REGIONAL AND SECTORAL STUDIES 21089 October 2000 Geographical Targeting for Poverty Alleviation Methodology and Applications DAVID BIGMAN HIPPOLYTE FOFACK __ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ IU Geographical Targeting for Poverty Alleviation WORLD BANK REGIONAL AND SECTORAL STUDIES Geographical Targeting for Poverty Alleviation Methodology and Applications EDITED BY DAVID BIGMAN AND HIPPOLYTE FOFACK THE WORLD BANK WASHINGTON, D.C. ©2000 The International Bank for Reconstuction and Development/The World Bank 1818 H Street N.W. Washington, D.C. 20433 All rights reserved Manufactured in the United States of America First printing October 2000 1 234503020100 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 organ- izations, or to members of its Board of Executive Directors or the countries they represent. The World Bank does not guarantee the acccuracy of the data included in this publication and accepts no responsibility for any consequence of their use. The material in this publication is copyrighted. The World Bank encourages dissemination of its work and will normally grant permission to reproduce portions of the work promptly. Permission to photocopy items for internal or personal use, for the internal or personal use of specific clients, or for educational classroom use is granted by the World Bank, provided that the appropriate fee is paid directly to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; telephone 978-750-8400, fax 978-750-4470. Please contact the Copyright Clearance Center before photocopying items. For permission to reprint individual articles or chapters, please fax a request with complete information to the Republication Department, Copyright Clearance Center, fax 978-750-4470. All other queries on rights and licenses should be addressed to the Office of the Publisher, World Bank, at the address above or faxed to 202-522-2422. Library of Congress Cataloging-in-Publication Data Geographical targeting for poverty alleviation: methodology and applications / edited by David Bigman, Hippolyte Fofack. p. cm. -- (World bank regional and sectoral studies) ISBN 0-8213-4625-3 1. Poverty--Developing countries--Data processing. 2. Poverty-Developing countries--Mathematical models. 3. Geographic information systems--Developing countries. I. Bigman, David. II. Fofack, Hippolyte, 1963- III. Series. HV29.5.D5 G46 2000 362.5'8'091724--dc2l 00-036817 Contents Acknowledgments .............................................. x Foreword .................................................... xi Preface .................................................... xiii Introduction and Overview ....................................... 1 David Bigman and Hippolyte Fofack Part One: Geographical Targeting and GIS: An Overview 41 1. Geographical Targeting: A Review of Different Methods and Approaches ............................................. 43 David Bigman and Uwe Deichmann 2. Applications of a GIS in Program Impact Evaluation: Lessons from USAID Experience ....................................... 74 Glenn Rogers Part Two: Alternative Methods of Geographical Targeting .......... 99 3. Combining Census and Survey Data to Study Spatial Dimensions of Poverty: A Case Study of Ecuador ......................... 101 Jesko Hentschel, Jean Olson Lanjouw, Peter Lanjouw, and Javier Poggi ZV Vi Geographical Targetingfor Poverty Alleziation 4. Community Targeting for Poverty Reduction in Burkina Faso . .125 David Bigman, Stefan Dercon, Dominique Guillaume, and Michel Lambotte 5. Applying Household Expenditure Survey Data to Improve Poverty Targeting: The Case of Ghana .............................. 154 Hippolyte Fofack 6. Spatial Indicators of Access and Fairess for the Location of Public Facilities ................................................ 181 David Bigman and Uwe Deichmann Part Three: Applications of a Geographical Information System (GIS) for Geographical Targeting ........................................ 207 7. Evaluation of Food Security in the Sahel: An Analysis Using the Demographic and Health Survey (DHS) Data with a Geographical Information System ...................................... 209 Mark McGuire 8. Using a GIS to Target River Blindness Control Activities in Guatemala ............................................ 235 Frank 0. Richards 9. A Geographical Information System Applied to a Malaria Field Study ............................................. 258 Allen W. Hightower, Maurice Ombok, Richard Otieno, Richard Odhiambo, and William A. Hawley 10. A Geographical Information System as a Component of the Animal Health Information System in Thailand ............... 277 Pramod Sharman and Angus Cameron 11. Location Criteria of Nongovernmental Organizations Providing Credit to the Poor: The Experience in Bangladesh . .... 292 Manohar Sharma and Manfred Zeller Figures 1 Targeting Performance under Optimal Selection of States. 15 1.1 Spatial Outliers .57 1.2 Illustration of Aggregation Problems .62 3.1 Standard Errors on Headcount Rates and Population Disaggregation ........................................ 116 Conlten-ts onii 4.1 Water Point Proximity .................................. 133 4.2 Schools Proximity and Characteristics ..................... 134 4.3 Poverty Index at Village Level ........................... 148 5.1 Distribution of Household Per Capita Expenditure by Quintile .............................................. 161 5.2 Spatial Distribution of Poverty across Agro-Climatic Regions . .171 5.3 Type I Error Probability Estimate across Agro-Climatic Regions .............................. ................ 172 5.4 Type II Error Probability Estimate across Agro-Climatic Regions .............................................. 173 6.1 Service Area Dehneation Source: United Nations (1997) ...... 185 6.2 Computation of Accessibility Indicators ................... 196 6.3 Shortest Distance Indicator for Two Provinces in Madagascar ........................................... 197 6.4 Number of Women in the Reproductive Age Group by Travel Time to the Closest Service Facility ............................ 199 6.5 Accessibility Indicators-Women in Reproductive Age Groups (WRA) ............................................... 201 7.1 DHS Cluster Locations and ICRAF Aridity Zones ........... 216 7.2 Child Malnutrition by Aridity Zone ....................... 217 7.3 Examples to Illustrate Spatial Filtering of DHS Clusters ...... 222 7.4 Composite Biophysical Index ............................ 224 7.5 Final Composite Vulnerability Index ...................... 224 7.6 Access and Nutrition Index Derived from DHS Data ........ 225 7.7 Example Queries with Final Indices ....................... 227 8.1 Central Endemic Zone for Onchocerciasis, Guatemala ....... 243 8.2 Central Endemic Zone for Onchocerciasis .................. 244 8.3 Spatial Analysis of the Horizontal Plain ................... 245 8.4 Horizontal and Vertical Distances from the Map Position ..... 246 8.5 Central Endemic Zone for Onchocerciasis .................. 248 8.6 Central Endemic Zone for Onchocerciasis .................. 249 8.7 Central Endemic Zone for Onchocerciasis .................. 251 8.8 Comparison of Three Data Sets for Community Positions in the Target Area ........................................... 252 8.9 Comparison of Two Sources for Community Inventory Data . .253 9.1 Map of the Bednet Study Area ........................... 266 9.2 Map of the Immunity Study Area ........................ 267 9.3 Average Numbers of Anopheles Gambiae Trapped per Continent ............................................ 269 9.4 Estimated Areas of High, Mid-High, Mid-Low, and Low Mosquito Abundance ...................... .................... 270 Viii Geographical Targetingfor Poverty Alleviation Tables 1. Population Shares and Poverty Measures in India, 1983 ....... 12 2 Performance Measures for Regional Targeting in India ........ 12 3 Performance Measures for Rural and Urban Targeting in India .13 4 Performance Measures for Targeted Programs in India ........ 14 1.1 Summary Statistics of the Three Different Realizations in Figure 1.2 ........................................... 63 3.1 Points by Services Included in the INEC BN Indicator ....... 104 3.2 Poverty Incidence under Alternative Welfare Definitions ..... 105 3.3 Distribution of Bottom 20 Percent under the BN Criterion across Consumption Expenditure Quintiles ................ 107 3.4 Distribution of Bottom 20 Percent Using Predicted Consumption across Actual Consumption Expenditure Quintiles .......... 109 3.5 Regional Poverty Rates for Ecuador ....................... 111 3.6 Ecuador Poverty Map: Urban and Rural Provinces .......... 114 A.3.1 Comparative Descriptive Statistics . .118 4.1 Data Sources .......................................... 132 4.2 Descriptive Statistics on Variables Used in the Estimation .... 135 4.3 Descriptive Statistics on Variables Used in the Prediction ..... 136 4.4 Poverty and Consumption in Burkina Faso: Estimates of the Priority Survey (1994) .................................. 138 4.5a Regression Results-Dependent Variable Is log (Consumption per Standard Adult) ....................................... 139 4.5b Regression Results: Estimated Variance with Multiplicative Heteroscedasticity ..................................... 140 4.6 Comparison of the Model's Predictions of the Headcount Measure of Poverty and the Direct Estimates for Villages in the Sample for Three Provinces .............. 144 4.7 Distribution of the Population in the Provinces of Burkina Faso into Four Poverty Categories according to the Classification of Their Communities (Percent) .............. 145 5.1 Summary Statistics Distribution of Per Capita Expenditure by Survey Type and across Region .............. 160 5.2 Indices of Extreme Poverty and Rate of Mistargeting across Regions: A Comparison of the Priority Surveys and Living Standards Measurement Surveys .............. 164 5.3 National and Regional Poverty Predictors for Poverty Analysis .................. 169 5.4 Indices of Extreme Poverty and Rate of Mistargeting across Regions: A Comparison of LSMS and Imputed Expenditures. .170 6.1 Accessibility Indicators-Women in Reproductive Age Groups (WRA) ................... 200 Contetiets ix 7.1 Study Site Summary of DHS Database .................... 214 7.2 Aridity Zones and Population Characteristics of WASAP Database ............................................. 216 7.3 Proxy Variables Used by UNEP to Develop a Human Development Index (HDI) from the WASAP Database ................... 217 7.4 DHS Cluster-Level Principal Component Analysis .......... 220 7.5 Spatial Filtering Examples for Soum, Burkina Faso, and Bakel, Senegal .............................................. 221 8.1 Examples of Repeated Names in the Community Inventory Data Set .............................................. 239 8.2 GIS Buffering Exercise Communities to Be Assessed or Treated .............................................. 247 8.3 Endemic Communities Identified in the Gazetteer Database . .252 9.1 Parasitemia Prevalence and Entomologic Measures by Household and Distance to the Nearest Mosquito Larval Habitat ........ 268 11.1 Descriptive Statistics of Regression Variables-Thana Level . . 299 11.2 Placement of NGOs: Estimated Fixed-Effects Logit Equation . .301 11.3 Outreach Equation: BRAC and ASA .303 Acknowledgments IN RECENT YEARS, the potential applications of spatial analysis were broadened significantly with the advances in geographical information systems (GIS) methods and the accumulation of a large volume of increasingly more reliable data that also contain spatial references. The interest in these applications for the design of development strategies in general and of poverty allevia- tion programs in particular was reflected in the large and lively participation in the Conference on Geographical Targeting for Poverty Alleviation organized by the Poverty Reduction and Social Development Division of the Africa Region of the World Bank in the fall of 1997. This conference led to a fruitful collaboration between the editors of this volume and a large number of researchers in the World Bank, in other multinational organizations, and in various research institutes in industrial and developing countries. Many researchers sent us their work for consideration and made it possible for us to select for this volume chapters that address most effectively the issues raised by the choice of spatial characteristics as criteria for targeting the allocation of public resources and the methods that can be used for geographical targeting. We wish to thank all these researchers for their interest in our work and to express our sincere appreciation to the contributors to this volume. We are also grateful to the three anonymous reviewers whose valuable comments on an earlier draft of the manuscript led to considerable improvements in some of the chapters. We are particularly indebted to Jack W. van Holst Pellekaan, formerly Lead Specialist Poverty in the Africa Region of the World Bank, for his invaluable guidance at the beginning of this project and for his continuous support which was essential in bringing this project to completion. Several other colleagues have helped us in the preparation of this volume, and we would like to extend our sincere gratitude to all of them, particularly to Ye Xiao and Francoise Genouille for their assistance during the organization of the conference. The completion of this project also profited from the working environment provided by the Poverty Reduction Group within the Africa Region of the World Bank under the leadership of Roger Sullivan, Sector Manager. We are also grateful to the Government of Norway for providing financing to sup- port this project through the Norwegian Trust Fund, managed by Antoine Simonpietri, and to the Government of the Netherlands, particularly Ms. Margreet Moolhuijzen, for financing the overall costs of printing this volume through the Dutch Trust Fund, managed by Lionel Demery. Product development, design, editing, production, and dissemination were directed and managed by the World Bank's Office of the Publisher. x Foreword RISING POVERTY AND DECLINING public resources present a major challenge for the majority of developing countries. In Sub-Saharan Africa (SSA), the problem is especially acute. After nearly two decades of low rates of eco- nomic growth that, in many countries, fell well below the rates of popula- tion growth, the size of the poor population in the subcontinent increased substantially, and at the end of the 1990s nearly two-fifths of the population in this region live in poverty. However, countries in SSA were not equally affected by these trends. A number of countries adopted sound macroeco- nomic policies that yielded encouraging results. In these countries, econom- ic growth rates gradually accelerated, and poverty was reduced. These countries, which represent a growing number in Sub-Saharan Africa, sug- gest that persistent efforts to improve economic policies can lead to signifi- cant poverty reduction. The task of renewing economic growth must also include concerted efforts to stabilize the government budget with the goal of breaking the vicious circle of rising public deficits, ensuing runaway inflation, and rising poverty. Successful implementation of these stabilization programs, how- ever, requires considerable sacrifices and painful tradeoffs in the short run. For poverty reduction programs to be successful, it is necessary to explore creative solutions that will allow the governments of these countries to tar- get their limited resources to the most needy, and to use these resources in the most effective way. Programs that cover a country's entire population, such as general food subsidies that were common in the 1960s and 1970s, li xii Geographical Targeting for Poverty Alleviation were of limited effectiveness in reaching the poor, led to bloated public deficits and debts, and are no longer sustainable. During the past decade, the need for effective targeted programs that provide significant support to the poor within the tightening budget con- straints has become more apparent than ever. The design of efficient pro- grams that are tailored to the specific conditions and needs of each country presents a challenge that government agencies and international develop- ment institutions must confront. This book is intended to contribute to this challenge by addressing the complex factors that need to be taken into account in the design of successful poverty alleviation programs. The book presents altemative methods of geographical targeting aimed at improving the living standards of the poor, and carefully evaluates their effect on social welfare and their implications for public resource allocation. The book demonstrates that the use of geographical information systems (GISs) makes possible a detailed mapping of the incidence of poverty in the country that can be used, in turn, for more precise targeting. The incorpora- tion of data from a wide variety of sources by means of GISs also reveals the multidimensional aspects of poverty and enables a more accurate identifi- cation of factors relevant for the design of effective poverty alleviation pro- grams within the tight budget constraints. Finally, the book also shows that GIS methods can be more broadly applied in support of targeting service delivery and access to the poor, for instance by optimizing the planning and location of health and education centers in poor and under-served areas. We welcome this contribution to our work. We are confident that our staff in the Africa Region and other groups in the World Bank, in other development institutions, and in developing countries themselves will ben- efit from the methods presented in this book. Use of these methods will improve the effectiveness of targeted programs in reaching the poor and will help to achieve the greatest poverty reduction impact within today's stringent limits on public resources. CALLISTO MADAVO JEAN-LouIs SARBIB VICE PRESIDENT VICE PRESIDENT AFRICA REGION AFRICA REGION THE WORLD BANK THE WORLD BANK Preface THIS BOOK GREW OUT OF TWO PROJECTS undertaken by the Poverty Reduction and Social Development Group of the World Bank's Africa Region: The first was a study on targeting public projects and allocating resources for health, education, and development across rural communities in Burkina Faso. The second was a conference on "Geographical Targeting for Poverty Alleviation" that was organized by the Division in the fall of 1997 in order to explore the potential contribution of GIS applications to improving the distribution of public resources across a country's geographical areas and securing more effective targeting on the poor. These two projects were by no means the World Bank's first endeavors to draw attention to the significance of the spatial dimension in project design and the implementation of antipoverty policies and programs. Nevertheless, for all too many projects and programs, a careful design of the general structure of the project and a meticulous evaluation of its total costs and benefits are still followed by much less scrupulous attention when it comes to the decision on where-that is, in which specific village or urban community-to implement the project. In rural areas in particular, this is often the most important decision that determines the benefits to the local population from the project. The widening availability and growing use of spatial data, organized in a computer system as a geographical information system (GIS), significant- ly advance the possibilities of analyzing the spatial impact of projects and programs and of achieving more effective targeting. Nevertheless, the actu- al use of GISs, both in development organizations and in developing coun- tries themselves, is still limited. The discrepancy between the extensive xiii liz, GeoaraphIical Tairecting for Pozertit Allev'iatio1n possibilities that this tool opens up and the very slow pace at which it is applied by economists and social scientists working in development, was the main motivation for the initiative of bringing together the papers pre- sented at the conference, as well as several other papers that were submit- ted for consideration at a later stage, in this volume. The book has three objectives. First, it introduces the basic concepts of GISs to readers who are not yet familiar with this tool, and demonstrate the forms of organizing geographic and nongeographic data in this system for potential users. Second, the book presents different methods for using the data from a Household Income and Expenditure Survey together with other surveys and the population census in order to provide estimates for the standard of living and the poverty incidence in different geographical areas of a country. The ultimate objective of these estimates is to establish guide- lines for targeting poverty alleviation projects and programs. Third, the book illustrates different applications of GISs for identifying the target pop- ulation of a program, determining the spatial "sphere of influence" of a proj- ect, or deciding on the location of public facilities. These illustrations are for a variety of projects or programs in health, nutrition, and education. The focus in this book is primarily on the methodology, and, in principle, the same method that is being presented in one of the chapters for calculating the sphere of influence of a disease, for example, can later be used to calcu- late the sphere of influence of a hospital. Although the potential contribution of a thorough analysis of the spatial effects of projects or policies to improve their effectiveness is obvious, the obstacles to a more extensive use of spatial analysis in general and of GISs in particular in most developing countries are still enormous. They include the difficulties in collecting and properly organizing all the data required for the analysis and the resources needed for acquiring the know-how and the hardware in order to use this system. It is clear that no single project, large as it may be, will be able to justify these expenses by itself. It is also clear, however, that, with only minor modifications, this system can serve all proj- ects over an extended period of time. To secure the availability of all the data and the resources necessary for this system, all projects will have to combine their efforts and share these expenses. This, in turn, will require considerable cooperation between the different units that work on develop- ment project in the country and a long-term perspective. a ALAN HAROLD GELB CHIEF ECONOMIST AND SECTOR DIRECTOR ECONOMIC MANAGEMENT AND SOCIAL POLICY AFRICA REGION, THE WORLD BANK Introduction and Overview David Bigman and Hippolyte Fofack LARGE INEQUALITIES IN THE STANDARD OF LIVING between geographic areas and "pockets of poverty" are common in all countries, rich and poor. The northern region of Nigeria; the Indian states of Bihar, Orissa, and West Bengal; the "inland" provinces in China; the southern regions in Italy; and the Deep South of the United States are just a few examples of geographic areas in which the incidence of poverty is much higher than in the other parts of these countries. The main reasons for such marked inequalities are the unequal distribution of natural resources (including water), differences in agro-climatic conditions, and differences in geographic conditions (pri- marily the distance to the centers of commerce, to the main transport routes, and to seaports.). Another factor that leads to income disparities between regions, districts, and communities is geographic bias in infrastructure pol- icy decisions, reflected in the poor quality of such local infrastructure as access roads and the availability of public services. Studies on income inequality and poverty generally use an individualis- tic, human-capital model that seeks to explain differences in income and consumption by individual and household characteristics. Spatial (that is, geographic) variables are added in some studies to explain these differ- ences, but usually in an ad hoc way.' The formal argument of the individu- alistic model is straightforward: In a country where internal migration is free and the economy is in equilibrium, the standard of living must be fully determined by characteristics of individuals and households other than geography. If spatial characteristics did have an effect on the well-being of 2 Geographical Targetingfor Poverty Alleviation individuals and households, so the argument goes, then they would tend to move to better locations.2 Nevertheless, large differences in the incidence of poverty between dif- ferent geographic areas do exist. Their magnitude is often far too large to be explained by differences in individual or household characteristics alone. One reason for the persistence of these differences is that internal migration is not really free-in some countries because of deliberate government poli- cies, and in all countries because of economic, demographic, and cultural obstacles. Migration between rural areas, for example, is often constrained by the lack of available land for cultivation, and rural-to-urban migration leaves behind the very young and the very old. Moreover, migration is cost- ly and risky, and frequently individuals do not have the necessary informa- tion to make such a decision. Even migration from rural to urban areas, while common, is often a gradual process, beginning with the move of a sin- gle household member to an urban center to look for work; it may take a long time for other members of the household to follow. And there are fur- ther barriers to migration that help account for large inter-regional income disparities. Particular aspects of poverty, such as large households, poor health conditions, low levels of human capital, and, in some countries, the "feminization of poverty," reduce the capacity of the poor to migrate. Wealthier individuals in poor regions are less restricted in their decisions to migrate, and when they leave, the standard of living in these areas declines still further.3 In addition, geographic areas with a low standard of living often have a much lower quality of public services, particularly education and health, which impedes their residents' accumulation of human capital, and therefore their earning capacity and prospects for migration. Binswanger and others (1993) evaluated the significance of rural infrastruc- ture for local investments and growth. Foster and Rosenzweig (1995) emphasized the varying conditions in different rural communities, and the resultant impact on the diffusion of new farm technologies; the low level of human capital in the poor communities demonstrably slows down the adoption of new technologies by the local farmers and reduces earning capacity. All these factors increase the likelihood that households in poor regions will retain individual and community characteristics that doom them and their offspring to continued poverty. The spatial dimension of the economic activities, decisions, and charac- teristics of individuals, despite its importance, has long been given only lit- tle attention in economic theory. Until the mid-1990s, the early writings on spatial economics of Harris (1954) and Myrdal (1957) had only a handful of followers. In a series of lectures published as Development, Geography, and Economic Theory (1995), Paul Krugman asked why spatial issues remained a blind spot for the economic profession, observing that "economic geogra- phy-the location of activity in space-is a subject of obvious practical introdtuction anid Oz'crziez'n 3 importance and presumably considerable intellectual interest. Yet it is almost completely absent from the standard corpus of economic theory."4 In recent years, though, a growing body of economic research seeks to explain the mechanisms and rules by which economic forces operate in geography, thus putting spatial considerations into the mainstream of economic theory. One aspect of this research analyzes the factors that determine the location of and the spatial interactions between industrial and market centers (see, for example, Krugman 1993, Ades and Glaser 1995, and Fujita and Mori 1996). Another branch of this research analyzes the geographic characteris- tics of cities, regions, or countries, the impact of these characteristics on the pace of economic development, and their implications for inequality among regions and among nations (Fujita and Mori 1996, Krugman 1991, Krugman and Venables 1995). These studies represent what has become known as the "new economic geography," and they give center stage to the evaluation of the effects of dis- tance on the economic growth of and income disparities between nations and regions. They emphasize the important, sometimes critical role that dis- tance-to and from sources of raw materials, main transport routes, sea- ports, and large population centers-plays in determining the division of production between industrial and primary-producing regions, and in aug- menting the divergence of incomes between North and South. In these stud- ies, the comparative advantage of regions and nations has been determined not only by their factor endowments, but also by their geographic attributes. Warner and Sachs (1997) argue that because Africa is more land-locked and has fewer navigable rivers than almost any other area in the world (except for Central Asia), distance is the main factor inhibiting its economic growth. Although technological progress has gradually reduced the importance of distance and has provided cheaper, alternative means of transportation- thereby changing the comparative advantage and disadvantage of geo- graphic areas-this change has been slow, with a noticeable impact only in the last 150 years. Until that progress arrived (and in many developing countries it has yet to come), the geography of a nation and a region deter- mined not only their historical and political lot, but also their economic fate. Climate, distance, and access are the primary components of economic geography, and their respective importance varies when explaining the large differences in the pace of development between regions, and even between communities. In most countries, climate is the primary factor deter- mining the prospects of development of the agricultural sector, and with it, the rate of growth of the entire economy. In industrial countries, this factor gradually became less significant, but in the majority of developing coun- tries, where agriculture remains the dominant sector, climate continues to be the most significant factor that determines the country or region's level of development. With the spread of industry and commerce, distance-to 4 Geographical Targetingfor Poverty Alleviation the sources of raw materials and energy on the one hand, and to the main population centers on the other-became more important to growth. Equally important are the ease of access to the ports and to the main centers, and the availability of passable and safe roads. At the local level of the village and the urban community, the relevant distances are to the nearest town, to the main transport route, and to public services. Access to the community year round and the absence of natural or man-made obstacles to access are equally important for development. In many developing countries, and particularly in Sub-Saharan Africa, villages located more than 5 kilometers from the main transport routes are likely to have significantly lower standards of living than closer villages, and villages without a passable access road during the rainy season are noticeably poor- er than villages with year-round access. The incorporation of these geographic factors can significantly enrich economic analysis and policy decisions. By identifying the spatial and envi- ronmental factors that affect the standard of living in a community, it is pos- sible to formulate the policies that are necessary in order to raise living standards. The information on distance and access needed for this analysis is difficult to obtain, and in many developing countries, the process of gen- erating and collecting this information is still at an early stage. In recent years, however, these geographic indicators have received much greater attention and their importance for the design of public policies has been widely recognized. This process took a large leap forward with the devel- opment of new and sophisticated methods for incorporating spatial data by organizing them as a geographic information system (GIS), suitable for computer analysis, and by a surge of technological innovations, particular- ly satellite imagery, which advanced the ability to collect spatial and cli- matic data. In a GIS, the database contains information not only on the value of social, economic, climatic, or environmental observations, but also on their location and spatial arrangement. This allows the presentation of data in the form of maps and overlaying interfaces for cross-comparisons, and the performance of spatial analysis assessing the relationships between these data according to their geographic location.5 In many countries, and at all levels of development, these systems have become the single most impor- tant analytical tool for the analysis of a wide range of geographic and socio- economic data and for the design of policy measures that account for space. The great interest in this subject was evident from the high attendance and the lively participation generated by a conference on "Geographical Targeting for Poverty Reduction," organized by the Institutional and Social Policy Division of the World Bank's Africa Region in the Fall of 1997.6 The conference highlighted the potential of GIS applications to improve project and program design and to provide guidelines for effective targeting of aid to selected geographic areas, with the goal of cost-effective reductions in introducti)! a)id C),vcrzi'ceu 5 poverty. In view of the mounting constraints on resources available for development projects, and the growing pressures to achieve better results on the ground, GIS applications gain increasing significance. However, current- ly available books on GIS methodology and its applications are not tailored to the needs and professional background of economists and social scien- tists working on development issues. By bringing together the papers pre- sented at the World Bank's 1997 conference, this volume aims to provide an introduction to the basic concepts of GISs and a sample of their different applications to spatial analysis. This introductory chapter also provides a brief overview of the economic motivation for targeting public projects in general, and of the pros and cons of geographic targeting in particular. Economic Criteria for Targeting Targeting poverty alleviation programs to a subgroup of the population has an intuitive appeal for policymakers and economists, but also considerable perils. The principle that guides policymakers in planning programs of this type is how to use the available resources in order to provide the greatest possible amount of assistance to those who need it most. The intuitive method of targeting subgroups identifies poor individuals and directs all benefits only to them, but this practice is marred with problems and diffi- culties. First, the costs of identifying poor individuals can be very signifi- cant, and requires information that is not available in most developing countries. Second, even in industrial countries where this information can be obtained, it is impossible to ascertain if targeted programs will indeed reach all the poor, and only the poor. The costs of obtaining this information can be very high, higher than the costs of implementing the first best solu- tion of reducing leakage to individuals who are not poor. Third, benefits to the poor provide incentives to non-poor households to change their behav- ior in order to qualify for the program, thus raising the costs of the program and reducing the impact of social welfare. Fourth, targeted programs tend to stigmatize the poor, both in their own eyes and in the eyes of their fellow citizens, potentially leading to reactions that can frustrate efforts to break the cycle of poverty and thus undermining the goal of the program. The negative effects of targeted programs are exacerbated when information necessary for identifying the poor cannot be obtained, causing governments to resort to second-best solutions of identifying the poor by means of indi- rect criteria. These difficulties have led some economists to conclude that targeted programs should be discarded altogether in favor of programs with universal coverage. Nevertheless, the constraints on public resources available for social welfare programs make targeting the only viable alternative for practically all developing countries. In these countries, growing budget constraints 6 Geographical Targetingfor Poverty Alleviationi prevent governments from providing universal coverage programs such as general food subsidies that were highly popular in earlier years. Difficulties in gathering information and the weak administrative capacity of the gov- ernment also dictate the form of intervention.7 There are two general alter- natives for managing social welfare. The first is the use of indirect criteria to determine eligibility, such as the household size, the number of children in the household, the size of the household's landholdings or other assets, or the place of residence. The second is the use of self-targeting programs, such as food-for-work, subsidies for commodities that are consumed primarily by the poor, or targeting research and extension services on the agricultural products of the poor Targeted programs that use indirect criteria are bound to involve con- siderable leakage of benefits to the non-poor, while excluding many of the poor not found eligible under the program's criteria, and the savings com- pared to nontargeted programs may therefore be quite low. For example, in family assistance programs implemented in a number of Latin American countries, eligibility was determined by the number of children in the household; the costs of these programs were pushed to intolerably high lev- els by slack entitlement conditions and large leakage of benefits, and they had, in addition, negative effects on the fertility rates of the poor. In Tanzania, difficulties in establishing clear eligibility criteria for the distribu- tion of food aid forced the government to delegate the distribution opera- tion to specialized local NGOs and village committees that drove up the costs. In Sri Lanka, nearly half the population has access to food stamps while less than 30 percent are eligible under the program's criteria (Subbarao and others 1997). The effectiveness of targeted programs for poverty reduction thus depends on the availability of an efficient and inexpensive mechanism for identifying the poor, on the cost reduction achieved by the exclusion of non- poor households from the program, and on the organizational capacity of the government to administer the program. Despite the leakage in targeted programs that use indirect indicators as eligibility criteria, the savings com- pared with universal coverage programs make targeting the choice by default in view of the mounting budget constraints. Grosh (1994) also noted the political feasibility of a program as one of the central factors that deter- mine its effectiveness, since the main obstacles to targeted programs have often been political. Unfortunately, targeted programs tend to isolate and stigmatize the target population, thus reducing the political support for the programs, while universal coverage may provide the political leverage to mobilize the support of the population not covered by the program, partic- ularly middle-income consumers. Anand and Kanbur (1990) reported that, after the introduction of a targeted food stamp program in Sri Lanka, the real value of the food stamps was allowed to fall quite sharply during peri- Inftroduiction anid Ovyerviewv 7 ods of high inflation as the interest of the middle class shifted to other issues and public support for the program declined. In some countries, particular- ly in Sub-Saharan Africa, targeted programs may also exacerbate ethnic ten- sions if the target group is perceived to be predominantly of a specific eth- nic origin. Income-based programs and means testing are common in industrial and middle-income developing countries; these programs are highly cost effective, and rank high in terms of the principal performance measures- that is, leakage and coverage (see further discussion below). Nevertheless, these programs are often more divisive than universal coverage, and raise political problems due to the stigma attached to beneficiaries (see Rainwater 1982; Besley and Coate 1992; and Smolensky and others 1995). Besley and Kanbur (1993) point out that stigmatizing beneficiaries of income-based programs can reduce the ability of welfare recipients to acquire skills and grow out of poverty. The stigma is particularly divisive when the costs of the program are being borne by a relatively small portion of the general population. Moffitt (1983) describes the stigma in terms of the "disutility arising from participation in the welfare program;" and Besley and Coate (1992) emphasize the "psychic costs of being on welfare." Smolensky and others (1995) distinguish between external and internal stigma; that is, the stigma created because the welfare program lowers the self-esteem of the recipients, and the stigma imposed by the society at large. Often, however, the most significant reason for political tensions and opposition to targeted programs is the leakage of benefits to ineligible households, and the per- ception that they are perceived to take a free ride on the back of the taxpay- ers. Improved targeting can therefore go a long way toward reducing such tensions. Rainwater (1982) pointed out, however, that more accurate target- ing may also have the opposite effect of further stigmatizing the poor by identifying them more accurately (p. 46). This will be the case particularly if more stringent eligibility conditions require the recipient to submit very personal information (for example, the name of the father of a child born out of wedlock). The next section of this chapter ("An Overview of Alternative Targeting Methods") provides an overview of the main targeted programs imple- mented in developing countries, and examines the pros and cons of geo- graphic targeting in comparison to other methods of targeting. Several examples that illustrate advantages of and problems with targeted pro- grams are discussed. While the subject of the overview is poverty alleviation programs, similar principles apply also to other targeted programs in which poverty alleviation is only one of several objectives. Other programs might include education for girls or children of a specific age group (irrespective of their parent's income), health programs for women of childbearing age or for households with a large number of children, nutrition programs for S Geographical Targetingo for Poverty Alleviationi mothers and young children, and so forth. As background to this overview, the remainder of this section summarizes the main measures that are com- monly used for evaluating the performance of targeted poverty alleviation programs. These measures include the following: * Type I errors-the error of inclusion, which denotes the number of non-poor individuals who are included in the program due to inaccu- rate specification of the criteria for entitlement and their proportion in the total number of the benefit recipients (also referred to as "vertical inefficiency"). * Type II errors-the error of exclusion, which denotes the number of poor individuals who are excluded from the program due to inaccurate specification of the criteria for entitlement and their proportion in the country's total number of poor (also referred to as "horizontal ineffi- ciency"). * The budgetary costs of the program-including the costs of collecting the data necessary for the design of the criteria of entitlement, as well as the program's administrative costs. * The effects of the program on the behavior of households and the implications for the households' welfare and the government budget. * The effects of the program on poverty reduction. The performance of the program, as indicated by these measures, depends on the criteria that are used to determine eligibility, and the instruments that are used to transfer benefits to the target population. Among the above criteria, the errors of inclusion and exclusion general- ly receive the greatest attention due to their intuitive appeal and their direct budgetary implications. Ravallion and Chao (1989) suggest a quantifiable performance measure for targeted programs which takes into account both of these errors: the gains from targeting are defined as the amount by which the budget for a nontargeted program would have to be increased in order to achieve the same reduction in poverty, as measured by the poverty gap ratio. They termed this measure the "equivalent gain from targeting." Clearly, the larger the type I error, the higher the costs of the targeted pro- gram, and the smaller the equivalent gain. Likewise, the larger the type II error, the smaller the cost increase with a nontargeted program that pro- vides the same reduction in poverty, and the smaller the equivalent gain. A complete specification of these performance measures must also include the choice of a measure for the reduction in poverty, and therefore a proper measure of poverty. The Headcount measure of poverty is not a proper measure for this purpose: if poverty is measured by the Headcount measure, then a program of income transfers would achieve the greatest reduction in poverty if targeted to the persons (or regions) who are the least poor, leaving the poorest people uncovered. The Headcount measure is also lbtrolductiow anid Ozenricci 9 not the proper measure for comparing programs: Datt and Ravallion (1993) evaluated the reduction of regional disparities through income transfers (while leaving intra-regional inequalities unchanged), and concluded that this program would yield only a marginal reduction in the Headcount measure of poverty. The methods of Datt and Ravallion fail, however, to measure the reduction in the poverty gap of the remaining poor population.8 If poverty in the target region(s) is well below the poverty line, the income transfer program may fail to lift the extreme poor out of poverty, whereas a nontargeted program will also reach the least poor, and by lifting them out of poverty, may generate an even larger reduction in the Headcount measure. When measuring poverty by the poverty gap ratio, the gains from geo- graphic targeting can be calculated as follows: Let Si be the share of the ith region's population in the country's total population, and let Hi be the Headcount measure of poverty in that region. Consider a targeted program that transfers income to the persons residing in that region. To simplify the illustration, assume that the income transfer does not change the number of poor in the region and, to simplify the notations, let this transfer be of 1 rupee per person. This transfer will raise the income of the poor population residing in this region-and thus reduce their income gap below the pover- ty line-by a total of (Si Hi N), where N is the total number of persons in the entire economy. The budgetary costs of the program will be (Si . N). Consider now the nontargeted program, and let w be the amount trans- ferred to all persons in the country under the latter program. If the number of poor does not change (much) with this transfer, then the poor's poverty gap is reduced with the nontargeted program by a total of (w. N * H), and the budgetary costs of this program are (N co). By equating the reduction in the poverty gap under the targeted program with the reduction in the gap under the nontargeted program, we can calculate the transfer ow that equates these two reductions, given by equation 1: S.= Si Hi (1) H Inserting this value into the formula for calculating the budgetary costs with the nontargeted program, and comparing these costs with the costs of the targeted program, determines the equivalent gain (EG), given by equation 2: EG= Hi . (2) H The target region is typically the one in which the incidence of poverty is higher than average, and the EG ratio is therefore larger than 1. Another performance criterion for evaluating a targeted program is the reduction in the poverty gap that can be achieved with a targeted program compared with the reduction in the gap that can be achieved with a nontar- 10 Geographical Targeting for Poverty Alleviation geted one, when the costs of the two programs are the same. By comparing the costs of a targeted program that transfers 1 rupee to each person in the target area with the costs of a universal coverage program that transfers e rupees to each person in the country at large, we can conclude that the trans- fer in this case would be E = Si; the expression of EG in equation 2 then meas- ures the ratio of the reduction in poverty with a targeted program relative to the reduction in poverty with a universal coverage program. In India, the percentage of the rural population in poverty was, in 1983, 43.9 percent, while that percentage in the state of Bihar was 60.8 percent (Datt and Ravallion 1993). With the same budgetary costs, a program targeted on the state of Bihar can therefore bring about a 50 percent larger reduction in the poverty gap compared with a universal coverage program. Equation 2 can also have a different interpretation. To see this, define the benefits from the targeted program as the average reduction in the poverty gap of the poor. Obviously, the larger the number of poor persons that are excluded from the program, the smaller the average reduction in the pover- ty gap of the poor and the smaller the benefits. When the total number of poor persons remains unchanged, the benefits per person would be equal to the share of the poor population covered by the program, given by: [(Si H1)/H], and the budgetary costs per person would be equal to Si. The benefit/cost ratio would therefore be given by the ratio (H1 /H), which is also equal to the value of the EG in equation 2. An alternative performance measure compares the costs and effective- ness of a targeted program with the costs and effectiveness of another tar- geted program. Here we consider two alternatives to a given targeted pro- gram: one is a program targeted on all other regions of the country; the other is a program targeted on another subgroup of regions in which targeting may be more effective. We can term these measures the "opportunity costs of targeting" (OC). The opportunity cost of targeting region R, is the value of the foregone alternative action of targeting any other region Ri with i + j. Using the same notation and the same procedure as above, the first alter- native is determined by equating the reduction in poverty in the two pro- grams, and comparing the respective costs of achieving that reduction with the targeted program and the program targeted on all other regions. After some algebra, that cost ratio is given by equation 3: OC=(1- Si) Hi Hi (3) H - HiSi HCi where HCi is the share of the poor individuals in the population of all other regions that are not covered by the target program under consideration. We provide two equations of opportunity costs of targeting and this expression can also be written as equation 4: hItrodtiction antd Overviec' 11 OC=EG 1- S (4) [1 - (SiHi / H)] The expression (1 - Si) measures the share of the general population that is not covered by the program; the expression 1 - (Si H1 /H) measures the share of the poor population that is not covered by the program (the error of exclusion). It is easy to verify that OC> EG