.a. the Coefficient except 0.63. of n.a. n.a. n.a. n.a. n.a. n.a. statistict design, deviation 1,601 No d sample VLSS. ed n.a. n.a. n.a. n.a. n.a. n.a. standar 1998 a oefficientC cluster the for and on 2.53 ) of based adjusted ors mean fects? a err continued( ooms ef d has calculations 's classr shifts fixed toilet of applicable. standar variable 13.10. size Author water All Not ce: ableT ariableV n.a. Note: Sour Electricity Clean Sanitary Library Multigrade Number Sample Community dependent 498 An Investigation of the Determinants of School Progress 499 Summary and Concluding Comments Vietnam's impressive economic performance in the 1990s was matched by very good performance in education. This chapter has reviewed changes in that decade and used recent household survey data from 1998 to explain school progress and academic performance. This brief concluding section highlights some of the main findings. A comparison of household survey data from the early and late 1990s shows increased enrollment in primary and secondary education. This in- creased participation in schooling is widespread: It is occurring in both urban and rural areas, in all regions, among all income groups, and among all ethnic groups. In fact, not only are ethnic minorities sharing in these educational outcomes, their entrance rates into lower and upper secondary schools are rising much more rapidly than those of the Kinh. Despite this excellent performance, Vietnam is still behind the East Asian "miracle" countries, and more needs to be done to improve educational out- comes. To assist policymakers in understanding the impact of household and school factors on school progress and academic performance, this chap- ter presents regression analyses of these two outcomes. The main findings of interest are: · Much, and perhaps almost all, of the weaker performance of ethnic minority children is due to community factors, as opposed to factors inherent in being an ethnic minority member. · In several regressions, the impact of providing the lessons in ethnic languages is negative, and in no case is the effect ever positive. · There is some evidence that shorter children, who presumably were malnourished in early childhood, do worse in school, but the effects are not strong and often lose statistical significance. Although the findings in this chapter should be of some use to policy- makers, much more remains to be learned. Indeed, the results here could suffer from a variety of econometric problems that can lead to biased esti- mates. Further research is needed before making major policy changes. Esti- mating the impact of government policies on education outcomes is a very difficult task, but a first step has been taken with these results. More steps are needed in the near future, but to do so will require even richer data than are available in the 1998 VLSS. Notes The author would like to thank Hai Anh Da. ng Hoàng for comments and for excel- ? - lent research assistance. 1. These numbers are from the Vietnam Living Standards Surveys (VLSSs), which are discussed in detail later in this chapter. The school entrance rates refer to children ages 11­12, 16­17, and 19­20, respectively. These ranges are most infor- mative about eventual entry into the corresponding levels of schooling, because some children start school at a relatively late age and others repeat certain years of 500 Economic Growth, Poverty, and Household Welfare in Vietnam schooling. For example, some children ages 14 and 15 are still in primary school, and thus it is not clear whether they will enter lower secondary school. 2. The 1992­93 VLSS spanned a full year, starting in October 1992; the 1997­98 survey began in December 1997 and also lasted one year. For brevity's sake, refer- ence is made to the surveys as the 1993 VLSS and 1998 VLSS, respectively, in this volume. 3. The seven regions are the Northern Uplands, Red River Delta, North Central Coast, Central Coast, Central Highlands, Southeast, and Mekong Delta. 4. Including children who no longer live with their parents implies that chil- dren in the sampled households who are not living with their parents must be dropped, because retaining them would lead to double counting for such children. Thus, the sample used "locates" children according to where their parents are currently living, not where the children currently reside. 5. Fixed effects logit estimates drop communities in which all children finished primary school or no children finished primary school. This reduced the sample size from 1,983 to 1,639. When an ordinary logit is estimated using this smaller sample, the results are very close to those shown in the first regression in table 13.5, which implies that any differences between the first and second set of results in that table are not due to differences in the sample. 6. In regressions that did include this variable (not shown here), it had no statis- tically significant impact on school progress. 7. The two variables on lessons in other languages were dropped because only one individual in the sample was an ethnic minority member who lived in a com- munity that provided lessons in ethnic languages. 8. The two sentences are: "The household living standards survey is of use to your country and benefits your family. Survey materials will be kept absolutely con- fidential and will not be used for other purposes except for serving as a basis for the government to research and build socioeconomic policies to stabilize and improve the lives of the people, including the interests of every family." 9. The statistical significance of the grade variable in table 13.9 is lower than in tables 13.7 and 13.8, but this difference is almost certainly due to the fact that the tests in tables 13.7 and 13.8 were the same across all grades, but the criteria used in the end-of-year examinations are presumably more advanced for higher grades. 10. The ethnic language variables were dropped because no ethnic minority members were found in the few schools that indicated that they provided instruction in ethnic languages. Bibliography Barro, Robert. 1991. "Economic Growth in a Cross-Section of Countries." Quarterly Journal of Economics 106(2): 407­43. Glewwe, Paul. 2002. "Schools and Skills in Developing Countries: Educa- tion Policies and Socioeconomic Outcomes." Journal of Economic Literature 40(2): 436­82. Glewwe, Paul, Hanan Jacoby, and Elizabeth King. 2001. "Early Childhood Nutrition and Academic Achievement: A Longitudinal Analysis." Journal of Public Economics 81(3): 345­68. An Investigation of the Determinants of School Progress 501 Glewwe, Paul, Stefanie Koch, and Bui Linh Nguyen. 2004. "Child Nutrition, Economic Growth, and the Provision of Health Care Services in Vietnam in the 1990s." In P. Glewwe, N. Agrawal, and D. Dollar, eds., Economic Growth, Poverty, and Household Welfare in Vietnam. Washington, D.C.: World Bank. Grosh, Margaret, and Paul Glewwe. 1998. "Data Watch: The World Bank's Living Standards Measurement Study Household Surveys." Journal of Economic Perspectives 12(1): 187­96. Lucas, Robert. 1988. "On the Mechanics of Economic Development." Journal of Monetary Economics 22(1): 3­42. Mankiw, N. Gregory, David Romer, and David Weil. 1992. "A Contribution to the Empirics of Economic Growth." Quarterly Journal of Economics 107(2): 407­37. UNESCO (United Nations Educational, Scientific, and Cultural Organiza- tion). Various years. Statistical Yearbook. Paris. Viet Nam Living Standards Survey (VNLSS). 1992­1993. Web site: www. worldbank.org/lsms/country/vn93/vn93bid.pdf. Vietnam Living Standards Survey (VLSS). 1997­98. Web site: www. worldbank.org/lsms/country/vn98/vn98bif.pdf. World Bank. 1993. The East Asian Miracle. Washington, D.C.: World Bank. ------. 2001a. World Development Report 2000/2001: Attacking Poverty. New York: Oxford University Press. ------. 2001b. "Vietnam Living Standards Survey (VLSS), 1997­98: Basic Information." World Bank, Poverty and Human Resources Division. Washington, D.C. ------. 2002a. World Development Indicators 2002. New York: Oxford University Press. ------. 2002b. World Development Report 2002: Building Institutions for Markets. New York: Oxford University Press. Part IV Other Topics 14 Child Labor in Transition in Vietnam Eric Edmonds and Carrie Turk Child labor1 is endemic in most of the world's poorer countries. As a re- sponse to chronic poverty and idiosyncratic shocks, poor children around the world are withdrawn from school, if they are attending, and required to make an economic contribution to the household. This may have a positive effect, in allowing the household and children within the household to maintain essential basic consumption in times of real hardship. A moderate amount of work in safe conditions can allow children to develop useful skills and a sense of responsibility. Child labor may also have negative effects, diminishing a child's human capital accumulation, creating an en- during poverty trap (potentially for generations), and exposing children to harmful situations that restrict their physical, psychological, and emotional development. There are clearly documented problems in outlawing all forms of child labor (Crawford 2000). Such prohibitions, if enforced, can cause severe hardship for households that are barely surviving, as well as drive children's economic contributions underground into illegal and exploitative areas of work. At the same time, there is also a need to safe- guard children from abuses, to protect them from harmful situations, to ensure their education, and to uphold their basic rights as children. The incidence of child labor appears to be negatively correlated with liv- ing standards. Using a cross-section of countries from 1995, Krueger (1997) shows that child labor virtually disappears once a country's gross domestic product (GDP) per capita reaches US$5,000. He finds that 80 percent of the international variation in child labor can be explained by GDP per capita alone. Vietnam does not appear to be an exception to this relationship. Driven by rural and other reforms in the late 1980s and early 1990s, Vietnam enjoyed rapid economic growth of more than 6 percent per year over the last decade. This in turn generated impressive reductions in the incidence of poverty, with the poverty headcount falling from 58 percent to 37 percent between 1993 and 1998 (World Bank and the Poverty Working Group 1999). Edmonds (2004) documents that the probability that a child (ages 6 to 15) 505 506 Economic Growth, Poverty, and Household Welfare in Vietnam works in agriculture, a family-operated business, or wage employment dropped by 28 percent between 1993 and 1998. He shows that most of the decline in child labor for rural households below the poverty line in 1993 can be explained with improvements in household economic status.2 Not all households benefit equally from improvements in living stan- dards. This chapter explores in detail the decline in child labor that Vietnam experienced during the 1990s and documents the heterogeneity across households in both levels of child labor and the incidence of this decline in child labor. The primary aim of this chapter is to develop a set of indicators to help direct policy toward children who remain vulnerable despite general improvements in living standards. Even after controlling for time-invariant household characteristics, sub- stantial heterogeneity across households is observed in the amount by which child labor declined in Vietnam in the 1990s. Decreases in the proba- bility that children participate in any type of economic activity have been largest in provincial towns, minor cities, the Southeast, and the rural Mekong Delta. Declines in the fraction of children working have been the smallest in urban areas, the Central Coast, and the Central Highlands. In addition to geographic indicators, other observable household charac- teristics are associated with variation in the decline in the probability that a child works. Children from ethnic minorities appear to work more than Kinh and Chinese children, but most of this additional work can be ex- plained by time-invariant household characteristics. Ethnic minorities con- stitute 14 percent of the population of Vietnam, but represent 29 percent of the poor. They have less access to infrastructure, social services, and other resources (Baulch and others 2004). Girls experience smaller reductions in child labor than do boys. Older children experience greater reductions, but that appears to be because older children were more apt to work in 1993. Both a past migration history and the actual departure of a household head are associated with higher levels of child labor, and migrant households experience smaller declines in child labor than do nonmigrant households. The creation of a new home enterprise seems to be associated with smaller declines in child labor, although households that had a home enterprise in 1993 experienced larger reductions in child labor than other households. Taken together, the results of this chapter paint an optimistic picture for child labor in Vietnam. Children were doing better in 1998 than in 1993. Al- though there is substantial heterogeneity across households and regions in the amount of reduction in child labor, this chapter does not find any identi- fiable group outside the Central Highlands that did not experience a decline in child labor between 1993 and 1998. There is still ample scope for policy to help improve the well-being of children, and there are groups of children who remain vulnerable even in the context of rising living standards. Some of the worst forms of child labor are not easily captured with household surveys. Nevertheless, for the average Vietnamese child represented in the Vietnam Living Standards Survey (VLSS), there is every reason to be opti- mistic about his or her future. Child Labor in Transition in Vietnam 507 The next section outlines recent trends in child labor in Vietnam and con- siders these trends in their economic and policy environment. The section fol- lowing examines how Vietnam's remarkable decline in child labor varies with observable household characteristics, including child age and gender, household location, living standards, migration history, enterprise status, and ethnicity. The chapter concludes with a summary of findings and a dis- cussion of what these findings imply for the future of child labor in Vietnam. The Child Labor Environment This chapter relies on the VLSS. There have been two nationally representa- tive rounds of the VLSS. The first round, conducted in 1992­93, interviewed 4,800 households, collecting data on a wide variety of household character- istics and activities. The second round took place in 1997­98 and followed a similar questionnaire and field design.3 The 1998 VLSS was designed to be a nationally representative, stand-alone, cross-sectional survey sampling 5,999 households, but it also revisited 4,305 households from the 1993 VLSS. When the analysis is based on nationwide comparisons, the two rounds of the VLSS are treated as separate, nationally representative (when appropri- ately weighted), cross-sectional surveys. The descriptive work is completed with regression analysis, where the sample is limited to the panel house- holds that appear in both rounds of the survey. Recent Trends in Child Labor There are limitations associated with using the VLSSs to investigate patterns and trends in child labor. First, some of the most exploitative forms of child labor, such as child prostitution, are likely to be hidden because they are illegal. Second, the VLSSs collected little information on working condi- tions. Whether or not work is harming the development of a child lies partly in the nature of the work and the exposure to physically hazardous or psy- chologically stressful conditions, or both. Because the VLSSs did not attempt to document working conditions and the data on hours worked in agricul- ture are not comparable between surveys, the quantitative analysis focuses on participation in work rather than working conditions. The analysis is supplemented by drawing on a growing body of qualitative studies that ex- amine issues around child labor. Also, some of the children who are laboring are likely to be hidden. Street children,4 for example, are often not part of households and are therefore likely to be omitted from household surveys. Households of unregistered migrants are less likely to be included in the VLSS (World Bank and the Poverty Working Group 1999), though studies suggest that children from such households are more likely to work for a living (Save the Children Fund [SCF] U.K. 1999). The VLSSs may also have missed the labor activities of children who have left their household. Children who have been "trafficked" overseas are 508 Economic Growth, Poverty, and Household Welfare in Vietnam very likely to be working, but because they no longer reside in Vietnam, they will not show up in household survey data. Similarly, case study liter- ature documents children leaving their parental home to stay and work with other families for certain periods, either in exchange for board, lodging, and education or to work for a wage as a domestic helper (SCF U.K. 1997; Vietnam- Sweden Mountain Rural Development Program [VN-Sweden MRDP] 1999). The work of these children may not be adequately captured by the data be- cause the children very often remain both unregistered in the host family (survey respondents may not consider the child when listing household residents or members) and absent from the family home (SCF Sweden and University of Social Sciences and Humanities [USSH] 2000). Fortunately, it is possible to assess the scale of this missing children prob- lem with the VLSS data. The 1993 survey collected a household roster of all individuals in the household at the time of the survey, and the 1998 survey asked about the location of each of those members. Table 14.1 reports the status in the 1998 VLSS of the 6,003 children ages 0 to 10 years who were in panel households during 1993. Children ages 10 and under in the 1993 survey were between 5 and 15 in the 1998 survey and are thus included in this chapter's data analysis. Ninety-two percent of children ages 0 to 10 years in 1993 reappeared in the 1998 round of the survey. Of the missing 8 percent, 42 children had died. Slightly more than half of the dead are boys, and 87 per- cent of the surviving, out-of-residence children departed the household when their family moved. Of the remaining 56 children, 31 (55 percent) are Table 14.1. The Out-Migration of Children from Panel Households Category Number of children Children ages 10 and under in 1993 VLSS (panel households): 100 percent 6,003 Children still in household in 1998 VLSS: 92 percent 5,540 Children in 1993 VLSS not in residence in 1998 VLSS ("missing"): 8 percent 463 Of missing, those who died: 9 percent 42 Missing who survived 421 Of missing who survived: children who moved with their families 365 Missing who survived but did not move with their families 56 Of missing who survived but did not move with their families: girls who married 11 Of missing who survived but did not move with their families: boys who married 9 Missing who survived but did not move with their families or marry 36 Source: Author's calculations using the 1993 and 1998 VLSSs. Child Labor in Transition in Vietnam 509 female, and 11 of these females left the household for marriage. Nine of the 25 males (45 percent) left the household for marriage. Thus, of 6,003 sampled children between the ages of 0 and 10 in 1993 in panel households, a total of 36 appear to have left the household for reasons other than death, parental movements, or marriage. The reasons given for migration of these 36 children are evenly split among employment, schooling, and other. Nine of the 20 girls report leaving home for employment, whereas only 3 of the 16 boys do. However, 8 of the 16 boys report leaving for "other" reasons (only 4 girls report "other"). Thus, although departing from one's household for work is undoubtedly an im- portant event in the lives of those children being sent away, this experience does not appear to be an integral part of the childhood experience of either the average boy or the average girl and is unlikely to substantively alter the conclusions. The VLSS surveys present several ways to define child labor. These defi- nitions are presented in table 14.2. For each household member age six and older, the VLSS asked whether the person works for pay outside the house- hold ("works for wages outside household"), works for the household in agriculture ("works in agriculture for household"), and works for the house- hold in self-employment or a household-run business ("works in business for household"). These three work categories are referred to collectively in this chapter as "traditional work." The survey also asked whether a person performs household work and chores such as cleaning, cooking, washing, shopping, collecting water or wood, and building or maintaining the house, its surroundings, or furniture. Collectively, this set of activities is referred to in this chapter as "household work."5 This chapter's emphasis on household work in addition to traditional work contrasts with much of the recent academic literature, which ignores household work and focuses on traditional work alone. Ignoring work in the production of nontraded household goods would raise three main con- ceptual issues. First, when a child works outside of his or her household as a paid domestic servant or a slave, that child is classified as a child laborer under the most stringent of definitions. It seems hard to defend reclassifying the child's production activities as something other than work if the child's employer changes. Second, treating the production of nontraded goods as something other than child labor makes it difficult to interpret the meaning of the state of "not working." For example, if home production is ignored in the definition of child labor, a child who stops limited work in a family busi- ness to take over extensive household responsibilities (say, because of the absence of a parent) would appear to stop working. Third, an assertion that child participation in the production of nontraded goods is not of academic or policy interest seems hard to defend, given the findings of this chapter. In particular, household work is the main activity children participate in other than schooling, and some of the most interesting differences between house- holds in child labor in the 1990s appear to take place in the household production of nontraded goods. 510 Economic Growth, Poverty, and Household Welfare in Vietnam Table 14.2. Participation in Child Labor in Past Seven Days, by Type of Work, for Children Ages 6 to 15 Years 1993 1998 Standard Standard Indicator Mean error Mean error Work (percent) Works for wages outside of household 2.3 0.3 1.3 0.2 Works in agriculture in household 25.6 1.6 19.3 1.7 Works in business in household 4.4 0.6 2.6 0.4 Works in traditional work 30.7 1.5 22.0 1.6 Works in household work 52.8 1.2 53.0 1.6 Works 62.1 1.3 56.8 1.5 Schooling (percent) Attends school 0.765 0.013 0.880 0.010 Works, attends school 0.448 0.014 0.481 0.018 Works, does not attend school 0.175 0.009 0.087 0.007 Hours for all Hours in traditional work 9.050 0.478 1.180 0.138 Hours in household work 5.973 0.196 4.431 0.190 Total hours 15.021 0.548 5.572 0.204 Hours for those who work Hours in traditional work 29.461 0.758 32.118 1.811 Hours in household work 11.310 0.317 8.352 0.264 Total hours 24.189 0.665 10.349 0.337 Note: Population means weighted to reflect sampling probabilities. Standard errors are corrected for clustered sample design. The 1993 data are from a sample of 6,071 children ages 6­15 representing a population of 16,340,704. The 1998 data are from a sample of 7,071 children ages 6­15 representing a population of 19,117,671. Hours for those who work refers to those who work in the specified category. Total hours for all do not add up precisely to the sum of hours for all in traditional and hours for all in household work because of some missing observations of hours worked in household work. Source: Authors' calculations from the 1993 and 1998 VLSSs. Table 14.2 documents the economic activities of children ages 6 to 15 in both rounds of the VLSS. The outstanding feature of table 14.2 is that a majority of children in Vietnam were engaged in some form of economic activity within the seven days before the household's interview. This is true Child Labor in Transition in Vietnam 511 in both the 1993 and 1998 rounds of the VLSS. However, participation rates declined by 9 percent between the survey years--from 62 to 57 percent. This decline is composed of a large (28 percent) decline in participation in tradi- tional work and a small (0.4 percent), statistically insignificant increase in participation in household work.6 Within the category of traditional work, children were most likely to be engaged in agricultural work within the household. The participation rate in agriculture within the household in 1993 is 26 percent. This declined to 19 percent in 1998, a 25 percent reduction rela- tive to the 1993 level. Work outside the household and work for a home enterprise were rare, with participation rates of 2 percent and 4 percent, re- spectively, in 1993. However, both of these categories experienced large, sta- tistically significant percentage reductions in the 1998 VLSS. Work outside the household declined by 44 percent. Work in a home enterprise declined by 42 percent. The Economic and Policy Context These changes in child labor are taking place in a rapidly evolving eco- nomic and policy environment. The rural reforms of the late 1980s returned responsibility for agricultural production to the autonomous farming household, and this reform is correlated with impressive growth in agricul- tural output. Over the 1990s, agricultural GDP grew by nearly 5 percent a year, prompting a rise of 60 percent in farm incomes between 1993 and 1998 (World Bank 2000). The industrial sector has also been expanding rapidly, growing at 13 percent a year between 1993 and 1998. Policies that promoted capital-intensive industries and protected domestic markets have meant that industrial employment over this period grew relatively slowly (at approximately 4 percent a year over the same period). The introduction of a new Enterprise Law in 2000 and recent announcements that the govern- ment of Vietnam intends to embark on further reforms to create a stronger environment for enterprise and international trade suggest that a more labor-intensive sector may develop rapidly over the coming years. Recent estimates based on General Statistical Office (GSO) data suggest that 300,000 new jobs were created in the private sector during 2000 (World Bank 2000). EDUCATION. Government policies in the post-Independence period have demonstrated a commitment to achieving universal primary education and protecting children from exploitative situations. By 1993, net enrollment rates in primary school reached 87 percent (World Bank [estimates based on VLSS 1993 data] 1999). Earlier emphasis on the provision of education was reinforced in 1991 by the introduction of the Law on the Universalization of Education and by the 1992 Constitution, which asserts that primary educa- tion is both free and compulsory. Although tuition fees are not charged for primary education, many sources have described the burdensome nature of a whole range of other costs associated with educating children (ActionAid 1999; Oxfam U.K. 1997; World Bank 1999). These studies suggest that the 512 Economic Growth, Poverty, and Household Welfare in Vietnam costs became more onerous over the 1990s and that they are an important cause of interrupted education. Recent estimates using VLSS data suggest that the costs of educating one student at the primary level are equivalent to nearly 5 percent of nonfood expenditure for a household in the lowest quin- tile of the population in the 1998 VLSS, and that the household's primary school costs rose between 1993 and 1998 (Government of Vietnam­Donor Working Group 2000). Households in the lowest quintile are well below the poverty line. As such, any nonfood expenditure diverts funds from basic consumption needs (World Bank and the Poverty Working Group 1999). Much of the qualitative literature on child labor and working children in Vietnam tracks a path from household economic difficulties to withdrawing children from school to, shortly afterward, scaling up the economic activity of children as a strategy for coping with hardship (SCF U.K. 1998, 1999; VN-Sweden MRDP 1999). Even though the costs of educating children can be considerable, enroll- ment rates in all levels of schooling rose over the 1990s. Table 14.3 contains school enrollment rates by quintile and level of schooling for 1993 and 1998. In 1998, net enrollment in primary education (grades 1­5) was 91 percent, up from 87 percent in 1993, with little difference between the enrollment rates of girls and boys. In 1998, enrollment in lower secondary school (grades 6­10) had climbed to 62 percent, up from 30 percent in 1993. How- ever, poor children have generally lower enrollment rates at all levels of Table 14.3. School Enrollment Rates, by Quintile (percent) Lower Upper Enrollment Primary secondary secondary Postsecondary rate/indicator 1993 1998 1993 1998 1993 1998 1993 1998 Net enrollment rate Vietnam 87 91 30 62 7 29 3 9 Poorest quintile 72 82 12 34 1 5 0 0 Most well-off quintile 96 96 55 91 21 64 9 29 Gross enrollment rate Vietnam 120 115 42 78 9 36 4 12 Poorest quintile 100 112 15 47 1 8 0 0 Most well-off quintile 130 104 77 105 24 75 13 37 Note: Net enrollment rates refer to enrollment rates of children in the relevant age range. Gross enrollment rates are the ratio of the number of enrolled students to the number of children in the relevant age range.Thus, gross enrollment rates can be over 100 if children who are not of the appropriate age enroll in a given level of schooling. Source: Nguyen 2004, based on estimates from the 1993 and 1998 VLSSs. Child Labor in Transition in Vietnam 513 schooling (table 14.3), and the quality of education services varies widely across the country. Moreover, Vietnam has one of the shortest primary school curricula in the world in terms of hours in the classroom (though this is currently under revision after the National Assembly's adoption of Re- solution Number 40/2000/QH on curriculum reform), and, particularly in rural areas, Vietnam's schools do not demand more than a few hours' atten- dance a day (Department for International Development 2001). For many children, progress through primary school is fully compatible with a mod- erate amount of work, either inside or outside the household, paid or unpaid. In fact, for some children, the costs of pursuing education may necessitate economic activity. A child has only so much available time, and time spent working may reduce time in school, time studying, or leisure time. A vast descriptive lit- erature suggests that low levels of work are compatible with continued school enrollment, but, as hours worked increase, schooling and work be- come incompatible. Even if school enrollment is compatible with child labor, work may still affect a child's human capital accumulation. First, a working child may be enrolled in school, but it is not clear that time spent in class is observed with enrollment information. Second, physically being in school is only a necessary, but not a sufficient, condition for learning. Work may limit the child's energy for school, or it may limit the child's ability to develop skills outside the classroom. Third, even if working has no effect on school- ing whatsoever, leisure and play are important in a child's development. Play enables a child to develop his or her social skills and creative thinking skills. It is possible that this cost to a child could be even greater than the lack of general skill accumulation. Of course, the types of general skills that a child learns in school are not the only types of skills that are useful. A child may use the skills developed while working throughout life. Thus, the rela- tionship between schooling and child labor is very difficult to analyze. This is further complicated because it is not possible to differentiate between whether children work because they do not attend school or they do not attend school because they work. With this in mind, school enrollment rates in 1993 and 1998 are examined for different work categories. Nothing can be said about the quality of time spent in school for working children, and the working child's consumption of leisure is not observable. With these caveats, however, it seems useful to consider school enrollment rates by the type of work performed by a child and examine how school enrollment rates vary through time in each work category. This is shown in table 14.4. Each cell in table 14.4 is calculated by stratifying the sample by each row. Hence, in the first row, school enrollment rates are computed for all children who do not work. In the third row, school enrollment rates are computed for all children who work in agriculture for their household. Any individual child can appear in multiple rows. For example, if a child works in agricul- ture and a home enterprise, that child is counted in both rows. The first two columns calculate school enrollment rates by year for all ages. The 514 Economic Growth, Poverty, and Household Welfare in Vietnam Table 14.4. School Enrollment by Age and Type of Work in Past Seven Days (percent) All (ages 6­15) Ages 6­11 Ages 12­13 Ages 14­15 Category 1993 1998 1993 1998 1993 1998 1993 1998 Does not work 83.3 92.3 83.7 92.9 88.3 93.3 67.0 85.8 Works for wages outside household 15.9 7.5 46.2 39.9 32.4 16.2 4.5 4.0 Works in agriculture in household 63.0 74.7 89.8 93.3 66.3 80.6 34.0 58.3 Works in business in household 48.9 59.7 86.8 90.1 58.8 62.9 28.1 51.4 Works in traditional work 59.3 70.2 89.5 92.6 63.2 76.9 30.9 53.3 Works in household work 72.5 86.3 88.8 96.0 72.1 88.2 43.0 72.4 Works 71.8 84.7 89.2 95.5 70.7 86.7 41.6 69.6 Source: Authors' calculations from the 1993 and 1998 VLSSs. remaining columns compute school enrollment rates for children ages 6­11, 12­13, and 14­15. Several interesting traits appear in table 14.4: · School enrollment rates were generally highest for nonworking chil- dren: 88 percent of 12- to 13-year-olds who did not work were en- rolled in school in 1993, but only 71 percent of 12- to 13-year-olds who worked attended school in 1993. The only exception to this is for pri- mary school­age children (ages 6 to 11). In this group, children who worked reported slightly higher enrollment rates, but this difference in enrollment rates for primary school­age children is not statistically significant. · In both 1993 and 1998, children were least likely to attend school if they worked outside the household (only 8 percent of children in this group enrolled in school in 1998) or if they worked in a household- run business. The majority of children were able to both enroll in school and work in agriculture or household work. For those above the age of 11, however, children who worked in any type of traditional work had enrollment rates that were (statistically) below enrollment rates for children who did not work. · Between 1993 and 1998, school enrollment rates increased across all rows of table 14.4 except for those children who worked outside the household. School enrollment rates were actually lower for children Child Labor in Transition in Vietnam 515 who worked outside the household in 1998, but this lower rate of school enrollment is statistically significant at the 10 percent level for only one age group: children ages 12 to 13. · Except for children ages 6 to 11, school enrollment rates increased more between 1993 and 1998 for working children than for nonwork- ing children. Part of this may be attributable to the fact that school en- rollment rates are bounded at 100 percent, and they started off very close to 100 percent for nonworkers in 1993. In addition, work could have been becoming more compatible with schooling in 1998. One mechanism for this increase in the compatibility between schooling and working might be a reduction in hours worked that accompanies the reduction in work participation rates observed in this chapter. Hence, the VLSS data demonstrate that older children who work are less likely to be enrolled in school than older children who do not work, and children who work are more likely to be enrolled in school through time. LEGISLATION. Vietnam was the second country in the world, and the first country in Asia, to sign the International Convention on the Rights of the Child (the Convention) in 1990. Article 32 of the Convention underscores the need for governments to "recognize the right of the child to be protected from economic exploitation and from performing any work that is likely to be hazardous to or interfere with the child's education, or to be harmful to the child's health or physical, mental, spiritual, moral or social develop- ment." The government of Vietnam has acted on this Convention through a number of legislative and regulatory measures, seeking to maintain an un- easy balance between allowing children to contribute to their own survival in times of hardship and safeguarding children's rights to physical and in- tellectual development. Among these measures, of particular importance are the Law on Child Protection, Care and Education (1991); the 1992 Constitution of the Socialist Republic of Vietnam (especially Article 65); the Labor Code (1994); many decrees and circulars that clarify specific issues in connection with child labor; and Decision Number 134/1999/QD-TTg, which approved the Program of Action to protect vulnerable children in 1999­2002. The outcome of these laws, decrees, regulations, and instructions is a regulatory framework that outlines the key definitions and priorities in re- lation to child labor. A child is a person younger than age 18 (according to the 1992 Constitution), but Articles 119­122 of the Labor Code specify con- ditions under which adolescents or juniors (15- to 18-year-olds) may work legally. Limitations that apply to the employment of 15- to 18-year-olds include restrictions on: · Working more than 7 hours a day or 42 hours a week · Working under dangerous conditions7 · Being forced to work or being involved in abusive or exploitative work. 516 Economic Growth, Poverty, and Household Welfare in Vietnam Junior employees between the ages of 15 and 18 are entitled by law to the same wages as adults, provided they are performing the same work. Children younger than 15 are allowed to work in a very restricted range of activities specified by Vietnam's Ministry of Labor, Invalids and Social Affairs (MOLISA) (Circular Number 21/1999/TT-BLDTBXH), but they are not permitted to work more than 4 hours a day or 24 hours a week, must be over the age of 12, and may work only with written consent of their parents or sponsors. The employer is obliged to ensure the child's school- ing. Children younger than 13 can be employed legally if they are being trained in certain occupations identified by the MOLISA (Decree Number 90/CP). The government of Vietnam ratified the International Labour Organisa- tion (ILO) Convention Number 182 on the Worst Forms of Child Labour in November 2000. By so doing, the government has indicated its commitment to eliminating "the worst forms of child labor" as defined in Article 3 of Convention 182 and is in the process of drafting a plan to implement the re- quirements of Convention 182 (MOLISA 2001). Vietnamese tradition accords an important role for children within the household and, in common with many cultures, a moderate amount of work within the household can be considered positive for the physical, intellec- tual, and personal development of children. This is legal as long as it is not harmful, dangerous, or exploitative and it does not interfere with the com- pletion of primary education (Institute of Labor Science and Social Affairs [ILSSA] and University of Wollongong 2000). Winners and Losers among Child Laborers The allocation of child time is an important component of a household's decisionmaking process. The household must weigh the value of child time spent in many activities, including schooling, wage work, work inside the household, and work in household chores or other components of house- hold production. The value of child time in any of these activities may de- pend on both child and household attributes. This section uses the existing qualitative literature on child labor to identify key child and household at- tributes that may be associated with heterogeneity across households in how child labor declined in Vietnam in the 1990s. The goal of this section is to consider several indicators that may be useful for directing policy toward children that have missed experiencing the dramatic declines in child labor that have occurred in recent history. The section begins with two key child attributes that are responsible for much of the variation across children in child labor: age and gender. Second, the variation in economic progress across regions in Vietnam has been a theme throughout this book, and the child labor literature is also concerned with differences in child labor trends across regions. Thus, regional patterns in child labor are discussed. Finally, several other household attributes that have drawn considerable attention in the qualitative literature are considered Child Labor in Transition in Vietnam 517 in the third part of this section. These household attributes include living standards, adult migration, enterprise ownership, and ethnicity. Child Attributes: Age and Gender The types of work that a child can perform vary with the child's age and may vary with the child's gender. A 6-year-old child is a less capable worker in most activities than a 15-year-old. Gender-typing of economic and house- hold activities can lead to different age-gender distributions of children's activities. If boys and girls perform different types of activities, it is possible that they have been differentially affected by the changes that Vietnam ex- perienced in the 1990s. In this section, changes in child labor are considered by gender, then gender differences by age are discussed. Table 14.5 presents participation rates in various types of economic ac- tivities by gender. Girls were more likely to work than were boys in both the 1993 and 1998 VLSSs. Higher participation rates in traditional work appear to be driven by greater participation by girls in the home enterprise. Also, girls are more apt to participate in household work. Most of the large gender differences in participation in any form of work (in table 14.5, the lines titled simply "Works") appear to be due to the substantially higher participation rates of girls in household work. The reduction (in percentage terms) in participation rates between the two rounds of the VLSS is larger for boys, Table 14.5. Participation in Child Labor in Past Seven Days, by Gender, for Children Ages 6 to 15 Years (percent) 1993 1998 Standard Standard Gender/category Mean error Mean error Boys Works for wages outside household 2.2 0.3 1.2 0.2 Works in agriculture in household 25.4 1.7 19.1 1.9 Works in business in household 3.6 0.5 2.5 0.5 Works in traditional work 29.9 1.6 21.6 1.9 Works in household work 45.2 1.6 46.8 1.8 Works 57.9 1.6 52.5 1.8 Girls Works for wages outside household 2.4 0.4 1.4 0.3 Works in agriculture in household 25.9 1.7 19.4 1.7 Works in business in household 5.3 0.7 2.7 0.5 Works in traditional work 31.6 1.5 22.4 1.6 Works in household work 60.7 1.3 59.6 1.6 Works 66.5 1.3 61.3 1.6 Source: Authors' calculations from the 1993 and 1998 VLSSs. 518 Economic Growth, Poverty, and Household Welfare in Vietnam although the declines in level of participation are nearly identical for boys and girls. However, the activities of boys and girls differ with their age. Thus, although there may be little difference between the way boys and girls as groups benefit from the growth in Vietnam in the 1990s, there may be important differences by age. Figure 14.1 presents child labor participation rates by age and gender.8 Throughout this chapter, figures similar to figure 14.1 are examined. Hence, it is important that the interpretation of figure 14.1 be clear. Figure 14.1a considers participation in all categories of work. Figure 14.1b considers only participation in traditional forms of work (the difference between the two being household work). Each line in figure 14.1 connects the participation rates by age for the group indicated in the legend. The vertical axis is labeled the "Probability of working." It is interpreted as being the fraction of children at a given age in a given group (for example, boys in 1993) who were working or being the probability, upon observing a child at the indicated age in the given group, of finding that the child works. When multiplied by 100, these are identical to the labor participation rates in the tables. The large drop in the probability that a child works, as well as the im- provements experienced by each gender, is evident in these figures. However, Figure 14.1. Participation in Work, by Age and Gender a. All work categories Probability of working (all categories) 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 6 7 8 9 10 11 12 13 14 15 Age Boys 1993 Girls 1993 Boys 1998 Girls 1998 Source: Authors' calculations from the 1993 and 1998 VLSSs. Child Labor in Transition in Vietnam 519 b. Traditional work Probability of working in traditional work 0.7 0.6 0.5 0.4 0.3 0.2 0.1 6 7 8 9 10 11 12 13 14 15 Age Boys 1993 Girls 1993 Boys 1998 Girls 1998 Source: Authors' calculations from the 1993 and 1998 VLSSs. there are some interesting age-gender differences evident in figure 14.1. In fig- ure 14.1a (all work categories), the largest drop in work probabilities appears between the ages of 8 and 10. These are primary school ages. The decline in work is smaller in older age groups, particularly for girls. However, for traditional work the decline in labor participation is greatest for older (post­primary school) ages. In figure 14.1b, girls older than 11 are more likely to engage in traditional work than are boys, although the differences are very small. Both boys and girls experienced a similarly large drop in participation rates in traditional work between 1993 and 1998. The magnitude of this (per- centage point) drop increases with age. Because older children were substan- tially more likely to do traditional work in 1993, it makes sense that they should have experienced the largest reductions over time. The interpretation of these gender differences is complex. Boys and girls may have different economic opportunities open to them, and the value of their time outside work may vary dramatically. Within the household, mem- bers may not agree on how to allocate child time. It is particularly important to recognize that the benefits from any particular decision may not accrue to those bearing the costs associated with that decision. This repeatedly emerges as a theme in studies on children in Vietnam. It is common, for example, to see one child (often a girl) withdrawn from school and put to work so that the other children can continue their educations (SCF U.K. 1999). 520 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 14.2. Distribution of Hours Worked in Nonagricultural Traditional Work Density 0.6 0.5 0.4 0.3 0.2 24 hours 42 hours 0.1 1.5 2 2.5 3 3.5 4 Logarithm of hours in nonagricultural work, conditional on working Boys 1993 Girls 1993 Boys 1998 Girls 1998 Source: Authors' calculations from the 1993 and 1998 VLSSs. An important part of the difference in work participation between boys and girls lies in their contribution to household work; thus, it is likely that the gender division of labor and gender-based inequities in decisionmaking within the household are important determining factors. This issue will be reexamined in the conclusion of this chapter. There also appear to be gender differences in hours worked. Figure 14.2 examines the distribution of hours worked in nonagricultural traditional work for children who work. The questionnaires from the 1993 and 1998 VLSSs are virtually identical with respect to child labor, except for a substantive change in the way hours worked in agriculture are collected. Consequently, only hours worked in wage work and work for the home enterprise ("nonagricultural traditional work") can be compared. Figure 14.2 contains nonparametric estimates of the density of the loga- rithm of hours worked in the past week.9 The densities for 1993 and 1998 and for boys and girls are estimated separately. In the 1993 density esti- mates, the density of time spent working for girls is more concentrated than for boys. This is made evident by the higher peak in excess of 42 hours of work in the week before the survey. Moreover, boys in 1993 were substan- tially more likely to work fewer than 24 hours than were girls. Girls were more likely observed working for more than 42 hours per week. Child Labor in Transition in Vietnam 521 The distribution of hours worked changed significantly in 1998. For both boys and girls, there was a drop in the mass of workers working in excess of 42 hours. For boys, there was an increase in the mass working close to 24 hours a week. For girls, two clear mass points emerged in the 1998 dis- tribution. The largest subset of girls worked more than 42 hours a week (although the fraction of girls working 42 hours declined between the two years). However, in 1998 there was a mass of girls who worked slightly fewer than 24 hours a week. The 1998 densities have not been corrected to reflect the fact that the probability of observing a child working in non- agricultural work is lower in 1998. Hence, the type of children pictured in the 1993 distribution might be different from the children remaining in the 1998 distribution. Nevertheless, to the extent that the children working in nonagricultural work in 1993 and 1998 are comparable, the densities in fig- ure 14.2 are consistent with many girls shifting from a large number of work hours in 1993 to relatively few work hours in 1998. Figure 14.2 shows that a considerable number of those children, particu- larly girls, who work outside agriculture are working hours above the legal limits set out in the Labor Code. Forty-five percent of these children are working in enterprises with five or fewer employees, which are not bound by the Labor Code, but these legal limits are still relevant as indicators of what Vietnamese society and legislators have decided is acceptable within the specific social, cultural, and economic context of Vietnam. The mean for the child doing nonagricultural traditional work in 1998 is still 34 hours of work a week, above the legal maximum set by MOLISA for children younger than 15 years. These children could be described as vulnerable in the sense that their working arrangements might be restricting their well- being and interfering with their basic rights as identified in Article 32 of the International Convention of the Rights of the Child. The concentration of children working more than 42 hours a week is especially worrying, because this exceeds by some margin the legal limits established for the 15­18-year age group. In the data used in this chapter, only 15 percent of the children who work more than 42 hours are even enrolled in school. A study of work- ing children in Ho Chi Minh City (Viet Nam Youth Institute 1999) corrobo- rates this pattern, indicating that working hours for girls peaked at a higher level than for boys and at levels above the maximum limit set by law. Residential Location The distribution of improvements in living standards has been different across rural and urban areas of Vietnam (Glewwe and Nguyen 2004). For that reason alone, differences might occur in improvements in the child labor situation across rural and urban areas (or in even greater geographic detail). However, children also engage in different types of economic activi- ties in rural and urban areas. Table 14.6 describes the types of activities undertaken by boys and girls of different ages in Ho Chi Minh City, and table 14.7 considers the activities of children in rural Vietnam. 522 Economic Growth, Poverty, and Household Welfare in Vietnam Table 14.6. Starting Age of Work of Children in Different Occupations, Ho Chi Minh City Children's ages (years) 6 7 8 9 10 11 12 13 14 15 16 17 Selling lottery tickets Peeling onions (at home) Making match boxes (at home) Weaving mats and baskets (at home) Scavenging at the dumpsite Making shoes (support workers--local) Filling bobbins (at weaving enterprises) Catching grasshoppers Making ball-point pens (boys) Making lanterns Classifying waste plastic (at home) Making operating parts of lamps Recycling glass Making ball-point pens (girls) Making chains Making silk-screen prints Making bag wheels Selling noodle soup Making nem chua (fermented pork) Making furniture and wooden art products Making scales Recycling glue Making chalk Recycling plastic (boys) Making shoes (migrant support workers) Making plastic sandals Making bicycle tires Sorting or recycling plastic (at factories-- girls) Dyeing materials Catching locusts 6 7 8 9 10 11 12 13 14 15 16 17 Source: SCF U.K. 1999. Both tables are taken from participatory research with working children in different locations in Vietnam. The scope of activities open to younger children in rural areas is much greater than it is in urban areas. Thus, be- cause of both the nature of the economic change in Vietnam and differences in the activities of children between rural and urban areas, it is important to Child Labor in Transition in Vietnam 523 Table 14.7. Starting Age of Work of Children in Different Occupations, Rural North Central Coast Region Children's ages (years) 4 5 6 7 8 9 10 11 12 13 14 15 Look after younger siblings Sweep house and yard Watch house Wash dishes Feed chickens Collect pig feed Catch crabs, shrimp, and snails Dig up worms for ducks Wash clothes Cook food for humans Cook food for pigs Feed pigs Fetch fuelwood Boil water Dry paddy Process cassava Tend cows and buffalos Collect grass Fetch water Collect cattle manure Harvest rice Transplant rice Weed and irrigate crops Plow and harrow fields Collect firewood from forest Mill and husk paddy Wage labor Fish Migrate 4 5 6 7 8 9 10 11 12 13 14 15 Source: SCF U.K. 1997. examine differences in child labor improvements by geographic area. The initial focus here is on rural and urban differences, followed by a discussion of differences across the 10 main geographic regions in Vietnam. (The 10 regions are major urban areas, minor urban areas, provincial towns, the Northern Uplands, the North Central Coast, the Central Highlands, the Red River Delta, the Southeast, the Central Coast, and the Mekong Delta.) Table 14.8 considers participation in each of the different work categories by geographic location (rural or urban). Participation rates are much higher in rural areas than in urban areas. This is true in both traditional work and household work. The extra participation in traditional work appears to be primarily in agriculture, because children seem to have similar levels of 524 Economic Growth, Poverty, and Household Welfare in Vietnam Table 14.8. Participation in Child Labor in Past Seven Days, by Residential Location, for Children Ages 6 to 15 (percent) 1993 1998 Standard Standard Area/category Mean error Mean error Urban Works for wages outside household 2.8 0.5 1.7 0.5 Works in agriculture in household 7.1 2.2 2.8 1.4 Works in business in household 5.4 1.1 1.6 0.4 Works in traditional work 14.9 2.5 6.1 1.5 Works in household work 44.3 3.3 40.5 2.6 Works 48.7 3.4 42.2 2.6 Rural Works for wages outside household 2.2 0.3 1.2 0.2 Works in agriculture in household 29.8 1.8 22.7 2.0 Works in business in household 4.2 0.6 2.8 0.5 Works in traditional work 34.3 1.7 25.3 1.9 Works in household work 54.7 1.3 55.7 1.8 Works 65.1 1.4 59.8 1.8 Source: Authors' calculations from the 1993 and 1998 VLSSs. participation in wage work or in home enterprises in both urban and rural areas. The reduction in work participation rates through time is slightly greater in percentage terms in urban areas; the reason may be that the prob- ability a child worked in urban areas was smaller in 1993 than in 1998. There is a notable difference in the changes in child labor through time between rural and urban areas; this can be seen in the probability that a child works in a home enterprise. Participation rates in a home enterprise for a child in an urban area drop by 70 percent to 0.02 between 1993 and 1998. In rural areas, the incidence of working in a home enterprise drops by 34 percent. This larger decline in urban areas occurs despite generally higher participa- tion rates in home enterprises in urban areas in 1993. This may be the result of increases in rural, nonfarm enterprises.10 This issue will be further ex- plored in the subsection on enterprise ownership. The description of child work in tables 14.6 and 14.7 suggests that there are important differences associated with age in the allocation of child time between urban and rural areas. Thus, work participation rates are first con- sidered by age. Figure 14.3 replicates the methodology used in figure 14.1 (by age and gender figures). Each line in figure 14.3 represents a separate regression for each urban and rural region, as well as for each round of the survey. Children in rural areas are much more likely to work, at every age, than are children in urban areas. Considering all forms of work (figure 14.3a), Child Labor in Transition in Vietnam 525 urban-rural differences in work probabilities appear largest for children ages 8 to 11. In this same group, the largest reductions in work probabilities for rural areas between the 1993 and 1998 VLSSs are found. Urban areas ap- pear to have experienced an approximately uniform reduction in work probabilities between 1993 and 1998. Consequently, for children ages 8 to 11, rural-urban differences appear to have decreased between 1993 and 1998, but for older children they appear almost unchanged. When traditional work is considered (see figure 14.3b), the evidence looks different. Here, the probability of working appears to decline more for older children in both rural and urban areas. Older children are much more likely to engage in traditional work, so it is not surprising that they would experience greater reductions in the probability of working in traditional work. As with all forms of work, children in rural households are more likely to engage in traditional work at every age past six years. However, the largest reductions in traditional work appear to take place among older, urban children. Recent work on poverty in Vietnam shows how per capita expenditures in urban areas have risen almost twice as fast (by 9.9 percent a year) over the 1993­98 period, as did the per capita expenditures in rural areas, which rose 5.4 percent a year (World Bank and the Poverty Working Group 1999). Given the strong correlation between household economic Figure 14.3. Participation in Work, by Age and Location a. All work categories Probability of working (all categories) 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 6 7 8 9 10 11 12 13 14 15 Age Urban 1993 Rural 1993 Urban 1998 Rural 1998 Source: Authors' calculations from the 1993 and 1998 VLSSs. 526 Economic Growth, Poverty, and Household Welfare in Vietnam b. Traditional work Probability of working in traditional work 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 6 7 8 9 10 11 12 13 14 15 Age Urban 1993 Rural 1993 Urban 1998 Rural 1998 Source: Authors' calculations from the 1993 and 1998 VLSSs. status and child labor participation documented by Edmonds (2004), these large reductions in traditional work for older, urban children are likely the result of the relatively high economic growth rates in urban areas. Comparing households in different settings is difficult--the argument can be made that households in rural areas are fundamentally different from households in urban settings. The consequences of improvements in stan- dards of living may be very different in the two environments. To compare improvements in child labor in one geographic region with another, or to compare improvements in cities to those in provincial towns or the country- side, a linear regression framework is used. In each regression, the depen- dent variable is an indicator with a value of one if a child works (in all work or in traditional work) and zero otherwise. Each regression is estimated using the linear probability model with controls for differences associated with a child's age, gender, and the year of observation in each regression.11 Table 14.9 considers how changes in child labor vary across the 10 main geographic regions of Vietnam. Minot and Baulch (2004) examine how other measures of poverty vary across Vietnam. In addition to the age, gender, and year controls, the regression in table 14.9 also controls for time-invariant household characteristics by including household fixed effects. These household fixed effects control for factors such as the remoteness of a Child Labor in Transition in Vietnam 527 Table 14.9. Regional Differences in the Decline in Child Labor, Linear Probability Model, Household Fixed Effects Results (multiplied by 100 gives percent) All work Traditional Dependent variable Coeff. work Coeff. Change in rural Mekong Delta -0.250** -0.181** (0.033) (0.023) Changes relative to the rural Mekong Delta Urban areas Major urban 0.191** -0.024 (0.078) (0.034) Minor cities 0.029 -0.027 (0.056) (0.050) Provincial towns -0.023 -0.026 (0.063) (0.036) Rural areas Northern Uplands 0.077* -0.020 (0.041) (0.047) Red River Delta 0.143** 0.066 (0.050) (0.041) North Central Coast 0.048 0.131** (0.051) (0.051) Central Coast 0.207** 0.073 (0.051) (0.061) Central Highlands 0.287** 0.079 (0.095) (0.074) Southeast 0.038 0.083** (0.059) (0.038) Adjusted R2 0.432 0.410 *Significant at 10 percent level. **Significant at 5 percent level. Note: Standard errors in parentheses. All regressions include a quadratic in age and gender, a constant, a year effect, and household fixed effects. Standard errors are corrected for arbitrary heteroskedasticity and the cluster/time design of the survey. The interpretation of -0.25 in the first row is that the probability a child works in the rural Mekong Delta declines by 25 percentage points, and so on through all later regressions. Source: Authors' calculations from the 1993 and 1998 VLSSs. household's location, its ethnicity, the household's size (but not changes in household size), the education of the head of the household, or any other household traits that do not change over time. Throughout this study, the regression results aim to document differences in declines in child labor across observable household characteristics. The results should not be inter- preted as the impact of any given characteristic on child labor. For example, the largest declines in child labor are in the Mekong Delta. If a household 528 Economic Growth, Poverty, and Household Welfare in Vietnam from the Central Highlands were picked up and moved to the Mekong Delta, there is no reason to believe that the Central Highlands household would experience the same decline in child labor as other households in the Mekong Delta. Table 14.9 shows the change in child labor that occurs in the rural Mekong Delta and the additional changes (relative to the Mekong Delta) that occur in other administrative regions. Thus, in the first column, the probability that a child works in any type of work declines by 0.25 (or 25 percentage points) in the rural Mekong Delta after household fixed effects and child attributes are controlled for. However, the probability that a child in major urban areas works in any type of work declines by only 0.059 (5.9 percentage points: -0.250 + 0.191). For all forms of work (first column), it can be seen that most areas of Vietnam decrease their child labor by less than the rural Mekong Delta. The particularly large declines in child labor in the rural Mekong Delta may stem in part from Vietnam's integration into world rice markets (Edmonds and Pavcnik 2002). The only region that experiences larger reductions than the Mekong Delta is provincial towns. However, for traditional work, major urban areas, minor urban areas, and provincial towns all decrease the probability that their children work by more than the rural Mekong Delta. The percentage point change can be calculated for each of the regions in table 14.9 in both all work and traditional work for every region. These changes can be seen graphically in figure 14.4. Provincial towns have experienced the largest reductions in both cate- gories of work. Nevertheless, there is an increase in the probability that a child works in the rural Central Highlands. The Central Highlands is the second poorest region in Vietnam, with more than one-half the population living below the poverty line in 1998.12 The incidence of "hunger poverty" barely fell at all in the period between the two surveys, and the poverty gap index shows poverty to be deeper here than elsewhere in the country (World Bank and the Poverty Working Group 1999). School enrollment rates are lower in the Central Highlands than elsewhere in the country for all levels of education (Nguyen 2004). The difference between traditional work and household work implies that this increase in work probabilities in the Cen- tral Highlands is driven by participation in household work, but there is an active labor market associated with the coffee plantations in the rural Cen- tral Highlands, and it is possible that this influences the results here. This could be the case if increased demand for adult labor shifts the burden of household work to children. Also, the rural Central Highlands is a destina- tion for migrant agricultural workers (including children) during the coffee harvest (SCF U.K. 1997), and this may also contribute to the unusual result here. There is also a concentration of ethnic minorities in the rural Central Highlands. There is a discussion in this chapter's section on ethnicity of the finding that the slight increase in children who work in the Central High- lands is not the result of the greater presence of minorities in the Central Highlands. Child Labor in Transition in Vietnam 529 Figure 14.4. Reductions in the Probability That a Child Works in the Past Seven Days, by Region, Household Fixed Effects Results Decline in probability that a child works ( 100) 30 25 20 15 10 5 0 5 10 Major urban Minor urban towns River Delta North Coast Coast Delta ovincial Central Pr Northern Uplands Mekong Red Highlands Southeast Central Central Region Work Traditional work Source: Authors' calculations from the 1993 and 1998 VLSSs. Other Household Characteristics LIVING STANDARDS. The effect of improvements in living standards on child labor has recently received substantial attention. Ample qualitative studies suggest that improvements in living standards in Vietnam have en- abled children to work less. Interviews with working children in a range of different situations and locations identify the strong causal link between poverty and child labor. Though these studies do not really permit authori- tative quantification, they do indicate that grinding poverty is the primary reason for leaving home to find work for at least three-quarters of the re- spondents. This is often tied in discussion to the need to assist families in paying debts and servicing debt interest payments. These debts have often been acquired in response to a health crisis or other shock in the household (Bond and Hayter 1998; International Labour Organisation­International Programme for the Elimination of Child Labour [ILO-IPEC] 2000) or to invest in housing (Youth Research Institute and Barnen 1999). 530 Economic Growth, Poverty, and Household Welfare in Vietnam The experiences of households in the VLSSs appear consistent with this qualitative literature. In the VLSSs, household living standards are mea- sured with the logarithm of per capita expenditure. There are two justifica- tions for looking at expenditure rather than income. First, most households do not participate exclusively in formal labor markets. Hence, calculating in- come is very difficult. Second, although income is variable, households tend to try to smooth consumption through time, and the expenditure measure in the VLSS is designed to approximate household consumption. In the extreme, the permanent income hypothesis suggests that households con- sume their permanent income so that consumption represents the house- hold's information about the income path before it. Edmonds (2004) studies the relationship between living standards and child labor in the VLSSs, and his results are reproduced here. First (using nonparametric regression techniques), he calculates participation rates in traditional work across the range of the per capita expenditure distribution. His estimates for children ages 6 to 15 are in figure 14.5. Actual participation rates in 1993 are at the top of the figure (labeled with "o"), and 1998 partici- pation rates are at the bottom of the figure (labeled with " "). Second, using the VLSSs, Edmonds calculates by how much living standards improve for Figure 14.5. Living Standards and the Decline in Child Labor Participation in traditional work 40 Poverty line 35 30 25 20 15 10 Exit Poverty by 1998 5 6.6 6.8 7 7.2 7.4 7.6 7.8 Logarithm of 1993 per capita expenditure 1993 1998 Predicted 1998 Source: Edmonds 2004. Child Labor in Transition in Vietnam 531 each household between 1993 and 1998. He then uses the cross-sectional relationship between child labor and living standards from 1993 (the "o" line) to predict what child labor participation rates should be in 1998, based only on improvements in living standards. This prediction is the middle line in figure 14.5 and is marked with " ." The rightmost vertical line in figure 14.5 is the 1993 poverty line, and the range between the poverty line and the left- most vertical line indicates the range of households that exit poverty between 1993 and 1998. Most of the decline in child labor experienced in Vietnam in the 1990s can be explained by improvements in living standards. For households that exit poverty between 1993 and 1998, living standards can explain most (80 percent) of the decline in child labor. Improvements in living standards explain less of the change in child labor for children in households far above the poverty line in 1993, or for children in households that persist in poverty between 1993 and 1998. Improvements in living standards vary a great deal across households in Vietnam; the remainder of this chapter considers how improvements in child labor vary across different subgroups of the population. MIGRATION. Many case studies highlight migrants as a particularly vulnerable group in Vietnam. A Participatory Poverty Assessment in Ho Chi Minh City (SCF U.K. 1999) illustrates the multiple disadvantages faced by migrants, particularly those who lack official registration. Official registra- tion can be important in determining the access of migrant children to main- stream education and the access of migrant families to subsidized health care and credit facilities (World Bank and the Poverty Working Group 1999). Migrants are often moving because of economic circumstances, so children in migrant households would be more apt to need to work. In addition, the process of migration itself may influence the likelihood that a child works because of the disruption in the child's life associated with the move. Figure 14.6 considers the relationship between work participation and whether the household head has ever moved. A household where the head has ever moved is defined here as a "migrant" household.13 Households that have moved were more likely to have children work in both traditional and all forms of work in both 1993 and 1998. When traditional work and household work (figure 14.6a) are considered, it is obvious that participation rates in child work were very similar in migrant and nonmigrant house- holds for all ages in 1993. In fact, for children ages 10 and older, the proba- bility that a child works appears virtually identical across migrant and non- migrant households. In 1998, however, much larger differences were seen between migrant and nonmigrant households. This difference is attributable to large declines in the probability that a child works in nonmigrant house- holds. Furthermore, the distinction between migrant and nonmigrant households appears to increase between 1993 and 1998, especially for chil- dren ages 11 to 14. In this age group, participation rates appear similar in 1993 in both migrant and nonmigrant households. 532 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 14.6b presents the probability that a child does traditional work, by age and migration. For both types of households, declines are observed in the probability that the child does traditional work--much as we have seen throughout this paper. In traditional work, we observe greater drops in participation in traditional work for nonmigrant children between the ages of 10 and 13. It is interesting that for 15-year-olds, migrant households actually have a slightly reduced probability of doing traditional work. Of course, household heads who have moved are likely to be substan- tively different from heads who have never moved. Thus, the regression methodology described in the context of table 14.9 is used again here. These results are shown in table 14.10. Columns (1)­(4) focus on participation in all forms of work, as in figure 14.6a, and columns (5) and (6) are for participa- tion in traditional work only. To gauge the relative changes in child labor for households whose heads have ever moved in 1993, a variable is included that is the interaction of head ever moved before 1993 and a variable to in- dicate that an observation is from 1998. Hence, the interpretation of the first row in table 14.10 is the change in child labor participation experienced by a household where a head ever moved before 1993 relative to the change in Figure 14.6. Participation in Work, by Age and Migration Status of Head a. All work categories Probability of working 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 6 7 8 9 10 11 12 13 14 15 Age of child Ever moved 1993 Never moved 1993 Ever moved 1998 Never moved 1998 Source: Authors' calculations from the 1993 and 1998 VLSSs. Child Labor in Transition in Vietnam 533 b. Traditional work Probability of working in traditional work 0.7 0.6 0.5 0.4 0.3 0.2 0.1 6 7 8 9 10 11 12 13 14 15 Age of child Ever moved 1993 Never moved 1993 Ever moved 1998 Never moved 1998 Source: Authors' calculations from the 1993 and 1998 VLSSs. child labor experience by households where the head has never moved. Thus, a positive regression coefficient indicates smaller declines in child labor for households where the head moved relative to the decline experi- enced in households where the head has never moved. After controlling for child characteristics, column (1) of table 14.10 shows that households where the head has ever moved experience smaller declines in child labor. However, households with a head who has moved may be in different areas from households where the head has never moved. For example, there may be more people who have moved in cities than in re- mote rural areas. Hence, column (2) controls for differences in the residential location of ever movers with commune fixed effects. The table also controls for differences across regions in the declines in child labor with region × time effects (these are the regression coefficients in table 14.9). After controlling for differences in the location of movers, the results change. Column (2) demonstrates greater reductions in ever-mover households. Controlling for household differences [in column (3)] attenuates the relationship between ever-movers and child labor further. Hence, most of the differences in figure 14.6 appear to owe more to differences in the location of households with heads who have moved than to something intrinsic in moving itself. A sim- ilar result is seen in traditional work in column (5). 534 Economic Growth, Poverty, and Household Welfare in Vietnam Table 14.10. Adult Migration History and Child Labor in the Past Seven Days, Linear Probability Results (multiplied by 100 gives percent) All work Traditional work (1) (2) (3) (4) (5) (6) Dependent variable Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Head ever moved × 1998 0.017 -0.011 -0.003 -0.028 (0.019) (0.013) (0.022) (0.020) Father resident 0.028 -0.036 (0.038) (0.028) Mother resident -0.016 0.038 (0.046) (0.037) Head moved 0.059** 0.034* (0.020) (0.020) Commune effects No Yes No No No No Household effects No No Yes Yes Yes Yes Region × time effects No Yes Yes Yes Yes Yes Adjusted R2 0.260 0.372 0.431 0.432 0.410 0.410 *Significant at 10 percent level. **Significant at 5 percent level. Note: Standard errors in parentheses. Standard errors are corrected for arbitrary het- eroskedasticity and the cluster-time design of the survey. All regressions include a quadratic in age and gender, a constant, and a year effect. "Head ever moved" is an indicator for if the head reported ever having moved in 1993. It is interacted with the year effect for 1998. Hence, it has the interpretation of being the extra change in child labor in households where the head had moved before 1993 in addition to the general decline in the population. Source: Authors' calculations from the 1993 and 1998 VLSSs. In general, the VLSSs did not capture households that moved between 1993 and 1998. However, households where one or more members have departed or returned are observed. In columns (4) (all work) and (6) (tradi- tional work), the effect on child labor of change in the residency patterns of parents or household head can be seen. A substantially smaller decline in child labor can be seen in households where the head has changed between rounds of the VLSS. This smaller decline in child labor appears to be in both household work and traditional work, because there is a significant, but slightly smaller, effect of changes in the household head on traditional work. It is obviously not clear whether this additional work in households that have migrant heads is directly attributable to the departure of the head or if there are common factors that cause both. ENTERPRISE OWNERSHIP. Whether or not a household owns a business is likely to influence the economic activities of children. It is easier for a child to work inside his or her home than for an outside employer, so more child labor Child Labor in Transition in Vietnam 535 would be expected in households with home enterprises. Furthermore, it is often easier for children to begin to contribute to home enterprises at an earlier age than it is for a child to perform the manual labor of agriculture. Nevertheless, generally better-off households can afford to start home enter- prises, and better-off households often enjoy better access to formal financial services and information, which are important in the establishment of household enterprises (Vijverberg and Haughton 2003; World Bank 1999). Three issues are important here. First, are households with home enterprises more likely to have children work? Second, are changes in child labor different in households with and without a home enterprise? Third, are changes in child labor associated with changes in home enterprises? Are households with home enterprises more likely to have children work? Figure 14.7 considers the probability that a child works by whether the household operates its own enterprise. As has been seen in every figure, there are uniformly lower probabilities that a child worked in 1998 for both traditional and all work participation rates. Two interesting characteristics are unique to figure 14.7. First, in both work categories and years, children Figure 14.7. Participation in Work, by Age and Household Enterprises a. All work categories Probability of working 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 6 7 8 9 10 11 12 13 14 15 Age of child Owns business 1993 Does not own business 1993 Owns business 1998 Does not own business 1998 Source: Authors' calculations from the 1993 and 1998 VLSSs. 536 Economic Growth, Poverty, and Household Welfare in Vietnam b. Traditional work Probability of working in traditional work 0.7 0.6 0.5 0.4 0.3 0.2 0.1 6 7 8 9 10 11 12 13 14 15 Age of child Owns business 1993 Does not own business 1993 Owns business 1998 Does not own business 1998 Source: Authors' calculations from the 1993 and 1998 VLSSs. in households without a home enterprise work more than children in house- holds with a home enterprise. This is difficult to explain, but the authors suspect (supporting evidence in the next paragraph) that this result is at- tributable to households with family businesses living in better-off areas and being more affluent, on average, than households that do not operate a home enterprise. Second, in 1998 the difference between households with and without a home enterprise was greater in traditional work than in all work. This suggests that children who are performing generally less tradi- tional work in households with a business must be contributing more household work in 1998 than are children in households without a business. This extra household work in households with businesses in 1998 appears especially large for children between the ages of 9 and 13. Are changes in child labor different in households with and without a home en- terprise? Much of the extra decline in child labor associated with the owner- ship of a home enterprise appears to come from the location of household enterprises. Table 14.11 returns to the regression approach used in table 14.10. In columns (1) and (4) for all work and traditional work, differ- ences among children in age, gender, and year of the survey have been controlled for. There are greater declines in child labor in nonfarm, rural households and in households that own a business. In fact, the probability Child Labor in Transition in Vietnam 537 Table 14.11. Enterprise Ownership and Child Labor in the Past Seven Days, Linear Probability Results (multiplied by 100 gives percent) All work Traditional work (1) (2) (3) (4) (5) (6) Dependent variable Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Nonfarm, rural household -0.031 -0.023 -0.074** -0.056** (0.023) (0.015) (0.018) (0.016) Owns business -0.035** 0.009 0.039** -0.040** 0.005 0.036* (0.014) (0.010) (0.017) (0.014) (0.011) (0.020) Commune effects No Yes No No Yes No Household effects No No Yes No No Yes Region × time effects No Yes Yes No Yes Yes Adjusted R2 0.263 0.372 0.432 0.199 0.353 0.410 *Significant at 10 percent level. **Significant at 5 percent level. Note: Standard errors in parentheses. Standard errors are corrected for arbitrary het- eroskedasticity and the cluster-time design of the survey. All regressions include a quadratic in age and gender, a constant, and a year effect. Source: Authors' calculations from the 1993 and 1998 VLSSs. that a child works in all work categories is a statistically significant 6.6 per- centage points lower in a rural, nonfarm household that owns a business than in a rural farm household. Children in these rural, nonfarm households with businesses experience even larger relative declines in traditional work relative to children in rural farm households. The probability that a child does traditional work declines by an additional 11.4 percentage points in rural, nonfarm households with a business. However, after controlling for commune differences and region-time differences (columns (2) and (5) of table 14.11), children in households owning a business experience slightly smaller declines in child labor than children in households without a busi- ness. However, the difference in declines in child labor associated with busi- ness ownership is not statistically significant. Are changes in child labor associated with changes in home enterprises? Al- though the hypothesis that the differences in child labor associated with enterprise ownership (figure 14.7) stem from differences in the location of enterprises is not rejected, there is strong evidence that changes in house- hold enterprise status are associated with changes in the economic activities of children. Columns (3) and (6) of table 14.11 include controls for household differences with household fixed effects and a variable that indicates whether a household owns a business. With the household fixed effects, the coefficient on this variable is interpreted as how a change in whether the household owns a business is associated with changes in child labor. In both traditional work and household work, creating a new home enterprise 538 Economic Growth, Poverty, and Household Welfare in Vietnam between rounds of the VLSS is associated with smaller reductions in the probability that a child works. The effect of owning a home enterprise is slightly larger for all work than for traditional work. Thus, the creation of a home enterprise seems to lead to more work (relative to a child in a house- hold that did not create a business) for children in both traditional and household work. The changes in home enterprises that took place between 1993 and 1998 involved both openings and closings. Thus, the finding that creating a home enterprise leads to more work relative to a child in a household that did not open a business also implies that closing a home enterprise is associated with a larger decline in the probability that a child works. Figure 14.8 com- pares children in households that open and close a family business between the 1993 and 1998 rounds of the VLSS to children in households that had no change in family business. Figure 14.8a compares children in households with businesses that closed between 1993 and 1998 (more precisely, the household reported at least one enterprise in 1993 and did not report any enterprises in 1998) with children in households that did not change their household enterprise Figure 14.8. Participation in Traditional Work, by Age and Household Enterprise Change a. Closed business Probability of working in traditional work 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 6 7 8 9 10 11 12 13 14 15 Age of child No change in business 1993 Business enders 1993 No change in business 1998 Business enders 1998 Source: Authors' calculations from the 1993 and 1998 VLSSs. Child Labor in Transition in Vietnam 539 b. Opened business Probability of working in traditional work 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 6 7 8 9 10 11 12 13 14 15 Age of child No change in business 1993 Business starters 1993 No change in business 1998 Business starters 1998 Source: Authors' calculations from the 1993 and 1998 VLSSs. status. Figure 14.8b compares children in households with businesses that opened (that is, no enterprise was reported in 1993, and at least one enter- prise was reported in 1998) to children in households that did not change. Two interesting trends emerge from figure 14.8. First, in figure 14.8a, households whose businesses ended between 1993 and 1998 experienced larger reductions in the probability that children ages 12 to 15 participated in traditional work than households without a change in enterprise status. Younger children appeared to experience approximately the same drop in households that close and do not close businesses. This large drop in child labor for older children in households that closed their businesses appears to come from the fact that these older children in 1993 were more likely to be working in traditional work. A possible explanation for this higher level of work is that these older children are working to help in the home enterprise. It is surprising that children in households that closed businesses have lower work probabilities than the general population. Households that owned businesses in 1993 were generally better off (in 1993 and 1998) than households that did not own businesses in 1993; it is possible that this ex- plains why they had lower work probabilities in 1998 than the general pop- ulation. This would then imply that the closure of the household enterprise was not a permanent, negative shock to household well-being. 540 Economic Growth, Poverty, and Household Welfare in Vietnam Second, it can be seen in figure 14.8b that children in households that opened home enterprises between 1993 and 1998 experienced smaller drops in child labor in that period. Households that opened enterprises between 1993 and 1998 and households that had no change in enterprises between 1993 and 1998 appear to have had very similar probabilities of having a child work in 1993. However, children in households that opened enter- prises worked more in 1998. This is true at every age, but it appears largest for ages 12 and 13. This trend takes place despite the fact that households that created new enterprises are generally better off than households that do not. Hence, households in new businesses appear to rely on family labor to help with the business. ETHNICITY. Recent analysis of poverty in Vietnam illustrates that con- sumption poverty among ethnic minority groups is declining far more slowly than for the majority (that is, ethnic Vietnamese [Kinh]) population (Baulch and others 2004; World Bank and the Poverty Working Group 1999). Social indicators for ethnic minority groups also lag behind. Because reduc- tions in child labor in general have been closely associated with improve- ments in per capita expenditures, it is important to examine how child labor has changed for those groups whose poverty appears to be particularly Figure 14.9. Participation in Work, by Age and Ethnicity a. All work categories Probability of working 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 6 7 8 9 10 11 12 13 14 15 Age of child Ethnic minority 1993 Nonminority 1993 Ethnic minority 1998 Nonminority 1998 Source: Authors' calculations from the 1993 and 1998 VLSSs. Child Labor in Transition in Vietnam 541 b. Traditional work Probability of working in traditional work 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 6 7 8 9 10 11 12 13 14 15 Age of child Ethnic minority 1993 Nonminority 1993 Ethnic minority 1998 Nonminority 1998 Source: Authors' calculations from the 1993 and 1998 VLSSs. intractable. The unusual trends observed in the rural Central Highlands (figure 14.4), where a concentration of ethnic minorities lives, raise the pos- sibility that child labor for ethnic minorities is not dropping as rapidly as for the majority. The question this section addresses is whether there is any evi- dence that children in minority households have reduced their child labor by less than the majority ethnic groups. In both traditional work and household work, ethnic minorities tend to work more than nonminority groups. In the "all work" category, a slight increase can be observed in the probability that ethnic minority children older than age 11 worked between 1993 and 1998. For traditional work in figure 14.9b, it can be seen that ethnic minority children in 1998 worked more than nonminority children in 1993. However, there is no observable increase in the probability that children worked between 1993 and 1998 in traditional work. Thus, part of the increase between 1993 and 1998 in the "all work" category must stem from increases in household work. The differences between ethnic minorities and others in the "all work" category appear to be largely the result of differences in the geographic location of ethnic minorities. However, in traditional work, there are differ- ences between ethnic minorities and others, even when household fixed effects are controlled for. The linear regression results are in table 14.12. 542 Economic Growth, Poverty, and Household Welfare in Vietnam Table 14.12. Ethnic Minorities and Child Labor in the Past Seven Days, Linear Probability Results (multiplied by 100 gives percent) All work Traditional work (1) (2) (3) (4) (5) (6) Dependent variable Coeff. Coeff. Coeff. Coeff. Coeff. Coeff. Ethnic minority × 1998 0.117** 0.028 0.017 0.124** 0.063** 0.084* (0.033) (0.022) (0.035) (0.044) (0.027) (0.046) 1998 -0.100** -0.195** -0.252** -0.135** -0.163** -0.189** (0.024) (0.028) (0.033) (0.024) (0.020) (0.024) Commune effects No Yes No No Yes No Household effects No No Yes No No Yes Region × time effects No Yes Yes No Yes Yes Adjusted R2 0.264 0.372 0.432 0.197 0.352 0.411 *Significant at 10 percent level. **Significant at 5 percent level. Note: Standard errors in parentheses. Standard errors are corrected for arbitrary het- eroskedasticity and the cluster-time design of the survey. All regressions include a quadratic in age and gender, a constant, and a year effect. "Ethnic minority" is a dummy variable that is 1 if the household was identified as a minority household in the VLSS in 1993. It is interacted (first row) with the 1998 indicator so that the reported coefficient has the interpretation of being the extra change in the probability a child works in a minority household relative to the decline experienced by nonminority households. Source: Authors' calculations from the 1993 and 1998 VLSSs. Two variables are reported in table 14.12. The 1998 indicator reports the average decline in child labor across all households between 1993 and 1998. The coefficient on "ethnic minority × 1998" reports the extra increment ex- perienced by ethnic minorities. Column (1) controls for child attributes and demonstrates that child labor appears to increase in the "all work" category for ethnic minorities between 1993 and 1998. In traditional work [column (4)], there is a 13.5-point decline in child labor in nonminority households be- tween 1993 and 1998 but only a 1.1-point decline for minorities. Columns (2) and (5) control for community fixed effects and region × time effects. In the "all work" category, the hypothesis that minorities experience the same de- cline as the rest of the population cannot be rejected. However, a statistically significant, smaller decline in child labor can be observed in traditional work for minority households. Columns (3) and (6) include household fixed effects. This further attenuates minority and nonminority differences in the "all work" category. However, significantly smaller declines in traditional work can be observed for minority households. It is interesting that in the re- gion × time effects (not pictured in table 14.12), for both traditional work and the "all work" category, controlling for a household's minority status does not change the fact that households in the Central Highlands experience Child Labor in Transition in Vietnam 543 smaller improvements in child labor than do households in the rural Mekong Delta. Conclusion This chapter demonstrates overwhelming evidence of a reduction in child labor over the 1990s. The survey of qualitative work presented here suggests that rising living standards have been important in driving this reduction in child labor, and the quantitative results of this and other studies are consis- tent with these qualitative findings. Children of both sexes and all ages in almost all rural and urban areas have experienced large declines in child labor during the 1990s. Ethnic minority children and the children of recent migrants appear to remain particularly vulnerable even by the late 1990s. They are more likely than nonminority children to work at all ages and in all work categories. Children of all ethnicities in the Central Highlands appear to have missed many of the improvements in the 1990s, and children in the rural Mekong Delta and in provincial towns experienced the largest declines in child labor. The evidence from qualitative and quantitative work presented here suggests that children still working in Vietnam are doing so because their families are too poor to support the basic needs of the family without the children's economic contributions. This link between poverty and child labor is clearly important in shaping appropriate policy responses and pub- lic action. It indicates, first, that a future development path that puts equi- table growth and poverty reduction at its core (such as the government of Vietnam's recently articulated Socioeconomic Development Strategy 2001­10) is likely to generate further reductions in child labor. Second, the link between poverty and child labor demonstrates that, at the household level, there should be concern surrounding the hardship that could confront poor families--including their children--if attempts are made to eliminate child labor without due consideration of the conse- quences for household income. This underscores the need for government responses to child labor to be formulated in a participatory way that in- volves all stakeholders at appropriate times, including working children and their families. Moreover, almost all child labor within Vietnam appears to take place within the homes of working children. Thus, active outreach to the families of working children may be the only way to address the cir- cumstances of most working children in Vietnam. Mechanisms to involve vulnerable children in planning are not well developed within government, though there have been some interesting innovations piloted by nongovern- mental organizations. These trends identified from the analysis of the VLSS data are undeni- ably positive in terms of child welfare. It would be misleading, however, to suggest that either the problem of child labor will completely evaporate as the economy continues to grow over the next decade or child labor has ceased to be a problem for policymakers. Economic growth over the 1990s 544 Economic Growth, Poverty, and Household Welfare in Vietnam has not delivered benefits evenly across all groups of children and house- holds, and a number of concerns remain despite the general pattern of improvement: · Though the trends indicate that working children reduced their number of working hours during the 1993­98 period, there is clear evidence (figure 14.1) that there is a group of child laborers, includ- ing many girls, involved in traditional work who are still working hours well in excess of the legal maximum set for their age group (24 hours a week). Indeed, those under age 15 are working hours well in excess of the legal maximum (42 hours) for an older category of 15- to 18-year-olds. The fact that they are working outside legal limits suggests that enforcement of child labor regulations is not influenc- ing their work patterns and must raise the question of whether other safeguards designed to protect young workers are effectively enforced. The VLSSs tell us little about this, but other studies suggest that these safeguards may not be enforced (ILSSA and University of Wollongong 2000). Moreover, there does not appear to be any mech- anism in place through which these safeguards could be applied to the activities of children within their own household. As the environ- ment for enterprise development improves and more competition is introduced, the issue of working conditions may become more important. Limited information on labor standards and working con- ditions is publicly available. MOLISA conducts regular labor force surveys, but these alone may be unable to pick up potential problems of deteriorating labor standards--particularly for children--as industrial growth continues. · At every age group, girls are more likely to work than boys (fig- ure 14.2). In particular, girls bear a greater burden of household work at every age than do their male counterparts. The literature on women in Vietnam clearly illustrates that this pattern of women shouldering heavier daily workloads continues into adulthood (Pop- ulation Council 1999; World Bank and the Poverty Working Group 1999); Although net enrollment rates in primary school are similar for boys and girls for the country as a whole, there is a disparity in the lowest expenditure quintile of the population (where 80 percent of girls are enrolled in school compared with 84 percent of boys). It ap- pears that girls may be more vulnerable than boys under situations of economic stress. Actions to address gender-based inequities in decisionmaking within the household are likely to be fundamental to reducing the domestic workload of girls. This, in turn, may require longer-term attitudinal change by both men and women to overcome gender-based stereotyping of roles and re- sponsibilities. In the short term, attitudinal differences may be circumvented by policies such as school subsidies or targeted inventions that lower the costs of schooling girls or mitigate households' need for the labor of girls. Child Labor in Transition in Vietnam 545 There is considerable need for further research into the vulnerabilities of girls in poor households and especially for an assessment of specific inter- ventions that might reduce their work burden. · Ethnic minority children work more than nonminority children at all ages. Qualitative studies suggest that concerns raised over differ- ences in work patterns for boys and girls may be particularly acute in ethnic minority areas, and the burden of work inside the house- hold for girls is likely to be more onerous and more likely to interfere with education for girls than for boys (Duong 1997; VN-Sweden MRDP 1999). With regard to traditional work, ethnic minority chil- dren have experienced smaller reductions in the likelihood of work- ing than have nonminority children. Other work shows how ethnic minority children suffer multiple disadvantages. They are more likely to live in poverty, have less access to health and education services (World Bank 1999; World Bank and the Poverty Working Group 1999), are more likely to be malnourished, and are less likely to survive childhood (Ministry of Health 2000). Their parents are less likely to have access to information and are more likely to be isolated from broader policymaking and decisionmaking processes (World Bank and the Poverty Working Group 1999). Addressing child labor among ethnic minority groups is unlikely to be effective if the many other deprivations they face are not simultaneously addressed. These are critical areas for public action that should form part of the ethnic minority development plans that the government of Vietnam has undertaken to formulate over the coming years (Socialist Repub- lic of Vietnam 2001). · Some of the patterns observed raise the question of whether the gov- ernment of Vietnam's development strategies for the next 10 years might exacerbate some forms of child labor. The government's Socioeconomic Development Strategy 2001­10 implies that rural- to-urban migration is likely to increase. (The urban population is pre- dicted to increase to 30­33 percent by 2010, a rate of increase beyond natural population growth.) Though VLSS data are likely to exclude much of the unregistered migrant community in urban areas, chil- dren of migrants are more likely to work, on average, than other children. This is strongly reinforced by other studies (Caseley and Buom n.d.; SCF U.K. 1999). Children of migrants will demand partic- ular attention in the future if a concerted program of support to child laborers is to be developed as Vietnam becomes more urbanized. In particular, it will be important to ensure that children of migrants are not denied access to basic services because of their residential status (SCF U.K. 1999). · Government strategies (Communist Party of Vietnam 2000; Ministry of Agriculture and Rural Development 2000) envisage a shift in the rural economy that places far greater emphasis on nonfarm activities and employment generation. This is widely accepted as being an 546 Economic Growth, Poverty, and Household Welfare in Vietnam important step in raising agricultural productivity and incomes and reducing rural poverty (World Bank 2000) and, by extension, child well-being. This analysis shows that children in households that start new enterprises work more than do children in households without home enterprises or with stable, long-term enterprises. Agencies con- cerned with child welfare must remain vigilant to possible changes in the profile of child labor as rural livelihoods become more dependent on nonfarm sources of income. · Although enrollment rates in primary education are high for a coun- try of this level of per capita GDP, and are good for both nonworking and most categories of working children (table 14.4), children who work outside the household emerge very clearly as a group who are educationally at risk. These children need to be targeted carefully under the government of Vietnam's Education for All initiative. Chil- dren's abilities to combine work and education may be undermined as full-day primary schooling is introduced over the next few years. Education may well become less compatible with working while si- multaneously becoming more expensive if the costs of extending the hours of education are borne privately. It is too early to anticipate what impact this change might have on child labor, but careful mon- itoring will be important. · Finally, there are categories of child labor that defy easy monitoring but that are both harmful--falling within the ILO description of the "worst forms of child labor"--and reportedly on the rise. Though there are no clear estimates of the number of children involved in the commercial sex industry, for example, some studies indicate that the sex industry is expanding rapidly, and as many as one-third of com- mercial sex workers are children (ILO-IPEC 2000). Street children and underage domestic workers are vulnerable to abuse because of the nature of their work, and they are likely to be neglected or underrep- resented in current data collection. The scale of child labor in these areas is very difficult to assess, given current information, but reports suggest that it is on the rise (Bond and Hayter 1998; SCF Sweden and USSH 2000; Youth Research Institute and Barnen 1999). There have also been reports that children working in gold mines are both un- registered (and unenumerated) and forced to continue working through the practice of withholding wages (SCF U.K. 1997). It will be important to generate more reliable indications of the extent of these very harmful and exploitative forms of child labor if effective action is to be designed and implemented. Notes The authors are grateful to Paul Glewwe and participants of the Development Strat- egy Institute, Ministry of Planning and Investment, and World Bank Conference on Economic Growth and Household Welfare, in Vietnam, for helpful comments. Child Labor in Transition in Vietnam 547 1. In much of the literature on child labor, distinctions are made between children "working" and child "labor." The former is often used to describe situations where children's economic contribution is not harmful to their overall development, and child "labor" describes situations where a child's opportunities for development are being constrained by his or her work. In this chapter, the terms "labor" and "work" are used interchangeably. 2. Glewwe and Jacoby (1998), in looking at retrospective school enrollment and labor market information in the 1992­93 Vietnam Living Standards Survey, argue that schooling declined and formal labor market participation rates increased in Vietnam from 1986 to 1991. Unfortunately, there are no data available to link these patterns to changes in household economic status. 3. The 1992­93 survey spanned a full year, starting in October 1992; the 1997­98 survey began in December 1997 and also lasted a year. For brevity's sake, reference is made to the surveys as the 1993 VLSS and 1998 VLSS, respectively. 4. The term "street children" is used here to describe children who are working on the streets and who live on the street (with or away from their families) or who live in basic shelter away from their families or who return at night to live with their fam- ilies off the street. This is a mixed group of children with different vulnerabilities. 5. Household work information is missing for 47 children (0.4 percent of the total sample). Six of these children report doing traditional work. Thus, throughout this chapter when participation in traditional work is considered, there will be 41 more children than when work participation across all categories is considered, and 47 more children than when work participation in household work is considered. Because household hours are missing for these 47 children, all "hours worked" obser- vations contain 47 fewer children than do "hours in traditional work" observations. 6. When changes in child labor through time are discussed in this chapter, ei- ther percentage point changes (calculated by subtracting the 1998 participation rate from the 1993 participation rate) or percentage changes (calculated by dividing the percentage point decline by the 1993 base) will be considered. With the first method, the fraction of children doing traditional work drops by 8.7 percentage points; in the second, there is a 28 percent decline in participation in child labor. 7. This is defined by Circular Number 09/TTLB, 13.4.95, issued by the Min- istry of Labor, Invalids and Social Affairs (MOLISA) and the Ministry of Health, which specifies 13 harmful situations and 81 forbidden occupations. 8. Though the sample sizes in the VLSSs are relatively large, when children are separated by age and by gender the number of children of a given age and gender becomes relatively small. Hence, estimates of child labor participation rates are smoothed using a nonparametric (local) regression smoother. The lines are local re- gression lines estimated with an Epanechnikov kernel and a bandwidth of 0.9. With such a small bandwidth, these regression lines look only slightly smoother than just the raw, by age, sample means. Later, when the sample is bifurcated by household characteristics where the number of children at a given age is very small, a larger bandwidth is used, and this regression procedure imposes more smoothing. 9. Density estimates are kernel densities estimated with a Gaussian kernel and a bandwidth chosen by Silverman's rule of thumb (1986). 10. Vijverberg and Haughton (2003) examine the growth and survival of house- hold enterprises in more depth. 11. Age and gender differences are controlled for by a quadratic in age and gender plus all interactions. Standard errors are corrected for commune (psu)/ survey round clustering and arbitrary heteroskedasticity. 548 Economic Growth, Poverty, and Household Welfare in Vietnam 12. The poverty headcount and the incidence of food poverty here are calcu- lated based on VLSS consumption data. 13. In practice, most households that report the head having ever moved report doing so within the past five years. Thus, using a more narrow definition does not affect these conclusions. Bibliography The word processed describes informally reproduced works that may not be commonly available through libraries. ActionAid Vietnam. 1999. Ha Tinh: A Participatory Poverty Assessment. Hanoi: ActionAid. Baulch, Bob, Truong Thi Kim Chuyen, Dominique Haughton, and Jonathan Haughton. 2004. "Ethnic Minority Development in Vietnam: A So- cioeconomic Perspective." In P. Glewwe, N. Agrawal, and D. Dollar, eds., Economic Growth, Poverty, and Household Welfare in Vietnam. Washington, D.C.: World Bank. Bond, Tim, and David Hayter. 1998. A Review on Child Labor, Street Children, Child Prostitution and Trafficking, Disability, the Family. Hanoi: UNICEF. Caseley, Jonathan, and Nguyen Van Buom. n.d. Survey on the Situation of Street Children in Hanoi. Hanoi: Youth Research Institute. Communist Party of Vietnam. 2000. "Draft Socio-Economic Development Strategy 2001­2010." Hanoi. Crawford, Sheena. 2000. "The Worst Forms of Child Labor. A Guide to Understanding the New Convention." University of Edinburgh. Processed. Department for International Development. 2001. "Providing Quality Basic Education for All." Paper prepared for the Poverty Task Force. Hanoi. Duong, Thanh Van. 1997. Girls' Work and Girls' Education in Vietnam. Hanoi: UNICEF. Edmonds, Eric. 2004. "Does Child Labor Decline with Improving Economic Status?" Journal of Human Resources, forthcoming. Edmonds, Eric, and Nina Pavcnik. 2002. "Does Globalization Increase Child Labor? Evidence from Vietnam." NBER Working Paper 8760. National Bureau of Economic Research, Cambridge, Mass. Glewwe, Paul, and Hanan Jacoby. 1998. "Schooling Enrollment and Com- pletion in Vietnam: An Investigation of Recent Trends." In David Dollar, Paul Glewwe, and Jennie Litvack, eds., Household Welfare and Vietnam's Transition. Washington, D.C.: World Bank. Child Labor in Transition in Vietnam 549 Glewwe, Paul, and Phong Nguyen. 2004. "Economic Mobility in Vietnam in the 1990s." In P. Glewwe, N. Agrawal, and D. Dollar, eds., Economic Growth, Poverty, and Household Welfare in Vietnam. Washington, D.C.: World Bank. Government of Vietnam­Donor Working Group. 2000. Vietnam: Managing Public Resources Better. Public Expenditure Review 2000. Hanoi: Vietnam Development Information Center. ILO-IPEC (International Labour Organisation­International Programme for the Elimination of Child Labour). 2000. "Children in Prostitution in Southern Viet Nam." Hanoi. Unpublished draft. ILSSA (Institute of Labor Science and Social Affairs) and University of Wollongong. 2000. A Study on Child Labor. Hanoi: Labor and Social Affairs Publishing House. Krueger, Alan. 1997. "International Labor Standards and Trade." In Michael Bruno and Boris Pleskovic, eds., Annual World Bank Conference on Development Economics 1996. Washington, D.C.: World Bank. Ministry of Agriculture and Rural Development. 2000. "Strategy of Agricul- ture and Rural Development 2001­2010." Hanoi. Ministry of Health. 2000. "Strategic Orientations on Health Care and Protec- tion for People in Mountainous and Remote Areas, 2001­2010." Hanoi. Minot, Nicholas, and Bob Baulch. 2004. "The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting." In P. Glewwe, N. Agrawal, and D. Dollar, eds., , Economic Growth, Poverty, and House- hold Welfare in Vietnam. Washington, D.C.: World Bank. MOLISA (Ministry of Labor, Invalids and Social Affairs). 2001. "Draft National Plan of Action on Implementation of ILO's Convention 182." Hanoi. MOLISA, UNICEF (United Nations Children's Fund), and Vietnam Com- mittee for the Protection and Care of Children. 2000. Analysis and Evaluation of Legislation and Policies on Care and Protection of Children in Especially Difficult Circumstances, 2nd ed. Hanoi. Nguyen, Nguyet Nga. 2004. "Trends in the Education Sector." In P. Glewwe, N. Agrawal, and D. Dollar, eds., Economic Growth, Poverty, and House- hold Welfare in Vietnam. Washington, D.C.: World Bank. Oxfam U.K. 1997. "The Way to School in Duyen Hai: Education Issues in a Mekong Delta District." Hanoi. Population Council. 1999. "Changes in Work and Fertility Patterns in Households during Vietnam's Post­Doi Moi Period, 1994­1999." Hanoi. 550 Economic Growth, Poverty, and Household Welfare in Vietnam SCF (Save the Children Fund) Sweden and USSH (University of Social Sciences and Humanities). 2000. Children in Domestic Service in Hanoi. Hanoi: National Political Publishing House. SCF U.K. (Save the Children Fund, United Kingdom). 1997. "From House- work to Goldmining: Child Labor in Rural Vietnam." Hanoi. . 1998. "Child Labor in Ho Chi Minh City." Hanoi. . 1999. "Ho Chi Minh City: A Participatory Poverty Assessment." Hanoi: World Bank. Silverman, Robert. 1986. Density Estimation. London: Chapman and Hall. Socialist Republic of Vietnam. 1999. "Overcoming Challenges to Achieve Ef- ficient and Sustainable Socio-Economic Development." Government Report to the Consultative Group Meeting. Hanoi. . 2001. "Interim Poverty Reduction Strategy Paper." Hanoi. Viet Nam Living Standards Survey (VNLSS). 1992­1993. Web site: www. worldbank.org/lsms/country/vn93/vn93bid.pdf. Vietnam Living Standards Survey (VLSS). 1997­98. Web site: www. worldbank.org/lsms/country/vn98/vn98bif.pdf. Viet Nam Youth Institute. 1999. "Children in Paid Work in Ho Chi Minh City." Hanoi. VN-Sweden MRDP (Vietnam-Sweden Mountain Rural Development Program). 1999. Lao Cai: A Participatory Poverty Assessment. Hanoi: World Bank. Vijverberg, Wim, and Jonathan Haughton. 2003. "Household Enterprises in Vietnam: Survival, Growth, and Living Standards." In P. Glewwe, N. Agrawal, and D. Dollar, eds., Economic Growth, Poverty, and Household Welfare in Vietnam. Washington, D.C.: World Bank. World Bank. 1999. Vietnam Voices of the Poor: Synthesis of Participatory Poverty Assessments. Hanoi. . 2000. Entering the 21st Century: Pillars of Development. Hanoi. World Bank and the Poverty Working Group. 1999. "Vietnam: Attacking Poverty." Joint report of the Government of Vietnam­Donor-NGO Poverty Working Group. Hanoi. Youth Research Institute and Radda Barnen. 1999. Possibilities of Reuniting Street Working Children with Their Families. Hanoi: National Political Publishing House. 15 Economic Mobility in Vietnam Paul Glewwe and Phong Nguyen Vietnam enjoyed high rates of economic growth in the 1990s. One conse- quence of this growth was a remarkable decrease in the rate of poverty, from 58 percent of the population in 1993 to 37 percent in 1998 (General Statistical Office 2000). Yet over the same period, inequality rose--the Gini coefficient of inequality for consumption expenditures increased from 0.330 to 0.354.1 This suggests that better-off Vietnamese households experienced greater increases in per capita consumption expenditures than did poorer house- holds. Indeed, the per capita expenditures of the poorest 20 percent (quin- tile) of the population was D (dong) 854,000 in 1993 and D 1,099,000 in 1998 (both in 1998 prices), which implies an increase of 29 percent. The analogous figures for the most well-off quintile of the population were D 3,911,000 and D 6,032,000, implying an increase of 54 percent. Yet this depiction of the consumption expenditures of the most well-off as growing at a much faster rate than the consumption expenditures of the poor is somewhat misleading. It is highly unlikely that all of the households that were in the poorest quintile of the population in 1993 were again in the poorest quintile in 1998; some of them may have moved up into more afflu- ent groups. This implies that, looking at the same households in both years, the households that were in the poorest quintile in 1993 experienced a gain in consumption expenditures greater than 29 percent. Similarly, some of the households that were in the top quintile in 1993 almost certainly were no longer in that category by 1998, so that looking at the same households would show growth in consumption expenditures of less than 54 percent among the most well-off quintile. The extent to which this movement of households' relative positions in the distribution of consumption expenditures tempers this scenario of in- creasing inequality is an important policy question. Another way to think about this issue is that the long-run distribution of consumption expendi- tures (and also of income) is more equally distributed than the short-run distribution if some individuals or households are poor in some years and 551 552 Economic Growth, Poverty, and Household Welfare in Vietnam others are poor in other years. Such economic mobility is a crucial aspect of the distribution of consumption expenditures and how that distribution changes over time. This chapter examines economic mobility in Vietnam in the 1990s, using recent household survey panel data from that country. Economic mobility is measured by comparing the incomes or expendi- tures of individuals or households over time. In practice, data are needed from a household survey that follows the same individuals or households over time. Recent examples of such studies are Fields and Ok (1999), Gardiner and Hills (1999), Gottschalk (1997), and Maasoumi and Trede (2001). A seri- ous problem with any empirical work on economic mobility is that house- hold income (and household expenditure) is likely to be measured with a large amount of error. This measurement error exaggerates both the amount of inequality at a given point in time and the degree of economic mobility over time. This chapter uses estimation methods that minimize the bias caused by measurement error. It begins with a brief discussion of the measurement of economic mobility, then shows how bias due to measurement error can be overcome in measures of mobility based on the correlation of individual or household income (or expenditure) over time. It then applies this method to a large panel dataset from Vietnam and finds that at least one-half of measured mobility is due to measurement error. Economic Mobility: Concepts and Measurement Economic mobility focuses on changes in an individual's or household's income over time.2 Yet the term "mobility" is often used in different ways. For example, an economy experiencing high economic growth that raises the incomes of all members may be characterized as having a large degree of mobility because everyone's income is increasing. However, there may be lit- tle change in individuals' income shares at each point in time, so that people do not change their relative positions in the distribution of income. In contrast, this chapter is interested in mobility in terms of its potential to reduce inequal- ity in the distribution of long-run income, which implies a focus on changes over time in the relative position of individuals or households in the distribu- tion of income. This concept of mobility is often referred to as relative mobility. The most common relative mobility measures are those based on corre- lation of functions of the income variable. Let y1 be the distribution of income in time 1 and let y2 be the distribution of income for the same house- holds or individuals in time 2. The simplest mobility measure can be defined as 1 ­ (y1, y2), where (y1, y2) is the correlation coefficient of y1 and y2. (The correlation coefficient is the covariance of y1 and y2 divided by the standard deviation of y1 and the standard deviations of y2.) If income shares do not change at all between the two time periods, then y1 and y2 are perfectly cor- related, so that (y1, y2) = 1 and the above mobility measure will be zero, sig- nifying no mobility. In contrast, if y1 and y2 are completely uncorrelated, so that any given household's income in the first time period has no relation- ship at all to its income in the second time period, then (y1, y2) = 0 and the Economic Mobility in Vietnam 553 above mobility measure equals one, which can be thought of as "full" mobility. Mobility measures based on the correlation coefficient range from zero (no mobility) to one (full mobility). In almost any data from any country, mobility will be somewhere between these two extremes.3 In fact, this ap- proach to measuring mobility can be generalized to include correlation between (monotonic) transformations of the income variable at two time points. For example, instead of examining the correlation between y1 and y2 one could use the correlation of the rank of y1 and y2, where the rank is one for the poorest person, two for the second poorest person, and so forth. Other mobility measures have been proposed using other transformations: For example, the Hart (1981) index uses the correlation of the logarithm of y1 and y2. In this chapter, several transformations will be used to check the ro- bustness of the findings. All of these mobility measures satisfy fundamental properties that a mobility index should have (see Glewwe [2003] for details). All mobility measures suffer from a serious problem--they exaggerate the extent of economic mobility when the income variable is measured with error. To see this problem, note that empirical studies of economic mobility typically use data from household surveys, which collect data on house- holds' incomes, expenditures, or both. Anyone who has observed how such data are collected understands that these variables are measured with error--in some cases, with a large amount of error (see Deaton [1997], Deaton and Grosh [2000]). By definition, virtually any measure of mobility will overestimate true mobility because fluctuations in measured income that are purely due to measurement error are mistakenly interpreted as actual income fluctuations. The simplest example of this is the case of no mobility at all. In this case, actual mobility should be zero, but random mea- surement error in the data will show some (spurious) mobility and thus will exaggerate the extent of actual mobility. The rest of this section explains in detail why errors in the measurement of income and expenditures exaggerate the extent of mobility and how instru- mental variables estimation can be used to correct for the bias caused by mea- surement error. Readers with an economics or statistics background should have no difficulty understanding the technical details of this argument. Read- ers with less technical backgrounds may find it somewhat challenging and thus may want to turn directly to the beginning of the next section. The bias caused by random measurement error can be demonstrated more formally with mobility measures that are based on the correlation of functions of the income variable. Let m(y1, y2) denote the simplest type of such a mobility measure, that is, m(y1, y2) = 1 - (y1, y2). In equation 15.1, the correlation coefficient is defined as: y1,y2 y1,y2 (15.1) (y1, y2) = = y1y2 2 2 y1y2 where y1,y2 indicates covariance and y1 and y2 indicate standard deviations. 554 Economic Growth, Poverty, and Household Welfare in Vietnam If the measurement errors in the two periods are uncorrelated with each other, (y1, y2) in equation 15.1 will be underestimated, implying that mobility--which equals 1 - (y1, y2)--will be overestimated. More specifi- cally, if random errors are added to y1 and y2, the numerator in equation 15.1 will be unchanged, but the denominator will become larger: me(y1, y2) = y1,y2 y1 + e1 y2 + e2 2 2 2 2 = (y1, y2) y1y2 2 2 y1y2 + y1e2 + e1y2 + e1e2 2 2 2 2 2 2 2 2 where me(y1, y2) is the observed correlation when measurement error is present and e1 and e2 are the random measurement errors added to y1 and y2, respectively. Intuitively, these random errors add "noise" to y1 and y2. The larger the amount of noise, the less correlated will be y1 and y2, moving (y1, y2) closer to zero and increasing m(y1, y2). Fortunately, there is a simple way to estimate (y1, y2) that avoids mea- surement error bias.4 All that is needed are instrumental variables that are correlated with y1 and y2 but uncorrelated with e1 and e2. To see this, recall that in a simple ordinary least squares (OLS) regression of a variable x1 on a constant term and one other variable, call it x2, the estimated coefficient for x2, call it b1LS, has a probability limit (plim) equal to x1,x2/x2. Similarly, a 2 regression of x2 on x1 produces an estimated coefficient, call it b2LS, that has a plim equal to x1,x2/x1. Thus, to estimate the correlation coefficient be- 2 tween y1 and y2, one can regress y1 on y2 and y2 on y1 and then take the square root of the products of the associated coefficients: (y1,y2)2 plim b1LSb2LS = y1y2 = (y1, y2) 2 2 where b1LS is the coefficient from an OLS regression of y1 on y2 and b2LS is the coefficient from an OLS regression of y2 on y1. Of course, if b1LS and b2LS are taken from simple OLS regressions, this estimate of (y1, y2) will still suffer from measurement error. It is possible, however, to use instrumental vari- ables to correct for this measurement error (assuming that credible instru- ments can be found) and then to use the two bLS coefficients to obtain a consistent estimate of (y1, y2). Although this method to overcome bias due to measurement error works perfectly well in theory, finding suitable instrumental variables is not a sim- ple task. Several problems can arise. Consider estimation of (y1, y2) by means of instrumental variables (the same reasoning applies for correlation between transformations of y1 and y2). If there were data on y1 and y2 with- out measurement error, it would be possible to consistently estimate (y1, y2) as the square root of the product of the OLS estimates of 1 and 2 from the following two regressions: (15.2) y1* = 1 + 1y2* + u1 (15.3) y2* = 2 + 2y1* + u2 Economic Mobility in Vietnam 555 where asterisks denote variables that are measured without error. The u terms are, by definition, uncorrelated with the regressors in each equation. Unfortunately, one never observes y1* or y2* but instead observes: (15.4) y1 = y1* + e1 (15.5) y2 = y2* + e2 where y1 and y2 denote observed values and e1 and e2 are random measure- ment errors. Substituting (15.4) and (15.5) into (15.2) and (15.3) gives the fol- lowing relationships between observed variables: (15.6) y1 = 1 + 1y2 + u1 + e1 - 1e2 (15.7) y2 = 2 + 2y1 + u2 + e2 - 2e1. For equation 15.6, an instrumental variable is needed that is correlated with y2* (and thus correlated with y2) but uncorrelated with u1 + e1 - 1e2, and for equation 15.7, an instrument is needed that is correlated with y1* (and thus with y1) but uncorrelated with u2 + e2 - 2e1. Turn to the requirement that the instrument for y1, denoted as z1, must be uncorrelated with u2 + e2 - 2e1. Consider an instrument for household per capita income or expenditure that has some causal relationship, such as land or capital stock or the education of the head of household. The first-stage equations for y1* and y2* are: (15.8) y1* = 1 + 1z1 + v1 (15.9) y2* = 2 + 2z2 + v2. Even if such an instrumental variable is completely uncorrelated with the measurement errors e1 and e2, it can be shown that z1 will be correlated with u1. Indeed, as explained in Glewwe (2003), attempts to estimate correlation of y1* and y2* using causal instrumental variables z1 and z2 will produce es- timates of the correlation of z1 and z2, not of y1* and y2*. One interesting ex- ample is the case where z does not change over time, so that z1 = z2; it can be shown that using this instrument will always yield a correlation coefficient of unity between y1 and y2. This problem with causal variables as instruments implies that repeated measurements of y1 and y2 should be used as instrumental variables. For example, income and expenditure could be treated as two separate mea- surements (with error) of an underlying "standard of living" variable. Thus, income could be used as an instrument for expenditures and vice versa. Glewwe (2003) shows that the estimated correlation coefficient does not de- pend on which is the instrument and which is the instrumented variable, and this method provides unbiased estimates of the correlation coefficient if the measurement errors at the same time point across the two measurements are uncorrelated (one must also assume that measurement errors are not correlated over time across the two different types of measurement). If the measurement errors are positively correlated between the two different measurements at one time point, instrumental variable (IV) estimates will 556 Economic Growth, Poverty, and Household Welfare in Vietnam overestimate mobility; if they are negatively correlated, IV estimates will underestimate mobility. To make these estimation issues more concrete, consider using income as an instrument for expenditures. This variable is constructed using different sections of the household questionnaire from Vietnam, so that random errors in recording data on the expenditure questions should have no effect on errors in recording data on the income questions. However, it is possible to imagine circumstances where observed income is positively correlated with measurement error in the expenditure variable. For example, suppose that some survey respondents are worried that the interviewer is a tax collector in disguise. These respondents may underreport both income and expendi- tures, so that the measurement errors in observed incomes are positively cor- related with measurement errors in the expenditure data.Another scenario is an interviewer who wants to finish the interview quickly. He or she may not ask probing questions about additional sources of income and additional types of expenditure, leading to the same problem. Finally, it is possible to have situations where the respondent is not the person most knowledgeable about household income and expenditure (perhaps because the most knowl- edgeable person is temporarily away) and thus does not report some types of income and expenditure. To summarize, this discussion suggests that any instrumental variable that has a causal role will yield inconsistent estimates of mobility, and an instrumental variable based on repeated measurements will tend to over- estimate mobility. The ideal instrument would be a repeated measurement variable for which there is a good argument that its measurement errors are uncorrelated with those of the variable of interest. Anthropometric mea- surements, particularly those based on weight, are probably the best variables of that type. A final point is that when there is more than one instrumental variable, the assumption that all the instruments are uncorre- lated with the (composite) error term can be tested using a standard over- identification test (see Davidson and MacKinnon [1993]). Mobility in Vietnam in the 1990s Vietnam provides an excellent case for studying mobility. In the 1980s, it was one of the poorest countries in the world. During the 1990s, its high rate of gross domestic product growth (8 percent) made it one of the most success- ful countries in reducing poverty and raising living standards. The reasons behind this success are currently under study. Despite these achievements, there is concern that the benefits of this economic growth are not being shared by all members of the population (World Bank 1999). Background and Data Another advantage of studying Vietnam is the availability of high-quality panel data. The data used in this chapter are taken from two household Economic Mobility in Vietnam 557 surveys conducted in the 1990s. The 1993 Vietnam Living Standards Survey (VLSS) was conducted from October 1992 to October 1993, collecting data from 4,800 households that made up a nationally representative sample. The 1998 VLSS was conducted from December 1997 to December 1998. It sam- pled 6,002 households, including about 4,300 of the households interviewed in the 1993 survey. Both surveys are patterned after the World Bank's Living Standards Measurement Study household surveys, which have been con- ducted in about 30 developing countries (Grosh and Glewwe 1998, 2000). The two VLSSs contain a large amount of data on many different topics. The appendix to chapter 1 of this volume provides a brief description of these surveys; for detailed information, see World Bank (2000, 2001).5 The focus of this chapter is on the overall economic welfare of households, par- ticularly the mobility of household welfare over time. In both surveys, the indicator of economic welfare is per capita household consumption expen- ditures. Although income data exist, they are likely to be less accurate than expenditure data. More important, standard economic theory measures in- dividual and household utility in terms of consumption expenditures, not income per se. However, income data can be useful. In particular, they can be used as an instrumental variable for per capita expenditures. A final issue to address regarding the data is the number of households in the panel dataset and whether these households are a representative sam- ple of Vietnamese households interviewed in the 1993 survey. This informa- tion is summarized in table 15.1. Of the original sample of 4,800 households, all but 96 (2 percent) were selected to be reinterviewed in the 1998 survey.6 Of the 4,704 selected households, 404 (8.4 percent) were not reinterviewed in the 1998 survey. More specifically, interviewers were instructed to return to the dwelling that the household inhabited in the 1993 survey. If the house- hold had moved within its village, interviewers attempted to find them and complete the interview. If the household had moved outside the village, no attempt was made to reinterview them. If some household members moved and others remained in the original dwelling, the interview was done using all the current inhabitants of the original dwelling (both original members and newcomers). Thus, of the 4,800 households interviewed in 1993, 4,300 were reinter- viewed in 1998, which is a retention rate of 89.6 percent. However, some of the households that remained may have rather tenuous links to the original household. First, households for which the head in 1993 is no longer a household member in 1998, and the new head in 1998 was not a member in the 1993 survey, should probably be excluded. Doing this eliminates 24 households, slightly reducing the retention rate to 89.1 percent. The remain- ing 4,276 households are the first sample used in this chapter. A stricter definition of household retention is to require that at least one-half of the individuals who were members in either 1993 or 1998 were members in both years. Doing this eliminates another 440 households, which leads to a reten- tion rate of 79.9 percent.7 The remaining 3,842 households are the second sample used in this chapter. 558 Economic Growth, Poverty, and Household Welfare in Vietnam Table 15.1. Panel Attrition Indicator Households Individuals Sample attrition over time Sample in 1993 survey 4,800 23,839 Excluded from 1998 survey 96 (2.0%) 421 (1.8%) Entire household moved 404 (8.4%) 1,786 (7.5%) Remaining households 4,300 (89.6%) 21,632 (90.7%) Among remaining 4,300 households Head is the same in both years 4,276 (89.1%) 21,538 (90.3%) 50 percent or more members are the same in both years 3,836 (79.9%) 19,100 (80.1%) 50 percent or more members are the same in both years, plus six "natural" cases 3,842 (80.0%) 19,119 (80.2%) Note: The six natural cases refer to households in which no one moved in or out of the house- hold in the past five years, but death or birth led to cases where the number of household mem- bers present in both years was less than 50 percent of the individuals who were members in either year. Examples are a household with three adults in 1993, of which two had died by 1998, and a household with a married couple in 1993 who had had three children by 1998. The figure of 19,119 includes 1,660 individuals in panel households who joined the house- hold after 1993. When those individuals are excluded, the number of individuals who were members in the 3,842 households in both years is 17,459, which is 74.5 percent of the individu- als originally surveyed in all 4,800 households in 1993. Source: Authors' calculations using the 1993 and 1998 VLSSs. Measured Mobility without Correction for Measurement Error By definition, mobility measures summarize in a single number the rela- tionship between the distribution of income at two points in time. These numbers do not always have intuitive appeal, so it is useful to start by de- picting mobility in the form of transition matrices. Table 15.2 presents (relative) transition matrices for Vietnam from 1993 to 1998, using the VLSS data. In each year, households are grouped by quintiles (poorest 20 percent, next poorest 20 percent, and so on, up to the most well-off 20 percent) to see how frequently they move across these groups. To check for robustness, both samples of panel households that were described above are used--one in which households are assumed to be the same if the head in one year was also a household member in the other year, and the other in which at least one-half of the individuals who were members in either 1993 or 1998 were members in both years. The results appear to display a substantial amount of mobility. Only 41 percent of the population remained in the same quintile in the two years; about 40 percent moved up or down by one quintile, and 19 percent moved up or down by two or more quintiles. These results are almost identical for the two samples. Thus, ignoring measurement error, one might conclude that the modest increase in inequality in Vietnam in the 1990s is not a major Economic Mobility in Vietnam 559 Table 15.2. Transition Matrix for Vietnam, 1993 and 1998 1998 quintile 1993 quintile 1 2 3 4 5 Row total Head of household is the same 1 2,186 1,143 689 332 45 4,395 (10.2%) (5.3%) (3.2%) (1.5%) (0.2%) (20.4%) 2 1,069 1,366 1,180 615 146 436 (5.0%) (6.3%) (5.5%) (2.9%) (0.7%) (20.3%) 3 501 936 1,169 1,244 501 4,351 (2.3%) (4.4%) (5.4%) (5.8%) (2.3%) (20.2%) 4 163 569 1,038 1,463 1,073 4,306 (0.8%) (2.6%) (4.8%) (6.8%) (5.0%) (20.0%) 5 48 148 440 929 2,536 4,101 (0.2%) (0.7%) (2.0%) (4.3%) (11.8%) (19.1%) Column total 3,967 4,162 4,516 4,583 4,301 21,529 (18.4%) (19.3%) (21.0%) (21.3%) (20.0%) (100.0%) 50 percent or more of household members are the same 1 2,007 1,054 620 242 33 3,956 (10.5%) (5.5%) (3.3%) (1.3%) (0.2%) (20.7%) 2 909 1,302 1,086 568 113 3,978 (4.8%) (6.8%) (5.7%) (3.0%) (0.6%) (20.8%) 3 463 874 1,077 1,127 402 3,943 (2.4%) (4.6%) (5.6%) (5.9%) (2.1%) (20.6%) 4 131 492 924 1,325 876 3,748 (0.7%) (2.6%) (4.8%) (6.9%) (4.6%) (19.6%) 5 36 106 385 792 2,160 3,479 (0.2%) (0.6%) (2.0%) (4.2%) (11.3%) (18.2%) Column total 3,546 3,828 4,092 4,054 3,584 19,104 (18.6%) (20.0%) (21.4%) (21.2%) (18.8%) (100.0%) Note: All numbers and percentages are in terms of individuals, not households. Source: Authors' calculations using the 1993 and 1998 VLSSs. concern, because low levels of household expenditures appear to be a tem- porary phenomenon for many households. In particular, about one-half of the population that was in the poorest quintile of the population in 1993 was no longer in that bottom quintile in 1998. How does this degree of mobility manifest itself in terms of mobility measures based on correlations of functions of the household expenditure variable? The answer is seen in table 15.3. As long as incomes are not nega- tively correlated over time, correlations will lie between zero (complete mobility in the sense that incomes in period 1 and period 2 are uncorrelated) and one (no mobility). Thus, all mobility measures based on correlation of 560 Economic Growth, Poverty, and Household Welfare in Vietnam Table 15.3. Estimated Mobility, Ignoring Measurement Error Sample in which head Sample in which 50 percent of household is the of members are the same Mobility index same in both years in both years 1 - (y1, y2) 0.309 0.299 1 - (y1, y2) (0.011) (0.012) 0.292 0.278 (0.011) (0.011) 1 - y1, y2 2 2 0.395 0.394 (0.012) (0.013) 1 - (rank(y1), rank(y2)) 0.331 0.316 (0.011) (0.012) 1 - (ln(y1), ln(y2)) 0.298 0.282 (0.011) (0.011) Number of households 4,281 3,845 Note: Standard errors in parentheses. Source: Authors' calculations using the 1993 and 1998 VLSSs. functions of the income variable will lie between one (complete mobility) and zero (no mobility). The mobility measures in table 15.3 range between 0.278 and 0.395, which in general indicates substantial mobility, although it is further from complete mobility than from complete immobility. The main point of table 15.3 is to show how the mobility seen in the tran- sition matrices of table 15.2 is measured by these mobility indexes. With one exception, the different mobility measures give similar results. Specifically, when the mobility index based on the correlation of the square of the income is excluded, the indexes range from 0.278 to 0.331. The highest value-- 0.395--occurs for the mobility index based on squaring the income variable. The mobility shown in tables 15.2 and 15.3 is almost certainly overesti- mated because it ignores measurement error. This issue is addressed in the next subsection, but before turning to that, it is useful to demonstrate that the regression approach is in fact an alternative way to estimate the correla- tion coefficient. This is seen in table 15.4 for the simple correlation coeffi- cient. The first line shows the correlation coefficients for per capita expendi- tures in the two years for both samples of households, which is simply one minus the associated mobility index given in table 15.3. The second line shows the OLS estimates of the parameter 2, the "slope" coefficient from a regression of 1993 per capita expenditures on 1998 per capita expenditures and a constant term. The third line shows the estimate of 1, the slope coef- ficient from a regression of 1993 per capita expenditures on 1998 per capita expenditures and a constant term. The fourth line demonstrates that the square root of the product of the estimates of these two coefficients yields the (estimated) correlation coefficient. Economic Mobility in Vietnam 561 Table 15.4. Correlation Coefficients without Correction for Measurement Error Sample in which 50 percent Sample in which head is the of members are the same Statistic same in both years in both years (y1, y2) 0.691 0.701 (0.011) (0.012) 2(OLS) 0.315 0.327 (0.013) (0.015) 1(OLS) 1.517 1.502 12 (0.107) (0.106) (OLS) 0.691 0.701 (0.028) (0.029) Note: OLS = Ordinary least squares. Standard errors in parentheses. Standard errors for OLS estimates account for clustered sample design. Standard errors for estimates of12 calculated using the delta method. Source: Authors' calculations using the 1993 and 1998 VLSSs. Estimates of Mobility Corrected for Measurement Error Once suitable instrumental variables are found, estimates of 1 and 2 that are free of attenuation bias can be obtained and then be used to calculate mobility. This was done for the mobility index 1 - (x, y) for three different types of instrumental variables. The first instrumental variable is simply household income per capita. Household income is collected in a different part of the VLSS questionnaire than the data used to calculate household expenditures, which reduces (but does not necessarily eliminate) the possibility that ran- dom errors in reported household expenditures spill over into the household income variable. Of course, household income is likely to be measured with random error as well, but as long as those errors are unrelated to the errors in the expenditure variable, it is still a valid instrumental variable. The first row of table 15.5 shows estimates of economic mobility when per capita expenditures are instrumented using household income. As ex- pected, the estimated mobility is much lower than the uncorrected estimates given in table 15.3. The figures in brackets show the IV-corrected estimates as a percentage of the uncorrected estimates. This figure is 56 percent for the "head same" sample and 53.8 percent for the "50% of members same" sam- ple. Recall that if measurement errors in income are positively correlated with measurement errors in expenditures, then these IV estimates will overestimate true mobility. This implies that these estimates can be thought of as upper bounds of the true amount of mobility. Thus, nearly half, and perhaps even more than half, of the mobility shown in table 15.3 is due to measurement error and is therefore spurious. 562 Economic Growth, Poverty, and Household Welfare in Vietnam Table 15.5. Estimated Mobility Using Three Different Types of Instrumental Variables Sample in which head Sample in which 50 percent is the same in of members are the same Instrument set both years in both years Per capita income 1 - (y1, y2) 0.173 [0.560] 0.161 [0.538] Durable goods 1- (y1, y2) 0.102 [0.330] 0.118 [0.395] 2(5) tests: 2 86.5*** 95.7*** 1 69.7*** 85.0*** Body mass index 1 - (y1, y2) 0.121 [0.392] 0.101 [0.338] Per capita income and body mass index 1 - (y1, y2) 0.167 [0.553] 0.153 [0.512] 2(1) tests: 2 0.4 0.4 1 2.6 3.5* *Significant at 10 percent level. ***Significant at 1 percent level. Note: Numbers in brackets are the estimated mobility as a fraction of estimated mobility (ignoring measurement error) given in table 15.3. Source: Authors' calculations using the 1993 and 1998 VLSSs. Because use of income as an instrumental variable is likely to over- estimate mobility, it is useful to estimate mobility by using other plausible instrumental variables. One possibility is the ownership of basic durable goods, such as televisions, bicycles, motorcycles, video cassette player- recorders (VCRs), and refrigerators. Households should make many fewer errors in reporting this information relative to reporting their incomes. If they make no errors at all, then there can be no correlation between errors in reported income and errors in the ownership of durable goods (because the latter type of error is always equal to zero). Estimates of mobility that correct for measurement error by using the ownership of durable goods as instrumental variables are reported in the second row of table 15.5. For both samples, the reported mobility is even lower than when household income is used as an instrument. Specifically, mobility is estimated to be 0.102 for the "head same" sample and 0.118 for the "50 percent of members are the same" sample. Taken at face value, these estimates suggest that almost two-thirds of the observed mobility in Vietnam seen in table 15.3 is purely due to measurement error in the expen- diture variable. Economic Mobility in Vietnam 563 Yet there are conceptual problems with durable goods as an instrumen- tal variable. First, it is possible that some durable goods are forgotten alto- gether (or deliberately omitted) during the interview. This could cause positive correlation in the measurement errors of expenditures and of durable goods, because the expenditure variable used here includes the estimated "use value" derived from the ownership of durable goods. Such correlation would lead to overestimation of mobility. Second, and more seri- ously, even if there were no measurement error in durable goods, it is possi- ble that this instrument is correlated with the u terms in equations 15.2 and 15.3. Because durable goods by definition last a long time, their production of use value in both time periods is similar to the impact of using causal vari- ables as instruments. Thus, these instrumental variables will not provide consistent estimates of mobility. The validity of durable goods as instrumental variables was checked using overidentification tests for the regressions corresponding to equations 15.2 and 15.3. This is possible because there were six durable goods used as instruments (color televisions, black and white televisions, bicycles, motor- bikes, VCRs, and refrigerators). The results are shown in the third and fourth rows of table 15.5. The overidentification tests strongly reject the as- sumption that the instrumental variables are uncorrelated with the compos- ite error terms in equations 15.4 and 15.5, so the estimates of mobility based on durable goods as instrumental variables must be discarded. A final instrumental variable considered in this chapter is the average body mass index (BMI) of adults ages 18 and over. The VLSSs collected height, weight, and arm circumference information from all household members. This can be used to calculate each adult's BMI, which is defined as the weight of an individual (in kilograms) divided by the square of his or her height (in meters). Very simply, this indicates how "heavy" a person is given his or her height. Poorer individuals have leaner diets and thus are less heavy. The key advantage of using BMI is that any measurement errors in it are extremely unlikely to be correlated with measurement errors in house- hold expenditures. First, this information was not collected by the inter- viewer who filled out the household questionnaire, but instead it was filled out by a completely different survey team member. Second, none of the sce- narios describing how income and expenditures may be correlated (such as households fearing tax collectors or interviewers wanting to finish the interview quickly) provides a coherent story as to why errors in the mea- surement of BMI should be correlated with errors in the measurement of household expenditures. Nevertheless, there is a potential that BMI is corre- lated with the u terms in equations 15.2 and 15.3. A "thin" person in 1993 may have a compromised ability to earn income not only in that year but also in future years, which implies that BMI in 1993 may have a direct causal relationship with household income and expenditures in 1998. This would lead to underestimation of the true amount of mobility. The fifth row of table 15.5 provides estimates of the mobility index 1 - (y1, y2) using household BMI (averaged over all adult household 564 Economic Growth, Poverty, and Household Welfare in Vietnam members) as an instrumental variable. Mobility is estimated at 0.121 for the "head same" sample and 0.101 for the "50 percent of members are the same" sample. As explained, this is a lower bound on the true value of mobility. This implies that true mobility may be only about one-third of the mobility measured without correcting for measurement error. As a final check on the regression results obtained from using household income and BMI as instrumental variables, both were used as instruments. The results are shown in the last three rows of table 15.5. Predicted mobility is slightly lower than it was when income alone was used. More interesting, because there are two instrumental variables, the exclusion restrictions can be tested using an overidentification test. In contrast to the case where durable goods were used, this specification easily passes the overidentifica- tion test in three of four cases, and in the fourth case the hypothesis that the instruments are not correlated with the compound error term can be rejected only at the 10 percent level. These estimates suggest that about one-half of measured mobility is spurious, which implies that true mobility is much lower than seen in tables 15.2 and 15.3. Conclusion Vietnam's rapid economic growth and relatively stable distribution of in- come suggest that all socioeconomic groups are benefiting from the boom- ing Vietnamese economy. Moreover, simple calculations using panel data suggest that there is a large amount of economic mobility within Vietnam, which is appealing because it suggests that the long-run distribution of in- come is more equal than the distribution at any given point in time. How- ever, such estimates of mobility may well overestimate true mobility be- cause there is a large amount of measurement error in the data. This chapter applies a simple method to estimate economic mobility that corrects for bias caused by measurement error in the variable of interest. When applied to the data from Vietnam, it shows that almost one-half, and perhaps even more, of economic mobility is an artifact of measurement error and is thus illusory. This implies that Vietnam's worries about increasing inequality cannot be dismissed by pointing to high economic mobility, be- cause such mobility is much lower than simple calculations suggest. Given the Vietnamese government's desire to minimize increases in inequality as economic growth continues, efforts to keep inequality from increasing must be in the forefront of the government's agenda. Notes The authors would like to thank Gary Fields and Andrew Foster for useful discus- sion and comments. 1. In this chapter, we focus on consumption expenditures instead of income be- cause expenditure data are, in general, more accurate (see Deaton and Grosh [2000]). 2. Following the literature on the measurement of mobility, this section refers to households' incomes, rather than to their consumption expenditures. However, everything in this section also applies to analyses based on household expenditures. Economic Mobility in Vietnam 565 3. In theory, negative correlation in incomes over time could exist; this would lead to mobility measures greater than one. But such a relationship, which would imply that households that are better off than average in the first time period would be poorer than average in the second time period, has never been found in data from any country. 4. The problem of measurement error and the use of instrumental variable methods to deal with it have been used in the literature on intergenerational mobil- ity. See, for example, Solon (1992). 5. These documents can be downloaded from the Web site http://www. worldbank.org/lsms/lsmshome.html. 6. The 96 excluded households were all from the Red River Delta region. They were excluded because the 1998 survey oversampled certain regions. The Red River Delta was not one of the oversampled regions, so fewer households were needed from it even though the sample size of the survey was increased from 4,800 to 6,002. 7. This retention rate includes six "natural cases" in which the number of house- hold members present in both years was less than 50 percent of the individuals who were members in either year; no one moved in or out of the household during the five-year time period because all changes were due to births or deaths. Examples are a household with three adults in 1993, of which two had died by 1998, and a house- hold with a married couple in 1993 who had had three children by 1998. Bibliography Davidson, Russell, and James MacKinnon. 1993. Estimation and Inference in Econometrics. New York: Oxford University Press. Deaton, Angus. 1997. The Analysis of Household Surveys: A Microeconomic Approach. Baltimore: Johns Hopkins University Press. Deaton, Angus, and Margaret Grosh. 2000. "Consumption." In Margaret Grosh and Paul Glewwe, eds., Designing Household Survey Question- naires for Developing Countries: Lessons from 15 Years of the Living Stan- dards Measurement Study. New York: Oxford University Press (for the World Bank). Fields, Gary, and Efe Ok. 1999. "Measuring Movement of Incomes." Economica 66(264): 455­71. Gardiner, Karen, and John Hills. 1999. "Policy Implications of New Data on Economic Mobility." Economic Journal 109(453): F91­F111. General Statistical Office. 2000. "Viet Nam Living Standards Survey 1997­98." Statistical Publishing House, Hanoi. Glewwe, Paul. 2003. "How Much of Estimated Economic Mobility Is Mea- surement Error? A Method to Remove Measurement Error, with an Application to Vietnam." University of Minnesota, Department of Applied Economics, St. Paul. Gottschalk, Peter. 1997. "Inequality, Income Growth and Mobility: The Basic Facts." Journal of Economic Perspectives 11(2): 21­40. 566 Economic Growth, Poverty, and Household Welfare in Vietnam Grosh, Margaret, and Paul Glewwe. 1998. "Data Watch: The World Bank's Living Standards Measurement Study Household Surveys. Journal of Economic Perspectives 12(1): 187­96. Grosh, Margaret, and Paul Glewwe, eds. 2000. Designing Household Survey Questionnaires for Developing Countries: Lessons from 15 Years of the Liv- ing Standards Measurement Study. New York: Oxford University Press (for the World Bank). Hart, Peter. 1981. "The Statics and Dynamics of Income Distributions: A Sur- vey." In N. Klevmarken and J. Lybeck, eds., The Statics and Dynamics of Income. Tieto, United Kingdom: Clevedon. Maasoumi, Esfanidar, and Mark Trede. 2001. "Comparing Income Mobility in Germany and the United States Using Generalized Entropy Mobil- ity Measures." Review of Economics and Statistics 83(3): 551­59. Solon, Gary. 1992. "Intergenerational Income Mobility in the United States." American Economic Review 82(3): 393­408. Viet Nam Living Standards Survey (VNLSS). 1992­1993. Web site: www. worldbank.org/lsms/country/vn93/vn93bid.pdf. Vietnam Living Standards Survey (VLSS). 1997­98. Web site: www. worldbank.org/lsms/country/vn98/vn98bif.pdf. World Bank. 1999. "Vietnam: Attacking Poverty." East Asia Region, World Bank, Washington, D.C. ------. 2000. "Viet Nam Living Standards Survey (VNLSS), 1992­93: Basic Information Document." Development Research Group, World Bank, Washington, D.C. ------. 2001. "Vietnam Living Standards Survey (VLSS), 1997­98: Basic Information Document." Development Research Group, World Bank, Washington, D.C. 16 Private Interhousehold Transfers in Vietnam Donald Cox This chapter investigates patterns of private, interhousehold income trans- fers using the 1992­93 and 1997­98 Vietnam Living Standards Surveys (VLSSs).1 Several questions will be explored, such as "Do private transfers help equalize incomes?" "Has Vietnam's rapid economic growth during the 1990s diminished the importance of private transfers?" "What are the so- cioeconomic and demographic factors most strongly associated with trans- fers?" "How much private transfer income flows from adult children to their parents?" "How much flows from parents to children?" "How do gifts differ from informal loans?" There are several reasons why private income transfers between house- holds are important, especially for a poor but rapidly growing country such as Vietnam. Private transfers can perform the same functions that public transfers do in richer countries. For example, private old-age support can function like social security for many elderly households. Furthermore, since the earliest analyses of private transfer behavior, economists have speculated that private and public transfers can interact. Most notably, Becker (1974) and Barro (1974) argued that expansions of public transfers could conceivably "crowd out" existing private transfers, leaving the distri- bution of living standards largely unchanged. Nevertheless, the specter of crowding out is not the only reason to be in- terested in private transfer behavior. Private transfers have been found to resemble credit markets in helping households overcome borrowing con- straints (Cox 1990), and they can assist households in dealing with risk (Cox and Jimenez 1998; Morduch 1995; Townsend 1994). Furthermore, they can help finance human capital investment by providing support to younger workers who have recently left home. Private income transfers could repre- sent one side of a transaction where in-kind help is exchanged between households (Bernheim, Shleifer, and Summers 1985; Cox 1987). 567 568 Economic Growth, Poverty, and Household Welfare in Vietnam The descriptive work in this chapter does not settle any of the deeper issues connected with crowding out or motivations for private transfers. Instead, it is a first step toward understanding the basics of private transfer be- havior in Vietnam. For example, for the problem of crowding out to have any policy relevance, private transfers need to be widespread and large enough to be supplanted by public transfers. Obviously, if there are few private trans- fers to begin with, not much would be crowded out by the expansion of public safety nets. Private transfers are indeed common and substantial in Vietnam, however, especially as a means of support for the elderly. Furthermore, much of the analysis in this chapter makes use of the panel aspect of the VLSSs. Despite the value of panel data for studying private transfer behavior, few true panel studies exist.2 After exploring the relation- ship between changes in private transfers and changes in household socio- economic and demographic variables, this study finds that private transfers appear responsive to changes in earning potential and life events, such as retirement or widowhood. The analysis is limited by the data in two ways. First, though private transfers can take many forms, such as time spent helping someone or the provision of moral support and companionship, this chapter's focus is on money transfers. The only in-kind transfers that are examined here are the money value of in-kind gifts included with monetary gifts. Second, though many transfers occur within rather than between households, almost all of this analysis is concerned with the latter. In addition, this work has a narrower definition of private transfers than the one used in earlier, related work that uses only the 1993 VLSS (Cox, Fetzer, and Jimenez 1998). There are two reasons for narrowing the defini- tion. First, the focus here is on transfer measures that contain information about the sources of transfers received and the destinations of transfers given, to analyze the directions of transfers according to generation. Not all private transfer measures in the earlier work provide information about generational directions of transfers. Second, this study concentrates on transfers that are measured consistently between the 1993 and 1998 surveys in creating a panel for private transfers. There is a further methodological limitation of this study. The analyses were limited to simple cross-tabulations to provide an overview of the data that is wide ranging and simple, rather than narrow and nuanced. This de- scriptive work is intended to stimulate interest in testing some of the more complex policy and behavioral issues, such as crowding out. Despite the simple methods, this chapter reaches several firm conclu- sions about private transfers: · Rapid economic growth has not diminished the importance of pri- vate transfers in Vietnam. · Private transfers are the main means of income redistribution in Vietnam; they are more than twice the size of public transfers. · Private transfers flow mostly from adult children to their parents, rather than the other way around. Private Interhousehold Transfers in Vietnam 569 · Those who give transfers are in better economic condition than those who receive them. · Inflows of private transfers increase upon retirement of the house- hold head. · Few gifts are given to nonrelatives, but half of all loans are made to nonrelatives. · Receiving private transfers in 1993 increased the chances of receiving them in 1998, but a nontrivial number of households changed from givers to recipients, or vice versa, between surveys. · Most private transfers flow between households sharing the same lo- cale, but many transfers cross regional boundaries, and a significant fraction of transfer income is received from foreign sources. · Victims of Typhoon Linda, a devastating storm that hit Vietnam's southernmost provinces just before the 1998 survey, appeared to re- ceive increased private transfers as a consequence of that disaster. · Though private transfers are widespread in Vietnam, they are not ubiquitous; half of the households neither gave nor received any gifts or loans in 1998, for instance. Some background is needed to help put the results in perspective. Vietnam experienced extraordinary economic growth in the 1990s, with living stan- dards a full two-thirds higher at the decade's end than at its beginning. Vietnam is still a poor, agrarian economy, but it has become a lot less poor-- and less agrarian--in recent years. Headcount poverty plunged from 58 per- cent to 37 percent from 1993 to 1998, thanks to broadly based growth (Glewwe, Gragnolati, and Zaman 2000). Agriculture accounted for a mere 25 percent of gross domestic product (GDP) at the end of the decade, com- pared with more than 40 percent at the beginning of the decade. Despite agriculture's dwindling share of GDP, farm productivity growth has been impressive. Increased rice yields have made Vietnam the world's second leading rice exporter. Vietnam's growth is due to two things. The first is a series of reform poli- cies (Doi Moi--"renovation") allowing free enterprise in farming, foreign direct investment, and elimination of price controls and trade barriers. The second, related to the first, is the start of a transition from agriculture to manufacturing. Despite recent, dramatic progress, Vietnam still has a severe poverty problem, which its public safety nets are ill equipped to handle (van de Walle 2004). An alternative to public safety nets is the system of informal, private safety nets in the form of interhousehold transfers. Cox, Fetzer, and Jimenez (1998) have explored the extent, magnitude, and patterns for these transfers in Vietnam using the 1993 VLSS. That study indicated that private transfers were large and widespread and frequently followed patterns simi- lar to means-tested public transfers, in that they appeared to flow from better-off to worse-off households. Cox, Fetzer, and Jimenez concluded by noting the difficulty of predicting the response of private transfers to eco- nomic liberalization on the basis of a single cross-section of data. 570 Economic Growth, Poverty, and Household Welfare in Vietnam This chapter extends that work by adding information from the 1998 VLSS. These two waves make it possible to track Vietnam's private transfers during a time of rapid economic growth and examine how they are related to changes in household incomes and life events. Another extension of the earlier work is to focus separately on familial giving versus lending; Cox, Fetzer, and Jimenez (1998) focused mostly on aggregated transfers. Conventional wisdom suggests that economic growth would weaken a household's ties with extended kin living elsewhere and would contribute to the ascendancy of the nuclear family.3 It also suggests that growth would alter the direction of private transfers, with less going from children to par- ents and more going from parents to children. It is important to know what growth did to Vietnam's interhousehold transfers. If, for example, extended familial networks do indeed begin to fall apart, growth might worsen income uncertainty and inequality. Further- more, the change in the direction of transfers--the so-called demographic transition--could threaten to leave a generation of elderly people deprived of familial support. Conversely, failure to attain demographic transition could leave younger persons short of the funds needed for acquiring human capital. Rapid economic growth in the region is, of course, not unprecedented; its impact on family networks in other countries has not gone unnoticed. Most notably, Lee, Parish, and Willis (1994) found that the rapid economic growth in Taiwan, China, did little to diminish children's support for their parents. Similar to Taiwan, Vietnam has a Confucian heritage that empha- sizes filial loyalty to parents. Also similarly to Taiwan, Vietnam's patterns of intergenerational support have changed little in the face of rapid economic growth. The first section of this chapter discusses patterns in private transfer and panel evidence. Then the geography of private transfers is investigated: how much is transferred within regions versus how much is transferred between regions, for example, and how much is transferred in faster versus slower growing areas. The final section describes private transfer inflows for house- holds affected by Typhoon Linda. Patterns in Private Transfers The description of private transfers in this chapter begins with separate dis- cussions of the 1993 and 1998 VLSSs, then proceeds to a description of panel evidence, which concentrates on the changes in household characteristics and their relationship to changes in private transfers. Cross-Sectional Patterns, 1993 Vietnam Living Standards Survey The 1993 VLSS was a nationwide household survey of 4,800 households. The VLSS is part of the World Bank's Living Standards Measurement Study, which collects information about household living standards for several Private Interhousehold Transfers in Vietnam 571 developing countries. The VLSS gathered data about the education, health, and employment of household members, as well as about household com- position, income, and expenditures. INTER-VIVOS TRANSFERS ("GIFTS") 1993. The 1993 VLSS measured private transfers in the form of money and goods transferred between households. The head of the household was asked questions about transfer inflows in the module for nonlabor income: "During the past 12 months has any member of your household received money or goods from persons who are not members of your household? For example, [has a member received] assis- tance sent by relatives working elsewhere or by children of household mem- bers, by friends or by neighbors?" The head was then asked to provide the names of those who sent trans- fers and their relationship to the person in the household who received them (father or daughter, for instance). The head was also asked to place a mone- tary value on in-kind transfers received. Transfer outflows were asked about in the module for household ex- penses. The question for outflows mirrors that of inflows. The head was asked: "During the past 12 months has any member of your household pro- vided money or goods to persons who are not members of your household? For example, [has a member provided assistance to] children or relatives living elsewhere or to other persons?" Paralleling what was asked about inflows, the head identified the person who sent each transfer and that person's relationship to the recipient. These transfers do not include remittances from someone temporarily away from home, because that person is still considered a household member, and the question is concerned only with transfers between households. Thus, a household is defined as a "recipient" if there is an affirmative answer to the question about transfer inflows, and a "giver" if there is an affirmative answer about outflows. About one-third of the households in the 1993 survey were involved with private transfers--as defined above--either as givers, recipients, or both (table 16.1).4 Table 16.1. Households Involved in Private Transfers, 1993 Category Number Percent Households involved in private transfers that 1,567 32.8 Only gave 597 12.5 Only received 780 16.3 Both gave and received 190 4.0 Households that neither gave nor received transfers 3,211 67.2 Total 4,778 100.0 Source: Author's calculations from the 1993 VLSS. 572 Economic Growth, Poverty, and Household Welfare in Vietnam Table 16.2. Transfers and Total Income, 1993 Indicator Private Public Transfers as a percentage of total income All households 7.9 2.3 Recipient households 32.0 11.7 Number of recipient households 970 1,014 Percentage of recipient households with pretransfer income in lowest quintile 25.0 21.8 Source: Author's calculations from the 1993 VLSS. Table 16.3. Household Economic Situation by Transfer Status, 1993 Indicator Net givers Net recipients Others Pre­private transfer income per year (thousand dong) 1,728 1,147 1,171 Post­private transfer income per year, (thousand dong) 1,633 1,689 1,171 Average percentage of economically active people in the household 57.8 50.5 55.0 Percentage with unemployed members 7.1 8.7 5.0 Percentage with educated household head 43.1 40.6 35.6 Number of households 646 913 3,219 Source: Author's calculations from the 1993 VLSS. For the whole sample, including those who did not receive anything, transfer receipts accounted for 8 percent of total household income (table 16.2). For only the sample of recipients, transfer receipts accounted for nearly one-third of income. Public transfers are as widespread as private transfers, but they are smaller, averaging slightly more than 2 percent of income for the whole sample (table 16.2). How do private and public transfers compare in their ability to reach the poorest households? First, consider the distribution of income before public or private transfers (that is, pretransfer income) and focus on the 20th per- centile. Twenty-five percent of private transfer recipients had pretransfer incomes that fell short of the 20th percentile, compared with 22 percent of public transfer recipients. Thus, at least by this crude measure, private trans- fers appear marginally better targeted to the poor.5 How do households giving private transfers differ from those receiving them? Table 16.3 contrasts the economic situation of givers, recipients, and those doing neither. Because some households both gave and received, net transfers are used to determine the excess of receipts over gifts and vice versa. Givers are in better economic condition than recipients. Consider house- hold income before private transfers, or pre­private transfer income. For Private Interhousehold Transfers in Vietnam 573 recipients, this is income minus net transfers; for givers, income plus net transfers. (Incomes are measured on an annual per capita basis and are expressed in thousands of dong per year [TDY].) Average pre­private transfer income of givers far exceeds that of recipients--1,728 TDY versus 1,147 TDY. At 1,171 TDY, the income of those neither giving nor receiving ("others") is in between these values but closer to that of recipients. Private transfers narrow the disparity between giver and recipient income, reducing the average income of givers to 1,633 TDY and raising that of recipients to 1,689 TDY. Note, too, that the posttransfer income of recipients exceeds that of the two other groups.6 Givers are better off than recipients in other ways besides pre­private transfer income. They have a larger proportion of economically active peo- ple in the household and experience a bit less unemployment. They are also better educated: Relatively more giver households are headed by someone with at least a lower secondary education. The figures in table 16.3 do not prove that private transfers flow from better-off to poorer households. Proof would require a dataset with matched donors and recipients. The VLSS records only one side of the transaction. It could be, for instance, that recipients have received their transfers from households even poorer than they. But the VLSS is a random sample of households, so the difference in the means of giver and recipient incomes is an unbiased estimate of the mean difference of giver and recipient incomes.7 Givers and recipients have different demographic characteristics, as well (table 16.4). Recipient households are more likely to be headed by an older person or a woman, and giver households are less likely to be headed by a younger person. Minority households, which make up 13.2 percent of the whole sample, are underrepresented among both givers and recipients.8 Interhousehold transfers and migration obviously have a lot to do with one another. An adult child making an interhousehold transfer to parents must have already left home. But what about a son or a daughter who takes a distant but temporary job and remits to parents? The VLSS downplays these transfers because it treats temporary migrants as members of the household. This is probably why having a person temporarily absent from the household matters so little for transfers.9 Net recipients have only a Table 16.4. Household Demographics by Transfer Status, 1993 Variable Net givers Net recipients Others Headed by young (percent) 7.9 11.1 11.8 Headed by elderly (percent) 13.2 26.1 15.1 Headed by female (percent) 20.9 35.9 21.5 With absent members (percent) 10.6 12.6 10.2 Household size (number of members) 5.8 5.4 5.9 Number of households 646 913 3,219 Source: Author's calculations from the 1993 VLSS. 574 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 16.1. Generational Directions of Private Transfers, 1993 4% Between other 3% Between nonrelatives relatives 6% Spouse to spouse 17% Old to young 41% Young to old 29% Sibling to sibling Source: Author's calculations using VLSS. slightly higher percentage of absent members than the other households (table 16.4). Table 16.4 shows that the elderly (defined as ages 60 and older) are over- represented among transfer recipients, but the young (defined as ages 30 and under) are not. These figures suggest that transfers tend to flow from young to old; more detailed calculations reinforce this result. Givers were asked to name the relationship of the recipient (his or her father, sister, son, father-in-law, and so on). Similarly, recipients were asked to name the rela- tionship of the donor. Transfers were classified by generational direction, using information about transfers received and transfers given. For instance, transfers given to older people were summed with transfers received from younger people to get total transfers from young to old. Transfers from old to young, sibling to sibling, and so forth were computed the same way.10 Fig- ure 16.1 displays this breakdown of private transfers. Figure 16.1 illustrates the importance of private old-age support in Vietnam. The value of transfers from young to old is more than twice as large as that from old to young (41 percent and 17 percent, respectively). This is exactly the opposite of what is observed in developed countries. (In the United States, for example, financial transfers from young to old are rare; most transfers go in the opposite direction, according to Cox and Raines [1985].) Figure 16.1 is also striking in the importance of transfers between siblings, which account for 29 percent of transfer flows. Most (92 percent) of what are called "young-to-old" transfers are trans- fers from children to their parents or parents-in-law, and nearly all (98 per- cent) of what are called "old-to-young" transfers are transfers from parents to their children or children-in-law.11 Private Interhousehold Transfers in Vietnam 575 LOANS, 1993. In addition to gifts, the 1993 VLSS contains information on interhousehold borrowing but little about lending. This discrepancy was fixed in the 1998 VLSS, so a detailed discussion of loans is deferred until the next section. But the reasonably detailed information in the 1993 VLSS about borrowing is nonetheless useful, because it shows that loans were wide- spread in 1993. Including loans in the definition of transfers received would almost double the percentage of recipient households, from 23 percent to 43 percent. Adding loans to gifts in the definition of transfers nearly doubles the percentage of private transfers in total income, from 8 percent to 15 per- cent. These issues are explored further in the next section, which analyzes the more comprehensive data on loans in the 1998 VLSS. Cross-Sectional Patterns, 1998 Vietnam Living Standards Survey One of the reasons for conducting the 1998 VLSS was to create a panel by rein- terviewing the 1993 VLSS households. Before analyzing the panel, though, two simpler issues are explored, using only the 1998 cross-section of the VLSS. The first issue concerns the stability of the cross-sectional private transfer patterns over time. They are indeed quite stable: The patterns found in 1993 are mostly repeated in 1998. The second issue concerns changes in the 1998 survey. Households were asked more detailed questions about inter- household loans and public transfers, and they were asked what their gifts and loans were spent on (that is, to finance a consumer durable, to buy food, and so on). The 1998 VLSS is larger than the 1993 VLSS: 1,200 new households were added to facilitate disaggregated analyses. The new households are not a self-weighted sample; urban areas and certain regions were oversampled.12 For this reason, the survey weights are used in the tables below. A comparison of the two cross-sections shows that Vietnam's economic growth has not reduced its private transfer activity; transfers were just as large and widespread in 1998 as they were in 1993. Table 16.5 classifies households according to their involvement with private transfers in 1998. Table 16.5. Households Involved in Private Transfers, 1998: A Comparison of Gifts and Gifts Plus Loans Gifts Gifts plus loans Category N Percent N Percent Households involved in private transfers that 2,208 37.2 3,112 52.4 Only gave 830 14.0 895 15.1 Only received 1,091 18.4 1,677 28.2 Both gave and received 287 4.8 540 9.1 Households that neither gave nor received 3,732 62.8 2,828 47.6 Total 5,940 100.0 5,940 100.0 Source: Author's calculations from the 1998 VLSS. 576 Economic Growth, Poverty, and Household Welfare in Vietnam Table 16.6. Transfers and Total Income, 1998 Private Private Indicator (gifts) (gifts and loans) Public Transfers as percentage of total income All households 6.8 12.2 3.1 Recipient households 25.3 32.7 17.6 Number of recipient households 1,379 2,217 1,178 Percent of recipient households with pretransfer income in lowest quintile 24.9 22.0 25.9 Source: Author's calculations from the 1998 VLSS. The first two columns of table 16.5 replicate what table 16.1 does for the 1993 households. The percentage of households participating in private transfers (as givers, recipients, or both) is found to be slightly higher in 1998 than it was in 1993--39 percent versus 35 percent, respectively. The next two columns in table 16.5 are based on an expanded definition of private transfers, which includes interhousehold borrowing and lending. (This was not possible for the 1993 VLSS, which had only limited informa- tion about household lending.) Expanding the definition of transfers to in- clude loans raises the percentage of households involved with transfers to 52 percent from the 37 percent based on only gifts (first row, table 16.5). The loans are large. Adding them to gifts raises the proportion of private trans- fers in total income to 12 percent from the 7 percent figure based on gifts (table 16.6, second and third columns). Public transfers were undercounted in the 1993 survey because social subsidies were not specified clearly. The 1998 survey gathered more detail about social subsidies and added questions about government poverty alle- viation and nongovernmental organization assistance. Despite these changes, public transfers are still only 3 percent of total income, a good deal less than that of private transfers, regardless of how the latter are defined (table 16.6).13 The 1993 survey results suggested that private transfers were slightly better targeted than public transfers. Table 16.6 overturns that conclusion. Among the households receiving public transfers, 26 percent were from the lowest income quintile (where income is measured before the private or public transfers are made). The equivalent figure for households receiving private transfers is either 25 percent or 22 percent, depending on whether loans are counted as part of private transfers. So it appears that, at least by the crude measures in table 16.6, public transfers are slightly better than pri- vate ones in reaching the poorest households.14 As with the 1993 VLSS, private transfers in the 1998 VLSS appear to flow from better-off to worse-off households. Table 16.7 contrasts the economic characteristics of net givers and net recipients. The entries in table 16.7 marked "Gifts only" replicate for the 1998 VLSS what was done in table 16.3 Private Interhousehold Transfers in Vietnam 577 Table 16.7. Household Economic Situation by Transfer Status, 1998 Indicator Net givers Net recipients Others Pre­private transfer income per year (thousand dong) Gifts only 4,677 2,925 2,634 Gifts plus loans 4,937 2,824 2,622 Post­private transfer income per year (thousand dong) Gifts only 4,231 4,035 2,634 Gifts plus loans 4,213 3,940 2,622 Average percentage of economically active people in the household Gifts only 57.9 49.4 55.9 Gifts plus loans 57.9 51.5 56.0 With unemployed members (percent) Gifts only 2.1 5.3 3.2 Gifts plus loans 2.2 4.5 3.2 With educated household head (percent) Gifts only 54.4 44.8 37.1 Gifts plus loans 55.1 41.9 36.0 Number of households Gifts only 906 1,292 3,742 Gifts plus loans 1,054 2,041 2,844 Note: Two different criteria for transfer status are used: gifts only and gifts plus loans. Source: Author's calculations from the 1998 VLSS. for the 1993 VLSS. As in 1993, the 1998 pre­private transfer income of net givers greatly exceeds that of recipients. Unlike in table 16.3, however, the pre­private transfer income of recipients in table 16.7 is slightly higher than that of "others" (those not involved with private transfers). Another differ- ence in table 16.7 is that it repeats the analysis with transfers defined as loans plus gifts. Table 16.7 shows that, regardless of how private transfers are de- fined, givers are in better economic condition than recipients (although again, "others" are the poorest of the three groups of households). They also are the least educated of the three groups (table 16.7). The inclusion of loans does not matter for demographic patterns, either, which are contrasted for givers, recipients, and "others" in table 16.8. The patterns are similar whether or not loans are counted. Note that the patterns for gifts in table 16.8 are similar to their 1993 counterparts in table 16.4. SOURCES OF LOANS VERSUS GIFTS, 1998. Though the inclusion of loans with gifts matters little for contrasting the characteristics of givers and recipients, the two forms of transfer do differ markedly in one respect. Gifts flow al- most exclusively between relatives, but loans do not. Half of all loan money 578 Economic Growth, Poverty, and Household Welfare in Vietnam Table 16.8. Household Demographics by Transfer Status, 1998 Variable Net givers Net recipients Others Headed by young (percent) Gifts only 3.2 4.3 6.5 Gifts plus loans 4.4 5.6 5.8 Headed by elderly (percent) Gifts only 12.2 30.2 16.0 Gifts plus loans 12.5 23.6 17.0 Headed by female (percent) Gifts only 22.1 33.4 21.5 Gifts plus loans 20.6 29.8 21.5 With absent members (percent) Gifts only 12.6 10.8 9.4 Gifts plus loans 11.8 10.9 9.1 Household size (number of members) Gifts only 5.3 5.0 5.6 Gifts plus loans 5.2 5.2 5.7 Number of households Gifts only 906 1,292 3,742 Gifts plus loans 1,054 2,041 2,844 Note: Two different criteria for transfer status are used: gifts only and gifts plus loans. Source: Author's calculations from the 1998 VLSS. flows between nonrelatives, described by survey respondents as "friends" or "neighbors." These informal loans made up one-third of total lending. The remaining two-thirds came from formal or quasi-formal sources such as banks, government credit programs, cooperatives, revolving credit associa- tions, or moneylenders, and these are not counted as interhousehold loans.15 People who borrowed from other households reported their relationship to the creditor (that is, parent, child, friend); those who lent money reported their relationship to the borrower. Half of the value of these informal loans occurs among nonrelatives (figure 16.2). The equivalent figure for gifts is a mere 2 percent (figure 16.3). Another innovation in the 1998 survey was the inclusion of questions about how gifts were used--whether for general consumption or for some investment-related purpose, such as schooling, investments in a farm or family business, or payment toward a house. A similar question was asked about borrowing, though the choices were different. One of these, to "buy food before harvest" clearly designated consumption, so it was lumped with "general consumption" to classify consumption loans. Several other choices, such as "working capital," "basic investment," "build or buy house," and "schooling," clearly represented investment and are classified here as such. The response "buy consumer durables" is also classified as investment.16 Private Interhousehold Transfers in Vietnam 579 Figure 16.2. Flows of Informal Lending, 1998 4% Young to old 19% Sibling to sibling 50% Between friends and neighbors 11% Old to young 17% Between other relatives Source: Author's calculations using VLSS. Figure 16.3. Generational Directions of Private Transfers, 1998 8% Between 2% Between nonrelatives other relatives 6% Spouse to spouse 39% Young to old 22% Old to young 23% Sibling to sibling Source: Author's calculations using VLSS. Still others--for example, to "repay a loan" or to "relend"--were harder to classify, so they were ignored in constructing the breakdown of loans by purpose. Gifts and loans are used differently. Nearly 75 percent of gifts--but fewer than 10 percent of loans--are spent for consumption (table 16.9). One might argue that the distinction between gifts and loans is little more than 580 Economic Growth, Poverty, and Household Welfare in Vietnam Table 16.9. Uses of Gifts versus Loans, 1998 (percent) Use Gifts Loans Consumption 71.6 9.3 Investment 28.4 90.7 Total 100.0 100.0 Source: Author's calculations from the 1998 VLSS. semantics--a gift, for example, could be reciprocated or a loan made below market interest. But the evidence suggests that there is more to the difference between loans and gifts than just labeling. They are used for different things and flow between different pairs of households. INTRAHOUSEHOLD TRANSFERS AND CORESIDENCE. The ideal study would track transfers between everyone, not just people from different households, but this would require elaborate measurements dealing with individual consumption and contributions to incomes of family farms and businesses that are beyond the scope of the VLSSs. Nonetheless, it is possible to learn something about intrahousehold transfers from the data. After all, the fact that a household contains persons who are not doing market work is prima facie evidence that some sort of transfer is occurring within the household. Rough estimates of the intrahousehold transfers can be calculated condi- tional on simplifying assumptions. The purpose is not to pinpoint exact intrahousehold transfers, which is not possible--instead, it is to demonstrate that intrahousehold transfers can, under plausible assumptions, far exceed interhousehold transfers. The calculations are based on 1993 data, but using 1998 data would not alter the conclusions. Imagine that total household income is divided for equal consumption among those doing market work and other persons. A market worker is someone reported to be economically active as a wage earner or participant in the family farm or business. Because most income (about five-sixths on average) comes from work, market workers implicitly transfer money to persons not engaged in market work. Assume for simplicity that all con- sumption is private, so there are no complications from nonexcludability or economies of scale. Finally, count children ages 0 to 4 years as 0.4 of an adult and children ages 5 to 14 years as 0.5 of an adult. (See Deaton [1997, p. 259] for a discussion of these equivalence scales.) These assumptions imply an average intrahousehold transfer of 187 TDY. Most of it, 102 TDY, goes to children age 14 or younger, but that still leaves a substantial 85 TDY being transferred between adults. These crude calculations show that intrahousehold transfers are potentially much larger than interhousehold transfers. Even counting only transfers to adults, this crude estimate of intrahousehold transfers is about double that of inter- household transfers.17 Private Interhousehold Transfers in Vietnam 581 Another, and in some ways complementary, intrahousehold transfer is the value of shared living arrangements for parents. These are difficult to measure in the VLSSs because it is hard to identify the persons responsible for making mortgage or rent payments. But the proportion of households headed by adult children living with nonworking parents, in-laws, or grandparents gives a rough idea of how widespread these shared living arrangements might be-- and 8 percent of the households in the 1993 VLSS fit this description. Still another form of transfer that occurs within the household is the ex- change of time-intensive, in-kind services between household members. The VLSS contains information about time each individual spends in house- work: preparing meals, washing clothes, cleaning house, and the like. With some assumptions about how such services are produced and shared, it can be inferred how large these implicit, time-intensive intrahousehold transfers are. For example, suppose such services are excludable, and suppose too that the same equivalence scales apply to consumption of these services as apply to other forms of consumption. Assume also that adults are more effi- cient at producing services than children are, and, for convenience, assume that these productivity differentials are the same as the consumption equiv- alence scales. Finally, suppose, again for simplicity, that there are no economies of scale in household production. (To illustrate, if a grandmother spends two hours cooking for herself and three other adults, those other adults each receive one-half hour of in-kind time transfers from her.) When this method is applied to the 1993 VLSS, using the information about house- work generates an average of 14 hours of time transfers per household per week. If this time is exchanged for the consumption provided by household members who work, then net intrahousehold transfers would be much lower than the figures cited above. Discussions of intrahousehold transfers are necessarily speculative because they are based on assumptions about unobservables such as household-sharing rules. They are intended only to illustrate the potential for intrahousehold transfers to exceed interhousehold transfers. A full accounting of transfers between all individuals is beyond the scope of this chapter. Panel Evidence Cross-sections leave several questions unanswered because they provide only a snapshot of private transfer patterns and reveal nothing about a household's experience over time. Does receiving transfers now make it more likely that they will be received later? Does transfer behavior respond to changes in the household's socioeconomic status? A panel is needed to address questions such as these. Panel evidence suggests some hysteresis in private transfer patterns, but many households also changed from recipients to givers, and vice versa, between surveys. Furthermore, changes in private transfers do indeed ap- pear responsive to changes in household characteristics, such as pretransfer 582 Economic Growth, Poverty, and Household Welfare in Vietnam income, demographic changes, and life-course events. Transfer inflows rise upon retirement and widowhood, for example, and are positively associ- ated with increases in health expenditures. The next sections provide more detail about these and other patterns. INCOME CHANGES. A leading issue in the literature on private transfers is how responsive they are to changes in household income. Indeed, such responsiveness is the key to the problem of crowding out, in which, for example, the introduction of public transfers would tend to supplant private transfers. Income responsiveness is critical, too, for determining whether private transfers insure households against income shortfalls or redistribute income to the less fortunate. Most empirical evidence on the income effects of private transfers is based on cross-sections. But cross-sectional evidence is of limited use for measuring the responsiveness of transfers to income, because contrasting transfer receipts for high- versus low-income households is not the same as looking at changes in transfers for the same household that was once well off but is now poor. Settling the issue of crowding out is beyond the scope of this simple de- scriptive analysis. Still, the descriptions that follow are illuminating in sev- eral respects. They show, for example, that there is enormous heterogeneity in private transfer responses. Furthermore, events such as retirement seem to matter a lot for transfer changes, but others, such as marriage of a son or daughter, do not appear to matter much at all. The first step is to explore how transfer status changes between surveys. Ninety percent of the households in the 1993 survey were reinterviewed in 1998. Most of those not reinterviewed had moved; many others were dropped deliberately. Only a small number were refusals.18 Eliminating the few others with missing information leaves a panel of 4,269 households. How many of the households that were recipients in 1993 changed into givers by 1998? How many remained recipients? Table 16.10 provides an- swers to questions such as these. Many households changed transfer status Table 16.10. Transitions in Transfers between 1993 and 1998 Transfer status, 1998 Transfer status, 1993 Net giver Other Net recipient Total Net giver 155 317 99 571 27.2 55.5 17.3 100.0 Other 391 1,933 480 2,804 13.9 68.9 17.1 100.0 Net recipient 98 372 424 894 11.0 41.6 47.4 100.0 Total 644 2,622 1,003 4,269 15.1 61.4 23.5 100.0 Source: Author's calculations from the 1993 and 1998 VLSSs. Private Interhousehold Transfers in Vietnam 583 Table 16.11. Increases versus Decreases in Real Private Transfers Per Capita by Windfalls versus Shortfalls in Pre­Private Transfer Income Per Capita Percentage of households whose excess of receipts over gifts Subsample Increased Decreased Stayed the same Total Households whose real pre­private transfer income Increased 25.8 28.8 45.4 100.0 N = 2,703 (64% of sample) Decreased 32.5 22.7 44.8 100.0 N = 1,518 (36% of sample) Decreased more than 50% 35.9 24.7 39.4 100.0 N = 587 (14% of sample) Source: Author's calculations from the 1993 and 1998 VLSSs. between surveys. Nonetheless, the data do indicate some inertia in transfer status. For example, only 11 percent of the net recipients in the 1993 survey became net givers in 1998, which is less than the unconditional 1998 figure of 15 percent (table 16.10). Nearly half of net recipients in 1993 remained so in 1998, even though the unconditional 1998 figure is less than one-fourth. Households that did not change their transfer status between surveys, and thus are located on the diagonal of table 16.10, represent 59 percent of the sample. What variables are correlated with changes in transfers? One way to address this question is to look at changes in some of the variables that were already examined in the cross-sections to see how they are related to changes in transfers. For example, changes in transfers for households that experienced shortfalls can be compared with windfalls in pre­private trans- fer income. These calculations are provided in table 16.11. The first two rows of table 16.11 split the sample by whether household income rose or fell between surveys. Income is measured before private transfers.19 It is also measured on a per capita basis, as are transfers.20 Both income and transfers are adjusted for inflation. For each survey, transfers are calculated as receipts minus gifts, or net transfer inflows. These inflows are positive or negative depending on whether receipts or gifts are larger. Changes in private transfers are: T = (receipts 1998 VLSS ­ gifts 1998 VLSS) ­ (receipts 1993 VLSS ­ gifts 1993 VLSS) which is the difference in net transfer inflows between survey years. Pre­private transfer income and net inflows of private transfers tend to move in opposite directions. Households with income shortfalls are more likely to experience increased transfer inflows. For example, 32.5 percent of 584 Economic Growth, Poverty, and Household Welfare in Vietnam households whose pre­private transfer income fell had increases in net trans- fer inflows between surveys, compared with 25.8 percent for households whose pre­private transfer income rose. Households that had particularly severe shortfalls in pre­private transfer income--decreases of 50 percent or more--were even more likely to have had a boost in transfer inflows. Nearly 36 percent of these households had increases in transfer inflows, compared with only 26 percent among households whose pre­private transfer, per capita incomes increased between surveys.21 However, although table 16.11 indicates that pretransfer income and private transfer inflows tend to move in opposite directions, there are many households for which the two move in the same direction. For example, 22.66 percent of the households experiencing shortfalls in pre­private transfer income also experienced shortfalls in net transfer inflows between surveys. Table 16.12 repeats the same calculations as table 16.11, but only for those households whose 1993 pretransfer incomes were less than the median, to see if the responsiveness of private transfers was more pronounced for households whose incomes were already low. The results, presented in table 16.12, support this idea. For example, 45.5 percent of low-income households whose incomes fell more than 50 percent had increases in net transfer inflows, compared with 25.5 percent of low-income households whose incomes increased. How large are these changes in transfers? The variation is enormous. For example, consider households that had income shortfalls. Define a household's transfer derivative as T/I, where I denotes the household's change in per capita, pre­private transfer income. A transfer derivative of ­1 means the entire shortfall was offset by increased private transfers. A Table 16.12. Increases versus Decreases in Private Transfers Per Capita by Windfalls versus Shortfalls in Pre­Private Transfer Income Per Capita, Restricted Sample: Households with Below-Median Per Capita Income in 1993 Percentage of households whose excess of receipts over gifts Subsample Increased Decreased Stayed the same Total Households whose pre­private transfer income Increased 25.5 27.8 46.7 100.0 (N = 1,677) Decreased 33.9 21.2 44.9 100.0 (N = 434) Decreased more than 50% 45.5 25.0 29.5 100.0 (N = 132) Source: Author's calculations from the 1993 and 1998 VLSSs. Private Interhousehold Transfers in Vietnam 585 Table 16.13. Changes in Private Transfers Per Capita for Households That Have Shortfalls in Their Pre­Private Transfer Income Per Capita Percentage of households that are Category Insured Destabilized Neither Total All households with shortfall 14.9 9.3 75.8 100.0 (N = 1,518) Poor households with shortfall 23.7 10.8 65.4 100.0 (N = 434) Source: Author's calculations from the 1993 and 1998 VLSSs. positive transfer derivative indicates that income changes are exacerbated by changes in private transfers. The values of the transfer derivatives at the 10th and 90th percentiles were ­0.75 and 0.29, respectively--an exceedingly wide range. Apply the following, admittedly arbitrary rule for identifying protection against income shortfalls. Classify as insured a household whose increases in transfers offset one-third or more of its income shortfall--that is, one with a transfer derivative of ­0.33 or lower. At the other end of the spectrum, des- ignate households with transfer derivatives larger than 0.33 as destabilized, because they simultaneously experience a reduction in both income and transfers. By these definitions, about 15 percent of the households were in- sured, and about 9 percent destabilized, by private transfers (table 16.13). The calculations were repeated for households with less than the median 1993 pretransfer income. For this subsample, nearly 24 percent were insured and nearly 11 percent destabilized. ECONOMIC AND DEMOGRAPHIC EVENTS. How are significant events--such as retirement, the loss of earners, the birth of children, and so on--related to private transfers? One way to explore this issue is to contrast the percent- ages of households experiencing increases versus decreases in per capita private transfers for various subsamples (table 16.14). The first row of table 16.14 provides figures for the entire sample, as a benchmark; these are then contrasted with the other rows, which pertain to select subsamples. Calculations for the first six subsamples in table 16.14 explore the role of private transfers as old-age support. They suggest that life events such as re- tirement and widowhood increase inflows of private transfers. For example, consider the second row, which gives the percentages of households whose private transfers increased, decreased, and remained unchanged for house- holds whose head retired between surveys.22 Nearly 41 percent of them had an increase in private transfer inflows, compared with only 28 percent for the whole sample. Conversely, only 19.7 percent of them had a decrease in 586 Economic Growth, Poverty, and Household Welfare in Vietnam Table 16.14. Increases versus Decreases in Per Capita Private Transfers, by Economic and Demographic Events Percentage of households whose private transfers Category Increased Decreased Stayed the same Total Entire sample 28.2 26.6 45.2 100.0 (N = 4,221) Subsample Household head retired 41.0 19.7 39.4 100.0 (N = 315) Nonhead retired 36.4 25.3 38.4 100.0 (N = 198) Widowed after 1993 33.0 28.9 38.1 100.0 (N = 97) Elderly person died 27.1 30.4 42.6 100.0 (N = 336) Son(s) left home 31.2 29.6 39.1 100.0 (N = 746) Daughter(s) left home 31.5 27.3 41.2 100.0 (N = 816) Loss of earner 31.3 25.1 43.6 100.0 (N = 1,476) Gain of earner 25.6 27.7 46.6 100.0 (N = 1,400) Son(s) married 29.2 25.3 45.5 100.0 (N = 672) Daughter(s) married 29.5 25.8 44.7 100.0 (N = 662) New child or children 25.1 23.9 51.1 100.0 (N = 854) Note: Rows may not total to 100% due to errors introduced by rounding. Source: Author's calculations from the 1993 and 1998 VLSSs. net inflows, compared with 26.6 percent for the entire sample. The third row in table 16.14 shows similar though slightly less dramatic results for the re- tirement of non­heads of household. Widowhood is also associated with in- creases in private transfers. The death of an elderly person between surveys is associated with decreases in private transfer inflows, further evidence that private transfers resemble old-age support. A child leaving home has a less pronounced association with private transfers. The subsample with one or more sons leaving home between sur- veys had a larger than average percentage with increases in private transfer inflows. This subsample, however, also had a larger than average percentage Private Interhousehold Transfers in Vietnam 587 with decreases in private transfer inflows. This finding is consistent with the idea that some sons are sources of increased receipts for parents and others are targets of increased gifts. The same pattern holds for daughters who left home (table 16.14). Changes in private transfers are not just restricted to economic and de- mographic changes associated with the aging of households. Losing an earner, regardless of his or her age, is associated with higher percentages of households with increases in net private transfer inflows. Conversely, gain- ing an earner is associated with lower percentages with increases in net private transfer inflows. The arrival of children is associated with mixed results for changes in private transfers. The sample with new children had a slightly smaller percentage with increases in transfer inflows, but a slightly smaller percent- age with decreases in transfer inflows, too. Marriage appears to have little effect on changes in transfers. Compared with the entire sample, the percentages of households whose transfers in- creased, decreased, and remained unchanged differ little for households whose children married between surveys. This is not to say, however, that marriage has little impact on total transfers. There are one-time expendi- tures and gifts that are related to the marriage ceremony, but these are not counted in the transfers variable. They are recorded in separate sections of the survey.23 Information from these sections in the 1998 survey was used to approxi- mate wedding-related expenditures and gifts--"approximate" in the sense that the VLSS lumps together funeral-related gifts with wedding gifts. The value of these one-time gifts is substantial. Among households that had one or more sons marrying between surveys, the average wedding expense, expressed as a fraction of their average total expenditures, was 8.1 percent. This number is all the more striking because the base includes not only households with sons who married in the previous year but also households with sons who married within the five years between surveys. In contrast, the wedding expenditure and gift data refer only to the previous 12 months. Part of the expense is defrayed by the receipt of gifts. Those same house- holds received wedding-related (but also possibly funeral-related [see above]) gifts equal to 4.4 percent of their total income. The comparable figures for households having at least one daughter marry are 5.7 percent (expenses) and 4.3 percent (gifts). Changes in private transfers are strongly related to changes in health expenditures. Consider the sample of households that increased the frac- tion of total spending on health by 5 or more percentage points between sur- veys (those marked "Health expenditures up" in table 16.15). Forty percent of these households had increases in net transfer inflows. The causality likely goes both ways--illness could prompt increases in private transfers, which in turn could help finance increased health expenditures. Similarly, a reduction in spending for health is associated with reductions in private transfers. 588 Economic Growth, Poverty, and Household Welfare in Vietnam Table 16.15. Increases versus Decreases in Per Capita Private Transfers, by Health-Related Events Percentage of households whose private transfers Category Increased Decreased Stayed the same Total Entire sample 28.2 26.6 45.2 100.0 (N = 4,221) Subsample Health expenditures up 40.6 22.2 37.3 100.0 (N = 604) Health expenditures down 24.5 29.9 45.6 100.0 (N = 882) Subtracted ill 27.5 27.4 45.0 100.0 (N = 1,888) Added ill 27.9 27.0 45.1 100.0 (N = 2,791) Only subtracted ill 27.6 25.9 46.5 100.0 (N = 815) Only added ill 28.1 26.0 45.9 100.0 (N = 1,718) Note: Rows may not total to 100% due to errors introduced by rounding. Source: Author's calculations from the 1993 and 1998 VLSSs. But changes in illness per se appear to have little impact on changes in transfers (final four rows, table 16.15). The proportions of households expe- riencing increases versus decreases in transfers between surveys differed lit- tle among subsamples with changes in the number of people who were ill. Like other life events, the effects of illness and other health-related events on private transfers merit further, separate study; the relationship between the two is likely to be complex. For example, becoming ill could raise someone's marginal utility of income if the illness is treatable, but reduce it if it is not. A SIMPLE REGRESSION. What are the partial correlations between the vari- ables discussed in tables 16.12­16.15? Are they statistically significant? To get a sense of this, changes in net transfers received are regressed on the economic and life-course variables from tables 16.14 and 16.15. But the goal is still descriptive, and it is important to note that the intention is not to at- tempt to estimate a causal model. For example, recall that health care ex- penditures are likely to be caused by private transfers as well as vice versa. The same could be true of several of the other right-hand-side variables in table 16.16. Table 16.16 conveys two messages. The first is that the regression results mostly reinforce what was shown in the cross-tabulations. For example, Private Interhousehold Transfers in Vietnam 589 Table 16.16. Regression of Differences in Log Net Transfer Receipts Per Capita on Economic and Demographic Events Estimated Estimated Variable Indicator coefficient t value mean Explanatory variable in per capita log income -0.192 -5.82 0.33 Loss of earner 0.314 2.78 0.35 Gain of earner -0.290 -3.22 0.33 Household head retired 0.661 4.13 0.07 Nonhead retired 0.361 1.83 0.05 Widowed after 1993 -0.070 -0.18 0.03 Elderly person died -0.035 -0.20 0.08 Household head died 0.280 0.68 0.03 Child left home -0.043 -0.30 0.31 Child married -0.046 -0.34 0.21 New child or children -0.026 -0.24 0.20 Health expenditures up 0.564 4.66 0.14 Health expenditures down -0.253 -2.42 0.21 Added ill person -0.106 -1.18 0.66 Subtracted ill person 0.060 0.07 0.45 Number of observations 4,221 Dependent variable mean 0.06 R2 0.03 F statistic 8.843 Source: Author's calculations from the 1993 and 1998 VLSSs. income increases are associated with reductions in private transfers. Second, it appears that economic events bear a more significant relationship to trans- fers than life-course events per se. For instance, situations such as losing an earner or suffering an income decline are both significantly related to changes in transfers, whereas widowhood is not. Of course, to move beyond merely describing correlations would require attention to problems of endogeneity, something that is beyond the scope of this chapter. Private Transfers, Regional Boundaries, and Economic Growth Vietnam experienced tremendous economic growth during the 1990s, but some places, such as cities in the south, grew much faster than others, such as rural areas in the northern mountains. Uneven growth increased divergence between the living standards of the better-off and the poor. Is it possible that private interhousehold transfers helped to narrow the widen- ing gap? For this to happen, private transfers would have to cross regional boundaries. This section investigates regional patterns of private trans- fers and contrasts transfer behavior in Vietnam's fastest-growing areas, the so-called "growth poles," with transfer behavior in the rest of the country. 590 Economic Growth, Poverty, and Household Welfare in Vietnam Although much money is being transferred between locales, most of it stays within the vicinity. In addition, economic growth is associated with more transfer activity, not less. PRIVATE TRANSFERS AND REGIONAL BOUNDARIES. For this section, Vietnam has been divided into 13 locales, using the VLSS regional definitions and distinctions between urban and rural areas. The 1993 VLSS divided the country into seven regions: the Northern Uplands, the Red River Delta (which includes the Hanoi-Haiphong corridor), the North Central Coast, the Central Coast (which includes Danang), the Central Highlands, the South- east (which includes Ho Chi Minh City), and the Mekong Delta. Only the Central Highlands region is completely rural. For the other regions, urban and rural locales are counted separately, generating 13 separate locales in all (that is, six urban and seven rural). Respondents report where transfer receipts came from and where gifts went, so it is possible to identify transfers within and between locales. In addition, some transfers came from outside Vietnam and, in a few cases, were sent outside Vietnam. Fifty-three percent of the transfers in 1993 were between households in the same locale (figure 16.4). Thirty-five percent crossed locale boundaries, and the remaining 12 percent crossed international boundaries. Figure 16.4 is constructed from transfer events, with no adjustment for the amount of money transferred. Figure 16.5 is based on the monetary value of transfers, provides quite a different picture of regional patterns, and shows that the largest transfers occur internationally. The breakdown of the money value of transfers in figure 16.5 departs markedly from figure 16.4. International transfers are more than 10 times larger than domestic ones. Thus, although only 12 percent of all transfers are international, they represent 63 percent of the money transferred. Figure 16.4. Regional Incidence of Private Transfers, 1993 12% International 53% Within 35% Between the locale locales Source: Author's calculations using VLSS. Private Interhousehold Transfers in Vietnam 591 Figure 16.5. Regional Value of Private Transfers, 1993 23% Within the locale 63% International 15% Between locales Source: Author's calculations using VLSS. Figure 16.6. Urban-Rural Incidence of Private Transfers, 1993 19% Urban to urban 51% Rural to rural 22% Urban to rural 8% Rural to urban Source: Author's calculations using VLSS. In contrast, there is not much difference in average domestic transfers that occur within versus between locales.24 The ratio of within-locale to between-locale transfers is about 3:2, whether measured in terms of events or money. URBAN-RURAL TRANSFER FLOWS. Another way to characterize the geo- graphic patterns of transfers is by urban-rural status. Most transfers do not cross urban-rural boundaries. The ones that do are mostly urban to rural transfers. For example, figure 16.6 is based on domestic transfer events, and it shows that 70 percent of all transfers were either rural to rural (51 percent) or urban to urban (19 percent). Among transfers that cross urban-rural 592 Economic Growth, Poverty, and Household Welfare in Vietnam boundaries, about three transfers flow from the city to the countryside for every one flowing in the opposite direction. Figure 16.7 repeats these calcu- lations but tracks the money value of transfers instead of events. The num- bers are different because more money is transferred among the urban households, who are better off on average. But the conclusions are the same--a little more than 70 percent of the money transferred does not cross urban-rural boundaries. THE 1998 VIETNAM LIVING STANDARDS SURVEY. The same figures were con- structed using data from the 1998 VLSS (figures 16.8 through 16.11).25 The Figure 16.7. Urban-Rural Value of Private Transfers, 1993 33% Rural to rural 40% Urban to urban 6% Rural to urban 21% Urban to rural Source: Author's calculations using VLSS. Figure 16.8. Regional Incidence of Private Transfers, 1998 14% International 49% Within the locale 37% Between locales Source: Author's calculations using VLSS. Private Interhousehold Transfers in Vietnam 593 regional patterns in private transfers are similar to those in 1993, except for the somewhat diminished importance of the money value of international transfers in 1998 (figure 16.9). Most domestic transfers still occur within rather than between locales, and the ratio of domestic transfers within ver- sus between locales is still 3:2. And, as in 1993, about 70 percent of transfers do not cross urban-rural boundaries. GROWTH POLES. Industrialized urban areas grew the fastest between sur- veys. These places--mainly the Hanoi-Haiphong corridor, Danang, and Ho Chi Minh City--attracted a disproportionate share of public investment Figure 16.9. Regional Value of Private Transfers, 1998 31% Within the locale 48% International 20% Between locales Source: Author's calculations using VLSS. Figure 16.10. Urban-Rural Incidence of Private Transfers, 1998 32% Urban to urban 39% Rural to rural 7% Rural to urban 22% Urban to rural Source: Author's calculations using VLSS. 594 Economic Growth, Poverty, and Household Welfare in Vietnam Figure 16.11. Urban-Rural Value of Private Transfers, 1998 22% Rural to rural 50% Urban to urban 9% Rural to urban 19% Urban to rural Source: Author's calculations using VLSS. Table 16.17. Households Involved in Private Transfers in 1993, by Growth-Pole Status Growth-pole Non-growth-pole households households Category Number Percent Number Percent Households involved in private transfers that 319 56.1 1,361 32.3 Only gave 87 15.3 500 11.9 Only received 183 32.2 710 16.9 Both gave and received 49 8.6 151 3.6 Neither gave nor received transfers 250 43.9 2,848 67.7 Total 569 100.0 4,209 100.0 Source: Author's calculations from the 1993 VLSS. (World Bank 2000b). How does private transfer activity in these growth poles compare with that of the rest of the country? The sample of households was split into those residing in growth poles and those not. The growth-pole households exhibited much more transfer activity in both surveys. More growth-pole households were involved in transfers, as givers, recipients, or both (tables 16.17 and 16.18). Furthermore, the panel evidence in table 16.19 indicates that, while a large percentage (40 percent) of growth-pole households experienced a decline in net trans- fer receipts between surveys, over one-third of them had increases in net Private Interhousehold Transfers in Vietnam 595 Table 16.18. Households Involved in Private Transfers in 1998, by Growth-Pole Status Growth-pole Non-growth-pole households households Category Number Percent Number Percent Households involved in private transfers that 544 54.5 1,843 37.3 Only gave 146 14.7 667 13.5 Only received 316 31.7 935 18.9 Both gave and received 81 8.2 242 4.9 Neither gave nor received transfers 455 45.5 3,098 62.7 Total 999 100.0 4,941 100.0 Source: Author's calculations from 1998 VLSS. Table 16.19. Increases versus Decreases in Private Transfers Per Capita, by Growth-Pole versus Non-Growth-Pole Residence Percentage of households whose excess of receipts over gifts Subsample Increased Decreased Stayed the same Total Households in growth-pole areas 33.9 40.1 25.9 100.0 (N = 451) Households in non-growth-pole areas 27.5 25.0 47.5 100.0 (N = 3,770) Source: Author's calculations from the 1993 and 1998 VLSSs. transfer receipts. Growth-pole households are clearly more active in private transfers. Part of the reason for the higher transfer activity in growth-pole areas could have to do with income inequality. Inequality, as measured by the coefficient of variation of log income, is higher in growth-pole areas than in non-growth-pole areas (table 16.20). This is true for both survey years, and it is also true whether income is measured before or after private transfers. If transfers help equalize incomes within these areas, as the numbers in table 16.20 suggest they do, then more inequality in income before transfers means more scope for income redistribution through private transfers, and hence more of them. 596 Economic Growth, Poverty, and Household Welfare in Vietnam Table 16.20. Income Inequality: Growth-Pole versus Non-Growth-Pole Households, 1993 and 1998 Coefficient of variation in Growth-pole Non-growth-pole household log income households households Income measure 1993 income before transfers 0.28 0.19 1993 income after transfers 0.24 0.18 1998 income before transfers 0.24 0.20 1998 income after transfers 0.20 0.18 Source: Author's calculations from the 1993 and 1998 VLSSs. Typhoon Linda In early November 1997, the southernmost provinces of the Mekong Delta were hit with a devastating typhoon, the worst it had seen since 1904. Typhoon Linda resulted in the death of more than 600 people and destroyed thousands of homes and one-half million hectares of rice fields. Hardest hit were the provinces of Ca Mau, Kien Giang, Bac Lieu, and Soc Trang. How did private transfers respond? This question is difficult to answer because of the structure of the VLSS. Respondents had a 12-month reference period for income and private trans- fers, making it difficult to pinpoint the storm's impact. In fact, the docu- mentation cites the long time frame as an advantage, which would "help to even out the impact of this natural disaster" (World Bank 2000a, p. 37). Still, this is a warning that survey results might have been affected for house- holds interviewed not long after the typhoon. Because of the ambiguity introduced by the time frame, it is difficult to gauge Typhoon Linda's effect on private transfer behavior. With this caveat in mind, the information from the community questionnaire was used, along with information about the path of the storm to identify the com- munes affected.26 The panel analysis of the impact of events on private transfers was replicated, and these households were compared with the rest of the sample (table 16.21). For the 340 panel households affected by the typhoon, evidence of pri- vate transfer effects is mixed. Compared with the rest of the sample, a slightly smaller percentage of these households had a reduction in net transfer receipts, but a slightly smaller percentage had an increase, too. Consistent with the documentation's assertions, however, timing of the interview appears to matter for gauging the typhoon's effects. For the subsample of affected households interviewed between December 1997 and March 1998, there does appear to be a boost in transfer inflows. A much larger than average percentage of them had increases in net transfer inflows between surveys, and a lower than average percentage had reductions in net transfer inflows. But this result must be interpreted with caution, because Private Interhousehold Transfers in Vietnam 597 Table 16.21. Increases versus Decreases in Private Transfers Per Capita, by Typhoon-Prone versus Non-Typhoon-Prone Percentage of households whose excess of receipts over gifts Subsample Increased Decreased Stayed the same Total Households in typhoon areas 25.59 22.35 52.06 100.0 (N = 340) Earlier interview 34.33 17.91 47.76 100.0 (N = 67) Later interview 23.44 23.44 53.11 100.0 (N = 273) Households in non-typhoon- prone areas 26.38 24.61 49.01 100.0 (N = 2,885) Source: Author's calculations from the 1993 and 1998 VLSSs. this subsample contains only 67 households. In addition, the Tet holiday was celebrated on January 28, 1998. Even with a 12-month time frame, in- terviews conducted around this time could contain distorted responses because of holiday-related increases in expenditures and perhaps transfers as well. Conclusion Typhoon Linda is just one of several issues addressed in this chapter that merits its own individual study and raises several questions for future re- search. Were households with links to non-typhoon-prone provinces better able to maintain their consumption in the face of declining incomes? How did the storm affect transfers over and above its impact on income? For in- stance, did it matter that incomes were affected by an unexpected storm instead of some other reason? Were risks well diversified, or did some typhoon victims find it necessary to provide support to nearby relatives or neighbors who were affected even more adversely? Just as many questions could be posed concerning the association of private transfers with health expenditures, retirement, and several of the other patterns uncovered by the simple descriptive analyses in this chapter. Of all the patterns that warrant further investigation, one of the most pressing is Vietnam's striking age patterns of private transfers. Unlike nearly all industrial countries--and many developing countries too--Vietnam's private transfers tend to flow from young to old, rather than the other way around. But one hallmark of a developed economy is its preponderance of transfers from older to younger people. Will Vietnam's age pattern of private transfers eventually reverse itself? If so, how will its elderly be 598 Economic Growth, Poverty, and Household Welfare in Vietnam provisioned in the new regime? If not, how will Vietnam's progress in edu- cation continue? Why Vietnam's age patterns in private transfers are the way they are, how they might be reversed, and what the likely consequences of such a reversal would be, are all critical research and policy questions for its future. Finally, it is worth keeping in mind that, with respect to Vietnam's pri- vate interhousehold transfers and loans, the "glass is half empty" in the sense that, in any given year, half of all households are not involved with private transfers or loans. This suggests strongly that there are likely many households that have no relatives or friends they can rely upon for money in times of need. The reach of private transfer networks in Vietnam is only partial, and their existence by no means obviates the potential for effective public income redistribution. Accordingly, readers are encouraged to con- sult this volume's analysis of the effectiveness of Vietnam's public safety nets by van de Walle (chapter 6) and the related study by Minot and Baulch (chapter 7) on Vietnam's spatial distribution of poverty and the prospects for geographic targeting of social safety nets. Appendix 16A Panel Response Rate Of the 4,800 households in the 1993 VLSS, 495 of them, or just over 10 per- cent, were not reinterviewed in 1998. Of these 495, 96 households were not interviewed because three communes had to be dropped from the Red River Delta as a result of the oversampling in 1998. (The extra 1,200 households in- cluded in 1998 were not representative of the population, but were chosen disproportionately from cities and other areas, especially the Central High- lands, to facilitate disaggregated analyses.) (World Bank 2000a, pp. 27­28). Of the 495 households, 281 were not reinterviewed because they moved away. Nineteen were missed because they were temporarily away from the commune, and 12 refused to be reinterviewed. One household was not rein- terviewed because it dissolved because of death. Sixteen other households of the 495 were not reinterviewed for some other reason. Forty-six house- holds were not reinterviewed for reasons unknown. A summary of the reasons households in the 1993 VLSS were not fol- lowed up in 1998 is provided below: Reason for attrition in the VLSS Panel Number Deliberately dropped from the sampling frame 96 Moved away 281 Temporarily away 19 Refused 12 Death 1 Miscellaneous reasons 36 Unknown reason 46 Reason not recorded 4 Total 495 Private Interhousehold Transfers in Vietnam 599 Notes This chapter has been prepared as part of the World Bank­funded project "Economic Growth and Household Welfare--Policy Lessons from Vietnam," directed by co- principal investigators David Dollar and Paul Glewwe. The author thanks Paul Glewwe for comments on earlier drafts and Emanuela Galasso for help with the 1998 Vietnam Living Standards Survey data. The support of the World Bank's research committee is gratefully acknowledged. 1. The 1992­93 survey spanned a full year, starting in October 1992; the 1997­98 survey began in December 1997 and also lasted a year. For brevity's sake, reference is made to the surveys as the 1993 VLSS and 1998 VLSS, respectively. 2. The best-known panel study for the United States (Altonji, Hayashi, and Kotlikoff 1997) uses only a cross-section of private transfer information. McGarry (2000) uses panel data on private transfers to test for parental altruism in the United States. Aside from Rosenzweig's study (1988) of private transfers in India, there are few other panel studies of private transfers for developing countries. 3. For an early discussion of this view, see, for example, Sussman (1953). 4. There are three other kinds of private transfers that are not counted in the survey questions but are available in the VLSSs: interhousehold loans, gifts related to ceremonies such as weddings or funerals, and inheritances. This chapter first focuses on the narrower definition (as defined by the two questions) for two reasons: to use measures that are consistent across the two VLSSs and use measures containing in- formation about generational directions of transfers. Loan information is incomplete in the 1993 VLSS, which has the flow of loans received but not loans given. This problem was remedied in the 1998 VLSS, so the discussion of loans is deferred until later in this chapter. Furthermore, the modules containing other forms of transfers, such as ceremonial gifts, do not provide the sources of gifts received or destinations of gifts given. Because the concern here is with the generational directions of transfers and developing a consistent definition of transfers over time for the panel analysis, a more restrictive definition of transfers has been used, and discussion of additional kinds of transfers has been deferred to a later part of this chapter. Applying the more inclusive definition of transfers, as analyzed in Cox, Fetzer, and Jimenez (1998), results in a much higher proportion of households involved in private transfers, though the patterns of these more inclusive transfers are similar to the narrower definition considered here. These more inclu- sive transfers are discussed briefly in a later section. 5. This result could have to do with the way public transfers are measured in the 1993 VLSS--similar calculations in this chapter for the 1998 VLSS, which has a better measure of public transfers, indicate little difference in how private and public transfers are targeted to the poor. 6. The reason for this apparent anomaly, where outflows and inflows of trans- fers do not balance, is because of transfers received from outside Vietnam, discussed later in this chapter. 7. Furthermore, a simple t statistic for testing the difference in means would be biased downward, because it would not take into account the (presumed) positive covariance between donor and recipient incomes. This simple t value (8.07) rejects the null hypothesis of equality of means at any popular level, which strongly sug- gests that private transfers do indeed, on average, flow from higher- to lower-income households. Note that the difference in means measures only differences between domestic givers and recipients. Taking into account the incomes of givers from abroad would likely strengthen this result. 600 Economic Growth, Poverty, and Household Welfare in Vietnam 8. Minority status is defined as in van de Walle and Gunewardena (2001): Eth- nic groups that are neither Kinh nor Chinese are defined as minorities. 9. A temporarily absent household member is defined as follows: the person is considered by the survey respondent to be a household member, and the person is reported to have been away from the household for 3 or more of the previous 12 months. 10. It would have been just as easy to concentrate only on either transfers received or transfers given alone to calculate generational directions. Aggregating information from both sides of the transaction is not double counting, but instead is averaging the two sources of transfer information--gifts and receipts. 11. These percentages--92 and 98 percent, respectively--are based on calcula- tions from the VLSS. The remaining 8 percent of transfers from young to old were given to grandparents, and the remaining 2 percent of transfers from old to young were given to grandchildren, nieces, and nephews. 12. In addition to urban households, rural households in the Central Coast, Central Highlands, and Southeast were oversampled. 13. See chapter 6 of this volume, by Dominique van de Walle, for a comprehen- sive analysis of Vietnam's public safety net. Her measure of public transfers includes a few more categories than are used in this chapter, such as educational scholarships, but these are miniscule compared with the largest public transfer, the social insur- ance fund. Thus, the value of aggregate public transfers used in this chapter is nearly identical to the one used by van de Walle. 14. Changing the cutoff to the lowest quartile does little to alter this conclusion, and the same goes for changing the cutoff to the poorest 15 percent of households. In each case, public transfers appear slightly better targeted than private gifts plus loans. 15. Labeling a source of credit "informal" is arbitrary to some extent. Here, in- formality is defined conservatively by including only relatives, friends, and neigh- bors. Obviously, credit cooperatives, moneylenders, and the like could be counted as informal sources, as well. 16. Sometimes purchases of durables are treated as consumption (that is, the United States National Income and Product Accounts) and sometimes as investment (the United States Flow of Funds Accounts). The latter is closer to the economic con- cept of investment--the act of paying now and enjoying later--as in buying a radio or bicycle that generates service over many years. 17. The disparity between intra- and interhousehold transfers in Vietnam ap- pears a good deal smaller, though, than the one reported for rural Pakistan by Kochar (2000), who finds that the predominant form of transfer from young to old occurs in the form of coresidence rather than cash transfers between households. 18. For more detail on panel attrition, see appendix 16A. 19. Pre­private transfer income is defined as income from all sources except private transfers received. A further possible adjustment, which was not made here, would be to add private transfers given to pre­private transfer income. It is not clear whether this is the proper pretransfer income measure, however, because gifts might be financed out of household wealth rather than income. 20. Household size is adjusted for equivalence scales: Children ages 0 to 4 years count for 0.4 of an adult, and children ages 5 to 14 years count for 0.5 of an adult. 21. The estimated correlation between changes in per capita, pre­private trans- fer income and changes in per capita, private transfers is negative (though small) and significant at any popular level (r^ = -0.106; estimated t value, -6.93). To take Private Interhousehold Transfers in Vietnam 601 account of the effects of outliers, a hyperbolic sine transformation was applied for each variable. The hyperbolic sine function, h(z) = ln(z + z2 + 1), is similar to a logarithm, except that it can be applied to negative values. 22. Retirement is defined as being economically active in the 1993 VLSS but not in the 1998 VLSS, for someone age 50 or older in the 1993 VLSS. 23. These expenditures and gifts were not included as part of interhousehold transfers for several reasons. First, unlike the transfer measures described in this chapter, there is no information about the sources of transfers received or the desti- nations of transfers given. For example, the recipients of expenditures for the wed- ding ceremony are a diffuse group that includes wedding guests. Second, the kinds of transfers directly connected with life events such as weddings, funerals, or holidays are likely to be behaviorally quite different from other kinds of transfers. For instance, wedding expenditures might include a substantial signaling compo- nent, intended to demonstrate family cohesiveness or intentions to provide future re- sources, or both. This kind of behavior lies outside the realm of more conventional theories about private transfers, and as such deserves a completely separate analysis, which is beyond the scope of this chapter. 24. The average within-locale transfer was 570 TDY; the average between- locale transfer was 532 TDY. In contrast, the average international transfer was 6,056 TDY. 25. Sample weights were used for constructing the 1998 VLSS figures. 26. About a dozen communes outside the Mekong Delta region also reported typhoon damage, and these were assumed to have been affected by storms other than Typhoon Linda. The 1993 commune identification numbers for the communes affected by Typhoon Linda were numbers 92, 98, 99, 101, 105, 106, 112, 113, 117­120, and 150. Bibliography The word processed describes informally reproduced works that may not be commonly available through libraries. Altonji, Joseph G., Fumio Hayashi, and Laurence Kotlikoff. 1997. "Parental Altruism and Inter Vivos Transfers: Theory and Evidence." Journal of Political Economy 105(December): 1121­66. Barro, Robert J. 1974. "Are Government Bonds Net Wealth?" Journal of Political Economy 82(November/December): 1095­117. Becker, Gary S. 1974. "A Theory of Social Interactions." Journal of Political Economy 82(November/December): 1063­94. Bernheim, B. Douglas, Andrei Shleifer, and Lawrence Summers. 1985. "The Strategic Bequest Motive." Journal of Political Economy 93(December): 1045­76. Cox, Donald. 1987. "Motives for Private Income Transfers." Journal of Politi- cal Economy 95(June): 1045­76. ------. 1990. "Intergenerational Transfers and Liquidity Constraints." Quarterly Journal of Economics 105(February): 187­217. 602 Economic Growth, Poverty, and Household Welfare in Vietnam Cox, Donald, James Fetzer, and Emmanuel Jimenez. 1998. "Private Transfers in Vietnam." In D. Dollar, P. Glewwe, and J. Litvack, eds. Household Welfare and Vietnam's Transition. Washington, D.C.: World Bank. Cox, Donald, and Emmanuel Jimenez. 1998. "Risk Sharing and Private Transfers: What about Urban Households?" Economic Development and Cultural Change 46(April): 621­37. Cox, Donald, and Fredric Raines. 1985. "Interfamily Transfers and Income Redistribution." In M. David and T. Smeeding, eds., Horizontal Equity, Uncertainty, and Measures of Well-Being. Chicago: University of Chicago Press. Deaton, Angus. 1997. The Analysis of Household Surveys: A Microeconometric Approach to Development Policy. Baltimore: Johns Hopkins University Press. Glewwe, Paul, Michele Gragnolati, and Hassan Zaman. 2000. "Who Gained from Vietnam's Boom in the 1990s? An Analysis of Poverty and In- equality Trends." Policy Research Working Paper 2275. World Bank, Policy Research Department, Washington, D.C. Kochar, Angini. 2000. "Parental Benefits from Intergenerational Coresi- dence: Empirical Evidence from Rural Pakistan." Journal of Political Economy 108(December): 1184­209. Lee, Yean-Ju, William L. Parish, and Robert J. Willis. 1994. "Sons, Daughters, and Intergenerational Support in Taiwan." American Journal of Sociol- ogy 99(January): 1010­41. McGarry, Kathleen. 2000. "Testing Parental Altruism: Implications of a Dy- namic Model." Working paper. Economics Department, University of California, Los Angeles. Minot, Nicholas, and Bob Baulch. 2004. "The Spatial Distribution of Poverty in Vietnam and the Potential for Targeting." In Paul Glewwe, Nisha Agrawal, and David Dollar, eds., Economic Growth, Poverty, and Household Welfare in Vietnam. Washington, D.C.: World Bank. Morduch, Jonathan. 1995. "Income Smoothing and Consumption Smooth- ing." Journal of Economic Perspectives 9(summer): 103­14. Rosenzweig, Mark R. 1988. "Risk, Implicit Contracts and the Family in Rural Areas of Low-Income Countries." The Economic Journal 98 (December): 1148­70. Sussman, Marvin B. 1953. "The Help Pattern in the Middle Class Family." American Sociological Review 28(February): 22­28. Townsend, Robert M. 1994. "Risk and Insurance in Village India." Economet- rica 62(May): 539­91. Private Interhousehold Transfers in Vietnam 603 van de Walle, Dominique. 2004. "The Static and Dynamic Incidence of Vietnam's Public Safety Net." In Paul Glewwe, Nisha Agrawal, and David Dollar, eds., Economic Growth, Poverty, and Household Welfare in Vietnam. Washington, D.C.: World Bank. van de Walle, Dominique, and Dileni Gunewardena. 2001. "Sources of Ethnic Inequality in Viet Nam." Journal of Development Economics 65(June): 177­207. Viet Nam Living Standards Survey (VNLSS). 1992­1993. Web site: www. worldbank.org/lsms/country/vn93/vn93bid.pdf. Vietnam Living Standards Survey (VLSS). 1997­98. Web site: www. worldbank.org/lsms/country/vn98/vn98bif.pdf. World Bank. 2000a. "Vietnam Living Standards Survey (VLSS), 1997­98: Basic Information." Poverty and Human Resources Division, Washington, D.C. Processed. ----. 2000b. "Vietnam 2010: Entering the 21st Century." Poverty Reduction and Economic Management Unit. Joint report of the World Bank, Asian Development Bank, and United Nations Development Pro- gramme. Washington, D.C. List of Figures, Maps, and Tables Figures 2.1 Per Capita GDP Growth Rates, 1960s to 1990s 33 2.2 Convergence and Divergence in Per Capita GDP Growth Rates in the 1990s 34 2.3 Economic Growth and Income of the Poor 35 2.4 Increased Trade and Changes in Inequality 36 2.5 Indicators of Vietnam's Reforms: Mid-1980s to Late 1990s 37 2.6 Effect of Economic Reform on Poverty, 1988­98 39 2.7 Poverty Reduction and Growth Rate in India, Vietnam, and China, 1992­98 40 2.8 Conditional Convergence, 1995­2035 41 2.9 Governance Pentagons: Vietnam, India, Thailand, Myanmar, and China 43 2.10 Global Competitiveness Report Rankings: Vietnam, China, and India 44 2.11 Maritime Transport to the United States (West Coast): Garments 46 2.12 Foreign Direct Investment as a Share of GDP, 1998 46 2.13 Foreign Direct Investment in Vietnam in the 1990s 47 3.1 Economic Growth in Vietnam 54 3.2 Levels of Schooling versus Labor Market Returns to Schooling by Region, 1998 68 3.3 Agricultural Labor Force Compared with Income Level 81 4.1 Household Choices in 1993 and 1998 101 5.1 Composition of Income, Vietnam 142 5.2 Trends in Rice Production 156 605 606 List of Figures, Maps, and Tables 5.3 Trends in Total Agricultural Output 157 5.4 The Distributional Impact of Changes in Rice Prices, Rural North 172 5.5 The Distributional Impact of Changes in Rice Prices, Rural South 174 7.1 Provincial Poverty Headcounts Estimated Using Urban-Rural and Stratum-Level Regression Models 251 7.2 Receiver Operating Characteristic Curves for Selected Targeting Variables 254 8.1 Kernel Densities of Per Capita Expenditure for 1998 280 8C.1 Results of the Multiple Adaptive Regression Spline Model for Kinh-Hoa Households 305 8C.2 Results of the Multiple Adaptive Regression Spline Model for Minority Households 307 9.1 Indirect Estimates of the Infant Mortality Rate 316 9.2 Indirect Estimates of the Under-Five Mortality Rate 317 9.3 Direct Estimates of Infant and Under-Five Mortality Rates 320 9.4 Trends in Under-Five Mortality, by Consumption Quintile 321 9.5 Concentration Curves for Under-Five Mortality 322 9.6 Decomposing the Sources of Changes in Child Mortality 325 10.1 Diarrhea, Stunting, and Wasting, 1993 354 10.2 Diarrhea, Stunting, and Wasting, 1998 355 12.1 Trends in School Enrollment, 1987­2000 427 12.2 School Net Enrollment Rates, by Quintile and Level, 1993 and 1998 429 12.3 School Net Enrollment Rate, by Gender and Education Level, 1993 and 1998 432 12.4 School Net Enrollment Rate, by Ethnicity and Level of Education, 1993 and 1998 433 12.5 Per Student Public and Private Spending on Primary Education, by Region, 1993 and 1998 446 12.6 Lorenz Distributions of Public Education Expenditure, by Level of Education, 1993 and 1998 450 12.7 Monthly Earnings of Private and Public Sector Workers, 1993 and 1998 457 14.1 Participation in Work, by Age and Gender 518 14.2 Distribution of Hours Worked in Nonagricultural Traditional Work 520 14.3 Participation in Work, by Age and Location 525 14.4 Reductions in the Probability That a Child Works in the Past Seven Days, by Region, Household Fixed Effects Results 529 14.5 Living Standards and the Decline in Child Labor 530 14.6 Participation in Work, by Age and Migration Status of Head 532 List of Figures, Maps, and Tables 607 14.7 Participation in Work, by Age and Household Enterprises 535 14.8 Participation in Traditional Work, by Age and Household Enterprise Change 538 14.9 Participation in Work, by Age and Ethnicity 540 16.1 Generational Directions of Private Transfers, 1993 574 16.2 Flows of Informal Lending, 1998 579 16.3 Generational Directions of Private Transfers, 1998 579 16.4 Regional Incidence of Private Transfers, 1993 590 16.5 Regional Value of Private Transfers, 1993 591 16.6 Urban-Rural Incidence of Private Transfers, 1993 591 16.7 Urban-Rural Value of Private Transfers, 1993 592 16.8 Regional Incidence of Private Transfers, 1998 592 16.9 Regional Value of Private Transfers, 1998 593 16.10 Urban-Rural Incidence of Private Transfers, 1998 593 16.11 Urban-Rural Value of Private Transfers, 1998 594 Maps 7.1 Incidence of Poverty by Province 245 7.2 Incidence of Rural Poverty by Province 249 8.1 Ethnicity and Expenditures 275 Tables 1.1 Vietnam's Economic and Social Performance 5 1.2 Changes in Inequality in the 1990s 8 2.1 Estimated Growth Effect of Vietnam's Reforms 38 3.1 Labor Force Participation, Ages 16­60 55 3.2 Unemployment Rates 56 3.3 Sectoral Composition of Employment 57 3.4 Rural Composition of Employment, by Sex 57 3.5 Urban Composition of Employment, by Sex 57 3.6 Wage Levels and Growth, by Region 59 3.7 Skilled, Private, Nonagricultural Wages, by Region 60 3.8 Average Annual Hours Worked in Wage Employment 61 3.9 State-Owned Enterprise Employment 62 3.10 Wage Regressions: Estimated Effect of Education and Experience 63 3.11 Wage Regressions: Broader Group Correlates 65 3.12 Wage Regressions, by Region, 1998 66 3.13 Wages, by Quintile 73 3.14 Changes in Wages, by 1993 Quintile 73 3.15 Changes in Wages, by 1998 Quintile 74 3.16 Inequality Measures of 1993 Annual Wages, by Region 75 608 List of Figures, Maps, and Tables 3.17 Inequality Measures of 1998 Annual Wages, by Region 75 3.18 Inequality of Household Expenditure Per Capita 76 3.19 Per Capita Household Expenditure, Broken Down by Main Income Source, 1998 76 3.20 Decomposition of Household Income Inequality, by Source, 1993 77 3.21 Decomposition of Household Income Inequality, by Source, 1998 77 3.22 Inequality Measures of Household Expenditure, Per Capita, by Main Income Source, 1993 79 3.23 Inequality Measures of Household Expenditure, Per Capita, by Main Income Source, 1998 79 3.24 Agriculture, in Relation to Income, across Countries 82 3.25 Inequality Projections, 2003 and 2008 83 4.1 Labor Market Participation, by Residence and Gender 97 4.2 Labor Market Participation by Quintile, Region, and Ethnicity 98 4.3 Percentage of Households with a Nonfarm Household Enterprise, 1993 and 1998 99 4.4 Labor Market Participation, by Age and Schooling Level 101 4.5 Logistic Model of Operation of an Enterprise, 1993 103 4.6 Accounting for the Panel Enterprises 106 4.7 Comparison of Panel Enterprises, Nonpanel Enterprises, and Enterprises in Attrited Households 109 4.8 Determinants of the Attrition Process: A Logistic Model 113 4.9 Enterprise Survival: A Logistic Model 114 4.10 Probability That a 1993 Enterprise Survived until 1998 116 4.11 Enterprise Startup: A Logistic Model 118 4.12 Dynamics in Enterprise Income 121 4.13 Determinants of Growth in Enterprise Income 122 5.1 Household Incomes by Source 138 5.2 Selected Price Indexes for 1998 Relative to 1993, Rural 144 5.3 Rice Prices across Time and over Space 146 5.4 Urban Food Demand Patterns and Expenditure Elasticities 149 5.5 Rural Food Demand Patterns and Expenditure Elasticities 150 5.6 Crop Output, Acreage, and Sales: Shares 153 5.7 Changes in Crop Output, Land Use, and Fertilizer Use Per Household 154 5.8 Decompositions of Output Growth 161 5.9 Rice Marketing, per Household 164 5.10 Rice Marketing Per Household, by Region, Rural Only 166 5.11 Decompositions of Rural Income Inequality, by Source of Income 179 List of Figures, Maps, and Tables 609 6.1 Incidence of Social Welfare Income 198 6.2 Population Receiving Social Welfare Income 200 6.3 Total Spending on Social Welfare in 1998 as Reported in the Vietnam Living Standards Survey 202 6.4 Incidence of School Fee Exemptions 203 6.5 Incidence of Household Contributions 206 6.6 Rural Population Living in a Commune with Poverty Programs and Other Programs 208 6.7 Small-Town Population Living in a Commune with Poverty Programs and Other Programs 209 6.8 Changes in Incidence of Social Transfers over Time 211 6.9 Incidence of Changes in Transfers by Initial Consumption and Changes in Consumption over Time 213 6.10 Baseline Discrete Joint Distribution 215 6.11 Joint Distribution without Transfers 215 6.12 No Change in Transfers between 1993 and 1998 216 6.13 Actual 1998 Distribution versus Uniform Allocation of 1998 Transfers 217 6.14 Actual 1998 Distribution versus 1998 Transfers Targeted on Equal Per Capita Basis to the Poor 217 6.15 Incidence of Proportionate Changes in Social Incomes 218 6.16 Incidence of Poverty-Related Programs and Beneficiaries by Rural Poor, Middle, and Rich Communes 220 6.17 Incidence of Social Transfers across the Rural Population by Terciles and Poor, Middle, or Rich Communes 221 6.18 Incidence of Social Transfer Amounts across the Rural Population by Terciles and Poor, Middle, or Rich Communes 222 7.1 Household Characteristics Common to the Census and the Vietnam Living Standards Survey 232 7.2 Determinants of Per Capita Expenditure for Rural and Urban Areas 237 7.3 Tests of Significance of Groups of Explanatory Variables in Urban-Rural Regressions 239 7.4 Comparison of Original and Census-Based Poverty Headcounts 242 7.5 Differences in Regional Poverty Headcounts and Their Statistical Significance 243 7.6 Provincial Poverty Headcounts Estimated with Urban-Rural Regression Model 246 7.7 Accuracy of Different Variables in Targeting Poor Households 256 7A Descriptive Statistics for Variables Used in Regression Analysis 259 7B Determinants of Per Capita Expenditure of Each Stratum 262 610 List of Figures, Maps, and Tables 7C Tests of Significance of Groups of Explanatory Variables in Stratum-Level Regression Model 264 7D Poverty Headcounts Estimated with Stratum-Level Regression 266 8.1 Characteristics of Majority and Minority Households, 1993 and 1998 276 8.2 Key Indicators for Major Minority Groups, 1993 and 1998 278 8.3 Primary School Enrollment Rates by Ethnic Group and Sex, 1999 281 8.4 Lower Secondary School Enrollment Rates by Ethnic Group and Sex, 1999 282 8.5 Intermarriage of Household Heads, 1999 283 8.6 Distribution of Ethnicity and Religion 285 8.7 Endowments of Majority and Minority Households, 1993 and 1998 289 8.8 Community Remoteness Variables for Majority and Minority Households, 1998 291 8.9 Decomposition of the Sources of Ethnic Inequality, 1998 296 8A Regression Estimates of Expenditure Equations, Full Sample of Households, 1998 300 8B Regression Estimates of Expenditure Equations, Mixed Commune Sample of Households, 1998 302 8C.1 Basis Functions for MARS Model of Log of Real Per Capita Income, Kinh and Chinese Households 304 8C.2 Basis Functions for MARS Model of Log of Real Per Capita Income, Minority Households 306 9.1 Life Table from the 1998 Vietnam Living Standards Survey 318 9.2 Variable Definitions, Means, and Standard Deviations 327 9.3 Weibull Parameter Estimates 330 9.4 Predicted Infant and Under-Five Mortality Rates 332 9.5 Means of Determinants of Child Survival, by Poverty Group 333 9.6 Decompositions of Changes in Survival, Based on Equation 9.4 335 9.7 Decompositions of Changes in Survival, Based on Equation 9.5 336 9.8 Decompositions of Changes in Survival, Based on Equation 9.6 337 9.9 Assumptions Concerning Progress on Determinants of Survival 340 9.10 Decompositions of Changes in Survival to 2015 341 9.11 Reversing Declines in Determinants of Survival among the Poorest Quartile 343 9.12 Targeted Improvements in Health Services, Water, and Sanitation among the Poorest Quartile 344 List of Figures, Maps, and Tables 611 10.1 Stunting and Wasting by Region, Children 0 to 60 Months 357 10.2 Malnutrition by Expenditure Quintiles, Children 0 to 60 months 359 10.3 Determinants of Child Malnutrition in Urban Areas, 1993 365 10.4 Determinants of Child Malnutrition in Rural Areas, 1993 369 10.5 Determinants of Child Malnutrition in Urban Areas, 1998 371 10.6 Determinants of Child Malnutrition in Rural Areas, 1998 373 10.7 Determinants of Child Malnutrition: Panel Data Estimates 377 10.8 Role of Economic Growth in Reducing Child Malnutrition 379 10.9 Descriptive Statistics of Selected Community Variables in Rural Areas, 1998 382 10.10 Impact of Community Health Services on Child Malnutrition in Rural Areas, 1998 384 11.1 Definitions and Descriptive Statistics 394 11.2 Annualized Health Service Contact Rates, by Provider 395 11.3 Mean Number of Visits to Different Health Care Providers 397 11.4 Average Contact Rates with Providers, by Health Status 397 11.5 Medical Expenditure by Household Type (1998 dong) 398 11.6 Health Insurance and Income Status 400 11.7 Frequency Distribution of Health Service Contacts 401 11.8 Determinants of Probability of Health Insurance Enrollment 406 11.9 Models for Number and Probability of Commune Health Center Visits 408 11.10 Government Hospital Inpatient and Outpatient Use 410 11.11 Models for Probability of Use of Private Health Care 412 11.12 Models for Pharmacy Visits 414 11.13 Models for Positive Medical Expenditure for Individuals 416 11.14 Models for Positive Medical Expenditure for Households 417 12.1 Net and Gross School Enrollment Rates, by Quintile and Education Level 428 12.2 Repetition Rate by Level of Education, 1998 430 12.3 Age and Grade Matching, 1993 and 1998 431 12.4 Girls' Enrollment Rates as a Percentage of Boys' Enrollment Rates, by Education Level, 1993 and 1998 432 12.5 Net Enrollment Rates by Location, 1993 and 1998 435 12.6 Public and Private Shares of School Enrollment, by Education Level, 1993 and 1998 436 12.7 Trends in Government Expenditure on Education, 1992­98 438 12.8 Share of Public Spending on Education and of Enrollments, by Education Level, 1993 and 1998 439 12.9 Public and Private Shares of Financing, by Education Level, 1993 and 1998 440 12.10 Per Student Household Expenditure on Schooling, 1993 and 1998 441 612 List of Figures, Maps, and Tables 12.11 Composition of Consumption, 1993 and 1998 442 12.12 Composition of Private Spending on Primary Education, by the Poorest and Most Well-Off Quintiles, 1993 and 1998 442 12.13 Composition of Private Spending on Primary Education, by Region 444 12.14 Share of Public Education Spending and School-Age Children, for Poorest and Most Well-Off Quintiles, 1993 and 1998 447 12.15 Per Capita Public Spending for Education, by Level and Region, 1993 and 1998 451 12.16 Incidence of Public Spending for Education, by Region, 1993 and 1998 453 12.17 Incidence of Public Subsidies for Primary and Lower Secondary Education, by Region, 1993 and 1998 454 12.18 Mean of Selected Variables, by Sector of Employment and by Sex, 1993 and 1998 456 12.19 Basic Earning Functions Based on Years of Schooling for Private Sector Workers, 1993 and 1998 458 12.20 Extended Earning Functions Based on Level of Education for Private Sector Workers, 1993 and 1998 461 13.1 School Entrance Rates in Vietnam, 1993 and 1998 468 13.2 School Enrollment Rates in Vietnam, 1980­98 469 13.3 Entrance Rates for Lower and Upper Secondary Schools, 1993 and 1998 470 13.4 Descriptive Statistics for Child, Household, and School Variables 473 13.5 Determinants of Primary School Completion 476 13.6 Determinants of Lower Secondary School Completion 484 13.7 Determinants of Reading Test Scores in Primary School 488 13.8 Determinants of Math Test Scores in Primary School 491 13.9 Determinants of Examination Rank in Primary School 493 13.10 Determinants of Examination Rank in Lower Secondary School 497 14.1 The Out-Migration of Children from Panel Households 508 14.2 Participation in Child Labor in Past Seven Days, by Type of Work, for Children Ages 6 to 15 Years 510 14.3 School Enrollment Rates, by Quintile 512 14.4 School Enrollment by Age and Type of Work in Past Seven Days 514 14.5 Participation in Child Labor in Past Seven Days, by Gender, for Children Ages 6 to 15 Years 517 14.6 Starting Age of Work of Children in Different Occupations, Ho Chi Minh City 522 14.7 Starting Age of Work of Children in Different Occupations, Rural North Central Coast Region 523 List of Figures, Maps, and Tables 613 14.8 Participation in Child Labor in Past Seven Days, by Residential Location, for Children Ages 6 to 15 Years 524 14.9 Regional Differences in the Decline in Child Labor, Linear Probability Model, Household Fixed Effects Results 527 14.10 Adult Migration History and Child Labor in the Past Seven Days, Linear Probability Results 534 14.11 Enterprise Ownership and Child Labor in the Past Seven Days, Linear Probability Results 537 14.12 Ethnic Minorities and Child Labor in the Past Seven Days, Linear Probability Results 542 15.1 Panel Attrition 558 15.2 Transition Matrix for Vietnam, 1993 and 1998 559 15.3 Estimated Mobility, Ignoring Measurement Error 560 15.4 Correlation Coefficients without Correction for Measurement Error 561 15.5 Estimated Mobility Using Three Different Types of Instrumental Variables 562 16.1 Households Involved in Private Transfers, 1993 571 16.2 Transfers and Total Income, 1993 572 16.3 Household Economic Situation by Transfer Status, 1993 572 16.4 Household Demographics by Transfer Status, 1993 573 16.5 Households Involved in Private Transfers, 1998: A Comparison of Gifts and Gifts Plus Loans 575 16.6 Transfers and Total Income, 1998 576 16.7 Household Economic Situation by Transfer Status, 1998 577 16.8 Household Demographics by Transfer Status, 1998 578 16.9 Uses of Gifts versus Loans, 1998 580 16.10 Transitions in Transfers between 1993 and 1998 582 16.11 Increases versus Decreases in Real Private Transfers Per Capita by Windfalls versus Shortfalls in Pre­Private Transfer Income Per Capita 583 16.12 Increases versus Decreases in Private Transfers Per Capita by Windfalls versus Shortfalls in Pre­Private Transfer Income Per Capita, Restricted Sample: Households with Below-Median Per Capita Income in 1993 584 16.13 Changes in Private Transfers Per Capita for Households That Have Shortfalls in Their Pre­Private Transfer Income Per Capita 585 16.14 Increases versus Decreases in Per Capita Private Transfers, by Economic and Demographic Events 586 16.15 Increases versus Decreases in Per Capita Private Transfers, by Health-Related Events 588 16.16 Regression of Differences in Log Net Transfer Receipts Per Capita on Economic and Demographic Events 589 614 List of Figures, Maps, and Tables 16.17 Households Involved in Private Transfers in 1993, by Growth-Pole Status 594 16.18 Households Involved in Private Transfers in 1998, by Growth-Pole Status 595 16.19 Increases versus Decreases in Private Transfers Per Capita, by Growth-Pole versus Non-Growth-Pole Residence 595 16.20 Income Inequality: Growth-Pole versus Non-Growth-Pole Households, 1993 and 1998 596 16.21 Increases versus Decreases in Private Transfers Per Capita, by Typhoon-Prone versus Non-Typhoon-Prone 597 Index Figures, maps, notes, and tables are indicated by f, m, n, and t, respectively. Alphabetization is letter-by-letter (e.g., "Childbirth" precedes "Child labor"). academic achievement. See education income distribution, 133­185 Africa and household enterprises inequality and, 177­180, 179t5.11 panels, 107 projections of, 76t3.19, 80­85, agriculture, 133­185 82t3.24, 83t3.25, 92n22 changes in, 144t5.2, 146t5.3, 148­163 rice marketing and, 149t5.4, cropping patterns, 138t5.1, 149t5.4, 150t5.5, 163­169, 164t5.9, 150t5.5, 152­158, 153t5.6, 154t5.7, 166t5.10, 184n25 156f 5.2, 157f 5.3, 183nn16­19 institutional changes and potential food demand patterns, 136, implications, 134­137, 146t5.3, 148­152, 149t5.4, 150t5.5, 182nn3­6 183nn14 & 15 labor force compared with income output growth, 156f 5.2, 160­163, level, 81, 81f 3.3 161t5.8, 184nn21­24 land reform, 40 rice output, land, and fertilizer prices, 143­148 inputs, 153t5.6, 154t5.7, 158­160, increase in, 136 183n20 liberalization of, 40 data, 133, 182n1 overview of 1993 and 1998 decentralization of household indexes, 144­145, 144t5.2, decisionmaking, 136 183n13 employment research implications, 22 self-employment, 56, 57t3.3­t3.4, 58 rice prices, 145­148, 146t5.3, 149t5.4, shrinkage in, 58 150t5.5, 169­177, 172f 5.4, family farms, 53, 56, 57t3.3 174f 5.5, 184n26 farming household inequality, 76, structure of income, 137­143, 182n7 76t3.19­77t3.21, 77­78 income inequality, 142t5.1, 143 geographic divisions, 133, 182n2 levels of income, 137­142, 138t5.1, household enterprises, 102 142f 5.1, 182nn8 & 9, 183nn10­12 615 616 Index antibiotics. See drugs household characteristics, 529­543 appliances. See consumer durable goods enterprise ownership, 534­540, armed forces personnel and social 535f 14.7, 537t14.11, 538f 4.8 insurance programs, 191 ethnicity, 540­543, 540f 14.9, 542t14.14, 545 Ba-na. See ethnic minorities living standards, 529­531, 530f 14.5 banking systems, 31, 37, 44 migration, 527t14.9, 531­534, breastfeeding, 353­354 532f 14.6, 534t14.10, 548n13 Buddhism, 284­286, 285t8.6 household enterprises, 95, 129n11, See also religion 506, 534­340, 535f 14.7, 537t14.11, academic achievement and, 483, 538f 4.8 487, 490 household work, 509, 547n5 child nutrition and, 336t10.3, 367, legislation, 515­516, 547n7 369t10.4, 372t10.5 missing children, 508t14.1 community health services and, 383, overview, 17­18 384t10.10 recent trends, 507­511, 508t14.1, 510t14.2, 547nn4­6 Canada and household enterprises research implications, 21, 23 panels, 107 residential location of, 521­528, Cao Daism, 284­286 522t14.6­524t14.8, 525f 14.3, Catholicism, 284­286 527t14.9, 529f 14.4, 547nn10­11, child nutrition and, 336t10.3, 548n12 369t10.4, 372t10.5 sex industry, 546 community health services street children, 507, 547n4 and, 384t10.10 child mortality Central Highlands minorities. causes of recent changes in, 329­339 See ethnic minorities decomposition results, 330t9.3, childbirth and delivery care, 277, 329, 331­338, 333t9.5, 335t9.6­337t9.8 333­334, 339, 340t9.9, 343, 384 estimation results, 329­331, child labor, 505­550, 547n1 330t9.3, 332t9.4, 347n15 age and gender, 17­18, 517­521, key points, 338­339 517t14.5, 518f 14.1, 520f 14.2, changing inequalities under Doi Moi, 522t14.6, 544­545, 547n9 320­323 children "trafficked" overseas, data and methods, 320­321 507­508 key points, 322­323 decline in, 505­506, 530f 14.5, results on trends in socioeconomic 531, 547n2 inequalities, 321­322, 321f 9.4, economic and policy context of, 322f 9.5 511­516 child survival prospects to 2015, education, 511­515, 512t14.3, 339­342 514t14.4 key points, 342 legislation, 515­516, 547n7 methods and assumptions, effects of, 505, 544­546 339­342, 340t9.9, 341t9.10 enterprise ownership, 534­540 computation of indirect estimates of environment and, 507­516, 547n3 IMR and USMR, 345­346 ethnicity and, 340­343, 545 data and methods, 320­321, 346n8 gross domestic production and, 505 modeling child survival, 323­329 home enterprises and, 506 basic model, 323­324 hours worked, 17­18, 515­516, data and variable definitions, 520f 14.2, 521 327­328, 327t9.2, 347nn11­13 Index 617 decomposing sources of change, children 324­327, 325f 9.6, 347n10 diarrhea, 353­356, 354f 10.1, 355f 10.2 key points, 328­329, 347n14 oral rehydration, 21, 381, policy implications, 342­345, 382t10.9, 383 343t9.11, 344t9.12 education. See education recent trends, 313, 315­323, 346n1, health care. See health care 346n4 illness in, 329, 383­384 direct estimates of infant and immunizations, 329, 330t9.3, 331, under-five mortality, 316f 9.1, 333, 333t9.5, 334, 340t9.9, 384 317­319, 317f 9.2, 318t9.1, labor. See child labor 320f 9.3, 346nn5­7 mortality. See child mortality indirect estimates of infant and nutrition. See child nutrition under-five mortality, 315­317, street children, 507, 547n4 316f 9.1, 317f 9.2, 345­346 stunting and wasting of. See child key points, 319 nutrition research implications, 20 survival of. See child mortality child nutrition, 351­388 "trafficked" overseas, 507­508 data and analytical framework, 351, China 360­365, 388n1 capital-intensive agriculture, 183n15 economic growth and, 352­360 global competitiveness report economic performance in 1990s, rankings, 43­44, 44f 2.10 358­360, 359t10.2 governance, 42, 43f 2.9 measurement of nutritional status, household enterprises, 96 352­353 poverty reduction, 40, 40f 2.7 nutritional status of children, reform, 32 353­358, 354f 10.1, 357t10.1 secondary school enrollment, 467 education and, 368, 369t10.4, 479, Chinese, ethnic 487, 496 See also ethnic minorities ethnic minorities and, 277 marriage, 83­84 health programs and, 380­387 wage determinants, 64, 65t3.11 econometric estimates, 381­387, Christianity, 285, 285t8.6 382t10.9, 383, 384t10.10, 388n2 See also religion growth in, 380­381 child nutrition and, 367, 369t10.4, incidence of malnutrition, 351 372t10.5 income growth and, 364­380, clean water. See water 365t10.3, 369t10.4, 371t10.5, closed capital accounts, 45 373t10.6, 377t10.7, 378­380, commune health centers. 379t10.8 See health care cross-sectional estimates, 364­375, community variables, constructing, 365t10.3, 369t10.4, 371t10.5, 126­128, 130n13 373t10.6 competitiveness, 43­44, 44f 2.10 panel data estimates, 369t10.4, Comprehensive Student Insurance 375­378, 377t10.7 (CSI), 399 malnutrition, 277 conditional convergence, 32, 40­41, religion and, 367 41f 2.8 research implications, 20­21 Constitution of 1992, 511, 515 stunting and wasting, 351, 354­358, consumer durable goods, 237t7.2, 240, 354f 10.1, 355f 10.2, 357t10.1, 562, 578, 600n16 359t10.2, 378, 379t10.8, 479, See also television ownership 487, 496 consumption, government, 36, 38t2.1 618 Index Contingency Fund for Preharvest inflation, effect of, 31, 36, 37, 37f 2.5, Starvation and Natural 38t2.1 Disasters, 191 infrastructure and, 32 contingent liabilities, 44 maintaining high rate of, 40­41 contraceptive use, 277, 384 measurements of, 30­31 corruption, 22, 31­32, 42, 43f 2.9 overview in 1990s, 1­26 Côte d'Ivoire and household poverty and, 39­40, 39f 2.6 enterprises panels, 107, 129n6 private property rights and, 31, 32, crops. See agriculture 38, 39 CSI (Comprehensive Student rule of law and, 31, 32, 36, 37f 2.5, Insurance), 399 38t2.1, 39, 49n2 customs administration, 45 trade and, 31, 34­36, 35f 2.3, 36f 2.4, 38­39, 38t2.1 See also trade Dao. See ethnic minorities trends from 1984 to 2000, 4­7, 5t1.1 death of children. See child mortality economic mobility, 18, 551­566, 564n1 Decision Number 134/1999/ concepts and measurement, 552­556, Qd-TTg, 515 564n2, 565nn3 & 4 Decree Number 140/TTg, 135 household enterprises and, 98­99, diarrhea, incidence in children. 99t4.3 See children overview, 18 disaster relief, 209 research implications, 21 Doi Moi policy reforms, 2­4, 25n2, trends in 1990s, 556­564 313­350 background and data, 23­24, See also specific subjects 556­557, 558t15.1, 565n5 domestic transport, 45 estimates of mobility corrected for Dominican Republic and household measurement error, enterprises panels, 107 559t15.2­562t15.5, 561­564 drugs, 380­383, 382t10.9, 418, 422n8 measured without correction for See also self-medication measurement error, 558­560, durable goods. See consumer durable 559t15.2­561t15.4 goods economic performance in 1990s, 27­185 agricultural and income distribution earnings mobility, 73, 92n20 in rural Vietnam under East Asia financial crisis, 45, 47 economic reforms, 133­185 economic growth household enterprises, 95­132 conditional convergence, 32, 40­41, overview, 10­12 41f 2.8 reform. See economic reform corruption, 31­32 wage labor market and inequality, cross-country growth regression, 53­93 30­31 economic reform, 29­51 determinants of, 30­36 commercial bank assets and, 37 financial development and, 31, 36, experience with, 36­41, 37f 2.5 37f 2.5, 44 financial sector reforms, 37 foreign direct investment (FDI), 31 interest rates, raising, 37 globalization and, 34f 2.2, 36 poverty and, 39, 39f 2.6 "globalizers" and, 32­33, 33f 2.1 stabilization from high inflation, 37, government corruption, 31 37f 2.5 income of poor and, 34­36, 35f 2.3, trade policy, 37 39­40, 39f 2.6 See also foreign direct investment inequality and, 7­9, 33­34, 34f 2.2, 39 (FDI); trade Index 619 education school progress and academic child labor and, 486, 511­515, achievement, 467­501 512t14.3, 514t14.4, 546 academic achievement, 486­498 child nutrition and, 368, 369t10.4 child labor and, 486 child survival and, 335t9.6, data and methodological 340t9.9, 343 framework, 470t13.3, 472­475, community health services and, 383, 473t13.4 384t10.10 language and, 482 enrollment, 426­435, 462 nutrition, 479, 487, 496 ethnic groups' differences, See also child nutrition 433­434, 433f 12.4 poor quality classrooms, gender differences, 431­433, 480­482, 486 431t12.3, 432f 12.3, 432t12.4 primary and secondary school general trends, 426­427, 427f 12.1, outcomes, 468­672, 512­515, 512t14.3, 514t14.4 468t13.1­470t13.3, 499­500nn1­3 regional differences, 434­435, results, 468t13.1­470t13.3, 475­498, 435t12.5 476t13.5, 484t13.6, 488t13.7, trends by expenditure group, 491t13.8, 493t13.9, 497t13.10, 427­431, 428t12.1, 429f 12.2, 500nn4 & 8­10 430t12.2, 431t12.3, 464n2 school progress, 475­486 ethnic minorities, 279­283, 281t8.3, teacher training and experience, 282t8.4, 294, 295, 297, 308nn7 & 476t13.5, 481, 486, 490 8, 309n16, 433­434, 433f 12.4, 479 trends in education sector, 425­465, academic achievement, 471­472 464n1 financing, 435­455, 462­463 enrollment, 426­435 beneficiaries of, 446­449, 447t12.14, financing, 435­455, 462­463 450f 12.6 returns to education, 63­68, private, 440­446, 455­462, 464 440t12.9­442t12.12, 444t12.13, wages and, 63­68, 65t3.10­t3.12, 446f 12.5, 447t12.14, 511­512 67, 68f 3.2 public and private, 436­440, electrification 436t12.6, 438t12.7­440t12.9 of community health centers, regional benefit from public 382t10.9, 383, 384t10.10 spending, 449­455, 451t12.15, growth and, 32 453t12.16, 454t12.17 as predictor of household welfare, household enterprises, 100­102, 237t7.2, 240, 270n14 103t4.5, 125, 129n11 emerging markets, comparison infrastructure investments in with, 41­47 schools, 209, 209t6.7 factor markets, 43­45 poverty and, 237t7.2, 238, 239t7.3, openness, 34, 35f 2.3, 45­47 256t7.7 property rights and governance, pro-poor budget planning and 41­43 management, 463­464 employment research implications, 23 See also agriculture; household returns to education, 63­68, 455­462, enterprises; wage labor market 456t12.18, 457f 2.7, 458t12.19, children. See child labor 461t12.20, 464 health insurance and, 407 scholarship programs, 197, 198t6.1, social insurance programs 200, 217 and, 191 school fee programs, 201­202, endogenous growth theories, 30 203t6.4, 225n11 Enterprise Law, 511 620 Index ethnic minorities, 273­310 factor markets, 43­45 assimilation, 274, 308n3 family farms. See agriculture child labor and, 17­18, 340­343, 506, FDI. See foreign direct investment 540­543, 540f 14.9, 542t14.14 fertility rates, 275, 277, 345­346 child nutrition and, 366t10.3, fertilizers, 135, 136, 144t5.2, 145, 372t10.5, 375 159­160 child survival and, 329, 330t9.3 financial development, 31, 36, 37, 37f 2.5 community health centers and, 383, Finland and household enterprises 384t10.10 panels, 107 composition of population, 91n11, food-for-work, 35 274, 275m8.1 footwear industry, 40 data analysis, 300­307 foreign direct investment (FDI) decomposition analysis updated, growth and, 31 286t8.9, 292­297, 308n15, 309n16, integration in global economy 309n17 and, 45 differences among minority groups, reform and, 37 277­286 as share of GDP (1998), 46f 2.12, 47 expenditures, 277­279, 278t8.2, trade policy reform and, 37 280f 8.1A, 280f 8.1B trend in 1990s, 47, 47f 2.13 intermarriage, 283­286, 283t8.5, 308nn10 & 11 garment industry, 45, 46f 2.11 religion, 284­286, 285t8.6, 308n12 gender comparison schooling, 279­283, 281t8.3, child labor, 506, 517­521, 517t14.5, 282t8.4, 294, 295, 297, 308nn7 & 518f 14.1, 520f 14.2, 544­545, 8, 309n16, 433­434 547n9 education child nutrition, 364, 365t10.3, 367, See also education 368, 369t10.4, 383, 384t10.10 enrollment, 433­434, 433f 12.4 child survival, 344 school progress and achievement, community health services, 383, 471­472, 479, 480 384t10.10 government policy toward, 286­287, education 308n13 academic achievement, 483, household enterprises, 98, 98t4.2, 484t13.6 129n3 enrollment, 431­433, 431t12.3, living standards, gap in, 273­277, 432t12.4 275m8.1, 276t8.1, 307n2, ethnic minorities, 281­283, 281t8.3, 308nn4­6 282t8.4 explaining divergence, 220t6.16, as predictor of poverty, 288­292, 289t8.7, 291t8.8, 237t7.2, 238 308n14 returns from, 459 local funding and resources, household enterprises, 97t4.1, 98, 108 189­191 labor force participation, 55­56, population groups, 273, 274, 55t3.1, 57t3.4­t3.5, 58 275m8.1, 307n1 research implications, 23 as predictor of per capita state-owned enterprises (SOEs), 62, expenditure, 238, 269n8 62t3.9 public safety net and, 190 wage determinants, 63, 64, 65, research implications, 23 65t3.11, 66t3.12, 67, 68, 91n11 wage determinants, 63, 64, geographic targeting. See public 65t3.11­t3.12, 91nn11 & 13 safety net Index 621 Ghana and household enterprises 1998 VLSS, 393­395, 394t11.1, panels, 107, 129n6 422nn1 & 2 Gia-rai. See ethnic minorities pharmacy visits. See this heading: gifts, intrafamilial. See private pharmacy visits interhousehold transfers policy issues, 418­420, Global Competitiveness Report, 43 418­422, 422 global competitiveness report rankings, private health facilities, 396, 398, 43­44, 44f 2.10 411­413, 412t11.11, 420, 422 globalization, 33­34, 34f 2.2, 36 results, 394t11.1, 405­415 "globalizers," 32­33, 33f 2.1 self-medication, 395, 396, 415, governance, 42­43, 43f 2.9 418­419, 420­421 government corruption. See corruption statistical issues in data analysis, government effectiveness, 22, 42, 43f 2.9 400­405, 401t11.7, 422nn4­6 government spending, 31, 32, 36 survey of main issues, 395­400, graft. See corruption 395t11.2 pharmacy visits, 380­381, 382t10.9, Hanoi 383, 384t10.10, 395­396, hours worked in, 61, 61t3.8 397t11.3­t11.4, 413­415, poverty estimates, 241­244, 242t7.4, 414t11.12, 420­421, 422n7 243t7.5 prenatal care and births, 277, 329, regional wage differences, 59­60, 333­334, 339, 340t9.9, 343, 59t3.6, 60t3.7, 62, 91nn7 & 8 344, 384 wage and income inequality, 74, private interhousehold transfers, 75t3.16­t3.17 587­588, 588t16.15, 589t16.16 wage determinants, 64, 65t3.11, 67 research implications, 20­21, 23 health care health insurance, 399­400, 400t11.6, child mortality. See child mortality 405­408, 406t11.8, 419 child nutrition. See child nutrition VHI program, 391, 399­400, 422n3 community health centers, 382t10.9, HEPR (Hunger Eradication and 383, 384t10.10 Poverty Reduction program), community health services, 329, 381 192, 223, 252 contraceptive use, 277, 384 Hmong. See ethnic minorities distance to facilities, 382t10.9, 383, Hoa. See ethnic minorities 384t10.10 Hoa Hao religion, 285­286 drugs. See drugs Ho Chi Minh City ethnic minorities and, 277 child labor, 521, 522t14.6, 531 facilities, 380­381 hours worked in, 61, 61t3.8 health insurance. See health poverty estimates, 241­244, 242t7.4, insurance 243t7.5 patterns of use, 391­423 regional wage differences, 59­60, access and costs, 396­439, 59t3.6, 62, 64, 65t3.11, 91n7 397t11.3­t11.4, 398t11.5 wage and income inequality, 74, commune health centers, 381, 75t3.16­t3.17 408­409, 408t11.9, 419, 420 wage determinants, 67 expenditure analysis, 415­418, hospitals. See health care 416t11.13, 417t11.14 household enterprises, 95­132 government hospitals, 396, 398, activities by, 95 409­411, 410t11.10 affluence, 95, 99­100, 99t4.3, health insurance. See health 125, 128n1 insurance capital and, 126, 130n12 622 Index household enterprises (continued) overview of household welfare in child labor, 95, 129n11, 506, 534­540, 1990s, 1­26 535f 14.7, 537t14.11, 538f 4.8 size and composition, 236, 237t7.2, constructing community variables, 238, 262­263, 269nn7 & 8 126­128, 130n13 housing and basic services as predictor dynamics, 100­124, 101f 4.1 of poverty, 237t7.2, 239­240, attrition of, 101f 4.1, 109t4.7, 112, 270nn12­14 112t4.8, 129n8 Hunger Eradication and Poverty constructing panel of enterprises, Reduction (HEPR) program, 97t4.1, 105­112, 106t4.6, 109t4.7, 192, 223, 252 129n6 operated by, 99t4.3, 100­104, illegal migrants, 74 101t4.4, 103t4.5, 129n5 immunizations, 329, 330t9.3, 331, 333, performance over time, 106t4.6, 333t9.5, 334, 340t9.9, 384 117­124, 121t4.12, 122t4.13, See also children; health care 129nn9 & 10 Index of Economic Freedom 2000 startups, explaining, 101f 4.1, 117, (Heritage Foundation), 29, 48n1 118t4.11 India survival of, 112­117, 114t4.9, global competitiveness report 116t4.10 rankings, 43­44, 44f 2.10 education and, 100­102, 103t4.5, 125, governance, 42, 43f 2.9 129n11 poverty reduction in, 40, 40f 2.7 family history and, 104 reform, 32 geography and, 102 industries investment climate factors, 125 footwear, 40 living standards and, 96­100, garment, 40, 45 97t4.1­99t4.3, 128n2, 129n3 growth, 40 number of, 95­96 shipping costs, 45, 46f 2.11 poor households and, 99­100, 99t4.3, inflation. See economic growth 125, 129n4 informal loans. See private regional variables, 102, 103t4.5, interhousehold transfers 125, 129n5 infrastructure, 32, 47, 135, 209, 209t6.7, role in future, 126 225n13 self-employment, 104 insurance "shooting stars," 99­100, 99t4.3, health. See health insurance 124, 125 social insurance programs, 191, "sinking stones," 99­100, 99t4.3 197, 217 survival of, 105­117, 107t4.7, 114t4.9, unemployment insurance, 35 116t4.10, 125 interest rates, 37 wage labor market compared, interhousehold income transfers. 76­79 See private interhousehold households transfers See also household enterprises intermarriage, 283­286, 283t8.5, income, 96 308nn10 & 11 See also wage labor market International Convention on the estimating, 71­72, 91nn17­19 Rights of the Child, 515 inequality, 76­79, 76t3.18­t3.19, International Labor Organization 77t3.20­t3.21, 79t3.22­t3.23 Convention Number 182 on interhousehold transfers. See private Worst Forms of Child Labor, interhousehold transfers 516, 546 Index 623 investment. See foreign direct Muong. See ethnic minorities investment (FDI) Myanmar, 42, 43f 2.9 Khmer. See ethnic minorities National Development Programs, 192 Kinh. See ethnic minorities National Target Program on Poverty Alleviation, 192 labor nonfarm household enterprises See also agriculture; household (NFHEs). See household enterprises; wage labor market enterprises children. See child labor Northern Uplands minorities. manufacturing sector, 40 See ethnic minorities participation rates, 55, 55t3.1 Nung. See ethnic minorities service sector, 40 nutrition wage labor market. See wage labor body mass index (BMI) and mobility, market 550t15.3, 559t15.2, 562t15.5, land use and reform, 40, 135 563­564 See also property rights children. See child nutrition languages, 81, 275, 277, 294, 308n8, 482, 500n10 Law on Child Protection, Care and occupation as predictor of poverty, Education, 515 237t7.2, 239, 269n11 Law on Universalization of OECD. See Organisation for Economic Education, 511 Co-operation and Development legislation old-age support, 567, 574 child labor, 515­516, 547n7 openness to trade. See trade education, 511 oral rehydration, 21, 381, 382t10.9 loans Organisation for Economic informal. See private interhousehold Co-operation and Development transfers (OECD), 32 nonperforming, 44 overview of economic growth and household welfare in 1990s, 1­26 Malaysia, 32, 46f 2.12, 47 child labor, 17­18 malnutrition. See child nutrition Doi Moi policy reforms, 2­4, 25n2 manufacturing sector, 40 economic and social performance, maps 4­9, 10­12 ethnicity and expenditures, 275 inequality in, 7­9, 8t1.2 poverty maps of Vietnam, 241­251, trends from 1984 to 2000, 4­7, 245m7.1, 249m7.2, 258, 271n25 5t1.1, 8t1.2, 25n3 Vietnam, xvi economic mobility, 18 marriage among ethnic groups. See interhousehold transfers, 18­19 intermarriage See also private interhousehold medicines. See drugs; health care; transfers self-medication poverty reduction, 12­14 "middle class," 73t3.13­t3.14 research considerations, 19­23 Millennium Development Goals social sectors, 14­17 (MDGs), 314, 315, 339, 346n3 Ministry of Labor, Invalids and Social pensions, 191 Affairs (MOLISA), 191, 192 Peru and household enterprises mobility. See economic mobility panels, 107 mortality. See child mortality pharmaceuticals. See drugs 624 Index pharmacy visits. See health care Poverty Reduction Strategy Paper political instability and violence, 42 (PRSP), 314, 339, 346n2 polygamy, 308n10 prenatal care, 277, 329, 333­334, poverty, 34­46, 39f 2.6, 40f 2.7, 49n4, 340t9.9, 344, 384 229­272 price reform, 40 data and methods, 229­235, private enterprises 269nn1 & 2 See also household enterprises applying regression results to corruption and, 31­32 census data, 233­235, 269n6 emergence of, 96 data, 231­233, 232t7.1, 269nn4 & 5 private interhousehold transfers, estimating poverty with 567­603 household survey, 233 background, 569 ethnic minority development, coresidence and, 580­581, 600n17 273­310 cross-sectional patterns from 1993 factors associated with, 232t7.1, VLSS, 570­575, 236­240, 237t7.2, 239t7.3, 571t16.1­573t16.4, 574f 16.1, 259­286 575­581, 575t16.5­578t16.8, consumer durables, 237t7.2, 240 580t16.9, 592­593, 599nn4­7, education, 238, 256t7.7 600nn8­11, 600nn12­14 household size and composition, "crowding out" of, 567 236, 237t7.2, 238, 262­263, definition of, 568 269nn7 & 8 economic and demographic events, housing and basic services, 585­588, 586t16.14, 588t16.15, 237t7.2, 239­240, 270nn12­14 601nn22 & 23 occupation, 237t7.2, 239, 269n11 economic growth and, 589­595, region, 237t7.2, 239t7.3, 240 590f 16.4, 591f 16.5, 601n24 maps, 241­251, 245m7.1, 249m7.2 generational direction of, 574, maps reflecting, 241­251, 245m7.1, 574f 16.1, 600n11 249m7.2, 258, 271n25 growth poles and, 593­595, provincial poverty estimates, 594t16.17, 595t16.18­t16.20, 244­251, 246t7.6, 270nn15­16 597t16.21 regional poverty estimates, health expenditures and, 587­588, 241­244, 242t7.4, 243t7.5 588t16.15, 589t16.16 overview, 12­14 income changes and, 582­585, potential of geographic and 582t16.10­585t6.13, 600nn18­21 additional targeting variables, in-kind services, 568, 581 245m7.1, 251­257, 251f 7.1, inter-vivos transfers (gifts), 571­574, 254f 7.2, 270nn17­23, 271n24, 571t16.1, 572t16.2, 573t16.4, 276t7.7 574f 16.1, 575t16.5, 599nn4­7, Receiver Operating Characteristic 600nn8­11 (ROC) curves, 237t7.2, 253­255, loans, 575, 575t16.5 256t7.7, 270n20 sources of loans versus gifts, research implications, 23 577­579, 579f 16.2­f 16.3, spatial distribution of poverty and 600nn15 & 16 potential for targeting, 229­272 old-age support and, 567, 574 survival prospects of children and, overview, 18­19, 568­569 313­350 panel evidence and, 581­589, 598 See also child mortality patterns in, 570­597 welfare programs, 189­227 regional incidence, 592­593, 592f 16.8, See also public safety net 593f 16.9, 601n25 Index 625 research implications, 21 scholarship programs, 197, 198t6.1, simple regression analysis of, 200, 217 586t16.14, 588­589, 588t16.14, school fee programs, 201­202, 589t16.16 203t6.4, 225n11 Typhoon Linda and, 569, 596­597, small towns, 209, 209t6.7 601n26 targeting of groups and programs, as urban-rural transfer flows, 190, 208t6.6, 209t6.7, 210, 216, 591­592, 591f 16.6, 592f 16.7, 217t6.13, 224n1 593f 16.10, 594f 16.11 uneven coverage, 189 wedding expenses, 587 privatization of state-owned Receiver Operating Characteristic enterprises (SOEs), 91 (ROC) curves, 237t7.2, 253­255, property rights, 31­32, 38­43, 47, 135 256t7.7, 270n20 Protestantism, 284­286, 285t8.6 reform, 29­51 See also religion See also Doi Moi policy reforms; child nutrition and, 336t10.3, 367, economic reform 369t10.4, 372t10.5 banking systems, 44 PRSP (Poverty Reduction Strategy comparison with other emerging Paper), 314, 339, 346n2 markets, 41­47 public safety net, 189­227 estimated growth effect of, 39, 49n2 behavioral response to transfers, experience with, 32, 36­41, 37f 2.5 194­196, 225nn5­8 financial sector, 37, 44 central budget financing, 189, 192 growth effect of, 30­36, 38t2.1 communes, role of, 192­193, 210, indicators of reforms mid-1980s to 218­221, 220t6.16 late 1990s, 37, 37f 2.5, 38, 38t2.1 data, 194, 224nn3 & 4 poverty and, 34­36, 39­40, 39f 2.6, decentralization, 189­190, 223­224 40f 2.7, 49n4 funding, redistribution of, 189­193 research implications, 22 geographic targeting, 218­221, rural areas and, 133­185 220t6.16­222t6.18, 224, 225n18 See also agriculture HERP program, 192 regulatory framework, 42, 43f 2.9, 47 household contributions, 205­208, religion, 284­286, 285t8.6, 308n12 206t6.5, 210, 225n12 child nutrition and, 336t10.3, 367, infrastructure investments, 209, 369t10.4, 372t10.5 209t6.7, 225n13 community health services, 383, initiatives, 191­192 384t10.10 local resources, use of, 189, 191, 224n2 research implications, 19­23 NGO transfers, 197, 198t6.1, 200, Resolution 5, 135 200t6.2 rice policy implications, 221­224 See also agriculture poverty and, 189, 193, 194­196, inequality and, 149t5.4, 150t5.5, 213, 223 169­177, 172f 5.4, 174f 5.5, 184n26 poverty-related programs, 196­210, marketing, 149t5.4, 150t5.5, 163­169, 198t6.1, 200t6.2, 202t6.3, 203t6.4, 164t5.9, 166t5.10, 184n25 206t6.5, 208t6.6, 209t6.7, output growth, 161t5.8, 162­163 225nn9­13 land and fertilizer inputs, 153t5.6, protection versus promotion, 154t5.7, 158­160, 183n20 210­218, 211t6.8, 213t6.9, prices, 144t5.2, 145­148, 146t5.3 215t6.10­218t6.15, 225nn14­17 production, 135, 152, 156f 5.2, 183n16 rural population, 209, 209t6.7 trade, 135, 136 626 Index ROC curves. See Receiver Operating schools. See education Characteristic (ROC) curves self-employment, 56, 57t3.3­t3.5, 58 rule of law, 36, 39, 42, 43f 2.9, 49n2 household enterprises. See growth and, 31, 32, 37f 2.5, 38t2.1 household enterprises rural populations self-medication, 395, 396, 415, 418­419, See also agriculture 420­421 child labor, 521­528, 523t14.7, See also health care 524t14.8, 527t14.9, 529f 14.4 service sector, 40 child nutrition, 368, 369t10.4, sex industry and child labor, 546 373t10.6, 375 shipping costs, 45, 46f 2.11 drugs and, 380 small and medium enterprises (SMEs), education 95­96 enrollment, 434­435, 435t12.5 See also household enterprises as predictor of poverty, 237t7.2, Social Guarantee Fund for Veterans 238 and War Invalids, 191 school progress and achievement, social insurance programs, 191, 470­471, 470t13.3, 472, 482­483 197, 217 employment, 53­54, 55­56, 57t3.4, 58, Social Security System, 191 59­60, 60t3.7 social welfare programs. See public See also agriculture safety net hours worked in, 61, 61t3.8 SOEs. See state-owned enterprises schooling, effect of, 68, 68f 3.2 state banks, 44 wage determinants, 65, 65t3.11 state-owned enterprises (SOEs) wage growth, 59, 59t3.6 credit and, 44 factors associated with poverty, employment by, 55t3.1, 57t3.3, 61­62, 236­240, 239t7.3 62t3.9 food demand patterns, 148, 150t5.5, privatization, 91 151, 168 success of, 44 income inequality, 177­180, 179t5.11 stock markets and growth, 31 labor force participation, 55, 55t3.1 street children, 507, 547n4 medical expenditures by, 398, stunting of children. See child nutrition 398t11.5 survival of children. See child mortality poverty estimates, 241­251, 242t7.4, 243t7.5, 251f 7.1 targeting poverty programs, 209, 209t6.7 public safety net, 190, 208t6.6, private interhousehold transfers, 209t6.7, 210, 216, 217t6.13, 224n1 591­592, 591f 16.6, 592f 16.7, spatial distribution of poverty 593f 16.10, 594f 16.11 and potential for targeting, stunting and wasting of children, 229­272 356, 357t10.1, 358 tariffs, 32, 37 urban dwellers, gap between, 53­54 Tay. See ethnic minorities telecommunication deficiencies and sanitation facilities, 209, 346n9, 347n12, growth, 32 382t10.9, 383, 384t10.10 television ownership child nutrition and, 367 See also consumer durable goods child survival and, 330t9.3, 331, estimated mobility and, 562 344t9.12 as predictor of per capita as predictor of poverty, 237t7.2, 240 expenditures, 237t7.2, 240 scholarship programs, 197, 198t6.1, targeting and, 255, 256t7.7, 270n23 200, 217 Thai. See ethnic minorities Index 627 Thailand wage determinants, 64, 66t3.12 FDI relative to GDP, 46f 2.12, 47 wage growth, 59, 59t3.6 financial crisis in, 45 food demand patterns, 148, 149t5.4, governance, 42, 43f 2.9 151, 168 reform and, 32 labor force participation, 55, 55t3.1 trade medical expenditures by, 398­399, economic reform and, 37 398t11.5 growth and, 38­39, 38t2.1 poverty, 236­240, 237t7.2, 239t7.3 inequality and, 34­36, 36f 2.4 estimates, 241­251, 242t7.4, integration with global economy, 45 243t7.5, 251f 7.1 liberalization, 35, 37 private interhousehold transfers, nontariff barriers, 37 591­592, 591f 16.6, 592f 16.7, openness to, 31, 36 593f 16.10, 594f 16.11 poverty and, 34­36 rural dwellers, gap between, 53­54 reforms, 37 stunting and wasting of children, rice, 135, 136 357t10.1, 358 social protection programs and, 35 tariffs, 32, 37 vaccinations. See immunizations trafficking in children, 507­508 Vietnam Health Insurance (VHI) Typhoon Linda, 569, 596­597, 601n26 program, 391, 399­400, 422n3 See also health insurance unemployment Vietnam Health Sector Review (World insurance, 35 Bank), 391, 392 labor force participation and, 55­56, Vietnam Living Standard Surveys 55t3.1 (VLSSs), 2, 23­24, 25n1 rates, 55t3.2, 56, 56t3.2 1998 VLSS, 393­395, 394t11.1, United Kingdom and household 422nn1 & 2 enterprises panels, 106­107 United States wage labor market, 53­93 conditional convergence, 32 composition of employment, 56­58, household enterprises panels, 107 57t3.3­t3.5, 90n4 urban populations determinants of wages, 63­68, child labor, 521­528, 524t14.8, 91nn10 & 11 527t14.9, 529f 14.4 returns to education and child nutrition, 365t10.3, 368, experience, 63­68, 63t3.10, 371t10.5, 375 65t3.11, 66t3.12, 91nn12 & 13 drugs and, 380 hours worked, 60­61, 61t3.8, 91n9 education household enterprises. See academic achievement, 470­471, household enterprises 470t13.3, 472, 482­483 inequality enrollment, 237t7.2, 434­435, distribution of wages, 72­75, 435t12.5 73t3.13­75t3.17, 86­87, 92n20 as predictor of poverty, 237t7.2, 238 measures, 68­72, 86­90, returns from, 459 91nn14­19 employment, 53­54, 55­56, 57t3.5, 58, projections of future inequality, 59­60, 60t3.7 76t3.19, 79­85, 79t3.22­t3.23, 81, hours worked, 61, 61t3.8 81f 3.3, 83t3.24­t3.25, 92n22 household enterprises, 97t4.1, 98 wages and income inequality, See also household enterprises 76­79, 76t3.18­77t3.21, schooling, effect of, 68, 68f 3.2 79t3.22­t3.23, 87­90, 92n21 628 Index wage labor market (continued) unemployment, 55t3.2, 56 labor force participation and wage growth, 58­59, 59t3.6, 91n5 unemployment, 55­56, 55t3.1, water 90n3 child survival, 330, 330t9.3, 340t9.9, output per person, 53, 54f 3.1, 90n1 343, 344, 344t9.12, 345 regional wage differences, 59­60, clean, 209, 346n9 59t3.6, 60t3.7, 62, 91nn6­8 as predictor of poverty, 240 research implications, 21 wedding expenses, 587 social problems and, 53­54 welfare programs. See public safety net state-owned enterprise (SOEs) women. See gender comparison employment, 55t3.1, 57t3.3, World Trade Organization (WTO), 45 61­62, 62t3.9 transition to, 53­54 Xo-dang. See ethnic minorities Vietnam is an economic success story. It transformed itself from a country in the 1980s that was one of the world's poorest to a country in the 1990s with one of the world's highest growth rates. With the adoption of new, market-oriented policies in the late 1980s, Vietnam averaged an economic growth rate of 8 percent per year from 1990 to 2000. This economic growth was accompanied by a large reduction in poverty (from 58 percent in 1993 to 37 percent in 1998), which included dramatic increases in school enrollment and a rapid decrease in child malnutrition. Economic Growth, Poverty, and Household Welfare in Vietnam uses an unusually rich set of macroeconomic and household survey data to examine several topics. These include the causes of the economic turnaround and prospects for future growth; the impact of economic growth on household welfare, as measured by consumption expenditures, health, education, and other socioeconomic outcomes; and the nature of poverty in Vietnam and the effectiveness of government policies intended to reduce it. Although Vietnam's past achievements are quite impressive, future progress is by no means ensured. This book draws lessons for Vietnam and for other low-income developing countries. It is a valuable resource for anyone--including those in the devel- opment community, academia, and the media--who is interested in economic policy, poverty reduction strategies, health care, education, and social safety nets. TMxHSKIMBy355435zv":;:&:%:. THE WORLD BANK ISBN 0-8213-5543-0