WORLD BANK DISCUSSION PAPER NO. 373 \o/1DP3'"f3 Work in progress for public discussion Octoer 9 A Poverty Profile of Cambodia NIhoa. Prxtl Recent World Bank Discussion Papers No. 304 Putting Institutional Economics to Work: From Participation to Governance. Robert Picciotto No. 305 Pakistan's Public Agricultural Enterprises: Inefficiencies, Market Distortions, and Proposals for Reform. Rashid Faruqee, Ridwan Ali, and Yusuf Choudhry No. 306 Grameen Bank: Performance and Stability. Shahidur R. Khandker, Baqui Khalily, and Zahed Khan No. 307 The Uruguay Round and the Developing Economies. Edited by Will Martin and L. Alan Winters No. 308 Bank Governance Contracts: Establishing Goals and Accountability in Bank Restructuring. Richard P. Roulier No. 309 Public and Private Secondary Education in Developing Countries: A Comparative Study. Emmanuel Jimenez and Marlaine E. Lockheed with contributions by Donald Cox, Eduardo Luna, Vicente Paqueo, M. L. de Vera, and Nongnuch Wattanawaha No. 310 Practical Lessons for Africa from East Asia in Industrial and Trade Policies. Peter Harrold, Malathi Jayawickrama, and Deepak Bhattasali No. 311 The Impact of the Uruguay Round on Africa. Peter Harrold No. 312 Procurement and Disbursement Manual for Projects with Community Participation. Gita Gopal No. 313 Harnessing Information for Development: A Proposal for a World Bank Group Strategy. Eduardo Talero and Philip Gaudette No. 314 Colombia's Pension Reform: Fiscal and Macroeconomic Effects. Klaus Schmidt-Hebbel No. 315 Land Quality Indicators. Christian Pieri, Julian Dumanski, Ann Hamblin, and Anthony Young No. 316 Sustainability of a Government Targeted Credit Program: Evidence from Bangladesh. Shahidur R. Khandker, Zahed Khan, and Baqui Khalily No. 317 Selected Social Safety Net Programs in the Philippines: Targeting, Cost-Effectiveness, and Options for Reform. Kalanidhi Subbarao, Akhter U. Ahmed, and Tesfaye Teklu No. 318 Private Sector Development During Transition: The Visegrad Countries. Michael S. Borish and Michel Noel No. 319 Education Achievements and School Efficiency in Rural Bangladesh. Shahidur R. Khandker No. 320 Household and Intrahousehold Impacts of the Grameen Bank and Similar Targeted Credit Programs in Bangladesh. Mark M. Pitt and Shahidur R. Khandker No. 321 Clearance and Settlement Systems for Securities: Critical Design Choices in Emerging Market Economies. Jeff Stehm No. 322 Selecting Development Projects for the World Bank. Jean Baneth No. 323 Evaluating Public Spending: A Framework for Public Expenditure Reviews. Sanjay Pradhan No. 324 The Bangladesh Rural Advancement Committee's Credit Programs: Performance and Sustainability. Shahidur R. Khand- ker and Baqui Khalily No. 325 Institutional and Entrepreneurial Leadership in the Brazilian Science and Technology Sector:Setting a New Agenda. Edited by Lauritz Holm-Nielsen, Michael Crawford, and Alcyone Saliba No. 326 The East Asian Miracle and Information Technology: Strategic Management of Technological Learning. Nagy Hanna, Sandor Boyson, and Shakuntala Gunaratne No. 327 Agricultural Reform in Russia: A View from the Farm Level. Karen Brooks, Elmira Krylatykh, Zvi Lerman, Aleksandr Petrikov, and Vasilii Uzun No. 328 Insuring Sovereign Debt Against Default. David F. Babbel No. 329 Managing Transboundary Stocks of Small Pelagic Fish: Problems and Options. Max Agiiero and Exequiel Gonzalez No. 330 China: Issues and Options in Greenhouse Gas Emissions Control. Edited by Todd M. Johnson, Junfeng Li, Zhongxiao Jiang, and Robert P. Taylor No. 331 Case Studies in War-to-Peace Transition: The Demobilization and Reintegration of Ex-Combatants in Ethiopia, Namibia, and Uganda. Nat J. Colletta, Markus Kostner, Ingo Wiederhofer, with the assistance of Emilio Mondo, Taimi Sitari, and Tadesse A. Woldu No. 333 Participation in Practice: The Experience of the World Bank and Other Stakeholders. Edited by Jennifer Rietbergen-Mc- Cracken No. 334 Managing Price Risk in the Pakistan Wheat Market. Rashid Faruqee and Jonathan R. Coleman No. 335 Policy Options for Reform of Chinese State-Owned Enterprises. Edited by Harry G. Broadman No. 336 Targeted Credit Programs and Rural Poverty in Bangladesh. Shahidur Khandker and Osman H. Chowdhury No. 337 The Role of Family Planning and Targeted Credit Programs in Demographic Change in Bangladesh. Shahidur R. Khand- ker and M. Abdul Latif No. 338 Cost Sharing in the Social Sectors of Sub-Saharan Africa: Impact on the Poor. Arvil Van Adams and Teresa (Continued on the inside back cover) WORLD BANK DISCUSSION PAPER NO. 373 A Poverty Profile of Cambodia Nicholas Prescott Menno Pradhan The World Bank Washington, D.C. Copyright @ 1997 The International Bank for Reconstruction and Development/THE WORLD BANK 1818 H Street, N.W. Washington, D.C. 20433, U.S.A. All rights reserved Manufactured in the United States of America First printing October 1997 Discussion Papers present results of country analysis or research that are circulated to encourage discussion and comment within the development community. To present these results with the least possible delay, the typescript of this paper has not been prepared in accordance with the procedures appropriate to formal printed texts, and the World Bank accepts no responsibility for errors. Some sources cited in this paper may be informal documents that are not readily available. The findings, interpretations, and conclusions expressed in this paper are entirely those of the author(s) and should not be attributed in any manner to the World Bank, to its affiliated organizations, or to members of its Board of Executive Directors or the countries they represent. The World Bank does not guarantee the accuracy of the data included in this publication and accepts no responsibility whatsoever for any consequence of their use. The boundaries, colors, denominations, and other information shown on any map in this volume do not imply on the part of the World Bank Group any judgment on the legal status of any territory or the endorsement or acceptance of such boundaries. The material in this publication is copyrighted. Requests for permission to reproduce portions of it should be sent to the Office of the Publisher at the address shown in the copyright notice above. The World Bank encourages dissemination of its work and will normally give permission promptly and, when the reproduction is for noncommercial purposes, without asking a fee. Permission to copy portions for classroom use is granted through the Copyright Clearance Center, Inc., Suite 910, 222 Rosewood Drive, Danvers, Massachusetts 01923, U.S.A. The complete backlist of publications from the World Bank is shown in the annual Index of Publications, which contains an alphabetical title list with full ordering information. The latest edition is available free of charge from the Distribution Unit, Office of the Publisher, The World Bank, 1818 H Street, N.W., Wash- ington, D.C. 20433, U.S.A., or from Publications, The World Bank, 66, avenue d'Iena, 75116 Paris, France. ISSN: 0259-210X Nicholas Prescott is senior economist in the World Bank's East Asia and Pacific Region. Menno Pradhan is a researcher for the Economic and Social Institute at the Free University in Amsterdam. Library of Congress Cataloging-in-Publication Data Prescott, Nicholas M. A poverty profile of Cambodia / Nicholas M. Prescott, Menno Pradhan p. cm. - (World Bank discussion papers ; no. 373) Includes bibliographical references. ISBN 0-8213-4020-4 1. Poverty-Cambodia-Statistics. 2. Cambodia-Economic conditions-Statistics. 3. Cambodia-Social conditions-Statistics. I. Pradhan, Menno, 1965- . II. Title. III. Series: World Bank discussion papers; 373. HC442.P74 1997 362.5'09596-dc2l 97-29733 CIP iii Contents Page Foreword ......................................................................................................................v Acknowledgm entsw.............................................................................................................. v Abstract ................................................................................................................... vii Executive Sum m ary......................................................................................................... viii 1. Introduction ..................................... 2. Socio-Econom ic Survey of Cam bodia, 1993-1994 .................................................3 3. Per Capita Consumption, Distribution and Inequality.............................................7 4. Setting a Poverty Line for Cam bodia.....................................................................15 5. Poverty Com parisons for Targeting .......................................................................21 6. International Poverty Comparisons........................................................................33 7. Com parison with Official Poverty Estim ates ........................................................35 8. Characteristics of the Poor.....................................................................................39 9. Im proving Poverty Analysis and Policy ................................................................41 References ....................................................................................................................47 ANNEXES A . Civil Servants, Poverty and Earnings ....................................................................49 B. Analysis of Gender-Poverty Linkages...................................................................53 C. Poverty Tables .......................................................................................................57 D . Reference Food Bundle..........................................................................................61 E. Technical Guide for Program m ers.........................................................................71 TABLES 1. Coverage and Sam ple Size of SESC 1993/94..........................................................4 2. Sum m ary Indicators of Per Capita Consumption 1993/94 ......................................7 3. Distribution of Per Capita Consum ption by Province .............................................9 4. M easures of Inequality by Region .........................................................................13 5. Differential Energy Requirem ents .........................................................................16 6. Composition of Food Poverty Line by Food Group..............................................18 7. Estim ated Food Dem and Equation ........................................................................16 8. Distribution of Poverty by Region.........................................................................19 9. International Poverty Com parisons........................................................................33 10. Decomposition of Differences in Poverty Estim ates.............................................38 11. Distribution of Social Indicators by Quintile.........................................................40 iv FIGURES 1. Distribution of Expenditure in Phnom Penh..........................................................10 2. Distribution of Expenditure in Other Urban Areas................................................11 3. Distribution of Expenditure in Rural Areas...........................................................11 4. Distributions of Calorie Consumption by Region .................................................12 5. Lorenz Distributions of Consumption by Region..................................................13 6. Distribution of Poverty Lines by Region...............................................................20 7. Sensitivity to Changes in Poverty Lines................................................................24 8. Dominance Test of Robustness..............................................................................25 9. Poverty Rate by Occupation of Head of Household..............................................27 10. Poverty Rate by Employer of Head of Household.................................................28 11. Poverty Rate by Level of Education of Head of Household..................................29 12. Years of Schooling by Age Cohort........................................................................30 13. Poverty Rate by Sex of Head of Households.........................................................31 14. Comparison of Poverty Lines ................................................................................35 15. Comparison of Poverty Rates ................................................................................36 BOXES 1. Three Different Poverty Measures.........................................................................22 V FOREWORD Since 1993 the Royal Government of Cambodia has made important strides in reestablishing political and economic stability. These achievements are impressive given the suffering that the country and its people experienced during the past quarter of a century. Most people still lack access to health and education facilities, potable water, electricity and serviceable roads. The country's natural and productive capital have suffered great losses. Land mines render large portions of valuable agricultural land unuseable, and pose a serious threat to people. But perhaps Cambodia's greatest loss was the depletion of its human capital by mass genocide and large-scale exodus of the most educated citizens during the years of Khmer Rouge rule from 1975 to 1979 when institutions were dismantled, the legal system destroyed and money abolished. With this legacy of suffering and devastation Cambodia is now one of the poorest countries in the world with a per capita income of only US$ 260. Achieving rapid poverty reduction is therefore a central goal of Cambodia's First Socioeconomic Development Plan, 1996-2000. Better and up-to-date information about the poor is needed to assist the government in designing effective policies for attacking poverty to achieve this goal. Who are the poor? How many poor are there? Where do they live? What are their sources of income? Answering such basic policy questions on poverty requires a systematic information base on the distribution of living standards. This paper constructs a consistent nationwide poverty profile to support the government's effort to strengthen the design of poverty reduction policies. The poverty profile is based on the first Socioeconomic Survey of Cambodia (SESC) carried out in 1993/94 by the National Institute of Statistics, Ministry of Planning and cosponsored by the Asian Development Bank and the United Nations Development Programme. The SESC represents the first large-scale national household survey of living standards undertaken in Cambodia, although complete coverage of rural areas was not feasible because of security risks. As such it marks an important contribution to development of the information base for better poverty analysis and policy in Cambodia. Cambodia's poverty profile for 1993/94 provides policy-oriented poverty comparisons which can be used to help target anti-poverty programs, to make international comparisons of poverty incidence, and to set a baseline for future monitoring of Cambodia's development progress over time. Publication of this paper makes available the methodology and results to stimulate discussion and comment on Cambodia's poverty reduction agenda within the development community. Javad Khalilzadeh-Shirazi Director Country Department 1 East Asia and Pacific Region vi ACKNOWLEDGMENTS Grateful acknowledgements are due to the Ministry of Planning National Institute of Statistics, the Asian Development Bank (ADB) and the United Nations Development Programme (UNDP) for granting permission to use the dataset of the Socio-Economic Survey (SESC) 1993/94. The SESC 1993/94 was conducted by the Strengthening Macroeconomic Management and Training Project in collaboration with the National Institute of Statistics and with the cosponsorship of ADB and UNDP. Special thanks are due to Andre Klap, Raja Korale and Mathew Varghese for their collaboration and support. Note: Unless otherwise noted, the data sources for Tables and Figures are the authors' compilations based on the SESC 1993/94 vii ABSTRACT This paper uses the Socioeconomic Survey of Cambodia (SESC) of 1993/94 to estimate poverty measures for Cambodia. The SESC was administered over four rounds to 5,578 households in three domains: Phnom Phenh, Other Urban and Rural areas. The paper begins by describing the SESC questionnaire and sampling frame to help interpret representativeness of the empirical results because not all areas of Cambodia could be included in the sampling frame. Basic data are given on the level and distribution of living standards as measured by per capita household consumption expenditures. New poverty lines for Cambodia are estimated and used to make poverty comparisons for targeting purposes -- assessing differences among regions, sectors of employment, levels of education, gender and household size -- and to make international comparisons between Cambodia and other countries in East Asia. These consumption-based poverty comparisons are supplemented with an assessment of the distribution of various nonmonetary welfare indicators between the poor and the better off in Cambodia. The paper concludes with recommendations for improving institutional capacity for poverty analysis and poverty in Cambodia. viii EXECUTIVE SUMMARY Key Issues in Poverty Analysis and Policy Poverty Reduction. Poverty reduction is a central goal of Cambodia's First Socioeconomic Development Plan, 1996-2000. Better and up-to-date information about the poor is essential to assist the Government in designing effective policies for attacking poverty. Who are the poor? How many poor are there? Where do they live? What are their sources of income? Policies intended to help the poor cannot succeed unless the Government knows who the poor are and how they are likely to respond to public interventions. This report takes an initial step towards answering these questions, and sets out an approach to developing the household survey information base for future poverty monitoring and assessment in Cambodia. Poverty Profile. Answering basic policy questions on poverty requires a systematic information base on the distribution of living standards in Cambodia. This report constructs a consistent nationwide poverty profile to support the Government's effort to strengthen the design and targeting of poverty reduction policies during implementation of the First Development Plan. This poverty profile is based on the first Socio-Economic Survey of Cambodia (SESC) carried out in 1993/94 by the Ministry of Planning, National Institute of Statistics and cosponsored by the Asian Development Bank and the United Nations Development Programme. The SESC represents the first large-scale national household survey of living standards undertaken in Cambodia, although complete coverage of rural areas was not feasible. As such it marks a historic contribution to development of the information base necessary to support better poverty analysis and policy. Poverty Comparisons. This poverty profile is used to make policy-oriented poverty comparisons which can be used to help target antipoverty programs, to make international comparisons, and to set a baseline for future monitoring of development progress over time. Is poverty higher among certain population groups-such as rural areas, remote regions, ethnic minorities--than others? Is poverty lower in Cambodia than its neighboring countries? Has poverty decreased over time? Setting a poverty line is a prerequisite for measuring poverty in order to make these poverty comparisons. Poverty Line. Unlike most of its neighboring countries, the Royal Government of Cambodia has not yet established an official poverty line. This report sets new region- specific poverty lines for Cambodia based on standard analytical methods used by the World Bank in constructing poverty estimates for neighboring countries. These poverty lines are based on a benchmark per capita calorie requirement of 2,100 calories per day. The composition of the underlying food bundle is chosen to represent typical consumption patterns in Cambodia. The regional poverty lines take into account geographic price variations in the cost of the same food basket in different parts of the country. The resulting poverty lines in 1993/94 are 1578 Riels per person per day in ix Phnom Penh. 1.264 Riels per person per day living in Other Urban areas. and 1,117 Riels per person per day in Rural areas. Poverty Indexes. Comparing these poverty lines with the individual distribution of per capita consumption expenditure reported in the SESC data provides the basis for measuring the poverty indexes which are used in making poverty comparisons. The most commonly used index of poverty is simply the proportion of the population whose expenditure levels fall below the poverty line, often called the head- count index. Poverty Comparisons for Targeting Regional Targeting. Targeting the design and placement of antipoverty programs is essential to reach disadvantaged groups and backward areas effectively and efficiently. Probably the most important use of the poverty profile is to support efforts to target development resources towards poorer areas, aiming to reduce aggregate poverty through regional targeting. Which regions should command priority in targeting? This question can only be answered at a highly aggregated level by the SESC because of the limited number of geographic domains which were sampled. The survey only supports regional comparisons between Phnom Penh, Other Urban and Rural areas. While this provides a broad sense of the appropriate policy orientation in regional targeting, it is obviously of limited practical value for choosing the geographic placement of project interventions. Looking first at the incidence of poverty in different areas, the regional poverty profile shows that rural poverty is higher than urban poverty. The incidence of rural poverty averages 43 percent -four times higher than the 11 percent poverty incidence found in Phnom Penh, and significantly higher than the 36 percent poor in Other Urban areas. Looking instead at the magnitude of regional contributions to national poverty - which take into account differences in the group's relative share of the national population, as well as differences in the incidence of poverty - shows that at least 85 percent of all the poor are concentrated in rural areas. Government policies to reduce poverty must, therefore, focus primarily on rural areas where the vast majority of the poor live. Employment Targeting. The ability of the vast majority of households in Cambodia to escape poverty will depend on their earnings from employment. Thus it is important to examine the relationship between poverty and the types of employment of working-age household members. The most important income-earner is usually the head of household. Looking first at the distribution of poverty incidence, the highest poverty rate--46 percent--is found among people living in households headed by farmers. By contrast, households headed by someone working in the government are least likely to be poor: in these occupations the poverty rate is only 20 percent. Clearly policies which aim x at reducing poverty through enhancing income generating capabilities should be targeted towards the agricultural sector. Turning to the contributions made to national poverty, the strongest policy message for targeting purposes is that more than three-quarters of the poor are found among households in which the head has an agricultural occupation. This reflects both the high proportion of people living in agricultural households and their above average poverty rate. This means that policies to reduce poverty in Cambodia must reach agricultural households if any major reduction in poverty is to be achieved -- any policy that misses the farmers will bypass around 75% of the poor. By contrast, households in which the head works for the formal public or private sectors account for less than 10% of overall poverty. Moreover, households headed by government workers account for only 3 percent of national poverty. In other words, policies that focus on employee conditions -particularly civil servants--will miss more than 90% of the poor in Cambodia. Education Targeting. The relationship between poverty and education is particularly important because of the key role played by education in raising economic growth and reducing poverty. The better educated have higher incomes and thus are much less likely to be poor. Cambodians living in households with an uneducated head are more likely to be poor, with a poverty rate of 47 percent. But at higher levels of education, the likelihood of being poor falls considerably. The prevalence of poverty among households in which the head has completed secondary education falls to around 30 percent. Raising educational attainment is clearly a high priority in order to improve living standards and reduce poverty -at present around three-quarters of all the poor live in households headed by somebody who has either completed only primary school (44 percent) or has no education at all (28 percent). Gender Targeting. The status of women, who in most developing countries are disadvantaged in comparison with men, is an important policy concern. One indicator of the gender gap is whether female-headed households are worse off than those headed by males. This might be expected to be a major concern in Cambodia since nearly one- quarter of the population live in households headed by women. In fact, the SESC data show that female-headed households in Cambodia are less likely to be poor than male- headed households. The incidence of poverty averages only 35 percent in female-headed households. compared to 40 percent in male-headed households. Although people living in female-headed households account for nearly 23 percent of the population, they account for only about 15 percent of all the poor in Cambodia. Overall, it does not appear to be the case that female-headed households are generally more vulnerable to poverty than those headed by males; in fact the opposite seems to be the case. In this respect gender and poverty patterns by household headship are similar to those observed in other East Asian countries such as Vietnam and Indonesia. xi International Poverty Comparisons Is Cambodia poorer or better off than other East Asian countries in terms of the proportion of the total population living in poverty? Reliable comparisons between Cambodia and other countries cannot be made at the national level because a large part of the country was excluded from the SESC sample frame. Nevertheless it is possible to compare the regional poverty estimates for Cambodian with corresponding estimates from other countries for which poverty estimates have been made using comparable methodology. These comparisons suggest that the incidence of rural poverty in Cambodia (43 percent) is lower than among its Indochina comparators --Vietnam (47 percent) and Laos (53 percent). But rural poverty in Cambodia remains considerably higher than elsewhere in East Asia, taking Indonesia as an example (24 percent). However it must be borne in mind that rural poverty in Cambodia may have been underestimated by the exclusion of significant portions of the countryside from the SESC sample. On the other hand, urban poverty appears to be slightly higher in Cambodia (24 percent) than among its Indochina neighbors --Vietnam (20 percent) or Laos (24 percent)-and much higher than Indonesia (10 percent). These international comparisons emphasise the magnitude of the development gap which remains to be closed between the economies of Indochina and the rest of East Asia, while suggesting that Cambodia's starting point is not very different from its neighbors in Indochina. Poverty Estimates in the Plan The First SocioEconomic Development Plan reports official estimates of poverty in Cambodia. These estimates were constructed using summary tabulations of the household expenditure distribution from the SESC 1993/94. This report revises these preliminary estimates by using the detailed individual records from the full SESC dataset. The revised estimates are significantly different: * the revised poverty line for Phnom Penh is 25% lower, while the revised poverty lines are similar in both the Other Urban and Rural areas. * the revised poverty rate for Phnom Penh is lower, while the revised poverty estimates for Other Urban and Rural areas are much higher. The lower poverty estimate for Phnom Penh is primarily due to differences in the poverty line, which in turn results from the relatively high allowance for nonfood consumption incorporated in the Plan estimates compared to this report. On the other hand the large difference in the poverty estimates for Other Urban and Rural areas is almost completely driven by the difference in the ranking method. In this case, ranking the distribution of households on the basis of total household consumption in constructing the preliminary estimates in the Plan--instead of ranking household members by per capita consumption--turned out to be an unfortunate approximation. xii Institutionalising Poverty Analysis and Policy Constraints. The policy relevance of the poverty profile for Cambodia based on SESC 1993/94 illustrates the importance of adequate data from large-scale household surveys in analyzing poverty and designing appropriate policies to translate the Government's commitment to poverty reduction into action. Three significant limitations will need to be overcome in future development of the information base on living standards in Cambodia: The first is that the SESC was not designed to be a fully integrated multipurpose survey of the type needed to conduct a comprehensive poverty assessment for policymakers. For example, while the expenditure data collected by SESC 1993/94 generate a detailed picture of the distribution of per capita consumption, it is not possible to analyse levels or determinants of access to social services among the poor because the survey did not simultaneously collect data on education enrollments or utilisation of health services, nor did it collect data on the price of obtaining access to these services. The ongoing SESC 1996 goes to the other extreme. While extensive data are collected on nonmonetary indicators such as child nutrition or immunisation coverage, only limited information will be collected on household consumption expenditure so that interrelations with poverty status may not be clearly identified. Second, the SESC is not able to support nationwide geographic disaggregations of key variables such as per capita expenditure and poverty incidence at the provincial level. This imposes an important practical limitation on its potential use in project planning. The need to overcome similar limitations in other countries have led to the adoption of much larger sample sizes. For example, the annual core household survey in Indonesia is now administered to over 200,000 households. Similarly, Vietnam now uses a sample of 45,000 households for the annual multipurpose survey. These enhancements are intended to provide much finer identification of the geographic location of poverty problems, support the design of more efficiently targeted poverty alleviation programs and strengthen capacity for decentralized planning at the provincial level. Third, the SESC surveys have not yet been institutionalized as part of a systematic, long-term and regular effort to evaluate the extent and nature of the poverty problem in Cambodia, to monitor progress in poverty reduction over time, and to evaluate the effectiveness of specific targeted antipoverty interventions. This means there is a risk that the potential benefits of these surveys will be short-lived. If the investments which have been made in creating new skills in survey design, field procedures, data processing methods, policy analysis and program design are not maintained continuously then future surveys will become much more difficult to implement. Agenda for Institutional Strengthening. The importance of consolidating and sustaining these nascent efforts to strengthen the information base for policymaking on poverty is increasingly recognized in neighbouring countries. For example, in the early xiii 1990s Indonesia initiated a major collaborative effort between the planning and statistics agencies to redesign its national household survey system (SUSENAS) to provide better data to guide the country's poverty alleviation programs. China is planning to set up a Poverty Monitoring and Evaluation System with the same objective. Recently Vietnam introduced an annual series of large-scale multipurpose household surveys of living standards. The agenda for institutionalizing poverty analysis and policy in Cambodia embraces four key elements. First, the government should consider adopting routine implementation of a new national household survey which offers both multipurpose coverage and geographic disaggregation. This calls for a two-part "core/module" household survey design. The purposes of the core are to support monitoring of changes in key indicators over time, and identification of priority areas for geographic targeting of development programs. To serve these needs, the contents of the core questionnaire could be fixed on a small number of key welfare indicators, e.g. per capita consumption, education enrollment, health care utilization rates, and the questionnaire would be implemented relatively frequently -- probably every year -- on a large sample so as to give estimates of the indicators disaggregated at least to provincial level. At the same time the core could be supplemented every year with a rotating sector module. The purpose of the modules would be to support in-depth analysis of sectoral issues and policies, such as the effects on the poor of changing pricing policies in the social sectors. Given the focus of the modules on analysis rather than monitoring, individual sector modules need not be carried out every year or on a large sample. Instead they could rotate over a 3 year cycle on a subsample of the core - for example, a social sector module, followed by an income and employment module, and then a detailed consumption module. The core/module survey of households needs to be linked to a community survey conducted at the village level. The purpose of the community survey is to collect data on variables which affect all households in the community, such as public/private provision of economic infrastructure (e.g. land, irrigation, agricultural extension, roads and markets) and social services (e.g. availability and quality of schools and health services). Thus the community survey plays an essential role in analysing determinants of household behavior and welfare based on merged datasets using the community and household data. Ultimately the community survey can also play a role in poverty monitoring for targeting purposes. This requires implementation of the community survey on a census basis. For example, in 1993 the government of Indonesia prepared a nationwide poverty map identifying poor villages based on data collected in the community survey (Potensi Desa or PODES) which is administered to all villages every three years. This poverty map has become the operational basis for targeting a major poverty alleviation program, comprising decentralized grants to poor villages. In addition, the poverty map has focused geographic targeting of many other government programs in different sectors. xiv A second key element of the future agenda is to improve the institutional linkages between the National Institute of Statistics as the technical agency responsible for the quality of data production, and its client policymakers in the Ministry of Planning and the relevant line agencies. In order to make effective use of the household survey database, the clients need to perceive it to be useful by contributing to design of the content of the surveys so that they are responsive to key policy questions. A third factor which conditions the strength of institutional linkages is timeliness of turnaround from the surveys. Improving the speed of data processing, availability and dissemination to users in the government would probably require significant enhancement in PC-based computing capacity at the National Institute of Statistics. Finally, for Cambodia to benefit fully from these improvements in the design, regularity and turnaround in the information base on living standards will require continuing improvements in the analytical skills of policymakers and researchers in Cambodia through a combination of training and hands-on experience. Meeting this objective demands a new effort to train staff in methods of applied policy analysis in government agencies. Proposed Work Program. Development of an integrated household survey system using a core/module design would need to be sustained over a multiyear period. The full cycle should last three years to allow for the necessary expansion in sample size, refinements of survey design, and capacity building in operational procedures. At the end of this period the National Institute of Statistics would have implemented all the components of a fully integrated multipurpose household survey system, and should be prepared to maintain it on a routine basis. Ultimately the operational value of improving the poverty related information base provided by the integrated household survey system will depend on the capacity to analyse the data and interpret its practical implications for policymakers. This capacity needs to be located within the Ministry of Planning because of the complex cross-sectoral agenda involved in formulation of antipoverty policies, and the need to interact with policymakers involved in setting strategic priorities between and within different sectors. Several other countries have adopted this approach. For example, in Indonesia the World Bank is currently financing a Social Sector Capacity Building Project implemented by the planning ministry (BAPPENAS). The project establishes a technical advisory capacity managed by BAPPENAS and linked to the line ministries. International donors are committed to supporting the poverty reduction goal set forth in the First Socioeconomic Development Plan, and share a common interest with the government in strengthening the information base on living standards in Cambodia so as to improve the policy dialogue on sectoral priorities and project design. Accordingly, the donors need to work together in mobilising the external financing required to implement the joint work program on data collection and policy analysis which will be necessary to underpin the government's policy commitment to fight poverty. UNDP has taken the lead xv in coordinating this effort in the framework of a new project on Capacity Development for Socio-Economic Surveys and Planning.(CMB/96/019/A/01/42). 1. INTRODUCTION Poverty reduction is a central goal of Cambodia's First Socioeconomic Development Plan, 1996-2000. Better and up-to-date information about the poor is essential to assist the Government in designing effective policies for attacking poverty. Who are the poor? How many poor are there? Where do they live? What are their sources of income? Policies intended to help the poor are unlikely to succeed unless the Government knows who the poor are and how they are likely to respond to public interventions. This report develops a consistent nationwide profile of poverty to support the Government's effort to strengthen the design and targeting of poverty reduction policies. The poverty profile is based on the first Socio-Economic Survey of Cambodia (SESC) carried out by the National Institute of Statistics in 1993/94 and cosponsered by the Asian Development Bank and the United Nations Development Programme. The SESC collected data from about 5,600 households representing Phnom Penh, other urban, and rural areas in Cambodia. For policy purposes the most important reason for measuring poverty is not the need for a descriptive number, but to make poverty comparisons in order to target antipoverty programs and monitor development progress. Is poverty higher among certain population groups--rural areas, regions, ethnic minorities--than others? Has poverty decreased over time? Setting a poverty line is a prerequisite for measuring poverty in order to make these poverty comparisons. Unlike most of its neighboring countries, the Royal Government of Cambodia has not yet established a firm basis for setting an official poverty line. This report sets new poverty lines for Cambodia based on standard methods used by the World Bank in poverty estimates for neighboring countries. These poverty lines are based on a benchmark per capita calorie requirement of 2,100 calories per day -- with the composition of the underlying food bundle chosen to be representative of typical consumption patterns in Cambodia -- and they take into account geographic price variations in the cost of the same food basket. Comparing these poverty lines with the distribution of per capita consumption expenditure from the SESC yields poverty estimates for 1993/94. Section B begins by describing the SESC questionnaire and sampling frame. This is important in interpreting the representativeness of the empirical results because not all areas of Cambodia were included in the sampling frame for the 1993/94 survey. Section C presents basic data on the level and distribution of living standards as measured by per capita household consumption expenditures. Section D then sets new poverty lines for Cambodia using standard World Bank methods. Section E uses these poverty lines to make poverty comparisons for targeting purposes -- assessing differences among regions, 2 sectors of employment, levels of education, gender and household size. Section F switches to international poverty comparisons, comparing the poverty estimates for Cambodia with recent World Bank estimates for Laos, Vietnam and Indonesia. Section G concludes the assessment of consumption-based measures of poverty by evaluating the difference between the official poverty estimates presented in the First Socioeconomic Development Plan and the revised estimates presented in this report. Finally, Section H gives an overview of the distribution of selected non-monetary welfare indicators between the poor and the better off in Cambodia. 3 2. SOCIO-ECONOMIC SURVEY OF CAMBODIA, 1993-1994 The data analysis is based on the Socio-Economic Survey of Cambodia held in 1993-1994 (SESC 1993/94). The survey was carried out by the National Institute of Statistics with technical assistance provided by the Asian Development Bank and UNDP. The SESC included 3,2079 people in 5,578 households. The survey was administered over four rounds to capture seasonal patterns in consumption. All sampling units were sampled in every round. The first round of the survey was conducted in the fall of 1993, the remaining three rounds in the succeeding three quarters of 1994. The survey distinguishes three main strata: Phnom Penh, Other Urban centers and the Rural areas. Sample Design Sample selection was based on a stratified two-stage random sample design. Stratification took place at the level of three main geographic domains: Phnom Penh, Other Urban and Rural. Within each domain, villages were selected at random in the first stage. Households were sampled in the second stage. Villages with more inhabitants had more households sampled. Villages were sampled on the basis of the UNTAC frame. Households were sampled on the basis of a listing of all households in the selected villages. The design was self-weighting in the sense that within each of the geographical strata, each household had an equal probability of being selected into the survey. The survey data files include an expansion factor I which can be used to obtain estimates for the surveyed areas. The sampling frame was derived from the nationwide village population data file prepared by the United Nations Transitional Authority (UNTAC) in Cambodia. Sample Coverage It is important to note that the survey used a truncated sampling frame which did not cover all of Cambodia. The truncated frame excluded those areas which were unsafe at the time of the survey or which were sparsely populated and expensive to reach. For example, some of the northern provinces could only have been reached through Vietnam. Thus the geographic coverage of the survey was limited to only 15 out of Cambodia's 21 provinces. Within those provinces which were covered by the SESC, selected villages were also excluded for security reasons. Altogether, 90 urban villages and 5,093 rural villages were excluded from the list of primary sampling units covered by the truncated sampling frame. Table 1 summarizes the resulting coverage of the I The expansion factor is defined as one over the sampling probability for households. 4 truncated frame relative to the original UNTAC frame, together with the actual sample size selected within each stratum. Table 1: Coverage and Sample Size of SESC 1993/94 Sample Expansion Truncated UJNTAC Percent size factor frame /a frame coverage Phnom Penh Village 160 3 496 496 100 Household 1,708 71 121,134 121,134 100 Individuals 10,254 65 667,814 667,814 100 Other Urban Village 99 6 566 673 84 Household 1,151 108 124,012 136,277 91 Individuals 6,835 87 595,993 661,872 90 Rural Village 239 27 6,489 11,588 56 Household 2,719 348 947,147 1,457,149 65 Individuals 14,990 299 4,488,565 7,493,809 60 Cambodia Village 498 15 7551 12,798 59 Household 5,578 214 1,192,897 1,754,260 68 Individuals 32,079 179 5,752,372 8,823,495 65 /a Count based on UNTAC listings Source: National Institute of Statistics, Ministry of Planning Looking first at coverage in terms of villages, all of the 496 villages in Phnom Penh were covered. In Other Urban areas, the truncated frame covered 566 or 84% of the villages. By contrast, coverage of Rural areas was significantly incomplete, including only 6,489 or 56 percent of the villages. 5 Expressed in terms of coverage of households - taking into account variations in household density per village -- the SESC sample covered 68 percent of all households in Cambodia, ranging from 100% in Phnom Penh to 65% in Rural areas. Apparently the less populated villages have been excluded from the survey. Sampling was more dense in the urban areas. The average expansion factor for households is 71 in Phnom Penh and 107 in Other Urban areas. In contrast, for the rural domain the average expansion factor amounts to 348. Overall, the survey covered 65% of the individuals of Cambodia - slightly lower than the percentage of households (68%). Coverage ranged from 100% of the population living in Phnom Penh, to 90% of the population in Other Urban areas and only 60% of the Rural population. All the empirical results presented in this report are valid only for those areas which were included in the survey. SESC Questionnaire Since the main objective of the SESC was to generate expenditure weights for a new consumer price index, the survey questionnaire collected very detailed information on consumption patterns. The survey distinguishes 177 food items of food expenditure. For each item the respondent was asked to provide the quantity consumed in the past week and the value of this consumption. Next, the respondent was asked to separate this amount between cash expenditures and in kind consumption. Cash expenditures include all goods purchased at the market, in cash or on credit. In kind consumption includes gifts and consumption of home produced goods. Both for in cash and in kind consumption the survey collects both quantities and values. In case the respondent indicates that his consumption on a particular item includes both cash as well as in kind expenditures, the interviewer is instructed to use the market price to value the in kind consumption. The survey distinguishes 266 different categories of nonfood con- sumption. The reference periods for nonfood consumption differ depending on the item. For nonfood consumption only expenditure values were collected. Again, the survey distinguishes between in kind and in cash consumption. The information gathered in the survey is sufficient to construct a descriptive poverty profile using consumption-based measures of poverty. The distribution of per capita consumption expenditure is available, together with the detailed data on quantities of food consumption which are needed to construct a calorie-based poverty line. Prices for food items can be derived in the form of unit values from the survey data, since both quantity and expenditure data were collected. However, the SESC 1993/94 is not sufficiently detailed to undertake a thorough analysis of the causes and consequences of poverty. This would require a comprehensive multipurpose survey linking detailed information on household consumption behavior, incomes and employment, and social services at both the household and community levels. 6 Data Cleaning The results presented in this report were obtained after two minor data cleaning operations. In the first place, where quantities were missing for food consumption but a nonzero consumption value had been recorded, quantities were imputed using the estimated price for that particular food item. Second, if a quantity (value) was outside the 95 percent confidence interval, and the unit value for that observation differed by factor of more than five from the estimated price for that region, the quantity (value) was imputed on the basis of the value (quantity) and the estimated price for that food item. Note that the second data cleaning method does not automatically erase extreme values. If the value and quantity observation both indicate an extremely high consumption, and hence the unit value is close to the estimated price, the observation is retained. 7 3. PER CAPITA CONSUMPTION, DISTRIBUTION AND INEQUALITY Per Capita Consumption Per capita consumption is widely used as a basic indicator of welfare standards. Per capita consumption in the survey areas averaged about 1,314 Riels per day in 1993/94, which is roughly equal to about USD218 per year (see Table 2). The high share of food in total household consumption expenditures is another indicator of Cambodia's low standard of living: on average 67% of all consumption expenditures are devoted to food consumption. A further indicator is the low level of calorie consumption, averaging 2,261 calories per capita per day. These indicators of low average consumption mask wide differences in consumption standards between different population groups. Looking at differences in real consumption -- nominal consumption expenditure deflated to take into account spatial cost of living differences -- shows a fairly large disparity between the vast majority of the population that live in rural areas and those who live in urban areas. Real per capita consumption expenditure is more than twice as high in Phnom Penh, and 50% higher in Other Urban areas. Table 2: Summary Indicators of Per Capita Consumption 1993-94 Per capita consumption per day Nominal Real /b Calories Food share Quintile /a 1-poor 734 963 1,676 0.75 2 1,029 1,356 1,997 0.71 3 1,314 1,721 2,200 0.69 4 1,803 2,307 2,405 0.66 5-rich 4,281 4,962 2,846 0.57 Region Phnom Penh 4,367 4,367 2,156 0.56 Other'Urban 2,412 2,873 2,132 0.67 Rural 1,403 1,887 2,247 0.69 Total /c 1,832 2,262 2,225 0.68 /a Quintiles are constructed on the basis of real per capita consumption per day; aggregation of the quintile distribution is based on sample weights /b Real consumption data are expressed in Phnom Penh prices using the Laspeyres price deflator reflected in the food poverty lines for each region. /c Total refers only to mean values for the truncated areas covered by SESC. 8 Distribution of Per Capita Consumption by Province A true nationwide breakdown of the geographic distribution of per capita consumption between provinces is not available because 6 out of Cambodia's 21 provinces were excluded from coverage by the SESC (Preah Vihear, Koh Kong, Mondol Kiri, Ratanak Kiri, Stung Treng and Kratie). The estimated mean values of per capita consumption expenditure in the 15 provinces which were covered are shown in Table 3. Because of village exclusions within the covered provinces, the sample size on some of these provinces was small. Accordingly, the provincial means are shown together with the 95 percent confidence intervals. The main urban centers -- Phnom Penh and Sihanouk Ville -- are clearly the best off. The two provinces surveyed in the coastal region -- Kompong Som and Kom Pot -- have significantly higher per capita consumption levels than the other regions. Border provinces in the far West near Thailand, and in the East near Vietnam, have the lowest average consumption levels. Most of the Northern provinces along the border with Vietnam, Laos and Thailand were excluded from the survey. 9 Table 3: Distribution of Per Capita Consumption by Province (in Riels per day) Daily per capita consumption 95% confidence interval of mean Number of median mean Standard lower upper households in error of band band sample mean Phnom Penh 1708 1219 4367 116 4140 4594 Kandal 510 1381 1642 45 1553 1730 Kompong Cham 664 1195 1426 36 1356 1496 Svay Rieng 270 1092 1194 35 1125 1262 Prey Veng 594 1225 1465 32 1403 1527 Takeo 413 1226 1521 100 1325 1718 Plain regions 4159 1323 1878 36 1808 1949 Kompong Thom 97 1269 1779 143 1495 2064 Siem Reap 175 1219 1549 114 1323 1775 Banteay 206 1137 1412 111 1194 1631 Meanchey Battambang 293 1125 1392 54 1286 1497 Pursat 207 1333 1697 89 1522 1872 Kompong 136 1271 1585 90 1408 1762 Chhnang Tonle Sap Lake 1114 1219 1529 39 1453 1605 region Sihanouk Ville /a 70 4395 5162 378 4407 5917 Kom Pot 130 1147 2159 573 1025 3294 Coastal Region 200 1397 2803 431 1953 3652 Kompong Speu 105 1214 1296 54 1189 1403 Plateau and 105 1214 1296 54 1186 1403 Mountain Region Cambodia 5578 1300 1833 32 1770 1895 /a Kompong Som province 10 Individual Distribution of Per Capita Consumption The disparities in individual per capita consumption standards attained by the Cambodian population are illustrated by the cumulative distribution function showing the proportion of the population which is at or below a given consumption standard. The cumulative distributions of per capita expenditure on food, non-food and total items are shown separately for each of the survey regions in Figures 1, 2 and 3. The lines show the fraction of the population on the vertical axis whose consumption is less or equal to the amount indicated on the horizontal axis. Note that the consumption values on the horizontal axis are expressed in logarithms. In Phnom Penh, the upper tail of the individual consumption distribution is sufficiently well off that the food and non-food expenditure distributions cross over before reaching a cumulative 80% share of the population. In other words, for at least the richest 20 percent of the population per capita expenditures are high enough to allow nonfood spending to exceed food expenditures. By contrast the distribution of individual consumption levels in the Rural sample is sufficiently low that food consumption exceeds nonfood spending for everybody. 1 0.8 o.0.6 0 0.4 Food Non-food Total 0.2 - 01 4 5 6 7 8 9 10 Log of daily consumption per person (in Riels) Figure 1: DISTRIBUTION OF EXPENDITURE IN PHNOM PENH 11 0.8 0.6 0.4 Non-foodToa 0.2 - Food 0 I 1 I I I 4 5 6 7 8 9 10 Log of daily consumption per person (in Riels) Figure 2: DISTRIBUTION OF EXPENDITURE IN OTHER URBAN AREAS 0.8 - 0.6 O 0.4 Non-food Food 0.2 o 0 4 5 6 7 8 9 10 Log of daily consumption per person (in Riels) Figure 3: DISTRIBUTION OF EXPENDITURE IN RURAL AREAS 12 The cumulative distributions of per capita calorie consumption by region are compared in Figure 4. The disparities in individual calorie consumption are very wide, with around one- half of the population purchasing the equivalent of fewer than 2,100 calories per day. Note that daily calorie consumption in rural areas exceeds the level attained in urban areas, while the calorie distribution patterns within urban areas are very similar. This reflects the higher energy requirements of daily activities in rural areas. 1 0.8 ~.0.6 Phnom Penh Other Urban 0.4 0.2 - 0 500 1000 1500 2000 2500 3000 3500 4000 Daily calorie consumption per person Figure 4: DISTRIBUTIONS OF CALORIE CONSUMPTION BY REGION Inequality How does inequality in Cambodia compare with other countries? Lorenz curves of the cumulative share of consumption as a function of cumulative population shares show that inequality in consumption expenditure is higher in urban than rural areas, with the greatest disparity evident within Other Urban areas (see Figure 5). The richest 10 percent of the population accounts for more than 30 percent of total consumption expenditure in the urban areas, while the poorest 10 percent consume less than 3 percent. 13 100 80 0. o 60 RuralPhnom Penh 40 Other Urban 20- -40 0 20 40 60 80 10 Cumulative percent of population (%) Figure 5: LORENZ DISTRIBUTIONS OF CONSUMPTION BY REGION Table 4: Inequality Measures by Region Gini Ti/b T2/c LV/d Consumption Consumption share (%) of share (%) of poorest 10% richest 10% Phnom Penh 0.39 0.31 0.26 0.46 2.5 31.2 Other Urban 0.44 0.46 0.33 0.49 2.7 36.7 Rural 0.27 0.13 0.12 0.21 4.4 22.9 Total/a 0.38 0.33 0.25 0.37 3.4 32.8 /a Total refers to sampled areas only /b TI is the population weighted Theil measure of inequality /c T2 is the per capita consumption weighted Theil measure /d LV is the variance of the logarithm of per capita consumption 14 The Gini coefficient -- defined as the ratio of the area between the Lorenz curve and the 45 degree line of complete equality -- ranges from a maximum value of 0.44 in Other Urban areas to a minimum value of 0.27 in Rural areas (see Table 4). Gini coefficients for neighboring countries are 0.32 in Laos, 0.34 in Vietnam and 0.32 in Indonesia. Given the heavy weight of rural areas in the population, Cambodia appears to exhibit a similarly low degree of overall inequality as some of its neighboring countries. 15 4. SETTING A POVERTY LINE FOR CAMBODIA The starting point for developing an appropriate poverty line is the basic notion that food is the most fundamental need of human beings. Extreme lack of food leads to death, while chronic insufficiency of food leads to physical weakness, greater susceptibility to disease and, among children, impaired cognitive development. This suggests that any method for calculating a poverty line should be closely tied to sufficient food intake. A large biomedical literature exists which attempts to calculate the amounts of food needed for normal daily activities and long-term health. While nutritional needs encompass a large range of requirements (protein, energy, and many micronutrients), for purposes of assessing the extent of poverty it is best to focus on energy intake, which is probably the single most important indicator of adequate food consumption. Measuring energy intake in terms of calories, it is possible to estimate a poverty line by calculating how much money is just enough to allow a household to meet its daily calorie needs as estimated by biomedical studies. Following recent common practice in other East Asian countries, this report adopts the benchmark per capita calorie requirement of 2,100 goto For the poverty line to be realistic, it needs to allow households to consume a "typical" basket of foods reflecting local tastes, rather than requiring a household to spend all its money on the single food item which has the lowest "price per calorie." This report takes the composition of food consumption in the third quintile--which on average consumes nearly 2,100 calories--as the reference composition of an appropriate food bundle. The cost of this reference food bundle is then determined in different regions taking into account local variations in the cost of the same food basket. This yields the food component of the poverty line. An additional allowance needs to be made for consumption of non-food goods. Even households that are poor in the sense that they are consuming less than their recommended daily calorie requirement still spend some of their money on non-food items. This report incorporates a minimal allowance for non- food goods based on the typical non-food spending of those who can just afford the reference food requirement but actually displace some amount of food expenditures Food poverty line As noted above, the benchmark adopted for setting the food poverty line is a 2,100 calorie minimum energy requirement per person per day. In principle, one could allow the calorie requirement to vary by age, weight and the activity of the individual. Table 5 illustrates the calorie requirements for different types of individuals. The heavier the daily activity, the more calories are needed. Children generally need less calories 16 than adults. These examples suggest that the 2,100 calorie requirement used in this report is conservative and may be considered a minimum daily energy requirement. Table 5: Differential Energy Requirements Calorie requirement Subsistence farmer 2780 Male engaged in heavy work 3490 Rural woman in developing country 2235 10 year old boy in developing country 2080 10 year old girl in developing country 1915 Source: WHO (1985) Setting the poverty line requires specification of a basket of food items yielding exactly 2, 100 calories. The relative composition of the basket is obtained from observed dietary patterns. To determine a typical consumption pattern, a reference household has been constructed. The reference household is derived from observations in the middle (third) quintile in the per capita consumption distribution (appropriately weighted). The reference food bundle is constructed by taking average values of the reported quantities for every food item. Next, the calorie content of this basket is determined. The food basket for the poverty line is obtained by scaling all quantities by the same factors such that the basket has a calorie content of 2100. In the Cambodian case, the reference food bundle has a calorie content of 2298 calories. As a result all quantities were scaled down by a factor (2100/2298) = 0.9138. The quantities of food in the reference food consumption bundle are listed in Annex 2 together with the calorie values which were used to derive its contents. For some categories, such as meals eaten outside of home, no predetermined calorie values exist. The assumption underlying the calorie computations is that food eaten outside of home is typically twice as expensive in terms of Riels per calorie as food prepared at home. This assumption is implemented as follows: For all households the Riels per calorie value of food prepared at home is determined. Next, the median value is taken for each region. These, multiplied by two, are then used as the prices per calorie for food eaten outside of home. The results are shown under the price listings for each region. The choice of the third quintile as the reference household deserves further comment. The third quintile was used for two reasons. First, because the third quintile 17 was also used in constructing the poverty estimates which are reported in the First Five- year Socioeconomic Development Plan. And second, because the benchmark of 2,100 calories per day is attained among this group of the population. In general, one observes that the poorer the reference group, the more rice (and less meat) oriented the consumption basket. Since rice has a high calorie content per Riel, poverty lines based on the consumption pattern of poorer reference households will tend to be lower. Having constructed the reference food bundle, it was priced against local market prices in each region. Separate price estimates were obtained for the three geographical regions distinguished in the SESC. Prices are estimated by taking median unit values of cash purchases for each product. Medians were used in order to make the estimate less sensitive to measurement error. In case less than 10 purchases were observed in a region, the price estimate is based on the median unit value in the whole sample. Using data on cash purchases only ensures that the price estimates represent market prices. In general, the analysis shows that the self-assessed value of in kind consumption has a lower unit value. One possible explanation for this could be that farmers retain their bad quality products for own consumption and sell the good quality on the market. Since the food poverty basket should reflect similar qualities in urban and rural areas, it was decided to use market prices to value the food basket. This method yields a food poverty line of 1,185 Riel per day for Phnom Penh, 995 Riel per day for Other Urban areas and 881 Riel per day for Rural areas. These lines represent the minimum expenditure required per person to reach a daily calorie consumption of 2,100. The food poverty lines and the underlying breakdown by broad food categories are shown in Table 6. More than two-thirds (69%) of the calories are obtained from cereals, especially rice. Cereals are a cheap way of obtaining calories - at most 28 percent of the total food poverty line is allocated for purchasing cereals. Meat consumption is the largest expenditure category in all regions. The high calorie values for beverages mostly stem from the consumption of local wines. The calories in "sugar, salt and seasoning" are driven by the use of sugar and fish sauce. The resulting food poverty line is thus a Laspeyres price index for food, where the consumption bundle of the reference household provides the weights for each of the food items. 18 Table 6. Composition of Food Poverty Line by Food Group (in Riels per person per day) Phnom Penh Other Urban Rural Calories 1184.9 995.3 881.4 2100 Beverages 51.3 37.1 31.2 122.3 Cereal 289.0 247.3 246.8 1440.2 Dairy products 7.6 2.7 5.7 1.5 Eggs 20.6 20.8 20.7 7.8 Fruit 104.5 78.2 62.5 55.6 Meat 433.7 368.3 311.7 202.8 Oils and fat 13.1 12.8 12.5 50.3 Other food products 54.4 35.4 26.2 55.7 Sugar, salt, spices and seasoning 92.3 84.2 81.0 121.5 Vegetables 118.3 108.5 83.1 42.2 Non-food allowance The method used to construct the non-food allowance in the poverty line was originally developed by Ravallion & Bidani in poverty estimates for Indonesia. They defined basic non-food spending requirements in terms of how much is spent on non-food goods by households who are just capable of reaching their nutritional requirements. In other words it is the amount of non-food spending which people who are at the food poverty line will allow to displace basic food expenditure as reflected in the food poverty line. This amount is determined on the basis of observed non-food consumption of households whose total expenditures equal the food poverty line. Apparently these households consider spending part of their expenditure on non-food items welfare improving. The welfare derived from this amount of non-food expenditure is apparently higher than welfare derived from the foregone food expenditures. It can thus be considered a minimal allowance for non-food spending. Regression analysis is used to identify the typical value of non-food expenditures of households capable of reaching the food poverty line. The following food demand function, representing the food share as a linear function of the value of total spending relative to the food poverty line, has been estimated: 19 S;= a+plog(xi/zf)+E; i=1,..,N j = 1(PhnomPenh), 2(OtherUrban), 3 (rural) where s,= share of total expenditure of household i devoted to food x, = total expenditure of household i z = food poverty line in region j The log in the regression ensures that the equation will fit a pattern of a diminishing food share as total expenditures increase. By allowing the constant term to differ by region, this method yields different shares of the non-food allowances in the poverty line for each region. This is necessary because price differences between regions of non-food items may be different from price differences of food items. The regression results are given in Table 7. Table 7: Estimated Food Demand Equation (dependent variable: food share in total consumption) Estimated Standard coefficient error constant 0.729846 0.00196 dummy Phnom Penh -.0628002 0.0053 dummy Other Urban -.0019263 0.0049 beta -.1076779 0.0029 R squared 0.28 Using this approach, the estimated non-food allowance is 393 Riels per day in Phnom Penh, 269 in Other Urban areas and 236 in Rural areas. Poverty Lines The poverty lines are obtained by adding the non-food allowance to the food poverty line for each region. After some calculus, it can be shown that the poverty line equals z = z (2-aj) where z = The poverty line for regionj. 20 The resulting overall poverty lines for Cambodia in 1993/94 are 1,578 Riels per person per day in Phnom Penh, 1,264 Riels for Other Urban and 1,117 Riels for the rural areas. These poverty lines, and their underlying food and non-food allowances are summarized in Figure 6. 1600 1400 - 1200 b 1000 - e 800 - l 600 - 400 - 200 0 Phnom Penh Other Urban Rural a Food poverty line C Non food allowance Figure 6: DISTRIBUTIONS OF POVERTY LINES BY REGION (in Riels per capita per day) 21 5. POVERTY COMPARISONS FOR TARGETING Choosing a Poverty Index Having set the real poverty line for Cambodia, an index of poverty needs to be chosen which calculates the extent of poverty based on the distribution of household expenditure. The most commonly used index of poverty is simply the proportion of the population whose expenditure levels fall below the poverty line, often called the head-count index. One limitation of the head-count index is that it does not measure how far poor households' expenditure levels do fall below the poverty line. A second index of poverty which does take into account variations in how far poor households' expenditure levels fall below the poverty line is known as the poverty gap index. The poverty gap measures the average shortfall (gap) between the poor households' expenditure levels and the poverty line. A third measure, the poverty severity index, goes further and takes into account the distribution of living standards among the poor. This measure is sensitive to such inequality, indicating more poverty when the average poverty gap is more unequally distributed among the poor. All of these belong to the general class of Foster- Greer- Thorbecke poverty measures (see Box 1.1). In practice, comparisons of poverty across different groups, or over time, may not yield results that vary substantially over these different indices. However, just as it is important to use more than one poverty line to see whether important results are sensitive to the choice of the poverty line, it is also important to check the sensitivity of results to different poverty indices. Regional Poverty Comparisons Targeting the design and placement of antipoverty programs is essential to reach disadvantaged groups and backward areas effectively and efficiently. In practice, one of the most important uses of the poverty profile is to support efforts to target development resources towards poorer areas, aiming to reduce aggregate poverty. Is poverty higher or lower in certain geographic regions? This question can only be answered at a highly aggregated level by the SESC because of the limited number of geographic domains which were sampled. The survey only supports an urban/rural comparison between Phnom Penh, Other Urban and Rural areas. While this provides a broad sense of the appropriate policy orientation in regional targeting, it is obviously of limited practical value for choosing the geographic placement of project interventions. 22 Box 1. Three Different Poverty Measures The Foster-Greer-Thorbecke (FGT) class of poverty measures is defined as follows: 1 N X where x = x; if x; < z =0 if X > 0 where xi is the expenditure level of the individual, z is the poverty line and N is the number of individuals in the population. In this general form the FGT poverty index appears rather abstract. The reason it is often used is that for particular values of a it gives the poverty indices discussed above. In particular, if a equals zero the FGT index becomes the headcount index, i.e. the fraction of the population whose expenditure levels fall below the poverty line.2 If a equals one the FGT index becomes the poverty gap index, i.e. the average gap between individual's incomes and the poverty line (where non- poor persons are assigned a gap of zero), divided by the poverty line. Finally, if a is greater than one the FGT index becomes distributionally sensitive in that greater inequality in the gaps among the poor leads to higher estimates of poverty, other things being equal. The regional poverty comparisons are presented in Table 8. Based on the overall poverty line, the incidence of poverty is found to be lowest by far in Phnom Penh, where only 11 percent of the individuals live below the poverty line. The second poorest group lives in Other Urban areas where the incidence of poverty rises steeply to 36%. The highest incidence of poverty is found in the rural areas where 43 percent are poor. Aggregating over the total SESC sample - which has a lower rural population weight of 78% than the true population distribution (85%) - gives a total poverty incidence of 39%. How sensitive are these regional poverty comparisons to the choice of poverty line? Using the food poverty line yields considerably lower levels of poverty, but it does not alter the policy conclusion that poverty is highest in rural areas and lowest in Phnom Penh. The head count index falls to 22 percent in the rural areas, 20% in Other Urban areas and only 6% in Phnom Penh. The sample-weighted aggregate incidence of food poverty becomes 20%. This is still a high figure -- it suggests that around one-quarter of the population in Cambodia is food-poor in the sense that they cannot meet their daily basic calorie requirement even if they were to devote all of their consumption to the basic food basket. 2Note that xo = 1 for any number x not equal to zero. However, xo = 0 if x equals zero. 23 Table 8: Distribution of Poverty by Region N Head count index Poverty gap Severity Index Food poverty (%) index contribution to index contribution to index, contribution to line (%) total (%) (%) total (%) (%) total (%) Phnom Penh 10.7 6.2 3.3 1.3 3.7 0.4 4.0 OtherUrban 11.0 19.6 10.8 4.4 13.1 1.4 14.8 Rural 78.2 21.9 85.9 4.0 83.2 1.1 81.2 Total /b 100.0 20.0 100.0 3.7 100.0 1.1 100.0 Poverty line Phnom Penh 10.7 11.4 3.1 3.1 3.6 1.2 4.1 OtherUrban 11.0 36.6 10.4 9.6 11.6 3.6 12.6 Rural 78.2 43.1 86.5 10.0 84.9 3.3 83.3 Total /b 100.0 39.0 100.0 9.2 100.0 3.1 100.0 /a N denotes the number of observations in the sample expressed in percentage weights; /b Note that the "Total" figure is representative for the sampled regions only and not for Cambodia as a whole. The quantitative sensitivity of the poverty comparisons is illustrated in Figure 7. The graphs give an idea of how sensitive the poverty measures are to changes in the poverty line. The vertical lines in the graph denote the poverty lines for each region. For the rural areas, the intersection between the poverty line and the distribution graph is on a rather steep section of the distribution. This indicates that small changes in the poverty line will yield relatively large changes in the head count index - the number of individuals below the poverty line. By contrast, in Phnom Penh the intersection is on a rather flat part of the distribution which indicates that the results will be rather robust with respect to changes in the poverty line. 24 1 0.8 o Other Urban . 0.6 Rural . 0.4 Phnom Penh 0.2 0 6 7 8 9 10 Log per capita daily consumption (Riels) Figure 7: SENSITIVITY TO CHANGES IN POVERTY LINES The qualitative robustness of the regional poverty comparisons can be assessed by the first-order dominance test of differences in the real per capita consumption distributions across regions. The real per capita consumption distributions expressed in Phnom Penh prices -- using the implicit spatial price deflators given by the relative food poverty lines -- are shown in Figure 8. The graph shows that the distribution of per capita consumption in Phnom Penh lies entirely below and to the right of the distributions for Other Urban and Rural areas. The Phnom Penh distribution curve does not intersect with the other curves so that first-order dominance of Phnom Penh holds unambiguously. In other words, Phnom Penh will emerge as the least-poor region wherever the real poverty line is set. However, the same is not true for poverty comparisons between Other Urban and Rural areas. In this comparison the consumption distribution lines do intersect in the poorest quintile so that first-order dominance does not hold over the entire distribution. Thus a very low poverty line would evaluate the incidence of poverty to be higher in Other Urban areas than Rural areas, while a higher poverty line would switch the results of the comparison. 25 0.8 Rural 0.6 Phnom Penh Other Urban 0.4 U 0.2 0 6 7 8 9 10 Log real per capita daily consumption Figure 8: DOMINANCE TEST OF ROBUSTNESS Poverty Comparisons by Employment The vast majority of households in any country are able to escape poverty, or cannot do so, because of their earnings from employment. Thus it is important to examine the relationship between poverty and the types of employment of working-age household members. The most important income-earner is usually the head of household. Figures 9 and 10 examine the prevalence of poverty in two different ways: according to the occupation of the head of household and the type of employer of the head of household (detailed tabulations are given in Annex 1). Looking first at the distribution by sector of 33 employment , the highest poverty rate, 46 percent, is found among farmers. By Sector of employment is grouped according to : "Manufacturing and mining" = Mining, quarrying, manufacturing; "Construction and utilities" = Electricity, gas, water supply, construction; "Hotels and restaurants" = Wholesale and retail trade, repair of motor vehicles, motorcycles and personal and household goods, hotels and restaurants; "Transport" = Transport, storage and communications; "Government" = Public administration, defense and compulsory social security; "Education Health" = Education, health and social work; "Other" = Financial intermediation, real estate, renting, business activities, private households with employed persons, extra territorial organizations and bodies. 26 contrast, households headed by someone working in public administration are least likely to be poor: in these occupations the poverty rate is only 20 percent. Clearly policies which aim at reducing poverty through enhancing income generating capabilities should be targeted towards the agricultural sector. Turning to the distribution of poverty according to employer4, those living in households where the head is self-employed is most likely to be poor (42 percent), followed closely by those working in the formal private sector (36 percent). The least likely to be poor are families where the household head works for the government. Only 21 percent of these households are poor --- far less than any other group in Cambodia. Thus policies directed at the formal labor market, such as an official minimum wage, will be highly ineffective in reducing poverty. Most people work outside of the regulated labor market. The only way to increase their incomes is by increasing their income generating capabilities. Another way of looking at the distribution of poverty is in terms of the contribution of different employment groups to national poverty in Cambodia, taking into account both the prevalence of poverty and the group's share of the national population. Looking at the contribution of different sectors of employment to national poverty, the strongest policy message for targeting purposes is that more than three-quarters of the poor are found among households in which the head has an agricultural occupation. This reflects both the high proportion of people living in agricultural households and their above average poverty rate. This means that policies to reduce poverty in Cambodia must reach agricultural households if any major reduction in poverty is to be achieved -- any policy that misses the farmers will bypass more than 75% of the poor. The policy messages for targeting poor households based on their employer are very similar. Households in which the head works for the formal public or private sector account for less than 10% of overall poverty. In other words, policies that focus on employee conditions will miss nearly 90% of the poor in Cambodia. 4 "Family worker" includes the following categories: "Employer in own family operated farm or business", "Worked with pay in own family operated farm or business" and "Worked without pay in own family operated farm or business". 27 45.9 37.0 35.0 29.4 21.9 I. 3 * ilmill Agriculture Manufactur- Construction Hotels and Transport Government Education Other Not reported and fishing ing/mining and utilities restaurants Health CONTRIBUTION TO TOTAL POVERTY (%) 73.8 8.0 2.1 1.4 4.4 3.9 3.3 2.3 0.8 mm --MWHl MI mm - Agriculture Manufactur- Construction Hotels and Transport Government Education Other Not reported and fishing ing/mining and utilities restaurants Health FIGURE 9: POVERTY RATE BY OCCUPATION OF HEAD OF HOUSEHOLD 28 57.3 42.0 36.2 35.0 20l5 Household worker Private sector Public sector Selfemeployed Fam ilyworker CONTRIBUTION TO TOTAL POVERTY (%) 81.1 6.2 8.0 2.5 2.2 Householdworker Private sector Public sector Selfemr ploved Famly-orker FIGURE 10: POVERTY RATE BY EMPLOYER OF HEAD OF HOUSEHOLD (%) Poverty Comparisons by Level of Education The relationship between poverty and education is particularly important because of the key role played by education in raising economic growth and reducing poverty. The better educated have higher incomes and thus are much less likely to be poor. As shown in Figure 11, Cambodians living in households with an uneducated head are more likely to be poor, with a poverty rate of 47 percent. But at higher levels of education, the likelihood of being poor becomes much lower. The prevalence of poverty among ambodia, the education system offers primary, seconday and undergratuate training. ..imary education provides 5 years of education, lower secondary 3years and higher secondary 3 years. 29 households in which the head has completed secondary education falls to around 30 percent. Raising educational attainment is clearly a high priority in order to improve living standards. 47.1 40.6 40.6 31.7 30.1 0.0 None Primary Lowersecondary Highersecondary Graduate Missing CONTRIBUTION TO TOTAL POVERTY (%) 44.4 28.1 18.5 0.0M None P reni re Lowersecondary Hghersecondary Graduate Missing FIGURE 11: POVERTY RATE BY LEVEL OF EDUCATION HEAD OF HOUSEHOLD (%) Recent trends demonstrate progress towards this goal. As shown in Figure 12, the education levels of the younger generations are clearly improving: average years of education in the 20-24 years age group have risen to 5.8 compared to only 3.8 among the 30-34 year age cohort. 30 5 4 3' 0 15-19 20-24 25-29 30-34 35-39 40-44 45-50 50+ age FIGURE 12: YEARS OF SCHOOLING BY AGE COHORT Poverty Comparisons by Gender The status of women, who in most developing countries are disadvantaged in comparison with men, is an important policy concern. One indicator of the gender gap is whether female-headed households are worse off than those headed by males. This might be expected to be a major concern in Cambodia since nearly 20 percent of the population live in households headed by women. In fact, the SESC data show that female-headed 6 households in Cambodia are less likely to be poor than male-headed households , and this holds regardless of the poverty line and poverty index used. Although people living in female-headed households account for nearly 23 percent of the population, they account for only about 15 percent of the poor. Put another way, using the benchmark poverty line about 40 percent of individuals in male-headed households are poor, but only 35 percent 6 Households headed by women include a wide variety of living arrangements, such as households where a woman is simply given as the reference person in the household, multigeneration households where the oldest person is given as the head, and households consisting of women and children only. It is the latter category that is especially vulnerable to poverty because of the low ratio of workers to dependents. Since female headship covers all categories, it is not a very sensitive indicator of vulnerability to poverty. Further research into the structure of female headed households is needed to identify which are most at risk. 31 of individuals in female-headed households have this characteristic. Overall, while women and girls in Cambodia may be disadvantaged in a variety of ways, it is not the case that female-headed households are generally more vulnerable to poverty than those headed by males; in fact the opposite seems to be the case. POVERTY RAIE BY SEX OF HEAD OF HOUSEHOLD (%) 39.8 34.6 mwe Jnale CONHIBUHON TD TOTAL POVERTY (%) 85.3 14.7 o M male Emale Figure 13: POVERTY RATE BY SEX OF HEAD OF HOUSEHOLD 33 6. INTERNATIONAL POVERTY COMPARISONS Is Cambodia poorer or better off than other East Asian countries in terms of the proportion of the total population in poverty? Precise comparisons between Cambodia and other countries are precluded by the fact that a large part of the country was excluded from the SESC sample frame. Nevertheless it is possible to compare the regional Cambodian estimates with corresponding estimates from other countries for which poverty estimates have been made using a comparable methodology for setting poverty lines. Comparable poverty estimates for Vietnam, Laos and Indonesia are summarized in Table 9. These comparisons suggest that the incidence of rural poverty in Cambodia (43 percent) is lower than among its Indochina comparators -- Vietnam (47 percent) and Laos (53 percent). But rural poverty in Cambodia remains considerably higher than elsewhere in East Asia, taking Indonesia as an example (24 percent). However, it must be borne in mind that rural poverty in Cambodia may have been underestimated by the exclusion of significant portions of the countryside from the SESC sample. On the other hand, urban poverty appears to be slightly higher in Cambodia (24 percent) than among its Indochina neighbors -- Vietnam (20 percent) or Laos (24 percent) - and much higher than Indonesia (10 percent). These international comparisons emphasize that the magnitude of the development gap which remains to be closed between the economies in Indochina and the rest of East Asia, while suggesting that Cambodia's starting point is not very different from its neighbors in Indochina. Table 9: International Poverty Comparisons Head count index (%) Poverty Gap (%) Severity index (%) Urban Rural Total Urban Rural Total Urban Rural Total Food poverty line Vietnam 9.9 28.2 24.5 2.0 6.2 5.4 0.6 2.1 1.8 Laos 7.6 26.0 21.6 1.0 5.5 4.4 0.2 2.0 1.6 Indonesia 2.8 10.7 7.9 0.31 1.26 0.97 0.06 0.23 0.18 Cambodia 12.9 21.9 20.0 2.9 4.0 3.7 0.93 1.1 1.1 Poverty line Vietnam 19.6 46.5 41.2 4.5 12.4 10.8 1.6 4.7 4.1 Laos 23.9 53.0 46.1 4.5 14.4 12.1 1.2 5.6 4.6 Indonesia 10.2 23.6 19.6 1.67 4.25 3.5 0.40 1.08 0.87 Cambodia 24.2 43.1 39.0 6.4 10.0 9.2 2.4 3.3 3.1 Source: World Bank Documents. Vietnam estimate based on 1992-1993 household survey. Laos estimate based on 1992-1993 survey. Indonesia estimate based on 1990 Susenas household survey. 35 7. COMPARISON WITH OFFICIAL POVERTY ESTIMATES The First SocioEconomic Development Plan reports official estimates of poverty in Cambodia. These estimates were constructed using summary tabulations of the household expenditure distribution from the SESC 1993/94. This report provides a revised version of these estimates by using the detailed individual records from the full SESC dataset. This section compares the revised estimates with those presented in the Plan and explains why they are different. The main differences between the revised and the Plan estimates are highlighted in Figures 14 and 15. As shown in Figure 14, the revised poverty line for Phnom Penh is 25% lower, while the revised poverty lines are similar in both the Other Urban and Rural areas. The resulting poverty rates (head count indices) are given in Figure 15. These show that the revised poverty rate for Phnom Penh is lower, while the revised poverty estimates for Other Urban and Rural areas are much higher. The nature and implications of these two differences-- in the poverty lines and in the ranking method -- are explained below: 2500 2000 . 1500 5 000L 500- Preliminary Phnom Penh Preliminary Other Urban Preliminary Rural Revised Phnom Penh Revised Other Urban Revised Rural m Food poverty line o Non-food component Figure 14: COMPARISON OF POVERTY LINES 36 50 43 40 37 32 o 30 27 119 0 10 L 0 .._.._-~ _-- Phnom Penh Other Urban Rural g Revised estimates 0] Preliminary estimates Figure 15: COMPARISON OF POVERTY RATES Differences in Poverty Lines The food poverty lines are based on a similar methodology, using a basket of food items valued at local prices. The main technical differences lie in how the methodology is implemented in constructing the line: * benchmark daily calorie requirement - the Plan estimates used a higher calorie benchmark of 2,200 per day, * composition of the reference food bundle - the bundle used in the Plan estimates consisted of only 7 products including rice, vegetables, pork, fish and oil, with relative quantities based on expenditure patterns in the third quintile of the household (not individual) consumption distribution; * pricing the reference food bundle -- the starting point in the Plan estimates are the prices prevailing in Phnom Penh. The valuation of food bundles for Other Urban and Rural areas is done by assuming that food prices in Other Urban are 70 percent of those prevailing in Phnom Penh, while food prices in Rural areas are 65 percent. However, the revised estimates are based on the observed unit values of cash purchases in each region. For Phnom Penh the revised price estimates are very close to the Plan estimates, while prices in Other Urban and Rural areas are much higher. Overall the observed price differential between Phnom Penh and other areas is much less than was assumed -- in 37 the revised estimates, the spatial food price deflators -- with Phnom Penh at 100 -- are 84 in Other Urban and 74 in Rural areas. The methods used in constructing the non-food allowance were quite different. The Plan estimates were based on a basket of non-food items. The basket contained an allowance for firewood, medical expenses, education, clothing, transport and housing. The alowances for Other Urban and Rural were set to a lower level copared to Phnom Penh. The revised estimates were derived from the observed non-food consumption of those households at the food poverty line. In general, the Plan estimates yield much higher values for the non-food allowance. The difference is greatest in Phnom Penh, which was assigned a daily non-food allowance of 477 Riels per day higher than in the revised estimates. Differences in Ranking Method Given that the poverty lines are not very different for the Other Urban and Rural areas it seems surprising that the headcount indices for these regions turn out to be much higher in the revised estimates. This difference is simply due to the fact that the Plan estimates are derived from summary tabulations in which households were ranked on the basis of total household consumption -- instead of the distribution of per capita consumption over the individual records of household members as was used in the revised estimates. As a result, the poorest deciles of the household consumption distribution contained a relatively large number of small households with few income earners. Although these households appear to be relatively poor if judged on the basis of their total household consumption, this is not necessarily so when the distribution of household members is sorted in ascending order of per capita consumption. In fact, experience from other countries shows that usually the households which are ranked as poor in per capita terms tend to be relatively large. The Plan estimates did attempt to take this into account by dividing the means of total household consumption by the average family size per decile. However, this did not change the underlying ranking of households in the distribution across deciles. Decomposition of Differences relative effects of these two differences on the headcount index are decomposed in Table 10. This decomposition is shown in two stages. The first-stage decomposition takes the official poverty estimates -- using the Plan poverty line and the household ranking method - as the benchmark for comparison. The first-stage adjustment then switches the ranking method from a household basis to a per capita basis while retaining the Plan poverty line. This shows that the large difference in the poverty estimates for Other Urban and Rural areas is almost completely driven by the difference in the ranking method. Switching to per capita rankings with the Plan poverty line would raise poverty by 10% in Other Urban areas (to 37%) and by 12% in Rural areas (to 44%). In this case using household level data turned out to be an unfortunate approximation. In Phnom 38 Penh, however, the error made by using the summary tabulations at the household level is not that large since household size exhibits little variation in this city. Thus, the first- stage adjustment would have raised poverty incidence in Pnom Penh by only 3% (to 22%). The second-stage decomposition changes the benchmark to the poverty estimates generated using the Plan poverty lines combined with the correct per capita ranking method. The second-stage adjustment then switches from the Plan to the revised poverty lines in order to identify the poverty line effect. This shows that the lower poverty estimate for Phnom Penh is primarily due to differences in the poverty line. The estimated incidence of poverty in Phnom Penh falls in half, to only 11%. Meanwhile, there is no further change in the poverty estimates for Other Urban areas (at 37%), and only a 1 percent change in Rural areas (to 43%). Table 10: Decomposition of Difference in Poverty Estimates LINE/RANKING COMBINATIONS Poverty Line Ranking Headcount method index Phnom Penh Other Urban Rural Revised poverty line Per capita 11 37 43 Difference due to poverty lines /b -11 0 -1 Plan poverty line Per capita 22 37 44 Difference due to ranking method /a +3 +10 +12 Plan poverty line Household 19 27 32 /a Using plan line/household ranking as the benchmark for comparison; /b Using plan line/per capita ranking as the benchmark for comparison; 39 8. CHARACTERISTICS OF THE POOR While the consumption-based measures of poverty are a convenient yardstick for measuring the distribution of living standards in the Cambodian population, they do not fully capture other characteristics of the poor such as literacy, health, or access to clean water. This section gives a brief overview of the distribution of selected non-monetary indicators of household living standards, using data collected by the SESC (see Table 11). Household Composition. Household composition, in terms of the size of the household and the characteristics of its members, is often quite different for poor and non-poor households. Table 11 shows household size and the age of the family members by expenditure quintiles. The poor do tend to live in larger households, with an average family size of 6.6 persons in the poorest quintile compared to 4.9 in the richest quintile. The poor also tend to live in younger households -- with twice as many children under age 15 per family (3.4) in the bottom than in the top quintile -- and slightly fewer elderly people over age 60. Better-off households tend to have heads that are somewhat older, but the difference across quintiles is very small. Literacy and Schooling. Literacy and schooling are important indicators of the quality of life in its own right, as well as being the key determinant of the poor's ability to take advantage of income-earning opportunities. Cambodia has achieved a (self-reported) basic literacy rate averaging 67 percent of adults older than 15, implying a high degree of literacy among the poor. The literacy gap which remains is quite large, with literacy ranging from just over half of adults (58 percent) among the poorest 20 percent of the population to 77% percent among the richest. Much larger differentials appear in the distribution of school attainment. Years of schooling among adults aged over 15 average only 3.1 years in the bottom 20 percent of the population, increasing to 5.3 years of schooling among the richest 20 percent. Here there is a very large gender gap, with mean grade attainment among men of 5.1 years compared to 3.2 years among women. Housing Conditions and Assets. Housing conditions are another important element among different aspects of social well-being. Water and sanitation are especially important influences on health and nutrition status. The SESC shows that the poor are extremely disadvantaged in access to safe sources of water supply and sanitation. Only 4 percent of the poorest quintile have access to piped water, while more than 17% of the richest quintile do. Similar differences are apparent in access to sanitation. Few of the poor - 9% -- have access to a toilet in the home, while around half of the richest 20 percent do. Another indicator of housing standards is access to electricity. Here again the access of the poor lags far behind. Access to electricity from a generator or line connection --the most convenient energy source -- rises sharply with income, from a mere 40 1% among people in the bottom quintile to 37% of Cambodians in the richest quintile. Table 13 above also indicates the percentage of households that possess bicycles and motorized transport. Access to bicycles is quite evenly distributed with at least one half of households owning a bicycle in every quintile, even the poorest. However access to cars, jeeps or motorbikes is very rare among the poor and rises sharply with income. Overall, the shift from bicycles to motorized transport is a strong indicator of better off families with access to a wider variety of services and amenities. Table 11: Decomposition of Differences in Poverty Estimates Quintile /a Domain Total Poorest 2 3 4 Richest Phnom Other Rural Penh Urban Household size 5.6 6.6 6.0 5.7 5.0 4.9 5.9 5.9 5.5 Children per family 2.4 3.4 2.8 2.5 1.9 1.8 2.3 2.6 2.4 (age 0-14) Elderly per family 0.3 0.3 0.3 0.3 0.4 0.4 0.3 0.3 0.3 (age 60+) Dependency disabled 0.84 0.78 0.93 0.93 0.91 0.68 /b Age head of 44.6 44.1 43.0 44.1 45.2 46.1 45.0 45.3 44.5 household Female head of 21.2 18.6 17.3 18.9 23.3 26.6 25.8 23.4 20.4 household (%) Literacy 66.6 57.7 64.3 66.2 67.9 77.1 81.8 72.5 63.7 (% adult aged 15+) Years of education 4.0 3.1 3.7 3.8 4.2 5.3 6.1 4.7 3.7 (avg. adult aged 15+) Years of education /c 5.1 4.1 4.7 4.8 5.4 6.6 7.3 5.8 4.7 (avg. male aged 15+) Years of education /c 3.2 2.4 2.9 3.0 3.3 4.3 4.9 3.8 2.8 (avg. male aged 15+) Access to piped water 7.1 4.3 3.4 3.5 5.5 16.7 33.6 13.4 2.9 (%) Toiletinhouse(%) 22.1 8.7 11.3 13.9 21.1 48.7 78.0 46.7 11.7 Electricity from line 12.4 1.3 2.1 5.5 10.4 36.8 67.4 30.2 3.0 or generator(%) Radio(%) 27.7 23.0 25.1 25.8 29.8 33.0 38.7 31.4 25.9 TV (%) 13.8 3.0 4.8 7.0 12.7 35.9 57.0 21.6 7.2 Bicycle(%) 60.9 57.7 65.3 63.6 61.6 56.8 41.1 60.9 63.5 Car, jeep or motor- 18.0 6.4 6.7 11.7 18.8 40.5 58.8 27.1 11.6 cycle (%) /a Quintile distribution based on real per capita expenditure using the implicit food poverty line price deflators; /b Dependency defined as number of disabled in family divided by number of family members times 100. /c Averages over all individuals aged 15 and above. All other variables are averages across households. 41 9. IMPROVING POVERTY ANALYSIS AND POLICY Adequacy of the Information Base Poverty reduction is a central goal of the Royal Government of Cambodia. The policy relevance of the poverty profile for Cambodia based on SESC 1993/94 illustrates the importance of adequate data from large-scale household surveys in analyzing poverty and designing appropriate policies to translate the Government's commitment to policy reduction into action. These data are essential to help government policymakers to: * identify the regional location, employment, gender and other characteristics of the poor in order to support design of more efficiently targeted poverty alleviation programs; * monitor changes in poverty over time so as to assess the pace and pattern of development progress; * analyze the levels and determinants of access to economic infrastructure and support services, as well as to access to social services and safety nets - many of which are publicly provided or subsidised -- as a guide to improving the effectiveness of policy instruments intended to enhance the welfare of the poor. The Government has already taken important ad hoc steps towards developing a capacity to collect and analyze poverty-related survey data on a regular basis. The first Socioeconomic Survey of Cambodia 1993/94, which was originally designed to establish the weights for a new consumer price index, has demonstrated the capacity to implement a large-scale income and expenditure survey on a sample of around 5,000 households in 16 provinces. The ongoing second Socioeconomic Survey of Cambodia 1996 has adopted a completely different questionnaire design focusing on welfare indicators in selected areas such as child labor, nutrition status and coverage of child health programs, housing characteristics, land tenancy and credit behavior. This second survey was designed to accommodate the data needs expressed by users in a variety of sectors for which funding could be mobilised. It will be carried out on a larger sample of 9,000 households, with oversampling to give province-level estimates in 8 provinces. These initial efforts have made an important contribution to capacity building at the National Institute of Statistics and it is important that they should be sustained. Three significant limitations will need to be overcome in future development of the information base on living standards in Cambodia. The first is that neither of the SESC questionnaires implemented to date have been explicitly designed as a fully integrated multipurpose survey of the type needed to conduct a comprehensive poverty assessment for policymakers. For example, while the expenditure data collected by SESC 1993/94 42 generate a detailed picture of the distribution of per capita consumption, it is not possible to analyse levels or determinants of access to social services among the poor because the survey did not simultaneously collect data on education enrollments or utilisation of health services, nor did it collect data on the price of obtaining access to these services. The SESC 1996 goes to the other extreme. While extensive data are collected on nonmonetary indicators such as child nutrition or immunisation coverage, only limited information will be collected on household consumption expenditure so that interrelations with poverty status may not be clearly identified. Second, neither of the SESC surveys are able to support nationwide geographic disaggregations of key variables such as per capita expenditure and poverty incidence at the provincial level. In practice, this imposes an important limitation on its potential use in project planning. The need to overcome similar limitations in other countries have led to the adoption of much larger sample sizes. For example, the annual core household survey in Indonesia is now administered to over 200,000 households. Similarly, Vietnam now uses a sample of 45,000 households for the annual multipurpose survey. These enhancements are intended to provide much finer identification of the geographic location of poverty problems, support the design of more efficiently targeted poverty alleviation programs and strengthen capacity for decentralized planning at the provincial level. Third, the SESC surveys have not yet been institutionalized as part of a systematic, long-term and regular effort to evaluate the extent and nature of the poverty problem in Cambodia, to monitor progress in poverty reduction over time, and to evaluate the effectiveness of specific targeted antipoverty interventions. This means there is a risk that the potential benefits of these surveys will be short-lived. If the investments which have been made in creating new skills in survey design, field procedures, data processing methods, policy analysis and program design are not maintained continuously then future surveys will become much more difficult to implement. The importance of consolidating and sustaining these nascent efforts to strengthen the information base for policymaking on poverty is increasingly recognized in neighbouring countries. For example, in the early 1990s Indonesia initiated a major collaborative effort between the planning and statistics agencies to redesign its national household survey system (SUSENAS) to provide better data to guide the country's poverty alleviation programs. China is planning to set up a Poverty Monitoring and Evaluation System with the same objective. Recently Vietnam introduced an annual series of large-scale multipurpose household surveys of living standards. InstitutionalisingPoverty Analysis and Policy The agenda for institutionalizing poverty analysis and policy in Cambodia embraces four key elements. First, the government should consider adopting routine implementation of a new national household survey which offers both multipurpose 43 coverage and geographic disaggregation. This calls for a two-part "core/module" household survey design. The purposes of the core are to support monitoring of changes in key indicators over time, and identification of priority areas for geographic targeting of development programs. To serve these needs, the contents of the core questionnaire could be fixed on a small number of key welfare indicators, e.g. per capita consumption, education enrollment, health care utilization rates, and the questionnaire would be implemented relatively frequently -- probably every year -- on a large sample so as to give estimates of the indicators disaggregated at least to provincial level. At the same time the core could be supplemented every year with a rotating sector module. The purpose of the modules would be to support in-depth analysis of sectoral issues and policies, such as the effects on the poor of changing pricing policies in the social sectors. Given the focus of the modules on analysis rather than monitoring, individual sector modules need not be carried out every year or on a large sample. Instead they could rotate over a 3 year cycle on a subsample of the core - for example, a social sector module, followed by an income and employment module, and then a detailed consumption module. The core/module survey of households needs to be linked to a community survey conducted at the village level. The purpose of the community survey is to collect data on variables which affect all households in the community, such as public/private provision of economic infrastructure (e.g. land, irrigation, agricultural extension, roads and markets) and social services (e.g. availability and quality of schools and health services). Thus the community survey plays an essential role in analysing determinants of household behavior and welfare based on merged datasets using the community and household data. Ultimately the community survey can also play a role in poverty monitoring for targeting purposes. This requires implementation of the community survey on a census basis. For example, in 1993 the government of Indonesia prepared a nationwide poverty map identifying poor villages based on data collected in the community survey (Potensi Desa or PODES) which is administered to all villages every three years. This poverty map has become the operational basis for targeting a major poverty alleviation program, comprising decentralised grants to poor villages. In addition, the poverty map has focused geographic targeting of many other government programs in different sectors. A second key element of the future agenda is to improve the institutional linkages between the National Institute of Statistics as the technical agency responsible for the quality of data production, and its client policymakers in the Ministry of Planning and the relevant line agencies. In order to make effective use of the household survey database, the clients need to perceive it to be useful by contributing to design of the content of the surveys so that they are responsive to key policy questions. 44 A third factor which conditions the strength of institutional linkages is timeliness of turnaround from the surveys. Improving the speed of data processing, availability and dissemination to users in the government would probably require significant enhancement in PC-based computing capacity at the National Institute of Statistics. Finally, for Cambodia to benefit fully from these improvements in the design, regularity and turnaround in the information base on living standards will require continuing improvements in the analytical skills of policymakers and researchers in Cambodia through a combination of training and hands-on experience. Meeting this objective demands a new effort to train staff in methods of applied policy analysis in government agencies. An Integrated Household Survey System Development of an integrated household survey system using a core/module design would need to be sustained over a multiyear period. The full cycle would last at least three years to allow for the necessary expansion in sample size, refinements of survey design, and capacity building in operational procedures. At the end of this period the National Institute of Statistics would have implemented all the components of a filly integrated multipurpose household survey system, and should be prepared to maintain it on a routine basis. At this early stage of development of the National Institute of Statistics, scheduling of the proposed household survey series will need to take into consideration other operational commitments. Preparatory work on the population Census has already started with a view to field implementation in March 1998, before the General Elections which are scheduled to take place in May 1998. Surveys undertaken in 1998 and 1999 will benefit from the improved household listing frame developed for the 1998 Census. A feasible implementation schedule might be as follows: * First Survey, 1997 (July): * design: stratified sample in one round; * sample size: around 5,000 households to provide regional and urban/rural estimates for core variables; * scope: core questionnaire plus "social sector" module plus village survey administeredto the module subsample; * Second Survey, 1998 (after May): * design: stratified sample in one round; 45 * sample size: increased >6,000 households to allow selected provincial estimates of core variables; * scope: core questionnaire plus "income and employmenf module plus village survey in core sample; * Third Survey, 1999: * design: stratified sample in one round; * sample size: increased to 15,000 households to allow complete provincial estimates of core variables; scope: core questionnaire plus "consumption" module plus expanded village survey administered on a census basis. A Poverty Analysis and Policy Unit Ultimately the operational value of improving the poverty related information base provided by the integrated household survey system will depend on the capacity to analyse the data and interpret its practical implications for policymakers. This capacity needs to be located within the Ministry of Planning because of the complex cross-sectoral agenda involved in formulation of antipoverty policies, and the need to interact with policymakers involved in setting strategic priorities between and within different sectors. Several other countries have adopted this approach. For example, in Indonesia the World Bank is currently financing a Social Sector Capacity Building Project implemented by the planning ministry (BAPPENAS). The project establishes a technical advisory capacity managed by BAPPENAS and linked to the line ministries. The scope of work ranges from informal policy notes to research studies linking government policies to household behavior and welfare. The necessary technical skills place a strong emphasis on public finance, microeconomics and econometrics. Donor Coordination International donors are committed to supporting the poverty reduction goal set forth in the First Socioeconomic Development Plan, and all share a common interest with the government in strengthening the information base on living standards in Cambodia so as to improve the policy dialogue on sectoral priorities and project design. Accordingly, the donors are working together in mobilising the external financing required to implement the joint work program on data collection and policy analysis which will be necessary to underpin the government's policy commitment to fight poverty. UNDP has taken the lead in coordinating this effort in the framework of a new project on Capacity Development for Socio-Economic Surveys and Planning.(CMB/96/019/A/0 1/42). 47 REFERENCES Foster, James, J. Greer and E. Thorbecke, (1984),"A Class of DecomposablePoverty Measures", Econometrica Vol 52, pp 761-765. Psacharopoulos, G. (1985), "Returns to Education: A Further International Update and Implicationg', The Journal of Human Resources Vol 20, no 4, pp 583-604. Ravallion, Martin (1992) "Poverty Comparisons: A Guide to Concepts and Methods", LSMS Working PaperNo. 88, World Bank, WashingtonDC. Ravallion, Martin and Benu Bidani, (1994), "How Robust is a Poverty Profile?', The World Bank Economic Review, Volume 8, Number 1, pp. 75-102. Royal Government of Cambodia (1995), First Socioeconomic Development Plan, 1996- 2000, Phnom Penh. , (1995) Report on the Socio-EconomicSurvey of Cambodia 1993/94, Ministry of Planning / National Institute of Statistics, Phnom Penh. World Bank (1993), Indonesia, Public Expenditures, Prices and the Poor, Report No. 11293-IND Washington DC.. (1995), LAO PDR, Social DevelopmentAssessment and Strategy, Report No. 13992-LA Washington DC.. (1995), Vietnam: Poverty Assessment and Strategy, Report No. 13442-VN Washington DC. World Health Organization (1985), Energy and Protein Requirements; Technical Report Series No. 724, Geneva. t 49 ANNEX A CIVIL SERVANTS, POVERTY AND EARNINGS The poverty comparisons between different employer categories suggest that people living in households in which the household head works for the government are the least likely to be poor in Cambodia (see Section E, para 37). This finding reflects the distribution of per capita expenditure between government and other households. For example, per capita consumption in government households averages 56% higher than in private sector households (see Table 1). It is important to note that this differential is also seen on the income side: on average per capita income is 25% higher in government than private sector households. Note that self-employed households are excluded from these comparisons because individual earnings data are not available for these households. Table A.1: Comparisons Of Public/Private Sector Workers Private Sector Wage Worker Public Wage Worker all without with all without with second second second second job job job job Per capita household 2,134 2,649 1,354 3,247 3,634 2,601 consumption (in Riels/day) Wage in primary job of head 4,265 5,299 2,666 2,130 2,523 1,479 of household (in Riels/day) Per capita household income 1,476 1,780 1,014 1,839 2,135 1,347 from all sources Number of workers in 1.95 1.95 1.94 2.15 2.17 2.12 household (including head of household) Household size 5.8 6.0 5.5 6.3 6.2 6.4 This Annex seeks to explain why civil servants tend to be better off in terms of per capita consumption than other employment groups in Cambodia. The explanation is sought in understanding the structure of employment and earnings as reflected on the income side of the data collected by SESC 1993/94. Note, however, that the ability of the income data used in these comparisons to explain the poverty differentials implied by variations in consumption expenditure is quite limited. 50 The limitations of the SESC data are evident in the simple fact that reported household incomes are considerably lower than the levels of consumption expenditure reported in the survey. Thus, for people living in households headed by a government worker, reported income per capita is equivalent to only 57% of consumption per capita. Among 'private sector' households per capita income is only 69% of reported consumption (see Table 1). Obviously, the income data can only be used to explain the variations which are actually observed in the reported incomes. Since the variance in incomes apparently accounts for only part of the observed variation in consumption, any insights gained from analysing the reported income data will necessarily be incomplete. Sources of Income: Monetary and Nonmonetary In order to understand why income differentials are observed between public and private sector households it is useful to begin by identifying the main components of income which are actually responsible for this difference in aggregated income levels (see Table 2). Overall, the per capita income differential is Riels 363 per day: the difference between Riels 1,839 per day in government and Riels 1,476 in private sector households. Disaggregation by the main sources of income shows that this differential originates entirely on the nonmonetga income side: per capita income from the group of nonmonetary income sources is Riels 435 per day higher among government workers. What, then, are the sources of this difference in nonmonetary income? Further disaggregation shows that nearly all of the difference can be attributed to nonrental income. Nonrental income averages Riels 313 per day more among government workers. This alone accounts for over 85% of the overall difference in per capita income. Although the value of income generated by monetary sources is slightly lower among public sector workers, it is worth noting that actual wage income seems to be much lower - around one half as much as earned by private sector workers. However this adverse wage differential is substantially offset by relatively high incomes from business and other cash sources. Other Factors Moonlighting. Given that average daily earnings from the main job are only 50% as high in government employment than in the private sector (Riels 2,130 per day compared to Riels 4,265 per day), one might expect that civil servants would moonlight in order to compensate with earnings from secondary jobs. Surprisingly, however, the data do not give any support to the moonlighting hypothesis. First, sizeable differentials in per capita income are observed between government and private sector households whether or not the head has a secondary job (see Table 1). And second, the probability of holding a second job is the same-28%--in both the public and private sectors. In other words, multiple jobbing among 51 Table A.2: Structure Of Income By Source And Occupation (in Riels per capita per day) Private sector Public Sector All sources 1,476 1,839 Monetary 1,141 1,068 Wages 833 446 Pensions 3 3 Other cash incomes 20 138 Business 161 412 Agricultural 124 69 Non-monetary 335 770 Agricultural 21 67 Rental 240 553 In kind 38 22 non agricultural 37 129 household heads who work for the government does not appear to be a significant factor raising their per capita income and consumption. Educational Attainment. Another possible explanation of the observed income differentials lies in average educational attainment of households where the head works for government or the private sector. In contrast to the pattern of earnings differentials, however, the average years of education turns out to be significantly higher among working age members of government households, whether they are household heads or not (see Table 3). Table A.3: Educational Attainment In Public/Private Sector Households (average years of education) Government worker Private sector worker Household head 7.95 5.55 Other (non-head) adults 5.3 4.0 (15+) Other adults who are 5.2 4.3 working Other adults who are not 5.9 4.0 working Percent of heads with a 28 28 second job 52 The positive role of education in explaining per capita income differentials across households is reflected in the Mincerian earnings function shown in Table 4. This does show the expected positive coefficient on years of education, as well as the higher earnings levels associated with residence in urban areas. However the favorable effect of education is clearly counteracted by the negative effect of being a government worker. Table A.4: Log Earnings Regression Coefficient T-statistic Constant 6.690 40.83 Female -0.189 -4.86 Age 0.056 5.96 Age squared -0.068 -5.56 Years of education 0.019 3.91 /100 Government worker -0.730 -18.32 Phnom Penh 0.342 9.14 Other Urban 0.229 5.45 R squared 0.17 53 ANNEX B ANALYSIS OF GENDER-POVERTY LINKAGES The living standards of female-headed households are often used as an indicator of the status of women, and female-headship is commonly believed to be a useful targeting indicator for antipoverty programs. Female headship is relatively common in Cambodia compared to other developing countries -- female-headed households account for 21.3% of total households included in the SESC sample. In the great majority of these households -accounting for 75% of all female-headed households, or 16.1% of total households-the female head is unmarried and belongs to the oldest generation living in the household (see Table 1). Table B.1: Characteristics Of Family Headship (individuals aged > 18 years) Cell description Cell characteristics Contribution to total heads (%) sex generation married % share of family % head of /a total size household 1 no 1.2 5.6 58.0 2.0 1 yes 25.4 5.8 94.9 68.6 2 no 8.1 6.7 0.7 0.2 male 2 yes 5.3 7.2 53.6 8.0 3 no 2.7 7.5 0.0 0.0 3 yes 2.1 7.7 0.0 0.0 1 no 8.3 4.9 68.3 16.1 1 yes 25.6 5.8 3.8 2.8 2 no 10.2 6.4 6.9 2.0 female 2 yes 6.5 7.2 2.0 0.4 3 no 3.5 7.2 0.0 0.0 3 1.1 7.8 0.0 0.0 Total 100.0 100.0 /a The definition of generation is based upon the family relations prevailing in the household. The first generation is defined as the oldest generation living in the households. Individuals are assigned to the second generation if a previous generation-- their parents--is present in the household. Third generation members exist if the grandparents of children live in the households --in that case the children would be classified as third generation. The third generation is also used as an 'other' category -- it includes servants and non-family related household members. 54 As shown in the main text (section E, para 40) the evidence from SESC 1993/94 does not support the policy inference that female-headed households are a hish-poverty target group in Cambodia. On average, female-headed households are less likely to be poor than male-headed households. This finding reflects the fact that per capita consumption expenditure is generally about 19% higher among female-headed households: Riels 2,075 per day compared to Riels 1,748 per day (see Table B.2). This finding is only reversed in the special case where female heads are the only person working in the household; however, this constitutes only a small minority--12.6%--of all female-headed households (which, in turn, account for only 21.3% of al households). Table B.2: Per Capita Consumption And Household Headship (per capita household consumption expenditure in Riels per day) Characteristics of Male Female Household Head mean % share mean % share Not working 1,865 6.8 2,094 21.3 Only worker 2,153 0.3 1,793 12.6 With other 1,776 92.9 2,123 66.1 workers Total 1,748 100.0 2,075 100.0 In interpreting the policy significance of the female headship data, it is very important to recognise that female headship is an empirically narrow, and therefore potentially misleading, index of the living standards of women in general. The reason is simply that the great majority of Cambodian women-79%--do not live in live in female-headed households (see Table B.3). Thus focusing on female-headed households alone simply ignores the welfare of most women in Cambodia. Table B.3: Distribution Of Females By Family Headship (in percent) Females Headship Young Working Total age Male 39 40 79 Female 8 13 21 Total 47 53 100 For this reason, an adequate assessment of gender-poverty linkages requires going beyond the narrow concept of female headship to a broader index which reflects the welfare of the majority of females living in Cambodia. Using per capita expenditure as the basic indicator of living standards, the obvious step to begin with is to examine the correlation between per capita consumption standards and the "femaleness" of the 55 households in which people live. Here, "femaleness" is measured by the percentage of working-age (15-60 years) household members who are women. Simple cross-tabulations of the grouped data do not show any relationship between the index of femaleness and consumption standards (see Table B.4). Very wide variations in the level of per capita expenditure across quintiles coexist with virtually no variation in the femaleness of household workership. The femaleness of the household labor force averages 55% in the whole population - and hardly varies from 56% in the poorest quintile to 54% among the richest 20% of the population. Table B.4: Distribution Of Femaleness By Expenditure Quintile (females as % of all working age household members) Consumption quintile Index of Femaleness 1-Poor 55.8 2 56.4 3 56.3 4 55.0 5 -Rich 53.5 Total 55.4 Multivariate regression analysis provides a more robust test of the hypothesis that higher femaleness of the household labor force is associated with lower per capita expenditure, while controlling for the influence of other relevant household characteristics such as location of residence, the average level of education of working members of the household, and the dependency ratio. The results of the loglinear specification of this relationship are given in Table B.5. Table B.5: Determinants Of Household Welfare (Dependent variable: log per capita consumption) Coefficient T- statistic Constant 7.17 214.98 Dependency ratio -0.64 -17.88 Femaleness 0.15 3.97 Phnom Penh 0.86 45.65 Other Urban 0.23 11.63 Years of education 0.06 21.07 (average over all adults in family) R squared 0.47 As expected, these findings indicate that higher dependency ratios reduce living standards, while urban location (particularly Phnom Penh) and years of education exert 56 strong positive effects on consumption standards. It is noteworthy that these powerful results reject the hypothesis of a negative effect of femaleness--instead, greater femaleness has a positive and statistically significant on per capita consumption. The femaleness elasticity of per capita expenditure is 0.15-thus, a 10% higher female proportion of the household work force is associated with a 1.5% higher per capita consumption standard. These findings show that the indicators which matter in targeting households with lower living standards are rural residence and low educational achievement regardless of sex, not femaleness per se. 57 ANNEX C POVERTY TABLES Table C.1: Distribution of Poverty By Sector of Employment of Head of Household Head Count Index Poverty Gap Severity Index Sector distr. index contribution index contribution index contribution (%) (%) to total (%) (%) to total (%) (%) to total (%) Agricultureand 62.7 45.9 73.8 10.6 72.0 3.5 70.3 fishing Hotels and 7.1 24.3 4.4 6.8 5.2 2.7 6.1 restaurants Government 6.4 20.1 3.3 4.8 3.3 1.5 3.1 Transport 4.7 31.6 3.9 7.2 3.7 2.3 3.6 Education 3.9 22.7 2.3 4.8 2.0 1.7 2.1 Health Manufacturing 3.7 21.9 2.1 4.7 1.9 1.7 2.0 and mining Construction 1.5 37.0 1.4 9.4 1.5 3.2 1.5 and utilities Other 1.1 29.4 0.8 8.1 1.0 3.1 1.1 Not reported 8.9 35.0 8.0 9.6 9.3 3.5 10.2 Total 100 39.0 100 9.2 100 3.1 100 Table C.2: Distribution of Poverty By Employer Head of Household Head count index Poverty Gap Severity index Employer distr. index contribution index contribution index contribution (%) (%) to total (%) (%) to total (%) (%) to total (%) Self employed 75.2 42.0 81.1 9.7 79.5 3.2 78.3 Public sector 11.8 20.5 6.2 4.5 5.8 1.4 5.4 Family worker 8.9 35.0 8.0 9.6 9.3 3.5 10.2 Private sector 2.4 36.2 2.2 10.2 2.7 4.2 3.3 Household 1.7 57.3 2.5 15.2 2.8 5.3 2.9 worker Total 100 39.0 100 9.2 100 3.1 100 58 Table C.3: Distribution of Poverty By Education Head of Household Head count index Poverty Gap Severity index Education distr. index contribution index contribution index contribution (%) (%) to total (%) (%) to total (%) (%) to total (%) None 23.2 47.1 28.1 12.3 31.1 4.5 33.4 Primary 42.7 40.6 44.4 9.6 44.4 3.2 44.0 lower secondary 22.7 31.7 18.5 6.8 16.7 2.1 15.6 higher 8.0 30.1 6.1 6.1 5.3 1.9 4.7 secondary graduate 0.7 0.0 0.0 0.0 0.0 0.0 0.0 Not reported / 2.8 40.6 2.9 8.3 2.5 2.5 2.3 Unknown Total 100 39.0 100 9.2 100 3.1 100 Table C.4: Distribution of Poverty By Literacy Head of Household Head count index Poverty Gap Severity index Literacy distr. index contribution index contribution index contribution (%) (%) to total (%) (%) to total (%) (%) to total (%) Can read & write 72.3 36.4 67.5 8.2 64.5 2.7 62.0 Cannotread& 27.7 45.7 32.5 11.8 35.5 4.3 38.0 write Total 100 39.0 100 9.2 100 3.1 100 Table C.5: Distribution of Poverty By Sex Of Head of Household Head count index Poverty Gap Severity index Sex distr. index contribution index contribution index contribution (%) (%) to total (%) (%) to total (%) (%) to total (%) Male 83.5 39.8 85.3 9.4 84.9 3.1 84.6 Female 16.5 34.6 14.7 8.4 15.1 2.9 15.4 Total 100 39.0 100 9.2 100 3.1 100 59 Table C.6: Distribution of Poverty by Family Size Head count index Poverty Gap Severity index Household size distr. index contribution index contribution index contribution (%) (%) to total (%) (%) to total (%) (%) to total (%) 1 0.2 1.0 0.0 0.1 0.0 0.0 0.0 2 2.3 11.1 0.7 2.2 0.5 0.6 0.5 3 6.3 18.5 3.0 3.3 2.2 0.9 1.9 4 11.0 26.5 7.5 5.0 6.0 1.4 5.0 5 14.5 34.2 12.7 7.1 11.2 2.2 10.5 6 17.3 40.3 17.8 9.1 17.1 3.0 16.4 7 15.7 46.7 18.8 11.1 19.0 3.8 19.0 8 13.3 44.5 15.1 11.4 16.5 4.1 17.3 9 9.1 49.8 11.6 13.2 13.1 4.8 14.0 10 5.1 51.9 6.8 13.1 7.3 4.7 7.7 11 2.7 40.4 2.8 11.9 3.5 4.3 3.7 12+ 2.5 47.9 3.1 12.8 3.5 4.9 4.0 Total 100 39.0 100 9.2 100 3.1 100 61 ANNEX D REFERENCE FOOD BUNDLE Table D.1: The Reference Food Bundle Market prices Unit of Calorie Phnom Other Rural Quantity of measurement value per Penh Urban 3rd quintile unit (*1000) Beverages Instant coffee 100 gms 129(*) 733 700 560 0,301 Ground coffee 100 gms 129 (!) 700 817 600 0,193 Powdered tonic drinks 100 gms 350 (****) 1350 1100 1100 0,000 Processed cocoa 100 gms 392 (*) 800 800 800 0,061 Tea leaves/dust 100 gms 300(*) 800 500 600 8,692 Other tea tea bags 0 (*) 800 800 700 3,039 Canned soft drinks cans 142 (**) 1500 1300 1300 0,000 Bottled soft drinks bottles 90 (**) 1225 2750 1450 0,153 Other types of soft drinks 100(****) 1000 1000 1000 0,125 Fruit drinks cans 50(**) 1458 1325 1400 0,000 Other fruitjuices 50(**) 1300 1300 1300 0,039 Mineral/distilled/ liters 0 55 300 300 0,111 Bottled Angkor Beer cans 145 (****) 2500 2500 2500 0,129 Tiger Beer cans 145 (****) 1800 2000 1875 0,042 Other Brands of Beer cans 145 (****) 2000 1050 400 0,649 Wine bottles 7500 1200 1000 800 10,961 Distilled spirits bottles 2357 1500 1000 800 20,726 Cereals Rice (good quality) kgs 3530(!) 700 550 600 117,880 Rice (broken quality) kgs 3530(!) 650 580 571 313,380 Rice (sticky) kgs 3530 (!) 857 800 700 4,227 Whole grain maize 100 gms 364(!) 40 40 35 2,672 62 Table D.1: (Cont.) Market prices Unit of Calorie Phnom Other Rural Quantity of measurement value per Penh Urban 3rd quintile unit (*1000) Corn on the cob number 120(**) 100 88 67 33,218 Other grains 235 (****) 0,000 Pnum pan 100 gms 240 (**) 300 200 200 7,080 Other bread 100 gms 301.5(1) 400 300 300 0,637 Wheat 100 gms 332(*) 0,000 Other flour 100 gms 341 (*) 1,174 Fermented rice 100 gms 203 (*) 80 67 60 92,618 Noodles White rice/clear 100 gms 203 (*) 281 150 160 5,226 Noodles Yellow noodles 100 gms 203 (*) 333 300 250 12,641 Other noodles 100 gms 358 (!) 200 200 200 0,000 Biscuits/cookies 100 gms 407 (*) 500 500 500 0,629 Rice cakes 100 gms 235(*) 200 133 100 28,962 Other traditionalcakes 100 gms 300 (*) 200 150 100 7,446 Other cereals 100 gms 285 (***) 200 200 200 0,000 Dairy products Condensedimilk 100 gms 115(!) 1400 410 1100 4,250 (sweetened) Powderedmilk 100 gms 115(1) 817 500 800 0,067 Powdered milk (baby) 100 gms 477 (*) 1600 775 450 0,701 Ice cream 100 gms 140(*) 200 100 200 5,808 Butter 100 gms 729(*) 2300 2300 2300 0,000 Cheese 100 gms 390 (*) 2250 2250 2250 0,000 Other dairy products 100 gms 412 (***) 500 500 500 0,003 63 Table D.1: (Cont.) Market prices Unit of Calorie Phnom Other Rural Quantity of measurement value per Penh Urban 3rd quintile unit (*1000) Eggs Chicken eggs number 76 (**) 208 217 200 7,969 Duck eggs number 76 (****) 200 200 200 97,855 Other fresh eggs number 76 (****) 200 200 200 0,000 Boiled duck eggs number 76(****) 300 250 250 1,231 Fermented/saltedeggs number 100 (****) 250 300 300 3,859 Fruit Fresh bananas number 125 (**) 1000 700 500 44,581 Fresh oranges number 46 (**) 208 167 133 19,885 Fresh pineapple number 250 (****) 600 400 400 6,981 Fresh coconut milk number 150 (****) 400 333 73 2,793 mango Other fresh mangoes number 150 (****) 300 115 100 9,988 Fresh lemon number 10 (**) 80 80 60 4,676 Lime number 5 (****) 100 67 50 42,408 Freshrambutan 100 gms 64(*) 150 165 150 15,552 Fresh mangosteen number 20 (****) 500 500 500 0,576 Fresh papayas number 400 (****) 500 400 200 20,846 Fresh durians number 450 (****) 6000 6833 4325 0,010 Fresh breadfruits number 125 (****) 1000 500 467 0,485 Sugarcane 50(****) 200 200 108 9,743 Apricot number 33 (**) 300 200 550 0,955 Lotus fruit number 100(****) 100 87 50 2,136 Pomelo/grape fruit 100 gms 39 (**) 600 500 400 2,622 Watermelon number 420 (**) 1000 500 500 11,491 Calamansi number 100 (****) 167 188 110 0,560 64 Table D.1: (Cont.) Market prices Unit of Calorie Phnom Other Rural Quantity of measurement value per Penh Urban 3rd quintile unit (*1000) Chico 100 gms 100(****) 120 114 80 9,223 Jackfruit number 360 (****) 3250 1750 2000 1,049 Grapes number 4 (*) 500 600 550 0,000 Apples number 56 (**) 646 700 700 0,000 Blackberry 100 gms 30(**) 0,000 Other fresh fruits 100(****) 300 180 92 6,063 Canned pineapple 100 gms 92 (**) 533 533 533 0,000 Canned lychees 100 gms 71 (**) 1900 2000 2000 0,000 Canned fruit 100 gms 83(**) 1200 1200 1200 0,586 salad/fruit cocktail Dates 100 gms 143(**) 300 500 450 7,172 Tamarind 100 gms 214(**) 100 100 100 68,829 Other prepared, dried, 100 gms 260(****) 50 63 100 11,786 Coconut (young and number 1336 (***) 600 500 500 8,289 matured) Cashew nuts 100 gms 543 (*) 0,000 Lotus nuts 100 gms 334(*) 180 200 200 0,386 Peanuts 100 gms 314.2(!) 200 200 200 9,128 Gourd seeds 100 gms 400 (****) 400 400 400 0,015 Other nuts 100 gms 400(****) 200 200 200 0,156 Meat Pork without fat 100 gms 359.6(!) 600 500 400 41,052 Pork with fat 100 gmas 457(*) 450 400 400 142,040 Fresh beef 100 gms 273 (*) 500 447 400 36,176 Fresh buffalo meat 100 gms 123.3(!) 450 450 350 0,954 Fresh chicken 100 gms 302(*) 500 450 350 77,549 Fresh duck 100 gms 126(!) 380 342 300 9,742 65 Table D.1: (Cont.) Market prices Unit of Calorie Phnom Other Rural Quantity of measurement value per Penh Urban 3rd quintile unit (*1000) Other fresh meat 100 gms 171.2(!) 350 349 349 0,816 Mud fish (large) 100 gms 150(*) 400 350 300 32,943 Mud fish (small) 100 gms 100 (****) 342 300 250 375,880 Snake fish loogms 100(****) 300 300 200 4,657 Cat fish loogms 91 (*) 350 250 233 114,950 Sea fish (large) 100 gms 100 (****) 400 300 300 4,244 Sea fish (small) loogms 100(****) 260 150 200 24,121 Shrimps/prawns 100 gms 90 (!) 1000 700 250 1,092 Crabs 100 gms 100(*) 475 300 50 63,775 Other fresh sea food 100 gms 100(****) 200 200 200 161,850 Imported processed 100 gms 325.9 (!) 0,000 meat Roasted pork 100 gms 249 (*) 1000 800 900 0,234 Roasted/friedchicken 100 gms 229 (*) 1000 850 1000 0,000 Treated beef 100 gms 200(*) 800 800 800 0,233 Other locally 100 gins 325.9(!) 675 700 500 0,791 Processed meat Smoked fish 100 gms 145 (*) 455 400 400 26,983 Fermented/cheesefish 100 gms 66 (*) 200 200 150 141,540 Dried fish 100 gms 335(*) 600 400 350 42,824 Canned fish 100 gms 179 (***) 700 500 400 2,753 Dried prawns/shrimps 100 gms 240.9 (!) 1000 900 500 0,525 Other processed 100 gms 179 (***) 275 275 275 0,082 Marine products Oils and fat Rice bran oil mIs 7(*****) 4 4 4 49,371 Vegetable oil/soybean mIs 7 (*****) 4 3 2 402,470 66 Table D.1: (Cont.) Market prices Unit of Calorie Phnom Other Rural Quantity of measurement value per Penh Urban 3rd quintile unit (*1000) Cooking oil 100gms 7 3 3 3 0,131 Pork fat 100 gms 816(*) 200 200 200 63,431 Margarine 100 gms 723 (*) 267 200 200 0,114 Other food products Fried insects 400 (****) 300 300 300 0,346 Peanut preparation 585 (*) 200 200 200 0,682 Flavoredice 100 (****) 200 200 150 3,455 Ice 100 (****) 250 300 200 23,201 Other food products 100 (****) 130 130 130 0,584 Meals at work/ calories 1(***) 45471,500 school/restaurant Snacks, coffee, calories 1(***) 12211,200 softdrinks purchased and eaten outside the house Prepared meals bought calories 1(***) 0,000 outside and eaten at home 0,000 Sugar, salt, spices and seasoning 0,000 Granulated(refined) 100gms 376.7(!) 150 140 150 47,010 Sugar Brown sugar 100 gms 376.7(!) 120 100 100 200,100 Juggery 100 gms 350(****) 137 150 100 6,127 Chocolate candy bars 100 gins 402.6 (!) 500 500 500 0,245 Hard candy 100 gms 402.6(!) 400 300 410 1,026 67 Table D.1: (Cont.) Market prices Unit of Calorie Phnom Other Rural Quantity of measurement value per Penh Urban 3rd quintile unit (* 1000) Other sugar products 100 gms 493 100 50 45 2,033 Salt 100 gins 0 40 40 40 165,800 Garlic 100 gins 117 (*) .200 200 200 46,543 Coriander 100 gms 37 (*) 200 200 200 2,616 Ground black and 100gms 325(*) 317 400 300 6,769 White pepper Black/white 100 gms 220.5(***) 375 300 300 3,162 Peppercorns Red pepper spice 100 gms 116 (*) 300 200 200 8,506 Monosodium 100 gins 0 500 500 500 41,774 Glutamate Ginger 100 gms 46 (*) 200 200 200 3,349 Palm vinegar mis 1.4(****) 1 1 1 474,550 Soysauce mIs 1.44 2 2 2 2170,270 Fish sauce mis 1.4(****) 1 1 1 16316,100 Tomato sauce/tomato 100 gins 83 (*) 350 350 350 0,000 Catsup Other spices and seasoning 0 200 200 200 0,000 Vegetables Trakun 100 gins 20(****) 100 100 100 158,970 Onion/leeks 100 gins 48 (*) 200 200 192 9,259 Leaves/shallot Cabbage leaves 100 gins 37(!) 100 80 60 181,450 Lettuce, spinach 100 gins 21 (!) 100 100 60 36,078 Other leafy vegetables 100 gins 20(****) 150 100 100 40,474 Tomatoes 100 gins 20(!) 90 100 50 65,820 68 Table D.1: (Cont.) Market prices Unit of Calorie Phnom Other Rural Quantity of measurement value per Penh Urban 3rd quintile unit (*1000) Bell peppers 100 gms 30(****) 160 150 100 2,226 Ridge gourd 100 gms 17(*) 100 100 80 82,835 Bittergourd 100 gns 19(*) 120 100 73 13,520 White/yellow/green 100 gms 17 (*) 80 80 60 133,090 Gourd Cucumbers 100 gms 12 (*) 80 60 50 179,530 Squash 100 gms 50 (*) 80 70 60 23,596 Brinjals/eggplants 100 gins 26 (*) 100 100 50 73,419 Onion 100 gins 30(*) 200 200 150 1,713 Cauliflower 100 gms 29(*) 150 150 150 5,067 Radish/whiteradish 100 gms 26 (*) 155 155 155 0,000 Turnip 100 gis 21(*) 50 55 33 4,673 Carrots 100 gms 37 (*) 200 275 200 0,222 Other vegetables 100 gms 17.6(!) 135 100 80 55,873 Potatoes 100 gms 82 (*) 200 250 40 2,881 Sweet potatoes 100 gins 108.8(!) 40 50 30 93,619 Cassava 100 gms 156(!) 50 50 30 58,789 Traov 100 gms 20(****) 80 110 50 38,192 Other tubers loogms 20(****) 110 50 80 6,360 Green gram 100 gms 20(****) 180 120 140 14,728 Green dhall loogms 20(****) 180 200 120 2,988 Cowpea 100 gms 37(*) 120 152 120 1,458 Bean sprouts 100 gms 20(****) 100 80 100 28,983 Long green beans 100 gms 20(****) 100 100 75 30,799 Short green beans 100 gms 20(****) 130 160 80 0,767 Other pulses/legumes 100 gms 20 (****) 85 85 85 1,892 Cucumberpickles 100 gms 14 (*) 200 200 100 11,202 69 Table D.1: (Cont.) Market prices unit of calorie Phnom Other Rural Quantity of measurement value per Penh Urban 3rd quintile unit (* 1000) Other pickles 100 gms 20 (****) 200 167 100 6,804 Tomato paste 100 gms 20(****) 250 250 250 0,031 Other prepared and 100 gms 20(****) 120 129 160 2,590 Preserved vegetables Sources for calorie values: *:FAO **: CalorietableN. Duinker-Joustra ***: estimate ****: Indonesia table *****: S.O.W. (!): Vietnam 71 ANNEX E TECHNICAL GUIDE FOR PROGRAMMERS Introduction This annex gives a detailed account of how the poverty lines were constructed for Cambodia. The annex is aimed at computer programmers who wish to repeat the analysis. Detailed instruction, including the essential lines in computer programs, are given to guide the reader. It is assumed that the main text, in which the method is explained is familiar to the reader. The annex should be read as a recipe for doing the calculations. We will start from the data files as they were prepared by the National Institute of Statistics in Cambodia. First, we give a description of how the data files are constructed and which data files are needed to calculate the calculate poverty lines. You can read this in section A2. Paragraph A3 is concerned with how we can create our own dataset from the Dbase files. We will also explain how we construct variables needed for the analysis in paragraph A4 and how the data were cleaned in part A5. Section A6 will tell us something about computing total consumption per person per day and A7 will describe how to compute calorie intake. In part A7 you will find a description on how we constructed the food poverty line. Description of the Data Files In this part of the annex we will explain how the datasets are constructed. The data we received consist of 24 different datasets, all in Dbase format. You can find the names of the datasets and a short description in Table I and 2 of Annex B. For doing the price estimates we also relied on original files, not included in the standard package. All data files are on a household level and contain 5578 observations, which can be identified by the variable HHSN: the household-number. This does not do for the file pufdemo.dbf, because this file is based on an individual level. In this particular dataset, each individual in each household is interviewed. In this way we get 32709 observations, so all these 32709 interviewed individuals belong to one of the 5578 households. In the same 72 dataset pufdemo.dbf, we can trace to which household the observation belongs to by using the variable HHSN. There are four datasets, namely pufdemo.dbf, pufhaus.dbf, pufinco.dbf and pufexpdt.dbf, which contain general information about the household, such as demographic characteristics and consumption and income patterns. From these four datasets we will only use the data file pufdemo.dbf, because this is the only file that can tell us something about the number of persons in a household. The other twenty datasets contain information concerning food and non-food products consumed per household for some specified period. E.g. if we take a look at the dataset x beverag.dbf we will find that this file contains data on several products with relation to beverages on a weekly basis. In this dataset we can see that for each product i there is a variable representing the quantity consumed in each household per week, product_q, and there is a variable representing the corresponding amount in Riels paid for that quantity, product_v. E.g. the quantity of instant coffee consumed, is the variable coffl_q and the amount in Riels paid for that quantity is coffl_v. There is one exception: the dataset x_otfood.dbf contains the number of Riels per household spent on meals at work, school and restaurant, the amount of Riels spent on snacks, coffee and/or softdrinks purchased and consumed outside the house and the amount of Riels spent on prepared meals bought outside and eaten at home, but does not have the matching quantities for these three products. This is a problem we will solve later. See also paragraph A7. If we take a look at the Dbase file of beverages we will also find the unit of measurement for each item. For instance: instant coffee has a unit of measurement equal to 100 grams. Note that the non-food datasets do not have data on the quantities bought of each product; there is only data available on the amount in Riels spent on the specified products, thus the variableproducti_v. Creating an Analysis file from The Dbase Files Now that we have made an thoroughly analysis of the datasets available we can make the first step by creating a dataset we can use for our analysis. In this part we will explain how we can create our own dataset from the data files provided by the Cambodia Institute of Statistics. 73 As mentioned above, we will need the file pufdemo.dbf, because we need to count the number of persons in a household. We also want to use all the files containing necessary information about the consumption of various products, food and non-food. We now arrive at the point where we can use our statistical software to create our dataset, because we now know what variables we need and which datasets for these variables are necessary. So we can merge all these datasets, pufdemo.dbf and the twenty consumption files starting with x_, into one dataset by using the merge variable HHSN. Remember that at this stage we only use the dataset pufdemo.dbf for creating a variable which represents the number of persons in a household. Creating the Necessary Variables In section Al we described the available data, but we also need the prices paid by the households for each food product in order to get a complete overview of the consumption pattern. In this paragraph we will also explain why and how we compute median prices per region for each product properly. After we have computed the correct median prices according to the described procedure in this paragraph, we can use these in the data cleaning process we will discuss in the next section. Price estimates for pricing the food basket are constructed on the basis of cash expenditures only. Price estimates used for data cleaning are on the basis of total expenditure. As the method for obtaining the median prices is identical, we will explain the construction second price estimates only. We compute median prices because we want to leave out of consideration the outliers of the prices and we compute median prices per region because eventually we will construct poverty lines per region. We also need median prices of the whole country, because if we decide that we have a median price for one region based on too little observations, we will take the country median price as an estimate of the median price for that region. Also: if we do not have a median price of a certain product for one region (i.e. the median price is equal to zero or is missing for that region), we will also use the country median price as an estimate of the median price for that region. Now we will explain the procedure of creating prices and median prices per region step-by-step. I. First we have to compute the prices for each product i in each food category (with food category we mean the ten categories: beverages, cereals, etc.) according to the following formula: productLp = product_v / productL_q, 74 producti_q = the quantity consumed of product i per household per week, product_v = the amount in Riels spent on product i per household per week. II. The next step is to calculate median prices for each product i: product_m, and we will compute this by region: Phnom Penh, other urban regions and the rural regions. III. We also need to compute the median prices for each product for the whole country, product_1. IV. Next we will use the country median price as an estimate of the regional median price if the regional median price is equal to zero or is missing: producti_m = product1, if product_m = 0 or product_m = missing, where producti_m = median regional price of product i, product_1= country median price of product i. V. Further we will use the country median price as an estimate of the regional median price, if the regional median price is based on too little observation (we use 10 observations): product_m = product_1, if product_n < 10, where product_m = median regional price of product i, product_1= country median price of product i, product_n = number of observations the median regional price of product i is based on. The Data Cleaning Process 5. The median prices per region we now have computed, will be used in the data cleaning process which is described in this section. We will use the computed median prices in two different ways. First: For some observations we have a positive amount of Riels spent on some product, but the corresponding quantity is missing or zero. It can also be the other way around: we know a household uses a positive quantity of some product, but the amount of Riels paid for that quantity is missing or zero. However: we do not want to delete these observations, because if we do we will lose information. In this case we use the median prices per region we have computed. In the case the quantity is missing or zero for some product but the cost of that product is positive, we can estimate the quantity of that product as the division of the positive amount of Riels paid for that product by the median price of that product. We use the same approach if the value is missing or equal to zero and the corresponding quantity for that household is positive. Stepwise: VI. producti_q = product_v/ produc_m, if (producti q = 0 orproductL_q = missing) and (product_v > 0), where producti_q = the quantity of product i consumed per household per week, 75 product_v = amount in Riels spent on product i per household per week. product_m = median regional price of product i. VII. producti_v = productL_q * productj_m, if (product_v = 0 or product_v = missing) and (product,_q > 0), where product_q = the quantity of product i consumed per household per week, product_v = amount in Riels spent on product i per household per week. product_m = median regional price of product i. Second: We also face another problem. If the price of a product divided by the median price of that product is larger than or smaller than one fifth and if the quantity of that product lies outside the 95% confidence interval, the quantity is an outlier. The fact that the price divided by the median price for that product is larger than 5 or smaller than one fifth and the fact that the quantity lies outside the 95% confidence interval, means that only the quantity of that product is an outlier and the corresponding amount in Riels paid for this product is not. We now use the median price for the second time to repair this sort of outliers by defining the quantity equal to the amount in Riels paid for that quantity divided by the median price. We can use the same approach in the case the amount paid in Riels for a product lies outside the 95% confidence interval and the price divided by the median price for that product is larger than 5 or smaller than one fifth. Use the following steps: VIII. Compute the mean product;_aq and standard deviation productL_sq of the quantity, producti_q, of product i consumed per household per week, IX. Check the condition: Lproduct_aq - product_q I > product_sq * 1.96 and (productL_p / product_rm > 5 orproduct_p /product_m < 0.2). If this condition is true then: product_q = product_v I product_m, X. Compute the mean product_av and standard deviation product_sv of the amount, product,_v, in Riels paid for product i per household per week, XI. Check the condition: Lproduct;_av - product_v v > product_sv * 1.96 and (product_p /product_m > 5 orproducti_p / productm < 0.2). If this condition is true then: productL v=productL_q * product_m. Computing Total Consumption per Person per Day In this paragraph we describe how food consumption, non-food consumption and total consumption per person per day can correctly be computed by using our own dataset. We have already seen in paragraph A2 that we have a data file, called pufexpdt.dbf, which contains information about the monthly consumption of each household for several 76 food and non-food categories. This dataset also contains information about total consumption. But now that we have our own clean version of the data, abstracted from the twenty datasets containing information about the consumption of individual products, we will use this cleaned dataset to compute these totals again. We do it this way because we will also do our other analysis on this dataset and we want all to be based on the same dataset. When we compute food consumption, non-food consumption and total consumption, we first have to compute consumption per category. We have twenty different categories, because this time we will also look at the non-food categories; we also want to know non- food consumption and total consumption. First we take a look at the ten food categories and we see that each category is on a weekly basis per household. So to get a correct figure of total food consumption, we first compute the sum of the amounts spent in Riels per household per week for each food category. Then we compute total food consumption as the aggregate of the ten totals of these categories. This aggregate is still the total food consumption per household per week, so to get total food consumption per person per day we have to divide this total by the number of persons in the household and we have to divide this by seven, i.e. the number of days in one week. Step-by-step: XII. Compute the sum of the amounts spent in Riels per household per week for each of the ten food categories: beverages, cereals, dairy products, eggs, fruits, meat, oils and fat, other food products, vegetables and sugar, salt, spices and seasonings. E.g. we compute the weekly consumption of beverages for each household as: S, productL_v, where we sum over all products i in the category beverages, XIII. Compute the sum of these ten totals. This is total food consumption per household per week, XIV. Compute total food consumption per person per day by dividing total food consumption per week by the number of persons in the household and by 7. We want to do the calculations on the non-food categories on a monthly basis, because this is the period that appears mostly. Because not all the data in the non-food categories are based on weeks or months only. If we take a look at e.g. the household expenditures on personal care and effects, i.e. the data file x_pers.dbf, we see that some products cover a month and others cover a whole year! So when adding up the expenditures in one category we have to be aware of the fact that we first have to convert all annual and 6 monthly periods to one month and then adding the products in one category to get monthly expenditures. After we have done this for all ten non-food categories we can compute the sum of the total non-food consumption by aggregating these ten totals. Stepwise: 77 XV. Compute the sum of the amounts spent in Riels per household for each of the ten non-food categories: clothes, education, furniture, housing, medicines, miscellaneous items, tobacco, recreation, personal care and transportation. E.g. if we want to compute the total consumption per household per month in the category transportation we do: S1iproducti, v + (S;2 product2 _v) /12, where we sum over all products il, which refer to a monthly period, such as gasoline and tubes. Then we sum over all products i2 which refer to an annual period, such as a car and a bicycle. These amount have to be divided by twelve to get expenditures per month. XVI. Compute the sum of these ten totals to get total non-food consumption per household per month. XVII. Compute total non-food consumption per person per day by dividing total non-food consumption per month by the number of persons in the household and by 30. After we have computed the correct amounts spent on total food consumption per person per day and total non-food consumption per person per day we can compute total consumption, by adding total food and non-food consumption: XVIII. To get total consumption per person per day: add total food consumption per person per day and total non-food consumption per person per day. Computing Total Calorie Intake per Person per Day In this paragraph we will describe how we can compute the total number of calories a person consumes per day. But first we make some comments on the unit of measurement used for the quantities given in the dataset. In the original datasets the households are asked what the quantity of each product is they use per week. These quantities are measured in the same unit of measurement for each observation. We have used these to determine the corresponding calorie value. E.g.: the households are asked how much rice they consume per week in kilo&rams. So we need the calorie value of rice per kilogram. Thus if we take into account the correct measures for each product, we can compute how high the calorie intake of each product per person per day is by multiplying the quantity and the corresponding calorie value divided by the number of persons in the household and divided by seven (number of days per week). We can indeed do this for all food items, but not for some products in the category other food products. That is because we do not have the quantity consumed of three products in this category, only the amounts in Riels spent on those products per household per week. These three products are: (a) meals consumed at work, school and restaurant, (b) prepared meals bought outside the house and eaten at home, and (c) snacks, coffee and softdrinks purchased and eaten outside the house. But we also want to use the calorie intake of these three products, when computing the total amount of calories consumed. 78 We can solve this problem as follows: compute the total amount in Riels of food eaten per person per day, minus the amount in Riels of the three products mentioned above. Compute also the total number of calories consumed per person per day (of course without the three product, because we do not have the correct calorie values yet). We now compute a ratio equal to the computed total amount spent on food per person per day divided by the computed number of calories per person per day. This ratio acts as prices per calorie. We can use this ratio to compute the median prices of the three products, per calorie and per region. We assume that the three particular products cost twice as much than when prepared at home (so one calorie cost twice as much outside the house) so now we can compute the number of calories for those three products. After we have these calorie values, we can now finally add all calorie values together to get the total calorie intake per person per day. We can say that for the three specific products the quantities are measured in calories. This means that we now have a value for the quantity of these products: namely the number of calories. In this case we can set the calorie value of these three products equal to one, because we already have assumed that the quantities are measured in calories. If we do this step-by-step we get: XIX. Compute the number of calories per product i per day by using the formula: producti _k = producti _q * producti _c / (members * 7), where productk = calorie intake of product i per person per day, product,_q = quantity of product i consumed per household per week, product_c = number of calories in product i, members = number of persons in the household. XX. Compute the total number of calories per person per day, for all food items, except for the three above mentioned products in the category other food products: xtotacpp = S product, xk, where we sum over all products i and all categories x, except for the three food items we mentioned above (because we do not have these data), xtotacpp = the total number of calories per person per day minus the three food products, producti, k = calorie intake per person per day of product i of category x, XXI. Compute tvalmeal = the total consumption of food per person per day minus the amount in Riels for the three products, XXII. Compute: ratio = tvalmeal / xtotacpp if xtotacpp > 0, XXIII. Compute the median prices per calorie by region for the three products as the median of the ratio: mratio = median(ratio) by region, XXIV. We now can compute the number of calories in each of the three products according to the formula: product_c =productL_v / (members * 7 * mratio * 2), where 79 product;_c = number of calories in one of the three products i, product_v = amount spent on product i per household per week, mratio = the median price by region per calorie, and we divide by two because we have assumed that one calorie costs twice as much outside the house. XXV. With these new calorie values computed for the three products, we can add up all consumed calories per person per day to get the total number of calories consumed per person per day: tota-cpp = xtotacpp + Si=1,2,3 product_c, where we sum over the three products. XXVI. For the three products we can define producti _q = product _c, where product; _q = quantity of product i consumed per household per week measured in calories, product_c = number of calories in product i. XXVII. For the three products we can define product, _c = 1, because the quantities are measured in calories, so we can set the corresponding 'calorie' variable equal to one. We now have all the information available to compute a correct poverty line. This is described in the final section. The Food Poverty Line We have arrived at the point at which we have all information available to do the final intensive data step: computing the food poverty line. First we need a correct weight variable for doing this analysis. We will use the variable weight, which we can find in all the datasets. Now we want to select a group on which we want to base our poverty line, e.g. all observations in the third quintile of total consumption per person per day, which we have computed in section A5, and by using the variable weight mentioned above. This will be our reference group. The next step will be to compute the means of the quantities for each product per person per day of this reference group. With these mean quantities we compute the total number of calories per person per day in this reference group. We will assume that the recommended food energy intake in calories per day for Cambodia is equal to 2100 per day. We can now calculate alpha as: 2100 divided by the average food energy intake of the reference group, we just have computed. Next we can construct a food poverty line per region by calculating: the median market price per region of product i times the mean quantity of the reference group of that product per person per day times alpha, and this summed over all products. Step-by-step: XXVIII. Use the variable weight and get all the observations in the third quintile of total consumption per person per day. These observations are our reference group. 80 XXIX. Compute: product _g = mean[ product, q / (members * 7)], for the reference group created in step one, XXX. Compute the average food energy intake of the reference group per person per day: xcalo = S;( product_g * product;_c), (remark: you must also sum over the three products in the category 'other food products'!), XXXI. Set alpha = missing and compute alpha= 2100 / xcalo if xcalo > 0, XXXII. Construct the food-poverty line per region: alpha * Si (product _m * product, _g), where product, m = the median regional market price of product i (estimate based on cash purchases only); producti_g = the average quantity consumed of item i by the reference household; alpha = 2100 / xcalo (see also step 4). Data files Dataset Description Period No. of Unique data Record ID records pufdemo Demographic N.A. 32079 HHSN and .dbf Characteristics LINENO puthaus. Housing characteristics N.A. 5578 HHSN dbf pufinco. Households income N.A. 5578 HHSN dbf sources pufexpdt Household expenditures Monthly 5578 HHSN .dbf by major item grouping 81 Data Files (Cont.) Dataset Description: Household Period No. of Unique expenditures on: ... by data Record expenditures items records ID x beverg.dbf Beverages Weekly 5578 HHSN x_cereal.dbf Cereals Weekly 5578 HHSN x_dairy.dbf Dairy Products Weekly 5578 HHSN x_eggs.dbf Eggs Weekly 5578 HHSN x_fruit.dbf Fruits Weekly 5578 HHSN x meat.dbf Meat, poultry & fish Weekly 5578 HHSN x oilfat Oils and fat Weekly 5578 HHSN x_otfood.dbf Other food products and Weekly 5578 HHSN Food consumed away from home x season.dbf Sugar, Salt, Spices & Weekly 5578 HHSN Seasoning x_vege.dbf Fresh Vegetables, Tubers Weekly 5578 HHSN & Pulses x_cloth.dbf Clothing and Footwear Six 5578 HHSN months x_educ.dbf Education Monthly 5578 HHSN x_furn.dbf House Furnishings and Annual / 5578 HHSN Household Operation Monthly x_house.dbf Housing and Utilities Six months / Monthly x medic.dbf Medical Care Monthly 5578 HHSN x misc.dbf Miscellaneous Annual 5578 HHSN x_pers.dbf Personal Care and Effects Annual / 5578 HHSN Monthly x recre.dbf Recreation Annual 5578 HHSN x tabako.dbf Tobacco Products Weekly 5578 HHSN x_trans.dbf Transportation & Annual / 5578 HHSN Communication Monthly u Distributors of COLOMBIA GERMANY ISRAEL NEPAL PORTUGAL SWEDEN Infoenlace Ltda. UNO-Verlag Yozmot Literature Ltd. Everest Media Intemational Services (P) Ltd Livraria Portugal Wennergren-Williams AB W orld Bank Carrera 6 No. 51-21 Poppelsdorer Allee 55 PO. Box 56055 GPO Box 5443 Apartado 2681, Rua Do Carmo 70-74 P 0. Box 1305 Aparlado Aereo 34270 53115 Bonn 3 Yohanan Hasandlar Street Kathmandu 1200 Lisbon S-171 25 Solna Pubhcations Santafe de Bogota, D.C. Tel: (49 228) 949020 Tel Aviv 61560 Tel: (977 1) 472 152 Tel. (1) 347-4982 Tel: (46 8) 705-97-50 Prices and credit ternis vary fronr Tel: (57 1) 285-2798 Fax: (49 228) 217492 Tel: (972 3) 5285-397 Fax. (977 1) 224 431 Fax: (1) 347-0264 Fax: (46 8) 27-00-71 countryto countr. Consilt our Fax (57 1) 285-2798 URL: http://www.uno-vedag.de Fax: (972 3) 5285-397 E-mail: mail@wwi.se local distributor kfrplacing an E-mail: unoverlag@aol.com NETHERLANDS ROMANIA re p r COTE D'IVOIRE R.O.Y Intemational De LindeboominOr-Publikaties Compani De Librari Bucuresti S.A. SWITZERLAND order. Center d'Edition et de Diffusion Africaines GREECE PO Box 13056 PO. Box 202, 7480 AE Haaksbergen Sir Lipscani no. 26, sector 3 Librairie Payot Service Institutionnel ARGENTINA (CEDA) Papasotiriou S.A. Tel Aviv 61130 Tel: (31 53) 574-0004 Bucharest C6tes-de-Montbenon 30 04 B.P 541 35, Stournara Sir. Tel: (972 3) 5461423 Fax: (31 53) 572-9296 Tel: (40 1) 613 9645 1002 Lausanne cina Libro Inteaciona Abidjan 04 106 82 Athens Fax: (972 3) 5461442 E-mail: lindeboo@wordonline.n Fax: (40 1) 312 4000 Tel: (41 21) 341-3229 1120 Buevs Aires Tel: (225) 24 6510:24 6511 Tel: (30 1) 364-1826 E-mail: royil@netvision.net il URL: http://www.wordonline.nV-lindeboo Fax: (41 21) 341-3235 Tel: (54 1)815-8354 Fax: (225) 250567 Fax: (30 1) 364-8254 RUSSIAN FEDERATION Tel: (54 1) 815-8354 Palestinian Authority/Middle East NEW ZEALAND Isdatelstvo ADECO Van Dietmen EditionsTechniques Fax: (54 1)815-8156 CYPRUS HAITI Index Information Services E8SCO NZ Ltd. 9a, Lolpachnly Pereulok Ch.de Lacuez 41 AUSTRALIA, FIJI, PAPUA NEW GUINEA Center for Applied Research Culture Diffusion RO.B. 19502 Jerusalem Private Mail Bag 99914 Moscow 101831 CH1807 Blonay SOLOMON ISLANDS, VANUATU AND ' Cyprus College 5, Rue Capois Tel: (972 2) 6271219 New Market Tel: (7 095) 917 87 49 Tel: (41 21) 943 2673 WESTERN SAMOA 6, Diogenes Street, Engomi C.P 257 Fax: (972 2) 6271634 Auckland Fax: (7 095) 917 92 59 Fax: (41 21) 943 3605 RO. Box 2006 Port-au-Prince Tel: (64 9) 524-8119 D6A. Inomtieon Services Nicosia Tel: (509) 23 9260 ITALY Fax: (64 9) 524-8067 SINGAPORE, TAIWAN, THAILAND 648 Whitehorse Read Tel: (357 2) 44-1730 Fax: (509) 23 4858 Licosa Commissionaria Sansoni SPA MYANMAR, BRUNEI Central Books Distribution Victoria Fax: (357 2) 46-2051 Via Duca Di Calabria, 1/1 NIGERIA Asahgate Publishing Asia Paciic Pte. Ltd. 306 Siom Road To: (61(3 9210 7777 HONG KONG, MACAO Casella Postale 552 University Press Limited 41 Kallang Pudding Road #04-03 Bangkok 10500 Fee (61) 3 9210 7796 CZECH REPUBLIC Asia 2000 Ltd 50125 Firenze Three Crowns Building Jericho Golden Wheel Building Tel: (66 2) 235-5400 E-mal: service @ddirect.cm.au National Information Center Sales & Circulation Department Tel: (55) 645-415 Private Mail Bag 5095 Singapore 349316 Fax: (66 2) 237-8321 URL hf:trwvw.dadirect.com.au prodelna, Konviktska 5 Seabird House, unit 1101-02 Fax: (55) 641-257 lbadan Tel: (65) 741-5166 CS - 113 57 Prague 1 22-28 Wyndham Street, Central E-mail: licosa@ftbcc.it Tel: (234 22) 41-1356 Fax: (65) 742-9356 TRINIDAD & TOBAGO, AUSTRIA Tel: (42 2) 2422-9433 Hong Kong URL: http/www.ftbcc.itAicosa Fax: (234 22) 41-2056 E-mail ashgate@asianconnect.com AND THE CARRIBBEAN Gerold and Co. Fax: (42 2) 2422-1484 Tel: (852) 2530-1409 Systematics Studies Unit Weihburggasse 26 URL: http://www .nis.cz/ Fax: (852) 2526-1107 JAMAICA NORWAY SLOVENIA 9 Wats Street A- b Wi2n E-mail: sales@asia2000.com.hk Ian Randle Publishers Ltd. NIC Info A/S Gospodarski Vestnik Publishing Group Curepe Tel: (43 1(512-47-31-9 DENMARK URL: http-/www.asia2000 com.hk 206 Old Hope Road, Kingston 6 Book Department, Postboks 6512 Eterstad Dunajska cesta 5 Trinidad, West Indies Fa: (43 1) 512-47-31-29 SamfundsLifleratur Tel: 876-927-2085 N-0606 Oslo 1000 Ljubljana Tel: (809) 662-5654 URL h(p:/www.gerld.c/at.oline Rosencems Alld 11 HUNGARY Fax: 876-977-0243 Tel: (47 22) 97-4500 Tel: (386 61) 133 83 47; 132 12 30 Fax: (809) 662-5654 DK-1970 Frederiksberg C Euro Info Service E-mail: irpl@colis.com Fax: (47 22) 97-4545 Fax: (386 61) 133 80 30 E-mail: tobe@tlnidad.net BANGLADESH Tel: (45 31) 351942 Margitszgeti Europa Haz E-mail: repansekj@gvestnik.s Micra Industrien Development Fax: (45 31) 357822 H-1 138 Budapest JAPAN PAKISTAN UGANDA Assistance Sciety (MIDAS) URL: http://www.si.cbs.dk Tel: (36 1) 111 6061 Eastem Book Service Mirza Book Agency SOUTH AFRICA, BOTSWANA Gustro Ltd. Ase oce (D Fax: (36 1) 302 5035 3-13 Hongo 3-chome, Bunkyo-ku 65, Shahrah-e-Quaid--Azam Forsingle titles: PO Box 9997, Madhvani Building Honse5, Read 16 ECUADOR E-mail: euroinfo@mail.matav.hu Tokyo 113 Lahore 54000 Oxford University Press Southern Africa Plot 16/4 Jinja Rd. Dhakn R/Area Libn Mundi Tel: (81 3) 3818-0861 Tel: (92 42) 735 3601 Vasco Boulevard, Goodwood Kampala Tel: (869 2(326427 Libreti Intermacional INDIA Fax: (81 3) 3818-0864 Fax: (92 42) 576 3714 PO. Box 12119, Ni City 7463 Tel (256 41) 251467 Tel: (880 2) 326427 PO. Box 17-01-3029 Allied Publishers Ltd. E-mail: orders@svt-ebs.co.jp Cape Town Fax: (256 41) 251 468 Fax:(880 2) 611166 Juan Leon Mera 851 751 Mount Road URL: http-/www .bekkoame.orjp/-svt-ebs Oxford University Press Tel: (27 21) 595 4400 E-mail: gus@swiltuganda.com BELGIUM Quito Madras - 600 002 5 Bangalore Town Fax: (27 21) 595 4430 Jean De Lannoy Tel: (593 2) 521-606; (5932) 544-185 Tel: (91 44) 852-3938 KENYA Sharae Faisal E-mail: oxford@oup.co.za UNITED KINGDOM A a. du Ro 282 Fax: (593 2)504-209 Fax: (91 44) 852-0649 Afica Book Service (E.A.) Ltd. PO Box 13033 Microinfo Ltd. 1868 Brsisl2 E-mail: librimul @librimundi.com.ec Quaran House, Mfangano Street Karachi-75350 For subscription orders: PO. Box 3, Alton, Hampshire GU34 2PG Tel: (322)538-5169 E-mail: librimu2@librimundi.com.ec INDONESIA P0. Box 45245 Tel: (92 21) 446307 Intemational Subscription Service England Tel: (32 2) 5381 Pt. Indira Limited Nairobi Fax: (92 21) 4547640 PO. Box 41095 Tel: (44 1420) 86848 Fan: (322)(538-841 EGYPT, ARAB REPUBLIC OF Jalan Borobudur 20 Tel: (254 2) 223641 E-mail: ouppak@TheOttice.net Craighall Fax: (44 1420) 89889 BRAZIL Al Ahram Distribution Agency PO. Box 181 Fax: (254 2) 330 272 Johannesburg 2024 E-mail: wbank@ukminfo.demon.co uk Publicacies Tecicas Internacionais Lda At Galaa Street Jakarta 10320 Pak Book Corporation Tel: (27 11) 880-1448 URL: http://www.microinfo.co.uk Rua Peliet Gemide, 2 9 Cairo Tel: (62 21) 390-4290 KOREA, REPUBLIC OF Aziz Chambers 21, Queen's Road Fax: (27 11) 880-6248 01409 Seo Paulo, 20 Tel: (20 2) 578-6083 Fax: (62 21) 390-4289 Daejon Trading Co. Ltd. Lahore E-mail: iss@is.co.za VENEZUELA Tel: (55 11 259-6644 Fax:(20 2) 578-6833 RO. Box 34, Youida, 706 Seoun Bldg Tel: (92 42) 636 3222; 636 0885 Tecni-Ciencia Libros, S A. Tel: (55 11) 259-6644 IRAN 44-6 Youido-Dong, Yeongchengpo-Ku Fax: (92 42) 636 2328 SPAIN Centro Cuidad Comercial Tamance Fan: (55 11) 258-6999 The Middle East Observer Ketab Sara Co. Publishers Seoul E-mail: pbc@brain.net.pk Mundi-Prensa Libros, SA. Nivel C2, Caracas E-mail: postmaster@pti.uol.br 41, Sherif Street Khaled Eslamboli Ave., 6th Street Tel: (82 2) 785-1631/4 Castello 37 Tel: (58 2) 959 5547; 5035; 0016 Cairo Delafrooz Alley No. 8 Fax: (82 2) 784-0315 PERU 28001 Madrid Fax: (58 2) 959 5636 CANADA Tel: (20 2) 393-9732 PO. Box 15745-733 Editorial Desarrollo SA Tel: (34 1) 431-3399 Fax: (20 2) 393-9732 Tehran 15117 MALAYSIA Apartado 3824, Lima 1 Fax: (34 1) 575-3998 ZAMBIA 5369 Canolk Road Tel: (98 21) 8717819; 8716104 University of Malaya Cooperative Tel: (51 14) 285380 E-mail: libreria@mundiprensa.es University Bookshop, University of Zambia Oawa, Ontario KaJ 93 FINLAND Fax: (98 21) 8712479 Bookshop, Limited Fax: (51 14) 286628 URL: http://www .mundiprensa.es/ Great East Road Campus Tel: (613) 745-2665 Akateeminen Kirjakauppa E-mail: ketab-sara@neda.net ir PO. Box 1127 P0. Box 32379 Fax: (613) 745-7660 PO. Box 128 Jalan Pantal Baru PHILIPPINES Mundi-Prensa Barcelona Lusaka E-mail: rdr.dept@renou1bo3n co7 6 FIN-00101 Helsinki Kowkab Publishers 59700 Kuala Lumpur Intemational Booksource Center Inc. Consellde Cent, 391 Tel: (2680 1) 252 576 URL: hp:dewwwp.renoufbooks.com Tel: (358 0) 121 4418 RO. Box 19575-511 Tel: (60 3) 756-5000 1127-A Antipolo St, Barangay, Venezuela 08009 Barcelona Fax: (260 1) 253 952 Fax: (358 0) 121-4435 Tehran Fax: (60 3) 755-4424 Makati City Tel: (34 3) 488-3492 CHINA E-mail: akatilaus@stockmann.fi Tel: (98 21) 258-3723 Tel: (63 2) 896 6501; 6505; 6507 Fax: (34 3) 487-7659 ZIMBABWE China Financial & Econmic URL: http://www .akaleeminen.con/ Fax: (98 21) 258-3723 MEXICO Fax: (63 2) 896 1741 E-mail: barcelona@mundiprensa.es Longman Zimbabwe (Pvt.)Ltd. INFOTEC Toude Road, Ardbennie Publishing House FRANCE IRELAND Av. San Fernando No. 37 POLAND SRI LANKA, THE MALDIVES P.O. Box ST125 8, Da Fo Si Dong Jie Word Bank Publications Government Supplies Agency Col. Todello Guerra International Publishing Service Lake House Bookshop Southerlon Beijing 66, avenue d'lena Oifig an tSol6thair 14050 Mexico, D.E UI. Piekna 31/37 100, Sir Chittampalam Gardiner Mawatha Harare Tel: (86 10) 6333-8257 75116 Paris 4-5 Harcourt Road Tel: (525) 624-2800 00-677 Warzawa Colombo 2 Tel: (263 4) 6216617 Fax: (86 10) 6401-7365 Tel: (33 1) 40-69-30-56/57 Dublin 2 Fax: (525) 624-2822 Tel: (48 2) 628-6089 Tel: (94 1) 32105 Fax: (263 4) 621670 Fax: (33 1) 40-69-30-68 Tel: (353 1) 661-3111 E-mail: infotec@tn.net.mx Fax: (48 2) 621-7255 Fax: (94 1) 432104 Fax: (353 1) 475-2670 URL: htp://rtn net.mx E-mail: books%ips@ikp.atm corn pl E-mail: LHL@sri.lanka net URL: http://www .ipscg.waw.pVips/export/ Wi17 Recent World Bank Discussion Papers (continued) Hartnett No. 339 Public and Private Roles in Health: Theory and Financing Patterns. Philip Musgrove No. 340 Developing the Nonfarm Sector in Bangladesh: Lessons from Other Asian Countries. Shahid Yusuf and Praveen Kumar No. 341 Beyond Privatization: The Second Wave of Telecommunications Reforms in Mexico. Bjorn Wellenius and Gregory Staple No. 342 Economic Integration and Trade Liberalization in Southern Africa: Is There a Role for South Africa? Merle Holden No. 343 Financing Private Infrastructure in Developing Countries. David Ferreira and Karman Khatami No. 344 Transport and the Village: Findings from African Village-Level Travel and Transport Surveys and Related Studies. Ian Barwell No. 345 On the Road to EU Accession: Financial Sector Development in Central Europe. Michael S. Borish, Wei Ding, and Michel Noel No. 346 Structural Aspects of Manufacturing in Sub-Saharan Africa: Findings from a Seven Country Enterprise Survey. Tyler Biggs and Pradeep Srivastava No. 347 Health Reform in Africa: Lessons from Sierra Leone. Bruce Siegel, David Peters, and Sheku Kamara No. 348 Did External Barriers Cause the Marginalization of Sub-Saharan Africa in World Trade? Azita Amjadi Ulrich Reincke, and Alexander J. Yeats No. 349 Surveillance of Agricultural Price and Trade Policy in Latin America during Major Policy Reforms. Alberto Vald6s No. 350 Who Benefits from Public Education Spending in Malawi: Results from the Recent Education Reform. Florencia Castro-Leal No. 351 From Universal Food Subsidies to a Self-Targeted Program: A Case Study in Tunisian Reform. Laura Tuck and Kathy Lindert No. 352 China's Urban Transport Development Strategy: Proceedings of a Symposium in Beijing, November 8-20, 1995. Edited by Stephen Stares and Liu Zhi No. 353 Telecommunications Policies for Sub-Saharan Africa. Mohammad A. Mustafa, Bruce Laidlaw, and Mark Brand No. 354 Saving across the World: Puzzles and Policies. Klaus Schmidt-Hebbel and Luis Serv6n No. 355 Agriculture and German Reunification. Ulrich E. Koester and Karen M. Brooks No. 356 Evaluating Health Projects: Lessons from the Literature. Susan Stout, Alison Evans, Janet Nassim, and Laura Raney, with substantial contributions from Rudolpho Bulatao, Varun Gauri, and Timothy Johnston No. 357 Innovations and Risk Taking: The Engine of Reform in Local Government in Latin America and the Caribbean. Tim Campbell No. 358 China's Non-Bank Financial Institutions:Trust and Investment Companies. Anjali Kumar, Nicholas Lardy, William Albrecht, Terry Chuppe, Susan Selwyn, Paula Perttunen, and Tao Zhang No. 359 The Demand for Oil Products in Developing Countries. Dermot Gately and Shane S. Streifel No. 360 Preventing Banking Sector Distress and Crises in Latin America: Proceedings of a Conference held in Washington, D.C., April 15-16, 1996. Edited by Suman K. Bery and Valeriano F. Garcia No. 361 China: Power Sector Regulation in a Socialist Market Economy. Edited by Shao Shiwei, Lu Zhengyong, Norreddine Berrah, Bernard Tenenbaum, and Zhao Jianping No. 362 The Regulation of Non-Bank Financial Institutions: The United States, the European Union, and Other Countries. Edited by Anjali Kumar with contributions by Terry Chuppe and Paula Perttunen No. 363 Fostering Sustainable Development: The Sector Investment Program. Nwanze Okidegbe No. 364 Intensified Systems of Farming in the Tropics and Subtropics. J.A. Nicholas Wallis No. 365 Innovations in Health Care Financing: Proceedings of a World Bank Conference, March 10-11, 1997. Edited by George J. Schieber No. 366 Poverty Reduction and Human Development in the Caribbean: A Cross-Country Study. Judy L. Baker No. 367 Easing Barriers to Movement of Plant Varieties for Agricultural Development. Edited by David Gisselquist and Jitendra Srivastava No. 368 Sri Lanka's Tea Industry: Succeeding in the Global Market. Ridwan Ali, Yusuf A. Choudhry, and Douglas W. Lister No. 369 A Commercial Bank's Microfinance Program: The Case of Hatton National Bank in Sri Lanka. Joselito S. Gallardo, Bikki K. Randhawa, and Orlando J. Sacay No. 370 Sri Lanka's Rubber Industry: Succeeding in the Global Market. Ridwan Ali, Yusuf A. Choudhry, and Douglas W. Lister No. 371 Land Reform in Ukraine: The First Five Years. Csaba Csaki and Zvi Lerman � _ - _ �. _ _ _ :- г' � r� _ ! _ � _ `� � ► J � J г_ � f � � � � � G г = � _ � � �� � _ i_, - т ���� � _ = - - - г - 'I. �� i� _ " _ •' г: - Гв � � _. `_' �^ .. _� Г: Т i J �� -_. 1�J l , , _ о � � l 1 + �J �� � � + ,� + / � .. i - + ;� _ � 1J J1 �� �.� _ � �� � �� �_ � � - - f = � _ - � - -- ' !,� ,, 0 _ _ ! N _ - } r - � = г Ф т- _� _ .� = � � а � ' � г � ,- - , . % 7С 1= .� й ао г о � ro _ w ��г А р О rv N р О А