Research Observer Volume 20 Number 1 Spring 2005 Democracy, Public Expenditures, and the Poor: Understanding Political Incentives for Providing Public Services Philip Keefer and Stuti Khemani Evaluating the Impact of Conditional Cash Transfer Programs Laura B. Rawlingsand Gloria M. Rubio Reassessing Conditional Cash Transfer Programs Jishnu Das, Quy-ToanDo, and Berk 0zler Infrastructure Privatization and Regulation: Promises and Perils Ioannis N. Kessides The "Cotton Problem" John Baffes Research Observer EDITOR Shantayanan Devarajan. World Bank COEDITOR Gershon Feder, World Banlc EDITORIAL BOARD Paul Collier, Oxford University Susan Collins,Georgetown University Angus Deaton, Princeton University Barry Eichengreen, University of California-Berkeley Emmanuel Jimenez, World Bank Benno Ndulu, World Bank Howard Pack, University of Pennsylvania Luis Serven. World Bank Michael Walton, World Bank The World Bank Research Observer is intended for anyone who has a professional interest in development. Observer articles are written to be accessible to nonspecialist readers; con- tributors examine key issues in development economics, survey the literature and the lat- est World Bank research,and debate issuesof development policy.Articles are reviewed by an editorial board drawn from across the Banlcand theinternationalcommunity of econo- mists. Inconsistency with Banlc policy is not grounds for rejection. The journal welcomes editorial comments and responses, which will be consideredfor pub- lication to the extent that space permits. On occasion the Observer considers unsolicited contributions. Any reader interested in preparing such an article is invited to submit a proposal of not more than two pases to the Editor. Please direct all editorial correspon- dence to the Editor, The World Bank Research Observer,1818 HStreet, NW,Washington, DC 20433, USA. The viewsand interpretationsexpressed in this journal are those of the authors and do not necessarily represent the views and policiesof the World Bank or of its ExecutiveDirectors or the countries they represent. The World Banlc does not guarantee the accuracy of data included in this publication and accepts no responsibilitywhatsoever for any consequences of their use. When maps are used, the boundaries, denominations, and other information do not imply on the part of the World Bank Group any judgment on the legalstatus of any territory or the endorsement or acceptance of such boundaries. For more information, please visit the Web sites of the Research Observer at www.wbro.oupjournals.org, the World Bank at www.worldbank.org, and Oxford University Press at www.oupjournals.org. Subscriptions X subscription to Thi~World Bllr~kResearch Observer (ISSN0257-3032) comprises 2 issues. Prices include postage: for subscribers outside the Americas, issues are sent air freight. Annual Subscription Rate (Volume20.2 issues. 2005) Acadzrnic libraries Print edition and site-wide online access: US$103/f 70 Print edition only: USS981f67 Site-wideonline access only: USS931fh3 Corporate Print edition and site-wide online access: IJS$128/E85 Print edition only: US$122/f 8 1 Site-wide online access only: US$11(>If77 Prrsonnl Print edition and individual online access: US$38/f 29 Please note: f Sterling rates apply in Europe. US$ elsewhere. Readers with mailing otldresses in non-OECD countries arld in socialist econon~iesin transition are eligibleto receive con~plinlen- tary s~~bscriptionson ruquest by writing to the UICaddressbelow. There may be other subscription rates available: for a complete listing, please visit www.wbro.oupjournals.org/ subscriptions. Full prepayment in thecorrect currency is required for all orders. Ordersare regarded as firm,and payments are not refund- able. Subscriptions are accepted and entered on a complete volume basis. Claims cannot be considered more than four months after publication or date of order, whichever is later. All subscriptions in Canada are subjecttoGST.Subscriptionsin the EU may be subject to European VAT. Ifregistered,pleasesupply details to avoid unnecessary charges. For subscriptions that include online versions, a proportion of the subscription price may be subject to UK VAT. Personal rates are applicable only when a subscription is for individual use and are not available if delivery is made to a corporate address. Back issues: The current year and two previous years' issues are available from Oxford University Press. Previous volumes can be obtained from the Periodicals Service Company. 1 1 Main Street. Germantown. NY 12526. USA. E-mail: psc@periodicals.com. Tel: ( 518) 537-4700. Fax: (518) 537-5899. Contact information: Journals Customer Service Department, Oxford University Press. Great Clarendon Street. Oxford OX2 hDP. UK. E-mail: jnls.cust.serv@oupjournals.org. Tel: +44 (0)1865 353907. Fax: +44 (011865 353485. In the Americas,please contact:Journals Customer Service Department, Oxford University Press. 2001 Evans Road. Cary. NC 2751 3. USA. E-mail: jnlorders@oupjournals.org. Tel: (800) 852-7323 (toll-free in USAICanada) or (919) 677-0977. Fax: (919)677-1714. In Japan.please contact:Journals Customer Service Department. Oxford University Press. 1-1-17-SF,Mukogaoka, Bunkyo-ku, Tokyo, 113-0023. Japan. E-mail: okudaoup@po.iijnet.or.jp.Tel: (03) 3813 1461. Fax: (03)3818 1522. Postal information:The World Bar~kResearch Observer (ISSN 0257-3032) is published by Oxford University Press for the International Bank for Reconstruction and De~el~pment/THEw o ~ wBANK. Send address changes to The World Barlk Research Observer, Journals Customer Service Department. Oxford University Press. 2001 Evans Road,Cary. NC 275 13- 2009. Communications regarding original articles and editorial management should be addressed to The Editor. The World Bank Research Observer. The World Bank. 1818 H Street, NW. Washington. D.C. 20433. USA. Permissions: For information on how to request permissions to reproduce articles or information from this journal. please visit www.oupjournals.org/permissions. Advertising: Inquiries about advertising should be sent to Helen Pearson. Oxford Journals Advertising. PO Box 347, Abingdon OX14 1GJ.UK. E-mail: helen@oxfordads.com. Tel: +44 ( 0 ) U3 3 201904. Fax: +44 (018704296864. Disclaimer:Statements of fact and opinion in the articles in The World Bank Research Observer are those of the respective authors and contributors and not of the International Bank for Reconstruction and Developmentl~~~ WORU BANK or Oxford University Press. Neither Oxford University Press nor the International Bank for Reconstruction and Develop- m e n t / ~ ~ ~BA WOR1.U NK make any representation,express or implied, in respect of the accuracy of the material in this jour- nal and cannot accept any legal responsibility or liability for any errors or omissions that may be made. The reader should make her or his own evaluation as to the appropriateness or otherwise of any experimental technique described. Paper used: The World Bank Research Observer is printed on acid-free paper that meets the minimum requirements of ANSIStandard 239.48-1984 (Permanenceofpaper). Indexing and abstracting: The World Bank Reserlrch Observer is indexed and/or abstracted by . A ~ ~ / I N F ~ Rtan, . \ IAbstracts, Current ContmtslSocialanrl Behavioral Sciences, Journalo/Economic LiteraturelEco~~Lit,PAIS International. RrPEr (Research in Economic Papers). Social Services Citation Inder,and Wilson Business Abstracts. Copyright QThe International Bank for Reconstruction and Development/~ll~:wo~1.11 R.\?.IK2005 Allrights reserved: no part of this publication may be reproduced,storedin a retrievalsystem, or transmitted in any formor by any means. electronic,mechanical, photocopying. recording,or otherwise without prior written permission of the publisher or a licensepermitting restricted copying issued in the UK by theCopyright Licensing Agency Ltd. 90 TottenhamCourt Road. London W 1P9HE,or in the USA by the CopyrightClearanceCenter.222 Rosewood Drive. Danvers,MA 01923. Democracy, Public Expenditures, and the Poor: Understanding Political Incentives for Providing Public Services Philip Keefer · Stuti Khemani The incentives of politicians to provide broad public goods and reduce poverty vary across countries. Even in democracies, politicians often have incentives to divert resources to polit- ical rents and private transfers that benefit a few citizens at the expense of many. These dis- tortions can be traced to imperfections in political markets that are greater in some countries than in others. This article reviews the theory and evidence on the impact on political incentives of incomplete information for voters, the lack of credibility of political promises, and social polarization. The analysis has implications for policy and for reforms to improve public goods provision and reduce poverty. Governments can accelerate economic development through their decisions on public expenditures. Allocated appropriately, public expenditures can overcome market failures that exacerbate poverty, such as the inability of poor people to borrow for education or to learn about preventive health care or the existence of externalities that increase the public health hazards that disproportionately hurt poor people. Allocation is often inappropriate, however. Public spending goes to wage bills for bulky state administrations, farm subsidies absorbed by the wealthiest farmers, or public works projects with limited public utility, all at the expense of the quality of public services. The Millennium Development Goals and many efforts around the world to "empower" poor people, whether through devolved decisionmaking or through participatory budgeting, are all responses to these failures of government expenditure policies. Misallocation has persisted despite a sea change in the way in which govern- ments are selected and remain in office. From 1990 to 2000 the number of countries governed by officials elected in competitive elections rose from 60 to 100 (Beck and others 2001). Low-income voters make up a large share of the electorate in many © The Author 2005. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THEWORLD BANK . All rights reserved. For permissions, please e-mail: journals.permissions@oupjournals.org. doi:10.1093/wbro/lki002 20:1­27 countries, and democratization might be expected to benefit them. However, policy- makers in poor democracies regularly divert spending away from areas that most benefit the poor or fail to implement policies that improve the services that are known to disproportionately benefit poor people. In most countries, the distribution of income is skewed toward the higher end of the income spectrum. The median voter hypothesis (that the policy most preferred by the median voter can defeat any other policy proposal under majority rule) implies that when the income of the median voter is less than the average income of all voters, government should be larger and social services should be correspondingly more extensive (Meltzer and Richard 1981). This model evidently does not hold in many low-income countries, because in many countries where the median voter is poor, social services are especially deficient. Moreover, there is little evidence that redistribution is greater in countries with greater income inequality (Knack and Keefer 1997). Political market imperfections, which disrupt the translation of voter preferences into government policy, help explain this puzzle. This article focuses on three political market imperfections that are particularly important for understanding government incentives to serve the poor: lack of infor- mation among voters about the performance of politicians, social fragmentation among voters manifested as identity-based voting, and lack of credibility of political promises to citizens. Efforts to improve the quality of public policy for the poor may fall far short of their goals unless efforts are expanded to address those political market imperfections. One difficult conclusion of the analysis is that the broad public services most important to the poor--health and education--are the most vulnerable to these three distortions. Political pressures to improve these services are therefore weakest. With respect to information, citizens cannot easily evaluate the quality and effi- ciency of these services. Even when they can, they cannot easily determine whether service providers, higher-level ministry officials, or politicians are responsible for breakdowns or improvements. Similarly, politicians in many countries can make credible promises only to narrow groups of voters. It makes little sense for politicians to promise reforms that benefit large numbers of voters when only a few voters believe these promises. These politicians prefer to promise narrow targetable goods, such as infrastructure, rather than improvements in broad public services. Social polarization exacerbates these problems because it raises the possibility that one group may oppose a broad reform from which it would benefit if the reform would also benefit other groups. Low demand for social services by poor people is an unlikely explanation for pervasive failures of social service provision, as the next section argues. The following sections assess the impact of each of these imperfections in political markets, particularly on social service provision, and then demonstrate how these political market imperfections help explain one of the more striking contrasts in government policymaking--the 2 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) famous cases of the states of Uttar Pradesh and Kerala in India. The lack of voter infor- mation, greater social polarization, and the noncredibility of political party platforms in Uttar Pradesh explain its poor performance on social services compared with Kerala. The final section explores some implications of this political economy analysis for policy initiatives to improve the allocation of public resources.1 Politics and Social Services There is ample evidence that governments, particularly in poorer countries, prefer to spend more on targeted programs, such as government jobs or infrastructure invest- ment, than on improvements in broad social services. Feyzioglu and others (1998) find that governments receiving foreign aid targeted to education are likely to offset the funds by reducing their own financing of education. When aid is used to support public infrastructure investment (transportation and communications), however, governments tend to maintain their own fiscal efforts in the provision of these goods. Foster and Rosenzweig (2001) and Pande (2003) show that when disadvantaged groups in India--the landless poor and scheduled castes--were newly able to elect their own representatives, more government resources flowed toward these disad- vantaged groups. But the increased flows took the form mainly of increased access to government jobs and targeted welfare transfers, not of improved education or access to other broad-based social services--even though the resource cost of such improvements, relative to the likely welfare gains, might have been much smaller.2 The preference of politicians for "pork barrel" or targetable spending is not the exclusive province of developing economies. It has been widely documented in the United States (see, for example, Mayhew 1974). The problem for development is that many governments have exaggerated preferences for targeted expenditures. For example, a large and disproportionate share of expenditures in health and education typically flows to the salaries of teachers and health workers, yet absenteeism and shirking are rampant, so that no effective services are provided in many cases. In India, where recurrent expenditures account for 98 percent of government expenditure on primary education, salaries account for 96 percent of recurrent expenditures (Tilak 1993, p. 60). Yet field investigations in rural areas, particularly in the north, reveal that teacher absenteeism is endemic, with almost two-thirds of the teachers employed in the sample schools absent at the time of the investigators' unan- nounced visits (Drèze and Gazdar 1996; Weiner 1991; Prasad and Eswara 1987). Governments use resources to provide (targetable) jobs rather than (less targetable) high-quality services. Service delivery falls far below the levels that even poorer countries can afford, suggesting that the tradeoffs between targeted and nontargeted public expenditures are much steeper in developing economies than in developed countries. Philip Keefer and Stuti Khemani 3 The problem is not simply "capture" of the levers of government policymaking by the "rich." Even explicitly propoor programs frequently take the form of targeted (and often unsustainable) redistribution programs, such as free food and temporary employment in public works, even when cheaper broad-based programs to improve basic health and education services would have a bigger impact on welfare. At the same time, apparently broad-based programs benefit mainly special interests or the rich. Bates (1981) has shown for Africa that agricultural subsidies, such as price protection and subsidized electricity, disproportionately benefit middle-size and large farmers, who grow the protected crops and are more likely to engage in capital- intensive, power-driven private irrigation. India provides a wealth of examples, ranging from power subsidies to fertilizer subsidies, exhibiting the same characteristics. Once again, such biases are well-known in all democracies but are particularly extreme in many developing areas. It is similarly unlikely that failures of social service provision are rooted in lack of citizen demand. A common finding of village studies and household surveys in India is that members of disadvantaged groups widely view education as the most promising chance for a better life for their children (Drèze and Sen 1995). A thought experiment also makes clear that demand-side considerations alone cannot explain why the landless poor, even if they are fully informed about the value of education, might still prefer targeted jobs. Suppose that a village with 30 families must choose between a single job for a randomly chosen individual in the village or the provision of a teacher from outside the village who would instruct one child from each family. Assume that literacy raises the present value of each child's lifetime income by 30 percent, that literacy takes five years of education to achieve, and that the present value of each child's income, absent education, would be the same on average as that of the average wage earner in any of the village families. Then the total value to all 30 families of having the teacher (assuming the teacher is guaranteed to stay for five years) would be 0.3 times 30 times the present value of the average wage earner's lifetime income. To match this wealth effect, and even assuming that the job offered in lieu of a teacher was guaranteed for the life of the beneficiary, the job would have to pay 10 times the average wage to make the expected value of the job to the village the same as the expected value of the teacher. The demand for education would there- fore have to be extraordinarily low to explain the apparent preference of the poor in many developing economies for jobs and subsidies over schools and clinics.3 However, because the poor are often observed to make large sacrifices to educate their children despite imperfect credit markets and discount rates that are not par- ticularly low, it makes sense to look beyond demand-side issues in asking why voters in some countries do not pressure politicians to provide high-quality education. 4 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Why does Political Competition Fail to Induce the Optimal Provision of Public Goods? It is easy to show the conditions under which, in a simple median voter model with no political market imperfections, governments always provide the optimal level of public goods. In the real world, where such imperfections are endemic, political incentives to provide public goods fall far below the optimum. One key imperfection that upsets the relationship between citizen preferences and politician decisionmaking is that citizens cannot easily observe the contributions that a specific politician makes to policy nor the connection between policy and their own welfare. Imperfect information makes it difficult for citizens to assign credit or blame. Politicians therefore have heightened incentives to seek rents rather than to provide public goods. Another imperfection arises because citizens are heterogeneous with respect to income, ideology, religious belief, language, and the value they attach to different public goods, occupations, and locations. These differences affect citizen preferences for government policies and public goods. Moreover, if citizens prefer politicians from a particular ethnic group at all costs, these preferred politicians are not likely to face opposition if they perform poorly. The lack of credibility of political competitors is the final major disruption in the relationship between citizens and government. In the worst case, when political challengers cannot make credible promises to any voters, the threat of political com- petition imposes practically no constraint on incumbent politicians. Incumbents, knowing that voters do not believe challengers' promises to do a better job, feel freer to underperform and seek rents. These imperfections--information asymmetries, social polarization, and non- credibility of political promises--undermine the role of elections in guaranteeing accountable and responsive government. Each imperfection is explored in greater detail in the sections that follow. Substantial evidence suggests that all are important sources of distortion in political markets. Reform efforts designed to improve the access of poor people to better-quality social services need to address them explicitly to be successful. Impact of Imperfect Information on the Provision of Social Services to Poor People Information constraints reduce the ability of citizens to hold politicians accountable and encourage politicians to cater to special interests, thus distorting incentives to provide social services to poor people. Even if poor people actively participate in the political process, imperfect information about individual politician's contributions to social service improvements can distort incentives. Broad segments of the poor pop- ulation might be particularly disadvantaged in accessing information--because of Philip Keefer and Stuti Khemani 5 illiteracy, limited mobility, and underdeveloped media for mass communications-- creating space for organized and informed special interests to push for narrowly tar- geted policies at the expense of broad services. Social services have several characteristics that present substantial information challenges to citizens attempting to assess a politician's contributions and to structure punishments and rewards based on the quality of services available to them. Because of these information problems, politicians prefer to expend resources on constructing and staffing schools and clinics, even if they remain empty and unused, for example, than on improving the quality of services. Politicians get some credit for easy-to-observe buildings and jobs but little or no credit (or blame) for the quality of services available. Mani and Mukand (2002) show that if elections are intended to enable voters to select the most competent candidates, then resource allocation will be biased against public goods whose outcomes are more noisy and harder to use to assess politician ability, and politicians will have an incentive to provide goods that are better signals of high ability. First, the quality of services that actually affect citizen welfare, such as health and education, depends more on the day-to-day behavior of providers than on any easily observable actions of remotely located politicians. Although policy decisions by poli- ticians, such as in teacher management or preventive health care initiatives, can strongly influence provider incentives and behavior, citizens need specific and substantial information to determine which policies have contributed to better-quality services. Second, measurable benefits (or costs) may not emerge until several years after the policy action has been taken, making it difficult for voters to reward (or punish) political agents within a particular electoral cycle. The effects of education reforms, for example, are typically difficult to verify until a cohort of students has been exposed to them over an extended period. Third, providing service jobs for constituents is the sort of activity for which politi- cians in low-information societies can easily take credit. Like roads and buildings, jobs are easily targeted and highly visible. But it may be politically more rewarding to give jobs to unqualified people who have fewer other job opportunities and therefore owe a greater debt to politicians than do highly qualified people. Gazdar (2000) and the World Bank (1998, 2001) show that nonprofessional qualifications are key factors in the placement of teachers in Pakistan. Similar reports from the Dominican Republic suggest that the posting of teachers is highly discretionary and not clearly related to educational concerns (Keefer 2002b). Fourth, if poor people are particularly uninformed participants in the political process, broad social services can also suffer as politicians make policy concessions to organized and informed special interests. Uninformed voters may be unduly swayed by political campaigns and advertisements rather than by independent eval- uations of the performance records of competing candidates. Special interests--single 6 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) firms, groups of manufacturers, farmers, public sector employees, or subsidized retail store owners--purchase narrowly targeted policies by providing campaign finance (Baron 1994; Grossman and Helpman 1996). These special interests need not be rich, but they are, by definition, informed.4 There is little direct evidence on the information available to or used by voters who are poor. Indirect evidence demonstrates the general importance of information for political accountability, however. One piece of evidence concerns the impact of exogenousshocks on voter behavior. Voters regularly punish politicians when exogenous conditions are unfavorable and reward them when those conditions are favorable. In Indian states, governments tend to lose elections in years of poor rainfall. In the United States, voters in oil-producing states tend to reelect incumbent governors during global oil price booms andvotethemoutofofficewhenoilpricesdrop(Wolfers2002).Unobservableordifficult- to-observe shocks that disrupt the connection between electoral success and govern- ment performance can undermine government incentives to perform at a high level. A second strand of evidence on the importance of information comes from the literature on electoral cycles in monetary and fiscal policies. Large and systematic elec- toral cycles in policies are more likely to occur when voters are ill-informed about the competence of politicians and their contributions to a voter's welfare.5 Cross-country evidence indicates that electoral cycles in monetary and fiscal policy are signifi- cantly larger in developing economies (Block 2002; Schuknecht 2000; Shi and Svensson 2000). Shi and Svensson (2000) establish a direct link between electoral budget cycles and limited information available to voters. They show that some of the difference in the size of political budget cycles across countries is due to varia- tions in access to free media--the greater the access, the smaller the observed budget cycles. Evidence of election cycles in the composition of public spending in India, a long-standing democracy, is also suggestive of the impact of information constraints on service delivery. Khemani (2004) finds that as elections near, state governments increase expenditures on easily observable public investment projects and decrease them on more broad-based public services. State governments in India also increase targeted tax breaks to narrow groups of producers before elections, actions that may be intended to elicit special interest contributions to campaign financing. To mitigate information problems, the evidence suggests that improving account- ability does not require that all voters be deeply informed, only that most voters share some information about political responsibility for key policy outcomes. Studies in the United States have shown that voters adopt simple voting criteria based on very limited information about politics and public policies (Ferejohn and Kuklinski 1990). Fiorina (1990) emphasizes that the information that people use when they vote comes from the ordinary performance of social and economic roles and is therefore "free." Fiorina and Shepsle (1990) and Chappell and Keech (1990) argue that citizens can employ voting rules requiring very little information and still motivate politicians to pursue policies in their interest. Ferejohn (1990, pp. 8­9) captures this Philip Keefer and Stuti Khemani 7 process as follows: "Find a way to get the electorate to commit itself to act as though it is a simple principal with a one-dimensional set of rewards. In this way, incumbents will be prevented from taking advantage of the conflicting interests in the electorate." This suggests the potential for citizens to coordinate information about particular policies and thereby coordinate voter responses for greater political accountability. Coordination problems emerge when the information diverges greatly across the population or is unavailable. The content of "free" information can vary widely across the electorate depending on the differentiation of occupations and social set- tings in the economy and the reach of mass media. For example, do poor people listen to state-controlled radio while well-off groups read independent newspapers? Only when a large enough part of the electorate is exposed to the same information is there a threat of coordinated voter action in response to underperforming incumbent governments (Ferejohn 1990; Iyengar 1990; Ottati and Wyer 1990). There is some evidence for the role of mass media in spreading and coordinating information among the electorate and thereby improving political accountability. Strömberg (2001) finds that between 1933 and 1935 in the United States federal assistance to low-income households was greater in counties where more house- holds had radios and were thus more likely to be informed about government policies and programs. The spread of radio particularly improved information access for rural voters, accounting for as much as 20 percent more in social assistance funds to a rural county than an identical urban county. Besley and Burgess (2003) show that state governments in India respond more to declines in food production and crop flood damage through higher public food distribution and calamity relief spending where newspaper circulation, particularly in local languages, is greater. Such evidence confirms that when voters are informed about particular policies they are able to extract greater resources and better performance from political agents. But it does not tell us whether an increase in media access improves govern- ment incentives either to target the neediest citizens with transfers or to invest in public goods benefiting large segments of the population. It could be that the mass media make it easier for politicians to take credit for targeted payoffs to particular constituencies, leading them to reduce expenditures on public goods or on broad- based social programs. In countries where the reach of information media is limited and difficult transportation and communications systems divide the electorate, coor ination is less likely. The information base of poor rural citizens of developing economies might therefore be skewed in a way that diminishes their ability to hold elected officials accountable for the quality of public services. This evidence suggests that reforms to solve information problems should include loosening restrictions on the media, including dependency on government advertising, and reducing barriers to entry for media firms. But promoting competitive media markets may not be enough to ensure adequate coverage of government performance. Competitive information media might provide inadequate information if the markets 8 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) they serve consist largely of better-off groups that do not consume public services. They also might not be able to solve the problem of voter coordination, because dif- ferentiated products providing different types of information might appear when barriers to entry are low.6 External interventions, such as journalist training, subsidized information transmission, or information campaigns by civic society organizations, may be useful in promoting the diffusion of the information needed for political accountability, particularly in poorer countries. There are numerous examples of successful information campaigns on public ser- vices, but no rigorous evidence for what kinds of information dissemination strategies systematically alter the nature of political competition. A well-known example is the public service report cards developed by the Public Affairs Centre in Bangalore, India (www.pacindia.org). Still, some types of information and dissemination strategy would seem likely to make more of a difference in improving political incentives than others. Information about broad public sector performance aggregates, for example, whether based on surveys, budget studies, or report cards, is likely to be politically relevant only if it provides individual voters with some notion of how their own gov- ernment representatives have hurt or helped them. Information is more likely to promote reform when information providers are able to coordinate voter responses to poor service provision by moving beyond information collection and dissemination and taking a more aggressive stance toward political accountability. Thus information campaigns need to tell citizens not just how bad services are and which are worse than others--information that citizens tend to have already--but also how bad the services are in their neighborhood relative to others, who is responsible, and whether specific policy reforms have resulted in improved services. Finally, the information burden on voters is lightened if the key actors are political parties rather than individual politicians. It is difficult for voters in one electoral district to hold their elected officials responsible for problems in social service delivery that afflict the entire country. The influence of individual legislators on overall govern- ment policy is difficult to establish, and in any case voters know that individual legis- lators are likely to have little influence over public services. Where parties are institutionalized, however, and exercise substantial control over their members, voters can more easily assign responsibility for poor performance. Social Polarization and Provision of Services to Poor People Social services for poor people also suffer when societies are deeply divided, an acute problem in many developing areas. Studies of electoral politics in India show that identity characteristics along ethnic, linguistic, and religious lines dominate political behavior (Weiner and Field 1974). An experimental study in Benin confirmed that social cleavages along ethnic lines often play a prominent role in determining political behavior (Wantchekon 2003). In socially or ethnically fragmented societies voters Philip Keefer and Stuti Khemani 9 tend to vote for candidates that they identify with most closely, whatever their per- formance records. Political competition among parties also concentrates on identity issues, and candidates are nominated from constituencies largely on the basis of demographic calculations of ethnicity and religion. Like uninformed voters, polarized voters are less able to hold politicians accountable for their performance in office. There is compelling evidence to show that the provision of public goods suffers under these conditions because politicians in polarized societies rarely internalize the societywide costs and benefits of their policy decisions. In a cross-country study, Easterly and Levine (1997) find that ethnic diversity is neg- atively correlated with the quantity and quality of such public goods as measured by the percentage of paved roads, the efficiency of the electricity network, and years of school- ing of the population. For cities in the United States, Alesina and others (1999) show that shares of public spending on productive public goods--education, roads, sewers, and trash pick-up--are inversely related to the city's ethnic fragmentation. Miguel (2001) concludes thathigher levels of local ethnic diversity in Kenya are associated with sharply lower contributions to primary school funding and worse school facilities. The effects of identity politics are especially dramatic when minority groups lack political power. In India, Betancourt and Gleason (2000) find that districts with a higher proportion of traditionally disadvantaged groups--scheduled castes and Muslims--receive lower public inputs in health and education. Similarly, Banerjee and Somanathan (2001) find that districts with a higher share of scheduled tribes receive significantly fewer desirable public goods. They also find some evidence linking ethnic heterogeneity and public goods delivery to underlying political incentives-- districts that are ethnically fragmented are also likely to be politically fragmented, with elections in these districts having a larger number of contestants and a smaller vote share for the winning party. There is also substantial anthropological and anecdotal evidence that disadvantaged groups in India are systematically excluded from using public goods within their own villages by social processes of discrimination. Micro-level case studies and survey evidence from India show that inequality in education access and achievement is significant within some villages, with privileged castes having enjoyed nearly universal adult literacy for several decades, whereas disadvantaged castes in the same village still have literacy rates close to zero (Drèze and Sen 1995). Adverse policy effects do not emerge in all countries with profound caste, reli- gious, tribal, racial, or other differences. The affinity-based voting patterns observed in India and Nigeria are not evident in the United States or Western Europe, for example. Surveys of voting behavior in the United States in the 1950s and 1960s (Campbell and others 1966) find that an overwhelming determinant of voting behavior is the ideological identification of voters with political parties. In this case, though, ideology has significant policy roots in citizen identification with the perceived position of political parties on a wide range of social and economic issues. 10 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Devising interventions to soften the social divides in a country thus requires understanding why these divides drive politics in some countries but not in others. One possibility is simply that members of the same ethnic, religious, or social group may value each others' welfare more than they do the welfare of members of other groups. Historic antipathies between groups may even mean that reductions in welfare for one group improve welfare for others. In the end such arguments simply say that group preferences are buried in the utility functions of citizens. To the extent that this is the case, the design of appropriate interventions is best guided by psychological and sociological insights into what sorts of information about and interactions between different groups would best reduce the animosity that divides them. A second explanation, the focus of the next section, is that when political competitors are generally not credible, voters often believe only the promises of competitors with whom they have personal connections. In the United States parties have established credible positions on certain issues. Voters then appear to be "ideologically" committed to the party when they are simply committed to the policy positions with which those parties are credibly associated. In India and Nigeria, though, the problem is the reverse. Voters believe promises only from political candidates within their own ethnic or religious groups; those promises are therefore necessarily narrow and tar- geted to members of that ethnic group. Political promises for broad-based public ser- vices that cut across social groups are not credible. Examination of the effects of increasing the political power of minority groups suggests that both animus and credibility may explain the role of social polarization in political competition. India, for example, instituted political reservations for scheduled castes and tribes in national and state legislative assemblies. Pande (2003) finds that mandated reservations in state legislatures resulted in increased public sector job quotas for scheduled castes, but significantly lower resources allo- cated to education. The antagonism of the majority toward the scheduled castes and tribes explains why benefits (jobs) grew along with political influence for scheduled castes and tribes. But because the new representatives of the disadvantaged groups could make credible appeals only to their own ethnic groups, they favored targeted rather than broad public goods, even when broad public goods would have brought larger benefits to their groups per rupee expended. Credibility and the Provision of Social Services As discussed, politicians may be unable to make credible promises to citizens because of imperfect information--when citizens cannot easily verify whether public goods have been provided, politicians have an incentive to underprovide relative to what they promise. Promises about social service provision are particularly difficult to verify. Knowing this, citizens disregard promises to improve social services, and politicians stop making such promises. Another reason for the inability to make credible promises Philip Keefer and Stuti Khemani 11 is the absence of penalties for reneging on promises. In particular, citizens do not believe the promises of a politician who would experience no reputational loss from reneging, such as a politician who is not expected to seek office in the future or who has no history of fulfilling promises. When preelection promises are not credible, elections become less effective instru- ments for holding politicians accountable. Even in the best case, when voters can coordinate on ex post performance standards and turn out incumbents who fail to meet them, public goods are still underprovided, targeted transfers are not provided at all, and rent-seeking is high (Persson and Tabellini 2000). In the worst case, if voters cannot coordinate, elections are useless and rent-seeking prevails (Robinson and Verdier 2002). Clientist Governments. In low-credibility environments, elections would thus have little meaning and politicians would never authorize narrow, targeted expenditures. In countries lacking established political parties with clear policy reputations, how- ever, politicians typically feel threatened by elections and are deeply concerned about their ability to deliver targeted transfers to constituents. Keefer (2002a) suggests that these outcomes emerge when politicians can make credible promises only to voters with whom they have built a personal reputation. Such ties emerge most clearly in the "patron-client" relationships that a large body of literature identifies with the politics of developing economies. Credible politicians are, for example, locally influential people who have helped families with loans or jobs or assistance with legal or bureaucratic difficulties. In the absence of well-developed political parties or national party leaders who are more broadly credible to voters, the promises of such influential people are all that voters can rely on in making electoral choices. Patrons have relatively weak incentives to provide public goods because they benefit nonclients as well as clients. Promises of buildings and government jobs become the currency of political competition at the expense of universal access to high-quality education and health care. Narrowly targeted goods can be provided to individuals and small groups of voters and therefore are clearly seen as evidence of political patrons fulfilling their promises to clients--recalling that only clients believe promises in the first place. Universal access by definition is not easily targeted. Improvements in quality are also difficult to target, both because they tend to involve managerial improvements that spread throughout the system and because they are difficult for voters to attribute to politicians. For example, if teacher quality or attendance has improved, voters cannot distinguish whether that happened because of their own pressure on the teacher, because the teacher decided to do a better job, because of a generalized reform in teacher quality that is affecting all teachers, or because of the targeted intervention of a particular politician. Evidence supports the notion that clientelist governments have a stronger than average preference for targeted infrastructure provision, are more corrupt than 12 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) average, and underprovide education. For example, Keefer (2000a) argues that in young democracies political competitors, especially political parties, are less likely to have built up reputations through repeated electoral cycles. Political competitors in young democracies are therefore less likely to be able to make credible promises to all voters and are more likely to rely on clientelist promises. Targeted spending--public investment--is higher in young democracies than in old, secondary school enrollment is lower, and corruption is greater. Qualitative evidence from Pakistan supports this logic. Competitive elections were a regular feature of the political landscape of Pakistan during the 1990s, but credible promises by political parties or political leaders to voters were supported largely by clientelist relationships and related to targeted transfers rather than broad public policies. There is little evidence of political competition on the basis of broad policy promises, nor of distinctions among the major parties by their stances on broad policy issues. In this kind of environment the provision of broad-based public goods would be expected to be low and the provision of targeted goods to be emphasized. Indeed, compared with countries with similar incomes per capita and demo- graphic characteristics (age and proportion rural), primary school enrollment in Pakistan in 2000 was 20 percentage points less than would be expected. Access to potable water, however, was 25 percentage points higher than expected. Invest- ments in potable water, particularly in rural areas where they consist largely of well- drilling, are particularly easy to target and their benefits are immediately accessible and observable (World Bank 2002). Short Time Horizons. Pakistan also illustrates the importance of a long time horizon for ensuring the credibility of political promises. For decades Pakistan has been marked by political instability, with the tenure of incumbent leaders frequently cut short. During the 1990s, no elected government completed its term, so voters and elected officials had short time horizons. This would have generated a marked preference for govern- ment provision of goods that generate benefits immediately and would not be depen- dent on the decisions of future governments. Political competitors with short time horizons are unable to credibly promise to implement projects that require a long time to bear fruit. Promises that concern jobs or public works projects, by contrast, can be fulfilled with little delay after an election, and their benefits can be quickly realized and recognized by voters. Government services, such as education, become considerably less useful to politicians because these services have long time horizons. Schooling yields few benefits until students achieve literacy and numeracy, which may not occur before grade 4. In Spain, returns to education are low before grade 10, but returns spike after completion of grade 4 (Sanmartín 2001). Glewwe and Jacoby (1994), in examining the determinants of student achievement in Ghana, gave sub- jects a basic reading and mathematics test. Performance was so poor for students who had not advanced beyond grade school that they were excluded from the sample, and Philip Keefer and Stuti Khemani 13 the study focused on middle school achievers only. To the extent that education also serves a signaling function, the horizon problem worsens: Receipt of formal credentials generates a discrete jump in the returns to previous years of schooling, but the credential is received only after a certain number of years of schooling.7 The horizon problem can have a significant impact on voter decisionmaking. Consider again the hypothetical case of the 30 families, each with a school-age child. These families can choose between a politician who promises to provide a job for one member of the community and one who promises to staff the school with a qualified teacher who is unrelated to the 30 families. Before the election the families do not know which one will get the job, so the expected value of the job to each family is 1/30 of the salary. If they know that the teacher will stay for five years, long enough for each child to become numerate and literate, then each family benefits from a 30 percent increase in lifetime income. If the teacher will stay less than five years, there is no benefit to education. These families prefer the politician who promises the teacher only if the promise is credible over the entire five-year period. Institutionalized Parties. There is no direct evidence of the horizon effect on social ser- vice delivery, but there is documentation related to other long-term promises of govern- ments. Arguing that the older a regime (number of years as a democracy or dictatorship) the longer the horizon of the political decisionmakers, Clague and others (1996) con- sider the security of property rights, which is clearly sensitive to the horizons over which government actors can make credible promises. They present evidence suggest- ing that the security of property rights is higher in both older democracies and older dictatorships. Keefer and Knack (2002) find that productive public investments, whose payoffs are largely in the future, fall when property rights are weak, whereas unpro- ductive public investments (corruption and rent-seeking) rise. These results suggest that social service provision, particularly education, is likely to suffer when government officials can make credible promises only over short time horizons. There is nothing inevitable about the formation of policy reputations, however. Although young democracies are less likely to exhibit institutionalized political parties, institutionalization is far from an automatic development as democracies age. For example, though political parties could gain by attempting to build strong policy repu- tations, there are strong political incentives in the opposite direction. In socially polar- ized countries, for example, there are political rewards to developing a reputation for favoring one group at the expense of another. As Glaeser and Shleifer (2002) show for the city of Boston and as the experience of Zimbabwe and countless other countries throughout history has clearly shown, politicians can often gain political advantage by fanning ethnic divisions. In the limit, as in Boston or Zimbabwe, electoral advantage is cemented by simply pushing the out-of-power ethnic group out of the jurisdiction. The difficulties of developing political party reputations are more general than this, however. In an environment where no party is credible, the most successful 14 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) parties have leaders who are personally credible and have national reputations. If these leaders build up the party's apparatus and reputation separate from their own, they cement their party's hold on power but loosen their personal hold. Once the party is institutionalized, the barriers to entry for party leadership fall. Competitors no longer need the national profile that is required for success in the absence of insti- tutionalized political parties. Given this tradeoff, it is not surprising that so many parties, for example, in Pakistan or Bangladesh, have failed to institutionalize. Finally, reputation can develop for the "wrong" policies when there is uncertainty about the most effective instruments for achieving particular public objectives. Political competition may lead to a bad equilibrium in which only promises for inefficient pol- icies are credible. New democracies established in poor, agricultural economies have historically adopted poverty-reduction strategies focusing on subsidies for consumption and agricultural production, at the expense of broad public services that may have higher returns in terms of poverty impact and economic growth. Once political reputa- tions are established for particular policies, even if the policies are suboptimal, they will receive greater public resources than they would if all political promises were credible. Given the obstacles to reputation building, interventions to strengthen the credi- bility of preelectoral promises of political competitors are difficult to devise. Many organizations seek to improve the professionalism of political parties. This can be important as long as it advances the policy basis of the party. Professionalization can also enhance the development of party machines, however. This may not be entirely bad: Relative to a personalized political system, a patronage-based party is likely to have a longer horizon and is more likely to provide public goods. However, competition between credible, policy-based parties that engage in some patronage is almost cer- tainly better for public policy than competition between more purely machine-based parties. The Dynamics of Political Competition and the Provision of Social Services: Uttar Pradesh and Kerala Some of the most striking contrasts in basic health and education outcomes exist between neighboring countries or regions within the same country with comparable levels of economic development--between Sri Lanka, Thailand, and the southern states of India on the one hand, and Pakistan, Bangladesh, and the northern states of India on the other. Drèze and Sen (1995, 1996) have examined this contrast between the northern state of Uttar Pradesh and the southern state of Kerala in India. The two states have almost identical per capita income and poverty rates, but dramatically different outcomes in health and education (table 1). Human develop- ment outcomes are comparable to those in some of the richest nations of the world in Kerala but to some of the poorest in Uttar Pradesh. Philip Keefer and Stuti Khemani 15 Table 1. Contrasting Experience in Social Services, Kerala and Uttar Pradesh Indicator India Kerala Uttar Pradesh Per capita income in current prices, 5,583 4,618 4,012 1991­92 (rupees) Poverty headcount ratio, 1987­88 (%) 44.9 44 47.7 Real per capita public spending on education, 228 309 147 1985­92 (1992 rupees) Real per capita public spending on health, 70 82 49 1985­92 (1992 rupees) Literacy rate, ages 7 and older, 1991 (%) Female 39.3 86.2 25.3 Male 64.1 93.6 55.7 Share of rural children ages 12­14 never enrolled in school, 1986­87 (%) Female 51 1.8 68 Male 26 0.4 27 Share of people ages 6 and older who have completed primary education, 1992­93 (%) Female 28.1 60.5 21.4 Male 48.6 65.8 47.3 Share of children ages 12­23 months who have 30 11 43 never been vaccinated, 1992­93 (%) Share of recent births preceded by an antenatal 49 97 30 check-up, 1992­93 (%) Infant mortality rate (per 1,000 live births), 1990­92 80 17 98 Proportion of villages with medical facilities, 1981 14 96 10 Source: Drèze and Sen (1995, 1996); public spending numbers from annual publications of the Reserve Bank of India Bulletin on state finances in India; India-wide numbers on public spending only for 15 major states. Drèze and Sen attribute these stark differences in outcomes to differences in public action. Their analysis is reinterpreted here to emphasize the differences in political incentives between policymakers in the two states to provide social services to all. There are striking differences in real per capita public expenditures on health and edu- cation in the two states (figure 1). In each decade between 1960 and 1995, average real per capita spending in Kerala has been more than double that in Uttar Pradesh. Kerala allocated 45 percent of public expenditures to education and health services in the early decades of democracy, and less than 30 percent to state administration--largely for jobs for nonservice providers (figure 2). During the same period, Uttar Pradesh (and most other Indian states) did the reverse, allocating 16 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Figure 1. Per Capita Public Spending on Health and Education in Kerala and Uttar Pradesh 1960­95 (1992 rupees) 600 500 gnid 400 en sp a it 300 cap erp 200 eal R 100 0 1960 1965 1970 1975 1980 1985 1990 1995 Year Kerala Uttar Pradesh Note: Public spending data refer to the current account, which includes the bulk of spending on health and education. Source: Nominal public expenditures, Reserve Bank of India Bulletin (various issues); price deflator and population, Ozler and others (1996). Figure 2. Distribution of Public Spending in Kerala on Health and Education and State Adminis- tration (percent of current account spending) iclbu 0.6 Health and education 0.5 Pfo ng 0.4 State administration iontr 0.3 Spendi 0.2 opor P 0.1 0 196 196 197 197 1980 198 199 199 0 5 0 5 5 0 5 Year Note: Public spending data refer to the current account, which includes the bulk of spending on health and education. State administration includes interest payments. Source: Reserve Bank of India (various issues). Philip Keefer and Stuti Khemani 17 Figure 3. Distribution of Public Spending in Uttar Pradesh on Health and Education and State Administration (percent of current account spending) 0.6 Health and ublic 0.5 P education of gni 0.4 State administration tion 0.3 Spend0.2 0.1 Propor 0 1960 1965 1970 1975 1980 1985 1990 1995 Year Note: Public spending data refer to the current account, which includes the bulk of spending on health and edu- cation. State administration includes interest payments. Source: Reserve Bank of India (various issues). almost 50 percent of public resources to state administration and less than 25 per- cent to health and education (figure 3). The quality of public services is indisputably harder to target, but here the evidence is clear: The quality and effectiveness of public spending are higher in Kerala. Drèze and Sen review extensive evidence showing that public schools and health clinics in Kerala have high teacher and physician attendance and are well supplied, adequately maintained, and widely used. Many facilities in Uttar Pradesh simply lie abandoned. Why did these differences emerge? Why do they persist? Why, for example, can a village school in Uttar Pradesh be nonfunctional for as long as 10 years because of teacher absenteeism and shirking, without any collective protest, whereas in Kerala there are massive demonstrations at the nearest district office if a primary health center is unstaffed for a few days (Dreze and Sen 1996)? Both states are in the same sovereign nation, and both have the same formal political and legal institutions. Both have had regular elections to their legislative assemblies since the 1950s in accordance with constitutional procedures and monitored by the Election Commission of India. The answer must lie in the dynamics of political competition rather than in differ- ences in the political institutions themselves. These include differences between the two states in access to information, social polarization, and the credibility of political promises. Kerala's voters have been better informed and less polarized on nonpolicy dimensions than voters in Uttar Pradesh, and Kerala's political parties compete on credible platforms for delivering broad social services, rather than on clientelist platforms as in Uttar Pradesh. 18 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Kerala entered the era of democratic elections with a politically active electorate and a literacy rate of about 40 percent, whereas Uttar Pradesh did so with a population of largely illiterate voters (less than 18 percent even in 1961) with limited access to widely published information. Massive voter illiteracy suggests that voters were poorly informed about the connection between political decisionmaking and their well-being, undermining the ability of any party to make credible promises to voters. Reputation is difficult to establish if voters are poorly informed. Under conditions of visible and often coercive social discrimination, organization of the victims of dis- crimination was very costly. Active institutions of social discrimination effectively prevented so-called backward castes from accessing public services or participating in political processes. Although caste issues also existed in Kerala, poor and socially disadvantaged voters were more likely to participate in political processes to influence policy issues affecting their welfare (Chander 1986; Nossiter 1982). These differences affected political party competition and the likelihood that credible political competitors would emerge to contest elections. Another key political variable was political party competition. Following indepen- dence, the Congress Party was the dominant party in India. It had the widest national presence, the most solid reputation based on its association with the fight for independence, and the longest history as a political party. Congress espoused a common socialist platform throughout India, focusing on government-led develop- ment and redistribution to the poor. At the state level, however, there was substantial variation in voters' perceptions of the Congress Party, the extent of competition from other parties, and the types of credible promises to voters that the party could or had to make to be successful. In states like Uttar Pradesh the Congress Party did not confront vigorous competi- tion from other credible and well-organized parties. In other states, like Kerala, regional parties and emerging communist parties mounted credible challenges to the Congress Party. The Communist Party was particularly active in Kerala and invested substantially in mobilizing the poorest voters. By interacting repeatedly with these voters, the party was able to make credible promises to them (Chander 1986; Nossiter 1982). This was not inevitable. First, the ability of the Communist Party to mobilize large numbers of activists to establish personalized contact with voters is unusual. Without this asset, the party would have been less able to establish its credibility in competition with the Congress Party. Second, leaders of the Communist Party could have chosen to use these personalized contacts mainly to focus on jobs and targeted services to core constituencies or to strengthen their own positions in the party. Although these activities were certainly a part of the message to voters, these activities were also used to establish the party's policy credentials. Competition in Kerala was therefore between two credible political parties, the Congress and Communist Parties, both able to make promises to broad segments of society and both driven by elections to make promises that would best serve the Philip Keefer and Stuti Khemani 19 needs of broad segments of the population--such as high-quality social services. This diminished the importance of patron-client relationships to an extent that was unparalleled in any other state. In Uttar Pradesh, such competition did not develop. The Congress Party in Uttar Pradesh confronted opposition only from personalized parties rather than from more institutionalized parties. Because the opposing parties were not broadly credible, the Congress Party could win elections with exceedingly modest promises--winning on average more than 70 percent of the seats in the state assembly (Butler and others 1995). In Kerala, political credibility was reinforced as promises were kept. The first two decades of elected state government saw dramatic improvements in human develop- ment indicators. Infant mortality fell 43 percent between 1956 and 1966 (Krishnan 1991). Much of this decline has been attributed to high rates of female literacy (expanding through state education programs), but it was clearly facilitated by access to public primary health care and state-sponsored programs of infant and child immunization (Zachariah 1992). In Uttar Pradesh, meanwhile, absenteeism and other indications of widespread shortfalls in service delivery have persisted since the beginning, suggesting that there have been few political pressures to use resources for purposes other than political patronage. Perhaps as a reflection of more informed voting along policy dimensions in Kerala, and greater credibility of political parties, voter turnout has been higher than in any other state, averaging more than 75 percent even in the first few state elections. In Uttar Pradesh, turnout averaged 55 percent in the first few elections, suggestive of greater voter apathy on policy issues because of a combination of lack of information, greater social polarization, and lack of credibility of political parties. Literacy has slowly climbed in Uttar Pradesh, and democratic institutions have pen- etrated the social fragmentation bit by bit over the past 50 years, especially through the political mobilization of traditionally repressed lower castes. However, consistent with the earlier discussion of the obstacles to building reputation, social services have not improved with more competitive elections, the emergence of new parties, the increased integrity of elections, and the representation of lower castes. Instead, new parties have either been personal vehicles for party leaders or have built themselves up as vehicles for representing narrow segments of the population. In either case political incentives to improve broad-based access to quality social services remain weak. Ethnic or caste-based political parties are an important new development in Uttar Pradesh. In an otherwise noncredible environment there is political advan- tage in mobilizing voters along caste or ethnic lines, because the credibility of prom- ises is greater when directed to a single group. Among the three main parties competing in the state today, the Bharatiya Janata Party appeals to upper caste Hindus; the Bahujan Samaj Party to so-called backward castes, scheduled castes and tribes, and minority religion groups; and the Samajwadi Party to similarly 20 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) marginalized religious and caste groups. These parties compete on explicitly clien- telist platforms. The Bahujan Samaj Party, for example, does not print any election manifesto to explain its platform, but it does publicize the ethnic profiles of its candi- date list to demonstrate commitment to its single-point program of proportional representation for every ethnic group in the bureaucratic institutions of the state (Chandra 1999). The contrast between Kerala and Uttar Pradesh demonstrates that the same formal institutions of democracy can sustain very different forms of electoral competition with substantial impact on the quality of public services. It also demonstrates that the sheer endurance of democracy is no guarantee that political market imperfections will disappear. In fact, path dependency in public policies or persistence of differences in the nature of electoral competition is exactly what would be expected when there are credibility problems in political markets. The analysis here suggests that external interventions to break this path dependency, change the process of political competition, and hence affect the quality of social services should focus on the information processing capacity of voters, social fragmentation, and the capacity of politicians to make credible commitments. Further research is needed to evaluate the impact of ongoing political and institu- tional reforms in the developing world and draw lessons from the variations in insti- tutions across countries. Where poor voters are already politically active, the real issue is to bolster the credibility of political candidates to provide broad social services while reducing political pressures to pursue clientelist policies. Policy Implications and Further Research The Millennium Development Goals call for dramatic improvements in the condition of the poor. A large body of literature suggests that institutions are likely to matter a great deal in this endeavor. How they matter is the subject of this article. The specific circumstances of electoral competition and political decisionmaking have a large effect on social service outcomes for the poor. Information gaps, social polarization, and the absence of credible political competitors lead to the underprovision of government services to the least informed, the most polarized, and the vast majority who do not have personal connections with a powerful patron. There are many examples of reforms intended to improve the ability of poor people to secure their own interests in government decisionmaking (to empower them), including decentralization, legislative reservations for minority groups in India, decentralization in Pakistan, and participatory budgeting in Brazil. These are all effective to the extent that they overcome the underlying political market imperfec- tions that impair government services to the poor. However, as the discussion of the effects of reservations in India suggests, the improvements wrought by these reforms Philip Keefer and Stuti Khemani 21 may be only partial if they leave many important aspects of the political marketplace unchanged. Two types of decentralization in particular are often seen as responses to inade- quate central government attention to social service delivery. One is decentraliza- tion of responsibilities for the provision of local public goods to locally elected village and municipal governments. A second is greater autonomy of decisionmaking for service providers, such as schools and clinics, combined with greater participation of citizens through community-based organizations, such as parent-teacher associ- ations and health committees. The political economy analysis here has implications for both. Decentralization is likely to have a positive impact on social services if it enables voters to become better informed about political responsibilities for local public good provision, if local public good provision is more salient in local elections than in central government elections, and if promises by political competitors are more credible in local elections than in central government elections. This type of decentralization may not succeed if local voters are apathetic to local elections and have little or no information about the resource availability and capability of local governments, if social polarization is more intense at the local level, or if clientelist promises to a few voters are easier to make and fulfill at the local level because of closer social relations between the elected representatives and their clients (Keefer and others 2003 provide a more extensive discussion of these issues). Decentralization of authority to service providers and communities can solve both credibility and information problems, although again not necessarily. Remotely located political agents are not able to credibly promise to improve the quality of services in such transaction-intensive sectors as health and education, because agents are unable to closely monitor service providers. By shifting monitoring to communities, this form of decentralization reduces the information voters need to hold politicians accountable: They need only verify whether political agents have made resources available for schools and clinics.8 But such policies are difficult to enforce because they require politicians to rein in their claim on public employment as political patronage. Complementary efforts to generate and widely disseminate information about the impact of policy changes on social services would make it easier for politicians to take credit for successful policies that increase benefits to broad sections of the population, thereby reducing the need for targeted patronage. Reforms other than decentralization can also reduce political market distortions caused by incomplete information. Independent validation of the quality of public goods and the scope of accomplishment and failure of individual political decision- makers, by clearly communicating to citizens that the information is widely shared, could provide a vehicle for voter coordination. Not only can the press provide this type of validation and coordination but so can external agencies and civil society groups. Survey instruments (such as citizen report cards that consolidate public 22 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) feedback on the state of government services) are potentially powerful vehicles for mobilizing voters around the issue of quality of public services. Credibility, like information, is likely to be particularly problematic in developing economies. Reforms designed to directly improve social service delivery can strengthen credibility by making it easier for politicians to take credit for public goods improvements. Social service reforms are more sustainable if they help politicians build a reputation for broad-based service delivery, as in Kerala. This is challenging, however. It would require that external aid be provided specifically for education and health when politicians have made emphatic, public, and verifiable promises regarding health and education services. This is not common in developing democracies. Finally, social polarization can have harmful effects on social service provision, just as it does on other aspects of civic life. When considering the design of external assistance to support reform, a key consideration should be that criteria unrelated to social cleavages determine access to resources. This may not always be practical, however, as when one group is poorer than the other--redistributional transfers based on income will always favor the poorer group. When social cleavages are purely "taste-based" (a preference for one ethnic group or tribe and a dislike for others), a two-pronged, long-term approach is necessary: educational responses to moderate these tendencies and strong signals from the center that such preferences are inappropriate. Ample experience sug- gests that this approach can soften such cleavages over time. Where polarization is in response to grave imperfections in political and economic markets, however, causing people to retreat into groups with which they have the strongest personal links and the greatest hope of accessing resources, the appropriate response is to remove those imperfections. The profound role that political market imperfections play in development is just beginning to be understood. More research and practical experiments are needed to discover how best to alleviate these imperfections. The decentralization reforms that are taking root in a variety of contexts and with significant institutional variation provide an important opportunity to evaluate one strategy for addressing them. Sys- tematic research on experiences around the world with information campaigns for improving public services is likely to be feasible and promising. Further research would be valuable on information availability and processing by poor voters and on how information provision mechanisms can be institutionalized to enable these vot- ers to create stronger incentives for politicians to improve the performance of public services. Finally, more innovative research is required on credibility in politics. Under what conditions do politicians shift their promises from targeted transfers to broad and continuous improvements in service quality? The rewards to such admittedly dif- ficult research are likely to be enormous, because it would vastly improve our under- standing of the extensive differences in government performance around the world. Philip Keefer and Stuti Khemani 23 Notes Philip Keefer is lead economist in the Development Research Group at the World Bank; his e-mail address is pkeefer@worldbank.org. Stuti Khemani is an economist in the Development Research Group at the World Bank; her e-mail address is skhemani@worldbank.org. This article was originally prepared as a background study for World Development Report 2004: Making Services Work for Poor People. Support and comments from the World Development Report team, especially from Shekhar Shah, are gratefully acknowledged. 1. This focus on fundamental characteristics of electoral competition is complementary to another strand of the literature (reviewed by Persson and Tabellini 2000) looking at the impact of electoral and political institutions--presidential or parliamentary regimes, majoritarian or propor- tional electoral rules--on government decisionmaking and spending. Preliminary work suggests that the magnitude of the institutional results varies significantly with the underlying electoral characteristics of countries, the focus in this article. For example, in poorer or younger democra- cies, where voters are less likely to be informed and political competitors are less likely to be credi- ble, the difference in government spending between presidential and parliamentary systems is less than half the difference in richer or older democracies. The analysis here is therefore a logical first step in a full and comprehensive analysis of the political economy of government expenditures and the poor. 2. In a review of research on public spending and the poor, Van de Walle and Nead (1995) find that resources spent on basic health and education services (broadly targeted fiscal policies) have higher payoffs for the poor than finely targeted food subsidies or other redistribution schemes, in part because of the greater administrative costs and behavioral consequences of the targeted schemes. 3. Demand might be low because returns to education are low, discount rates are high, or severe credit market constraints block access to complementary inputs (books, forgone child labor). 4. Alderman (1988) describes protests in Pakistan by retailers of government subsidized wheat and flour when their livelihood was placed at risk by amendments to the subsidies program. 5. The most persuasive theoretical explanation for electoral cycles emerge from rational expecta- tions models in which asymmetric information and shocks to politician competence over time create conditions under which it is rational for voters to focus on performance closer to elections (Persson and Tabellini 2000; Rogoff 1990). 6. Several studies have examined the conditions of competition in political and economic markets and the resulting incentives for the kind and quantity of information disseminated by the media industry (Djankov and others 2002; Mullainathan and Shliefer 2002; Strömberg 2004). 7. The magnitude of the signaling effect is contentious, particularly for primary and secondary edu- cation. Layard and Psacharopoulos (1974) strongly dispute it, but Hungerford and Solon (1987) claim to find convincing evidence of it. 8. This solution is limited to the extent that central government intervention is needed to remedy disparities in performance or funding across jurisdictions, however. References Alderman, H. 1988. "Estimates of Consumer Price Response in Pakistan Using Market Prices as Data." Pakistan Development Review 27(2):89­107 Alesina, A., R. Baqir, and W. Easterly. 1999. "Public Goods and Ethnic Divisions." The Quarterly Journal of Economics 114(4):1243­84. Banerjee, A., and Somanathan, R. 2001. "Caste, Community and Collective Action: The Political Economy of Public Good Provision in India." Massachusetts Institute of Technology, Department of Economics, Cambridge, Mass. 24 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Baron, David P. 1994. "Electoral Competition with Informed and Uninformed Voters." American Political Science Review 88(1):33­47. Bates, R. 1981. Markets and States in Tropical Africa: The Political Basis of Agricultural Policies. Berkeley: University of California Press. Beck, Thorsten, George Clarke, Alberto Groff, Philip Keefer, and Patrick Walsh. 2001. "New Tools in Comparative Political Economy: The Database of Political Institutions." World Bank Economic Review 15(1):165­76. Besley, T., and R. Burgess. 2003. "The Political Economy of Government Responsiveness: Theory and Evidence from India." Quarterly Journal of Economics 117(4):1415­51. Betancourt, R., and Gleason, S. 2000. "The Allocation of Publicly Provided Goods to Rural Households in India: On Some Consequences of Caste, Religion and Democracy." World Development 28(12):2169­82. Block, S. A. 2002. "Political Business Cycles, Democratization, and Economic Reform: The Case of Africa." Journal of Development Economics 67(1):205­28. Butler, D., A. Lahiri, and P. Roy. 1995. India Decides: Elections 1952­1995. New Delhi: Books and Things. Campbell, A., P. Converse, W. Miller, and D. Stokes. 1966. The American Voter. New York: J. Wiley & Sons. Chander, Jose N., ed. 1986. Dynamics of State Politics in Kerala. New Delhi: Sterling Publishers. Chandra, K. 1999. "The Ethnification of the Party System in Uttar Pradesh and Its Consequences." In R. Ramashray and P. Wallace, eds., Indian Politics and the 1998 Elections: Regionalism, Hindutva, and State Politics. London: Sage. Chappell, H., and W. Keech. 1990. "Citizen Information, Rationality, and the Politics of Macroeco- nomic Policy." In John Ferejohn and James Kuklinski, eds., Information and Democratic Processes. Urbana and Chicago: University of Illinois Press. Clague, Christopher, Philip Keefer, Stephen Knack, and Mancur Olson. 1996. "Property and Contract Rights under Democracy and Dictatorship." Journal of Economic Growth 1(2):243­76. Djankov, Simeon, Caralee McLiesh, Tatiana Nenova, and Andrei Shliefer. 2002. "Who Owns the Media?" Journal of Law and Economics 46(2):341­83. Drèze, Jean, and Haris Gazdar. 1996. "Uttar Pradesh: The Burden of Inertia." In J. Drèze and A. Sen, eds., Indian Development: Selected Regional Perspectives. New Delhi: Oxford University Press. Drèze, Jean, and Amartya Sen. 1995. India: Economic Development and Social Opportunity. Delhi, India: Oxford University Press. ------, eds. 1996. Indian Development: Selected Regional Perspectives. New Delhi: Oxford University Press. Easterly, W., and R. Levine. 1997. "Africa's Growth Tragedy: Policies and Ethnic Divisions." Quarterly Journal of Economics 112(4):1203­50. Ferejohn, John. 1990. "Information and the Electoral Process." In John Ferejohn and James Kuklinski, eds., Information and Democratic Processes. Urbana and Chicago: University of Illinois Press. Ferejohn, John, and James Kuklinski. 1990. Information and Democratic Processes. Urbana and Chicago: University of Illinois Press. Feyzioglu, Tarhan, Vinaya Swaroop, and Min Zhu. 1998. "A Panel Data Analysis of the Fungibility of Foreign Aid." World Bank Economic Review 12(1):29­58. Fiorina, M. 1990. "Information and Rationality in Elections." In John Ferejohn and James Kuklinski, eds., Information and Democratic Processes. Urbana and Chicago: University of Illinois Press. Fiorina, M., and K. Shepsle. 1990. "A Positive Theory of Negative Voting." In John Ferejohn and James Kuklinski, eds. Information and Democratic Processes. Urbana and Chicago: University of Illinois Press. Foster, A., and M. Rosenzweig. 2001. "Democratization, Decentralization, and the Distribution of Local Public Goods in a Poor Rural Economy." Working Paper. Brown University. Available online at http://adfdell.pstc.brown.edu/papers/democ.pdf. Philip Keefer and Stuti Khemani 25 Gazdar, Haris. 2000. "State, Community and Universal Education: A Political Economy of Public Schooling in Rural Pakistan." Asia Research Centre, London School of Economics. Glaeser, E., and A. Shliefer. 2002. The Curley Effect.NBER Working Paper 8942. National Bureau of Eco- nomics Research, Cambridge, Mass. Glewwe, Paul, and Hanan Jacoby. 1994. "Student Achievement and Schooling Choice in Low-Income Countries: Evidence from Ghana." Journal of Human Resources 29(3):843­64. Grossman, G., and E. Helpman. 1996. "Electoral Competition and Special Interest Politics." Review of Economic Studies 63(2):265­86. Hungerford, Thomas, and Gary Solon. 1987. "Sheepskin Effects in the Returns to Education." Review of Economics and Statistics 69(1):175­77. Iyengar, S. 1990. "Shortcuts to Political Knowledge: The Role of Selective Attention and Accessibility." In John Ferejohn and James Kuklinski, eds., Information and Democratic Processes. Urbana and Chicago: University of Illinois Press. Keefer, Philip. 2002a. "Clientelism, Credibility, and Democracy." World Bank, Development Research Group, Washington, D.C. ------. 2002b. "The Political Economy of Public Spending Decisions in the Dominican Republic: Cred- ibility, Clientelism and Political Institutions." Report prepared for the Public Expenditure and Insti- tutional Review, the Dominican Republic. World Bank, Washington, D.C. Keefer, Philip, and Stephen Knack. 2002. "Boondoggles and Expropriation: Rent-Seeking and Policy Distortion When Property Rights Are Insecure." Policy Research Working Paper 2910. World Bank, Washington, D.C. Keefer, Philip, Ambar Narayan, and Tara Vishwanath. 2003. "The Political Economy of Decentrali- zation in Pakistan." World Bank, Development Economics, Policy Research Group, Washington, D.C. Khemani, Stuti. 2004. "Political Cycles in a Developing Economy: Effect of Elections in the Indian States." Journal of Development Economics 73(1):125­54. Knack, Stephen, and Philip Keefer. 1997. "Does Inequality Harm Growth Only in Democracies? A Rep- lication and Extension." American Journal of Political Science 41(1):323­332. Krishnan, T. N. 1991. "Kerala's Health Transition: Facts and Factors." Harvard University, Center for Population and Development Studies, Cambridge, Mass. Layard, Richard, and George Psacharopoulos. 1974. "The Screening Hypothesis and the Returns to Education." Journal of Political Economy 82(5):985­98. Mani, A., and S. Mukand. 2002. "Democracy, Visibility and Public Good Provision." Williams College, Williamstown, Mass. Mayhew, David. 1974. Congress: The Electoral Connection. New Haven, Conn.: Yale University Press. Meltzer, A., and S. Richard. 1981. "A Rational Theory of the Size of Government." Journal of Political Economy 89 (5): 914­27. Miguel, E. 2001. "Ethnic Diversity and School Funding in Kenya." Working Paper C01-119. University of California, Center for International and Development Economics Research, Berkeley. Mullainathan, S., and A. Shliefer. 2002. "Media Bias." Harvard Institute Research Working Paper 1981, and MITDepartment of Economics Working Paper 02­33. Cambridge, Mass. Nossiter, T. J. 1982. Communism in Kerala: A Study in Political Adaptation. Delhi, India: Oxford University Press. Ottati, V., and R. Wyer. 1990. "The Cognitive Mediators of Political Choice: Towards a Comprehensive Model of Political Information Processing." In John Ferejohn and James Kuklinski, eds., Information and Democratic Processes. Urbana and Chicago: University of Illinois Press. 26 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Ozler, B., G. Datt, and M. Ravallion. 1996. "A Database on Poverty and Growth in India." World Bank, Policy Research Department, Washington, D.C. Pande, Rohini. 2003. "Can Mandated Political Representation Increase Policy Influence for Disadvan- taged Minorities? Theory and Evidence from India." American Economic Review 93(4):1132­51. Persson, T., and G. Tabellini. 2000. Political Economics: Explaining Public Policy. Cambridge, Mass.: MIT Press. Prasad, K., and V. Eswara. 1987. Wastage, Stagnation, and Inequality of Opportunity in Rural Primary Education: A Case Study of Andhra Pradesh. New Delhi: Ministry of Human Resource Development. Reserve Bank of India. Various issues. Reserve Bank of India Bulletin. New Delhi. Robinson, James, and Thierry Verdier. 2002. "The Political Economy of Clientelism." CEPR Working Paper 3205. Centre for Economic Policy Research, London. Rogoff, K. 1990 "Equilibrium Political Budget Cycles." American Economic Review 80(1):21­36. Sanmartín, Marta. 2001. "Linearity of the Return to Education and Self-Selection." Applied Economics 33(1):133­42. Schuknecht, L. 2000. "Fiscal Policy Cycles and Public Expenditure in Developing Countries." Public Choice 102(1­2):113­28. Shi, M., and J. Svensson. 2000. "Political Budget Cycles: Do They Differ between Developed and Developing Countries?" World Bank, Development Research Group, Washington, D.C. Strömberg, David. 2001. "Radio's Impact on New Deal Spending." Stockholm University, Institute for International Economic Studies. ------. 2004 "Mass Media Competition, Political Competition, and Public Policy." Review of Economic Studies 71(1):265­84. Tilak, Jandhyala B. G. 1993. "Costs and Financing of Education in India: A Review of Issues, Problems, and Prospects." Center for Development Studies, Trivandrum, India. Van de Walle, Dominique, and Kimberly Nead. 1995. Public Spending and the Poor: Theory and Evidence. Baltimore: Johns Hopkins University Press. Wantchekon, L. 2003. "Clientelism and Voting Behavior: Evidence from a Field Experiment in Benin." World Politics 55(April):399­422. Weiner, M. 1991. The Child and the State in India: Child Labor and Education Policy in Comparative Perspective. Princeton, N.J.: Princeton University Press. Weiner, M., and J. O. Field. 1974. Studies in Electoral Politics in the Indian States. Delhi, India: Manohar Book Service. Wolfers, Justin. 2002. "Are Voters Rational? Evidence from Gubernatorial Elections." Research Paper 1730. Stanford University, Graduate School of Business, Stanford, Calif. World Bank. 1998. "A Framework for Civil Service Reform in Pakistan." Poverty Reduction and Economic Management, South Asia Region, Washington, D.C. ------. 2001. "Nepal: Priorities and Strategies for Education Reform." Human Development Unit, South Asia Region, Washington, D.C. ------. 2002. "Poverty in Pakistan: Vulnerabilities, Social Gaps and Rural Dynamics." Poverty Reduction and Economic Management, South Asia Region, Washington, D.C. Zachariah, K. C. 1992. "Demographic Transition in Kerala in the 1980s: Results of a Survey in Three Districts." Centre for Development Studies, Thiruvananthapuram, India, and Gujarat Institute of Area Planning, Ahmedabad, India. Philip Keefer and Stuti Khemani 27 Evaluating the Impact of Conditional Cash Transfer Programs Laura B. Rawlings · Gloria M. Rubio Several developing economies have recently introduced conditional cash transfer programs, which provide money to poor families contingent on certain behavior, usually investments in human capital, such as sending children to school or bringing them to health centers. The approach is both an alternative to more traditional social assistance programs and a demand-side complement to the supply of health and education services. Unlike most development initiatives, conditional cash transfer programs have been subject to rigorous evaluations of their effectiveness using experimental or quasi-experimental methods. Evaluation results for programs launched in Colombia, Honduras, Jamaica, Mexico, Nicaragua, and Turkey reveal successes in addressing many of the failures in delivering social assistance, such as weak poverty targeting, disincentive effects, and limited welfare impacts. There is clear evidence of success from the first generation of programs in Colombia, Mexico, and Nicaragua in increasing enrollment rates, improving preventive health care, and raising household consumption. Many questions remain unanswered, however, including the potential of conditional cash transfer programs to function well under different conditions, to address a broader range of challenges among poor and vulnerable populations, and to prevent the intergenerational transmission of poverty. Conditional cash transfer programs are an innovative approach to the delivery of social services. They provide money to poor families conditional on investments in human capital, such as sending children to school or bringing them to health centers on a regular basis. That conditionality makes this new generation of social programs an instrument for longer-term human capital investments as well as short-term social assistance. Additionally, along with school voucher programs and certain subsidized health insurance schemes, conditional cash transfer programs are part of a growing policy emphasis on the use of market-oriented demand-side interventions to directly support the poor. They complement traditional supply-side mechanisms, such as © The Author 2005. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THEWORLD BANK . All rights reserved. For permissions, please e-mail: journals.permissions@oupjournals.org. doi:10.1093/wbro/lki001 20:29­55 general subsidies or investments in schools, health centers, and other providers of social services. Conditional cash transfer programs aimed at improving children's human capital have been established in numerous countries in recent years. Six are reviewed here. The first large-scale program to incorporate both health and education components was Mexico's Education, Health, and Nutrition Program (Progresa), launched in 1997.1 Following a similar model, Colombia has the Families in Action (FA) program, Honduras has the Family Assistance Program (PRAF), Jamaica has the Program of Advancement through Health and Education (PATH), Nicaragua has the Social Protection Network (RPS), and Turkey offers the Social Solidarity Fund (SSF). Each program promotes long-term human capital accumulation as a primary objective, recognizing its role in breaking the intergenerational transmission of poverty (table 1). The programs focus primarily on children as the recipients of the human capital investments promoted by the programs and closely monitor compliance with conditions as a prerequisite for receiving the transfers. Traditional social assis- tance strategies have focused on short-term poverty alleviation through redistribution during times of crisis. Implementation of conditional cash transfer programs has been accompanied by systematic efforts to measure their effectiveness and understand their broader impact on households' behavior, a marked departure from the limited attention to rigorous impact evaluations in the past.2 This article reviews the experience to date of six countries in setting up and evaluating the impact of such programs. The programs were selected to include those that provide conditional cash transfer for both health and education because policy and evaluation experience exist for such programs, as well as for those that provide in-kind conditional transfers.3 This review draws from program documents provided by administrators and evaluation reports produced by research institutions. Evaluation results are analyzed to draw conclusions about the welfare impact of this type of program and about how the evaluations have been used to inform policy decisions. Expected insights from forthcoming evaluations are briefly considered, followed by some reflections on the future direction of evaluations of social sector programs. Conditional Cash Transfer Programs: A New Approach to Social Assistance Conditional cash transfers together with other social assistance programs constitute a country's formal, publicly provided safety net system.4 Conditional cash transfer programs represent a new approach to social assistance that explicitly addresses several criticisms often levied at more traditional social programs, including weak poverty targeting; high administrative or component costs, such as materials in workfare 30 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) n n re in ne re ildhc nued)i h patingi h wit wit ildrhcro ildhc h dna wit nt derlyle, utet (Cont tion sdlo ne with stied nutri eh dna particton sdlo mo sdlo eh w eh nagerp men ople dna wo us ho 6­0 us us 5, 3 ho ho pe,56 s,eitil 65rednu th ts or population Heal Po esga programsrehto or Po ntangerp re 0­ or se ctating und Po ag la ervo bias ul di ad Target with ds 17­7 with 4 with ool ds e ds ol se sch eh )11 12­6 17­6 ol se eh yetton ad ol gr se eh d n us ag in 2­ us ag us ag Programs er sf ucatiodE horooP en en tee des have en drlihc dell ra enro (g horooP drlihc who mploc horooP drlihc Tran n h n ; htl s itio er Cashla n tho ritio io hea thlaeh at t; ionitd nut subsidy andhealt uc supply for trun;s mr incentives fo and ed nter gran thl Con thla tioni );rehc ce re Hea Nutrition he Demand (nutr vou incentives ca nginiart ation alth uc He ed Selected Components of for ool ives voucher); sch ools al grant pulationoP incent sch n hlyt y incentives id rgetaT ucatiodE bs ducation Bimon su Demand (e supply primary Education and g ts, sa tion ; e lau k d ty rv en apitalc emelyrt la er se m amon brea an rty; r; na smaller rty serve ex cu poorest in muh ac in help ies; ovep labo t e the pove improve, omes, ten lies Compon e mi litap apitalc of tc ce povtnerr ild ne th n of ationcude e and e ou du cu ch eas fa ty eas cr or safe cr mauh circle eas re ce ce safety Objectives In investment po munici a In of ildrenhc e cr families th In attainment altheh us th duer duer a as Objectives, FA 1. as ur II nd Table Program Colombia Ho PRAF Jamaica PATH poor olds;hes gn l- 2­5sega olds;hes ner ild fors hou atit sega h ma ch: or n re or tion po lements:p lac re dna,s n hou es nitie wit ild po rvic 0­5 ortu ild se sdlo es nutri ts: sup and ch ts: eh opp dna nth ag anrg no ch de re n, mo anrg ca 5 us ngit th 0­ ho 4 population Heal shaC itir nut nantgerp ish ur sra wome 4­2 no ye shaC altheh se or ag Po drenlihc rae e-gen d mo Target d an yr in r with lleor inc 3 ;l 20 nda to ppeu s d ds en de tsn co hoo e age ni an in ol eh 18­8 se (gra d sc)3 6 create clud udets ool se to raised in sch ag n us yr ner an 1­ 4­1 11­1 ms ai ucatiodE horooP des mitli to yrad ildrenhc enrolled yra withsdlohesuoh en ild prima ra 01 con or m or ch in higher) (g age 20 se Po 6­13 pri grades Po drlihc rolledner ). olde grades 2002 program d, ; s 1c, ; alth ic n es of alth he renewed food 5; (200 he ritio bas; iontac rvic ;s food n.o for ive for se nut iontp tritionun edu for d sicab t supply supplement an nkaB The d rvice kage under d gran and care e; se incent tives. ucatidet grant thl um grant kag health ation; pac yl Worl Hea shaC ns alth alth co he pac and improve he nutrition shaC tritionun uc re alth ), objec adul ed ca children supp He its 001 and (2 ned Components ts; loo iontac asi ments, broade ly ly Skouf gran pp sch edufo loo pp a), and improve su yt grant; sch su grant n for ucatiodE pport s;laireta for ali es qu rvic pport s;laireta evitn (2000 dades I R usingoh ce IFP Educational su m and se Education su m in Education 01), Oportuni redit, to (20 croc l,a E l l es e s FIS mean mi ot iontac tritiona families, nerdli taip m lie in on its ca hou n in trexe na miaf sa d cess ac edu ch ment in or based e nu ma muh e angehc th poor ers po evres; t hu ng th and, of arlyl th et tional ation viil e invest ely ty usta mo mu s eas m Objectives Improve altheh rticu d omo cu ld prefe verty cr st pa an Pr ac ho po In capital trexe lylanoitan ne tialner safe compil ogresarP a 02 (Continued) 20 through a a F 1. SS Authors': o sa olds Table Program Mexic regorP ragua ey March rk Source In a Nic RPS Tu househ programs; lack of integration of disparate projects with a multiplicity of overlap- ping or unrelated goals; accusations of paternalism and clientelism; and excessive focus on reducing current poverty with little attention to long-term, structural poverty. The traditional role of social assistance programs has been to redistribute income and resources to the needy, helping them overcome short-term poverty during periods of crisis. Social policies and programs are changing, however, and are beginning to encompass objectives of longer term economic growth and human capital development. As Ravallion (2003) explains, by making insurance available, helping credit- constrained poor people become productive workers, and providing incentives for long-term investments in human capital, safety nets are now seen to have a potentially important role in compensating for the market failures that help perpetuate poverty, particularly in high-inequality settings. With an emphasis on human capital accumulation and long-term poverty reduction, conditional cash transfers are perhaps the clearest policy manifestation of this new thinking on social assistance programs. Conditional cash transfer pro- grams address both future poverty, by fostering human capital accumulation among the young as a means of breaking the intergenerational cycle of poverty, and current poverty, by providing income support for smoothing consumption in the short run. Conditional cash transfer programs have also introduced other key design features that depart from traditional social assistance programs. First, they provide grants directly to poor households, thereby changing accountability relationships among the national government, service providers, and the poor. The conditions required by the grants provide an incentive for poor households to use available health and education services, strengthening the link between ser- vice providers and the poor. Conditional cash transfer grants also allow national governments to forge a direct relationship with poor families, seeking to foster coresponsibility by requiring families to assume responsibility for schooling, health care, and the appropriate use of the cash grants. The programs reviewed here designate mothers as recipients of the grants in recognition of the inter- national evidence that suggests that women often make more optimal household spending decisions affecting children's welfare. Second, they seek to exploit com- plementarities between elements of human capital development through their inclusion of health, nutrition, and education components. Third, the use of cash is promoted as efficient and flexible. It gives households spending discretion and avoids the price distortions and creation of secondary markets that are often asso- ciated with in-kind transfers. Finally, many conditional cash transfer programs also incorporate good technical program design features, including explicit pov- erty targeting criteria, often based on proxy-means tests, and strong monitoring and evaluation systems. Laura B. Rawlings and Gloria M. Rubio 33 Education and Health Components The programs reviewed here have both an education component and a health and nutrition component. The education component consists of a cash grant conditioned on school enrollment and regular school attendance (usually 80­85 percent of school days). The size of the grants varies considerably across countries (table 2). In Honduras, Mexico, and Turkey, the education grant covers both direct costs (school fees, school supplies, transportation costs) and opportunity costs in lost income from sending children to school rather than work. In the other countries the grant generally covers only part of the opportunity cost. In Colombia and Mexico education grants are higher for secondary school than for primary school, to reflect the increasing opportunity cost of work as children grow older. In Mexico grants at the secondary level are higher for girls, to provide an added incentive for reversing a pattern of unequal gender participation in secondary education and to internalize the educa- tion externalities that accrue as they raise families of their own (Skoufias 2001). In Turkey the value of the grant decreases proportionally according to the number of children in the family. Health and nutrition grants are targeted to children up to the ages of 2 or 3 years and in some cases up to the time they enroll in primary school. In Honduras, Jamaica, and Mexico, pregnant and lactating women are designated as program beneficiaries, and their inclusion is being discussed in Turkey. This component consists of a cash transfer aimed at food consumption, as well as health care and nutrition education for mothers. In Mexico and Nicaragua this component explicitly stipu- lates the provision of a basic health care package for the target household members. Receipt of the cash transfer is conditional on compliance with a predetermined number of health center visits and health and nutrition workshops.5 Children's health care visits are linked to growth monitoring and often to vaccination protocols. Health care visits for pregnant and lactating women seek to ensure appropriate prenatal, childbirth, and puerperal care. In Mexico and Jamaica adult household members other than pregnant and lactating women are also required to get a check-up once or twice a year (see table 2). The value of the monthly cash grant for the health and nutrition component var- ies across countries (see table 2). In Honduras, for example, the value of the nutri- tion and health voucher is equivalent to the value of the time invested by the mother in the trip to the health center and waiting for care. Jamaica set the health grant per beneficiary per month at US$9, the same level as the education transfer and twice average monthly spending per person on health care and medicine in 1999. Colombia set the grant to the mean income required to allow an average indigent family to reach the extreme poverty line and so to consume a nutritiously adequate diet. Jamaica and Turkey provide health and nutrition grants to individuals rather than family-based allocations. 34 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) y r yl mil pe year ht farep .3) r supp 9991 tsn gra dilhc pe 46 monr foodrof oldhes r in S$ tyili pe pe nth nth hour on 20)$ per) (U ational pe $6)S (US 066 erageva facr eligiblerep mber ar; pe S$13 mor mor pe ucde s (U 34) n nutriti 0 L$ yer me pe L$87,315 r: 5 and m and ,50 pe ev 0)20 (U S$9) (U old old 057$ ntho lio (US$ he 6, 0 12 rt mil monthr 08 2 y Health $46loC uc 50 useh x$ useh Mex ppo er pe Vo family incenti (US$ J$ ho Me ho of su C$4 ev 1.8LT monthr pe size ild 65 er, ntho ch rials; for child, dl child 0 ar; terfa ­1 001$ m nd yearrep 10 ye ge ild rep rstfi 8.8LT chi Transfer per ,00 nth rep 0 0 Mex mate rep per x$80 d ch t: 6)$ 28 mo hstn C$: LT for l$ dli 30 an 57,94 rep Me J$ plus ildhcr ildhcr per mo pe ntive hstno (US Co gran l:o 2) 000 ho nth chrep)8 dna,dn ar ar 2 ce L$ ye of dlihc at e,d co n nth schoolrof pe m grade ye subsequent gra by yr se ch sc per r mo ivetn gaeb by mor dli per) yreve) 20) in 2 month (US$ pply school per for ea rams yra ol pe pe ch rep 8) S$5 r 20 evet 2). ho 9) ries ­352$ su; dary for S$ en ogrP Col$14, ndoces dlihc (U 8 ceni S$ va (U ildhcr per varies $17S S$ (US (U 572$C 9.50)$ rials con (U ool: (U stud n 200,d 0 (program se (US er r ool: pply pe) 56 rials 0 sch 200 te plus mate n lio 1c, sfn school: mil on yr h;tn pe L$82 screp)0 su 50 ar 00, J$ ye h)tn S$6.50) sch yr 17­ yearep)1 0­2 ma 24$C rep dna io 7). yr U( Tra mo 12)$ r her:cuo t: 24 Mex$ t: hool mill .8 (200 mo shaClan Educati Prima pe (US V erageva 8 S$4 an dno ool sc r (U Gr sec pe imarP S$8 S$1 daryno x$e (U (U sec M sulp an sch Gr dlohesuoh S$0 for (U imarP 10 12. TL million nkaB d d ate -i Worl ioti visits r ol d visits ta es for for dulee ), ts pe eh ire ce sch nd visi ring required required us qu urtcel ts anc skl ts vacc alth nd He center visi up-to-date ta visi a 0012( Co by re the the nterec and n dna of atte to asi ed iont reac dna sitsiv th th status/ hol al ten itio reac reac ing ct tri monito alth e eht wi wi le and,s trition stryi Skouf nu alth owthrg altheh varies yb alth of he ag e thi he Min Se he of ich w nc of ats'reh trun alth nu alth nitoring rdo mo 0a), of and rs dna he monitoring, he and accs a the ns alth owth I Size itio laruge er wh lia er lopment ncye mot th ination child's mp d al by (200 ions.t He R for veed Compliance frequ Compliance mbun ar,ey neficiaryeb Co membe mbun laruge an he R gr ccav thlaeh laruge R owthrg tionan R set IFP Cond condil d al d of 01), rm) se e e Transfer l month-2 an (20 of of ter with and, with nt in pe nt, and a ndanc dna fewer, en schoo e e cuxenu E forced in riod nt, FIS ndanc in mell %58 days aysd e pe hool y ment promotion lity ec 7 nc doirep of 9 nc mell mell ro sc nc se h ec ht ation m se ro atte hlt ofs se ro atte na 80% onit en ab nt ab en always dan ol dan ol mon llya enroll day ab grade en 6 itio least elc mumi mum mum ol ol mum ool ool ool ool ho 3-mo ho %, an two-mon ho %58 compil not e Educa At atten cy Sc max sch a Mini atten (maximu sch Sc mini 85 nnua Scho th sch a sch Sc mini of hav Cond a a F 2. as SS Authors': ur o sa II nd Table Program Colombia FA Ho PRAF Jamaica PATH Mexic regorP ragua ey rk Source Programs a Nic RPS Tu Supply-Side Support Conditional cash transfer programs can be interpreted as a response to the perceived failures of traditional supply-side interventions, such as schools and health clinics, which have been underutilized by the poor because of unmanageable out-of-pocket expenditures, high opportunity costs, difficult access, and a lack of incentives for investing in children's human capital. However, conditional cash transfers are not a substitute for the provision of high-quality supply-side investments. Rather, they complement such investments by directly addressing the problem of insufficient demand for health and education services from the poor. This makes these pro- grams' ultimate success dependent on access to high-quality health and education services. No program should be conditioned on the mandated use of poor-quality, ineffective services. Because of the critical role of good quality health and education inputs, some countries go beyond providing demand-side monetary incentives to families by strengthening the supply of these services. In Nicaragua teachers receive a modest bonus per child participating in the program, half of it intended to pay for school materials, and nongovernmental organizations (NGOs) are contracted to provide health services. Mexico sets aside resources to ensure an adequate supply of equipment, medicines, and material to meet the increase in health services demand arising from the program. Honduras provides grants directly to schools and health centers as part of an experiment to compare the effectiveness of three alternative interventions combining demand and supply incentives. Poverty Targeting Directing benefits to the poor or vulnerable is a critical feature of each reviewed conditional cash transfer program. Most rely on both geographic and household targeting, using targeting mechanisms appropriate to the type of data available (table 3). At the geographic level Jamaica uses annual consumption data to construct a scoring formula to identify poor households at the parish level for allocating PATH funds. In Mexico eligible communities in rural areas are selected using a marginality index based on census data, whereas in Honduras malnutrition data from the Height Census of First Grade School Children are used to select program municipalities. In most countries, the criteria applied to select communities to receive the conditional cash transfer program include consideration of the supply capacity to respond to the increased demand in health and education services. At the household level programs are experimenting with proxy-means tests that estimate household poverty levels as a criterion for program participation (table 3). In Nicaragua the results of household-level proxy means tests are being compared 36 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Table 3. Targeting Criteria of Selected Conditional Cash Transfer Programs Program Geographic Household Colombia FA Municipalities other than department capitals Level 1 families in the SISBEN with fewer than 100,000 inhabitants; (an information system based municipalities not participating in other national on a proxy means test for programs with adequate supply of education and identifying poor households) health services and a bank; municipalities with up-to-date SISBENdatabase Honduras PRAFII Municipalities with the lowest average height None for age z-scores Jamaica PATH All parishes participate in the program; funds A scoring formula with a are distributed across parishes depending predetermined cutoff point on their poverty incidence Mexico Progresaa Rural communities with a high marginality Within eligible localities, index with more than 50 and fewer than 2,500 eligible households are inhabitants and access within a certain distance identified using discriminant to primary and secondary school and health care analysis of household income center; urban areas with a high marginality and other characteristics index have been included since 2001 Nicaragua RPS Departments and municipalities with high Pilot phase 1: all households extreme poverty incidence, good access to in selected census areas with schools and health care centers, good transport less than 14.1 ha and no and communication infrastructure and local vehicle; pilot phase 2: capacity; within eligible municipalities census eligibility is determined areas were classified in two groups according to by a scoring formula a marginality index based on family size, access to basic sanitation and safe water, and literacy rates, with group 1 eligible in the pilot phase 1 and group 2 in the second pilot phase Turkey SSF National coverage; no geographic targeting Proxy-means test based on a scoring formula Source: Authors' compilation based on FISE(2001) IFPRI(2000a), Skoufias (2001), World Bank (2001c, d, 2002). with the results of geographic targeting alone. Turkey will use a proxy-means test being developed especially for the conditional cash transfer program to target the poorest of the poor at the national level. Other countries are taking advantage of economies of scale in the use of proxy-means tests. In Colombia household eligibility is based on an existing information system managed by municipalities, the System for Selecting Social Program Beneficiaries (SISBEN). The system classifies households according to an unmet basic needs index and other indicators, such as average house- hold schooling that serve as income proxies. Used primarily to identify eligibility for the subsidized health regime, SISBEN is now being expanded to other social sector Laura B. Rawlings and Gloria M. Rubio 37 initiatives, including the conditional cash transfer program. Jamaica is planning to expand the use of the PATH scoring formula to other safety net programs to avoid duplication of administrative systems and increase coordination across programs. Some countries periodically review beneficiaries' eligibility. In Mexico and Jamaica household poverty status is reevaluated every three years. Nicaragua's RPS is designed to last three years in a beneficiary community before the cash transfers are phased out. Only the supply interventions are retained for two more years without a reassessment of eligibility. An Expanding Role in Poverty Alleviation As reflected in budget allocations and the number of beneficiaries, conditional cash transfer programs are playing an increasingly important role in poverty reduction strategies. Mexico's Progresa went from covering 300,000 households when it began operations in 1997 to reaching more than 4 million families in 2002, some 20 percent of the population. The program's 2002 annual budget was around Mex$18 billion (US$1.8 billion). PATH is a key element of the Jamaican government's initiative to transform the social safety net into a fiscally sound and more efficient system of social assistance for the poor and vulnerable. It aims to consolidate three major income transfer programs, strengthen targeting measures, improve the cost-effectiveness of delivering benefits, and adjust benefit levels to meet assessed needs. Turkey's SSF was introduced as part of a handful of crisis-response mechanisms to ease the impact of the 2001 economic crisis on poor households. It has an annual operating budget of US$100 million. Colombia's FAis the flagship program of the three safety net programs introduced in 2001 to provide relief from the effects of an economic recession. The program, designed to run through 2004, has a budget of US$455 million and is expected to reach more than a million children. In Brazil, Bolsa Familia is being introduced as an overarching welfare program that will consolidate numerous smaller programs to become a mainstay of Brazil's poverty reduction approach. Evaluation of Design and Implementation The first generation of conditional cash transfer programs in Colombia, Honduras, Nicaragua, and Mexico prioritized the early use of robust evaluations as a key element for informing program design and expansion. All but Colombia's program used ran- domized control designs as the primary evaluation methodology underpinning a large-scale social experiment, carefully planned well in advance with strong support from program staff and policymakers. 38 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) The first generation of conditional cash transfer evaluations aimed at assessing program impact and operational performance by examining the programs' adminis- trative adequacy, the extent to which programs reached poor areas and poor house- holds, the presence and size of expected impacts, any unanticipated effects, stakeholders' perceptions about the program, and the cost-effectiveness of delivery mechanisms. The impact evaluations focused on measuring changes in short- and medium-term indicators of human capital accumulation rather than on the income redistribution effects of the grants. In education the evaluations assessed changes in school enrollment and attendance rates, and some also analyzed changes in promotion and repetition rates. Evaluation of Honduras's PRAFand Mexico's Progresa went beyond outcome indicators to measure changes in impact indicators, such as average test scores. In addition, given PRAF 's evaluation objective of comparing the impact of supply- and demand-side interventions, evaluators are examining changes in the availability and quality of education inputs (percentage of teachers trained, percentage of schools with basic teaching materials). In health and nutrition, the evaluations included a wide range of health care utili- zation and quality indicators. Program variations in target population are reflected in the diverse selection of child, maternal, and adult health indicators. Child health indicators typically include vaccination coverage, malnutrition rates, incidence of diarrhea, and participation rates in child growth and development monitoring. Maternal health indicators include utilization rates and satisfaction with pre- and postnatal care. Honduras's PRAFevaluation is measuring final program impacts by analyzing changes in maternal and infant mortality. Changes in consumption levels and patterns are also central to many evaluations. Total consumption per capita disaggregated by food and nonfood items, such as health and education spending, is frequently used as an indicator. Given the implicit objective of reducing current poverty, Mexico's Progresa evaluation investigates the impact of cash transfers on the poverty headcount ratio, poverty gap, and poverty severity index. Development programs often have unplanned direct and indirect effects, both positive and negative. Some of the conditional cash transfer evaluations have ana- lyzed such impacts. For example, the distribution of cash grants directly to mothers may have an effect on resource allocations within households and on power relations. Cash transfers may crowd out remittances and other private transfers to households or affect household work incentives. Household-level targeting may also affect community relations when not all members of a community receive program benefits. Evaluation Design Program impacts are measured by assessing whether a program changes the mean value of an outcome variable among participants compared with what the outcome Laura B. Rawlings and Gloria M. Rubio 39 would have been had they not participated. The central evaluation problem then is that program participants cannot be simultaneously observed in the alternative state of no participation (the counterfactual). Evaluators typically simulate the counter- factual by comparing program participants (the treatment group) with a control or comparison group with similar characteristics. Construction of the counterfactual determines the evaluation design, which can be broadly classified as experimental or quasi-experimental. These evaluation designs vary in feasibility, cost, and the clarity and validity of results. Experimental or randomized control designs involve the random assignment of individuals (or another unit of analysis) into the treatment group or the control group. Because participants are selected randomly, any differences between the groups is due to chance, not selection. For this reason, experimental designs are usually regarded as the most reliable evaluation method and the one yielding the easiest results to interpret (Freeman and Rossi 1993; Grossman 1994). When randomization is not feasible, a quasi-experimental design is used to gener- ate a comparison group through alternative means. Statistical matching on the basis of observable characteristics is commonly used to select comparison group members who are comparable in essential characteristics to participants. Because unobservable characteristics of beneficiaries, such as motivation or organizational capacity, can strongly influence program impacts but are generally not addressed or addressed only with difficulty in evaluations using quasi-experimental designs, these approaches are often considered less methodologically robust. The first generation of conditional cash transfer evaluations took advantage of the gradual implementation of these programs (because of logistical complexities, fiscal constraints, and uncertainty about the magnitude of program impacts) to randomly add beneficiaries as the programs expanded.6 This approach reflected pragmatism and a desire to rigorously explore the impact of these new programs. Experimental designs are usually maintained for only a few years, however, thus limiting their ability to provide rigorous evidence on longer term program effects. Most first-generation conditional cash transfer evaluation designs rely on random allocation of program benefits by geographic area rather than by household (table 4). The broad geographic nature of some of the program components and the difficulties arising from having treatment and control groups in the same community made randomization at the household level impractical. In Mexico's Progresa, evaluators randomly assigned localities to treatment and control groups. Treatment localities entered the program in November 1997, whereas control localities started receiving Progresa benefits in December 2000. Randomization was implemented at the municipal level in Honduras's PRAF and at the census level in Nicaragua's RPS . In Honduras the evaluation objectives required three different treatment groups to compare the impacts of different combinations of demand and supply incentives. Randomization by municipality was the preferred option because 40 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) ); ); b ds ds ) -es ol ol se eh se eh se olds group us us eh hou 0 ho ho (Continued) cipaliti cipaliti cipaliti size municipalities 66, s) ld 20= 00 20= 00 10= usoh 00 Treatment 50 (10 ho G1 muni (1,6 G2 muni (1,6 G3 muni (8 Sample ci-i y)til group olds)h ci-i olds,h chae pa rol cii mun tiesil Cont 50 pa (8,347 useoh mun tiesil 20 pa (1,600 useoh in 80 mun 2 2 G s d m ey; eys; rv su oolshcs an nters;ec dna ; 1 eyvrus two rag and test pro er surv plus ces sour dlohes ty of of erstnec G lities ool nt alth of eys he eys reac old -ups--1 sch ce us eh dardized es rv d aftersr rv Data Hou su an su mmunioc ns cipainu us wo aseline day Ce m ho (b foll yea start); thlaeh or diagnostic stan sc ams Progr ;ycne s on, ;tne dna n- ns );ecnad s ality nta ; ome qu s with ices co er sfern tc rs fici petitier puts; inf in rv atten and ef us; rollm patt ou no utilization se practice tcome Tra dicatoni gn at en st lityi atic ou tisfaction reac reac n Cashl Mai etigraT ol nutrition otion, d ho sc chil thlaeh cation scores,t m andlanret sa h mptionus d Edu est( edu alth alt alth pro availab of he (ma mortality); an he he vities lat ere thi om s 1 pply acti meni :s w ton ramg wl th G3, ntive 4 design pro ta ndar: into 2 p) oft G G G Conditiona tiona per d undor lities wi lities lities ta ps: s), su+s ce the er er es)v d in ourgl tion an valuE i-exs 2 imen of h grour ch ch tro alua nicipau nicipau er Selectefo Ev Qua wit participating m matched similar m part Exp dalenap men nicipau pply assign m fou (vou (vou incenti (su ),ylno (con on ionst uatil Impact eva X X lua ev Eva itati a Qual study X n Ongoing uatio and Operations eval X letedp cali ework Com Log amrf X X 4. as ur II nd Table Program Colombia FA Ho PRAF ies )s )s group . eligible old us old size localit eh ns 377( s eh us ce group us Treatment 320 (7,887 ho 12 ea ar ho Sample d ies treatment 12 group el olds)h us olds)h the rol localit ns (8s 1). igib Cont 186 (4,682 el useoh ce 21 area useoh th.la into (200 He cli n fias tioa old ; ; Pub eh us eline pu- detc data value lle school; rveys; question- scores; clinic ivet eyvrus eyvrus r Skou ipalitiesci of et ces ho; su d program enil and mun of bas( woll c co ery )s c se fo an of old old fotne ba m 1a), Institu sour us lities eys 5 eys evt test tras pu- ional 25 ntho clini ol eh ns rv rv wo eh itut ss ols of n Data Ce loca su uslp m d ires; ho nsus us us se ho 200 onal su abou 6 an community na sc admini Ce area; ho foll ho inst as sc b, dli n Nati es andt ch iotp ge tne s; y ; ization rvic e; ty; um dna tion;aco 0002( rporatioo the ra rn I llm R ro ; tion)a inc by IFP d se ns nc d ycne qualit us ol all ce tteap reac ding rs util status; co vailability;a ovec niz early fertili at fici eh d n ef and 99), ecte dicatoni rollmenne atustsl sfersn care dan alth ric incide in st time us an iotp the he lucni( coll ance; alth old tra gn e (19 n's m immu eh es on n ool nd altheh ho he calo rty a- atten dil d tritiona us meo ate Mai Sch atte of an nu ho and pove changes w trni relations; priv etigraT enloohcs; akag s)e d (le rat an nsuoc ch rvic Todd utilization of se melyit based samples and are vities thi om thi om blood to acti wl wl clei ins p hrman art and tiona design ta ndar: oft ta ndar: oft and Be ta into ta ea 2), this ures tion imen valuE er alua Ev Exp dalenap men tiesli ntroloc men p imen er eatment assign loca tr and grou Exp dalenap arsu ourgl ni ns eatment assign ce tr rotnoc d (200 meas rs c rit on othe presentest uatil and ropome Impact eva X X resul ev anth ary on cted. itati a Attanasio imin on data lleoc Qual study X X prel n omrf data based e ments. The. d lin uatio fite ation Operations eval X assess seab ry sample bene ia first compil fic also e cali ework gned th si on y de (Continued) Log amrf bene onl a a 4. Authors': gn o sa thesi udi valuatie udes Table Program Mexic regorP is ragua Source Incl Th The Incl a b c d Nic RPS of their well-defined borders and the feasibility of linking each household, school, or health center with a particular municipality. Program municipalities were selected using data from the School Height Census. A subset of municipalities was randomly assigned to one of four evaluation groups: those receiving demand vouchers, those receiving vouchers plus improvements in service quality, those receiving improve- ments in service quality, and the control group. RPS in Nicaragua followed a similar process, randomly allocating census areas into treatment and control groups. In contrast to the other programs Colombia's FA applied a quasi-experimental design. Program implementers ruled out randomization, instead targeting the program to medium-size municipalities able to provide adequate health and education services and with at least one bank to be able to set up family accounts. A comparison group consisted of municipalities similar to the treatment group in terms of population and infrastructure but not qualifying for the program, often because of lack of a bank. Although not originally planned, the program was launched in a few treatment municipalities before the baseline data were collected. This provided an opportunity for a preliminary evaluation of the unanticipated treatment group and a comparison group constructed through propensity score matching techniques using the baseline data (the comparison group is constructed from nonprogram households with a participation probability closest to program beneficiaries, as determined by the probability of program participation based on socioeconomic characteristics). This required adjustments to the sampling frame, the inclusion of retrospective questions in the survey questionnaires, and additional econometric techniques to control for possible nonrandom selection of early participating municipalities, but the exercise yielded valid (if less precise) impact estimates. Data Collection Early planning of most evaluations allowed for the collection of baseline data, thus permitting comparisons of households in the treatment and control groups before and after program implementation. In this way evaluations can account for charac- teristics that do not change over time within treatment and comparison households, as well as for those that do and that are common to both groups. Random assignment into treatment and control groups, combined with the collection of baseline and follow-up data, allows measuring program impact using difference-in-differences estimators. Except for Colombia, all first-generation conditional cash transfer evalu- ations have baseline data collected before program implementation.7 Household surveys are the main data collection instrument in each of the cases reviewed. Each questionnaire contains a core set of questions about the demographic composition of the household, household expenditures and remittances, and socio- economic status, education, health, migration, and labor market participation of household members. Some country questionnaires include additional modules, such Laura B. Rawlings and Gloria M. Rubio 43 as anthropometrics (height and weight), fertility, participation in other programs, and time allocation. Honduras also incorporates two modules on the quality of health services and schools to evaluate the supply-side component of PRAF .8 School and health center surveys and community questionnaires are also frequently used for evaluation. In Honduras and Mexico student achievement test scores were used to analyze program impact on academic performance. Beneficiaries and other stakeholders' perceptions about the program are often captured through qualitative studies. As part of the operational evaluation of the program, Progresa conducted semistructured interviews with secondary school and health clinic staff and focus group discussions with beneficiaries, nonbeneficiaries, and community mothers who serve as local contacts for Progresa. Qualitative studies have also been used in Nicaragua to complement impact eval- uations. A study on beneficiaries' perceptions of the program's impact on welfare included a beneficiary survey; focus group discussions with beneficiaries and com- munity mothers; key informant interviews with representatives from the Ministries of Health and Education, the mayor's office, health care providers, NGO s, and local program office staff; and six case studies of beneficiary families in different munici- palities. Another study assessed perceptions of the poverty targeting mechanism and included surveys, focus group discussions with beneficiaries and nonbeneficiaries, and key informant interviews. Implementation Issues Social experiments present challenges at each stage of implementation. Evaluations of conditional cash transfer programs reveal two particular issues: the difficulty of coordinating impact evaluations with the implementation schedule and the chal- lenge of winning the political support required for a successful impact evaluation. New, logistically complex programs, such as cash transfer programs, can run into implementation delays or, as in Colombia, may move ahead of the evaluation schedule. Likewise, political changes (such as an upcoming election or changes in program administration) may also affect implementation or even program design. Natural disasters, such as flooding in Jamaica, can also alter the implementation schedule. Such events can effect the evaluation in various ways. For example, in Nicaragua baseline data was collected during August­September 2000, and follow-up data collection was scheduled for the same time a year later. But when the health component was delayed until June 2001 because of difficulties coordinating the health care providers, evaluators had to postpone follow-up data collection until October. Although having a control group helps in this kind of situation, conducting panel surveys at different times of the year may lead to problems from the confounding nature of seasonal effects. 44 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Delays in setting up the program management information system may also cause problems. Deficiencies in the delivery of program benefits may go undetected and thus be unaccounted for in the evaluation. In Mexico, Progresa's payment records revealed that 27 percent of the eligible population in the evaluation sample had not received benefits after almost two years of program operation. This can result in a divergence between the "intention to treat" effect estimated by the evaluation and the mean effect of the program on those who actually participated.9 Finally, as Mexico's Progresa and Honduras's PRAF have revealed, implementing impact evaluations requires strong political support, particularly for a randomized control design. Incorporating a control or comparison group into the evaluation can generate strong criticism and lead to political and media pressure to extend program benefits to these groups. It is important to secure a solid commitment from policymakers to maintain the integrity of the program and evaluation designs and to communicate clearly the benefits of random allocation when budget constraints prevent reaching all eligible beneficiaries at once. Evaluation Results and Impact on the Ground Evaluation results are available for Progresa in Mexico, FA in Colombia, and the RPS pilot in Nicaragua. These evaluations reveal that conditional cash transfers can provide effective incentives for investing in the human capital of poor people. Impacts on Education, Health, and Consumption In education, conditional cash transfer programs have demonstrated a positive effect on enrollment rates for both boys and girls.10 In Mexico primary school enrollment rates before Progresa were 90­94 percent. Econometric estimates of program impact using a difference-in-differences model controlling for household and com- munity characteristics show an increase ranging from 0.74 to 1.07 percentage points for boys to 0.96 to 1.45 percentage points for girls (table 5). At the secondary level, baseline enrollment rates were 67 percent for girls and 73 percent for boys. Estimates of program impact show an increase ranging from 3.5 to 5.8 percentage points for boys to 7.2 to 9.3 for girls. In Nicaragua program impacts are even more impressive (table 6). From a low starting point of 68.5 percent, average enrollment rates in treatment areas increased nearly 22 percentage points. Colombia's FApro- gram seems to have had no effect on enrollment rates among the primary school age population (7­13 years old) while boosting secondary school enrollment rates (for 14­17 years old) 5.5 percentage points in rural areas and 14 percentage points in urban areas (table 7). Laura B. Rawlings and Gloria M. Rubio 45 Table 5. Impacts on Education, Health, and Consumption of Mexico's Progresa Baseline Net change/program impact Education Primary school enrollment Female 90­94% 0.96­1.45percentagepointsa Male 90­94% 0.74­1.07 percentage pointsa Secondary school enrollment Female 67% 7.2­9.3percentagepointsa Male 73% 3.5­5.8percentagepointsa Health Mean growth monitoring visits (in the month 0.22 0.054­0.133b prior to the survey), children ages 0­2 Illness rates (in the month prior to the survey), 0.40 -4.7percentagepointsb children ages 0­2 Consumption Mean consumption level per household -- 13.4%c per month Median food consumption per person -- 10.6%c per month Median caloric acquisition per person per day -- 7.8%c Source: Education, Skoufias (2001); health, Gertler (2000); consumption, Hoddinott and others (2000). aEconometric estimates using a difference-in-differences model controlling for household and community char- acteristics. bEconometric estimates of program impacts using a difference-in-differences model. cPercentage difference between beneficiary and control households at 20 months postbaseline. The evidence of impacts on attendance is mixed. The evaluation of Nicaragua's RPS indicates a larger impact on attendance than on enrollment rates, finding a 30 percentage point increase in the share of children with fewer than six unexcused school absences in a two-month period. The evaluation of Progresa showed more pronounced effects on enrollment than attendance. Evaluations have also found improvements in child health and nutrition. The Progresa evaluation shows a significant increase in nutrition monitoring and immunization rates. Econometric estimates from difference-in-differences models accounting for individual fixed effects found that children 0­2 years old partici- pating in Progresa increased their growth monitoring visits 25­60 percent with respect to the baseline value of .22 visits during the previous month. Progresa also lowered illness rates for the same group of children by 4.7 percentage points, or 12 percent lower than the baseline value (Gertler 2000). The data also suggest that Progresa has had a significant impact on child growth, lowering the probability of child stunting for children ages 12­36 months (Behrman and Hoddinott 2000). 46 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Table 6. Impacts on Education, Health, and Consumption of Nicaragua's RPS Treatment areas Control areas Baseline Follow-up Baseline Follow-up Net change/ Indicator 2000 2001 2000 2001 program impacta Education (%) Share of children ages 7­13 enrolled 68.5 93.2 72 75.1 21.7 (2.7) in primary school (grades 1­4) Health (%) Share of children less than 3 years old 55.9 91.8 60.6 67.4 29.1 (4.3) participating in growth monitoring Share of children ages 12­23 35.4 81.9 40.3 68.5 18.3 (7.8) months with complete, timely immunization Consumption (cordobas) Per capita annual total expenditures 4,310 4,498 3,929 3,300 817 (231) Per capita annual food expenditures 2,922 3,165 2,684 2,175 753 (154) Source: IFPRI(2002b). aUnconditional difference-in-differences estimator. Note: Numbers in parentheses are standard errors. Table 7. Preliminary Impacts on Education, Health, and Consumption of Colombia's Families in Action (FA) Comparison areas Effect of program in treatment areas Urban Rural Urban Rural Education Enrollment probability, ages 7­13 0.941 0.915 0.003 (0.022) 0.012 (0.028) Enrollment probability, ages 14­17 0.639 0.496 0.138* (0.066) 0.055 (0.053) Health Probability of suffering from acute 0.212 0.170 -0.102* (0.055) -0.054 (0.065) diarrhea, children under 6, over past 15 days Probability of suffering from acute 0.448 0.404 -0.032 (0.103) -0.021 (0.073) respiratory disease, children under 6, over past 15 days Consumption Number of days 2­6-year-olds ate eggs 2.4 2.67 0.705* (0.437) 0.774* (0.428) Number of days 2­6-year-olds ate 1.26 1.67 1.383*(0.437) 1.148* (0.488) vegetables Source: Attanasio and others (2003). *Significant at the 5 percent level. Note: Numbers in parentheses are standard errors. Laura B. Rawlings and Gloria M. Rubio 47 In Colombia's FAthe proportion of children under age 6 enrolled in growth mon- itoring rose 37 percentage points. The incidence of acute diarrhea in children under age 6 was reduced by 10 percentage points in urban areas, but there was no significant change in rural areas. The study applied various measures of malnutrition to children under age 6 and detected no impact on global or acute malnutrition in any of the program areas. It did find a positive impact on weight-for-height and weight-for-age in rural areas though not in urban areas (Attanasio and others 2003). Nicaragua's RPSprogram generated similar improvements. Before RPS was implemented, some 60 percent of children under age 3 participated in nutrition monitoring. After several months of program operation more than 90 percent of children in RPS areas participated in nutrition monitoring compared with 67 percent in control areas. Rates of timely immunization among children 12­23 months old rose by 18 percentage points in the treatment group compared with the control group (IFPRI 2002b). Conditional cash transfer programs have also resulted in higher consumption levels. In Mexico after just over a year of program operation the average con- sumption level was 13 percentage points higher, and the value of food consumption for the median beneficiary household was 11 percent higher in Progresa households than in non-Progresa households. Higher expenditures on fruits, vegetables, and animal products accounted for much of the increase in household consumption. Median caloric intake per person in Progresa house- holds increased by 7.8 percentage points (Hoddinott and others 2000). Dietary intake also improved in FA households in Colombia. In Nicaragua consumption levels remained unchanged in RPS areas despite worsened economic conditions related to low coffee prices and a drought. By contrast, consumption declined sharply in control households (IFPRI 2002b). The net program impact translates into a 19 percent increase in per capita consumption and suggests that condi- tional cash transfer programs may help poor people protect consumption in times of crisis. The evaluation of Progresa revealed that conditional cash transfer invest- ments can be delivered cost-effectively. The administrative costs of delivering cash transfers to poor households appear to be small (Mex$8.9 of every Mex$100 allocated to the program) relative to the costs of previous Mexican programs and to targeted programs in other countries (Coady 2000). The largest cost components are those associated with household targeting (nearly 30 percent), followed by those associated with conditioning the receipt of transfers (26 percent). The evaluation also found that the cost of generating an extra year of schooling using subsidies is around Mex$10,000 in secondary school and Mex$55,000 in primary school, compared with Mex$168,000 for extensive expansion by building additional secondary schools and thus reducing travel distances. 48 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Impacts on Program Expansion and Design These findings of encouraging human development impacts have been used to revise programs and influence policy. But the evaluations reveal little about which element of the intervention (the transfer or the conditionality) is responsible for the observed changes or whether the relatively short-term changes will be translated into long-run impacts on human capital formation and poverty. They also provide no comparative evidence on whether alternative interventions would have achieved comparable results. The impact evaluations have triggered some program modifications, guided program expansion, allowed the programs to survive changes in political admin- istrations, and generated interest in replicating the programs internationally. The positive impacts of Mexico's Progresa helped prompt its expansion from rural to urban areas, and the program has continued with few alterations despite a change in government. Likewise, the measured achievements of FA in Colombia contributed to the program's continuation and expansion despite a change in government. The evaluation findings for Nicaragua's RPSshowing that the program had met most of its targets and exceeded many of them trig- gered plans for program expansion. The demonstrated utility of the conditional cash transfer evaluations also increased political support for other impact evalu- ations. In Mexico a congressional mandate calls for the evaluation of all social programs, whereas in Colombia the national performance monitoring system is expanding to include impact evaluations of the country's principal social programs. The evaluation results have also contributed to the ongoing debate on the design of conditional cash transfer programs and their potential role in broader reform of social protection systems. One issue under discussion is whether condi- tional cash transfer programs should take on income-generating activities. Although this may be a necessary condition for the sustainability of human capital investment in future generations, whether conditional cash transfer pro- grams should take this on themselves or whether separate income-generation programs are preferable is unclear. In both Mexico and Nicaragua there has been a tendency to expand conditional cash transfer programs to include train- ing and other income generation activities. Both programs are planning impact evaluations to help inform the debate. Evaluation results of conditional cash transfer programs and national education-oriented cash transfer programs in Brazil have contributed to a reform that will consolidate an array of cash trans- fer programs and triple the average monthly benefit per family. The new program is expected to reach 11.4 million families by 2006, about a quarter of Brazil's population, making this the world's largest conditional cash transfer program. Laura B. Rawlings and Gloria M. Rubio 49 Upcoming Evaluations: Expected New Insights New conditional cash transfer programs have recently begun in Jamaica, Turkey, and urban areas of Mexico. This second generation of programs is being imple- mented under considerably different circumstances than the first. They have benefited from the experience of the first generation, making the logistical aspects less daunting. Evidence of program impacts from the first generation has reduced uncertainty about program results and thus the need for small pilots or phased implementation. Finally, the socioeconomic and political circumstances are particularly pressing in some cases, so implementation plans incorporate nationwide expansion almost immediately. Both the PATH program in Jamaica and the SSF in Turkey have had short pilots, mainly to test program processes, followed rapidly by nationwide expansion. Consequently, new methodologies are being tested. Program pilots include only a process evaluation, reserving impact evaluations for the full-scale program. Because experimental evaluation designs are more challenging when used to evaluate a nationwide program, the second generation of conditional cash transfer programs relies on quasi-experimental designs. The evaluation of Oportunidades, the follow-on to Progresa, takes advantage of the proxy-means test used for beneficiary selection to construct a comparison group from households that applied to the program but were not selected because they fell above the cutoff point. Presumably, households immediately above the cutoff point are similar on average to program beneficiaries and can serve as a comparison group. A second comparison group will be drawn from eligible households in nonin- tervention areas, selected through propensity score matching techniques. The evaluation of Turkey's SSFanticipates applying a quasi-experimental design using panel data with a baseline and two follow-up measures, as well as a qualitative study. Data from the first follow-up survey--to be conducted about one year after the program begins--will be used to assess poverty targeting, short-term welfare impacts, changes in utilization of health and education services, and stakeholder perspectives. A more comprehensive impact evaluation is contemplated for two years after program implementation, using the last round of panel data. Because of the reliance on quasi-experimental designs, second-generation evalua- tions are politically less sensitive and less demanding to implement. However, the results are likely to be less robust and straightforward than those generated by carefully planned experimental designs. In addition, given the rapid expansion of these programs to national scale, there is less control over the timing of the implementation schedule and a greater need for flexibility in evaluation plans. Potential contamination of the comparison group is another problem. The use of households just above the cutoff point for constructing a comparison group risks contamination of the sample from premature incorporation of comparison group households into the program should the cutoff point change. 50 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) These second-generation evaluations will address many of the same core questions about program impacts on school attendance, health care utilization, and consumption and so will help confirm the cross-program robustness of earlier results. Together with continuing evaluations in Colombia, Honduras, and Nicaragua, these evaluations will also analyze new questions prompted by program objectives in each country and by a desire to increase the global body of knowledge about conditional care transfer programs. In Nicaragua the evaluation will assess the sustainability of behavioral changes by measuring program impacts once cash transfers are phased out and only supply-side interventions remain. In Honduras the evaluation will focus on the relative importance of supply and demand factors in increasing human capital as well as program impacts on maternal and child mortality rates. In Colombia, implementation of the conditional cash transfer program as one of three emergency safety net programs will allow for a comparison of the relative effectiveness of conditional cash transfer, workfare, and train- ing programs in achieving particular outcomes. In Mexico the evaluation will examine the results of a new educational savings program for Oportunidades students. Conclusions and Recommendations for the Future In contrast to many development programs, the recent expansion of conditional cash transfer programs is based on fairly solid evidence of program impact. Evalua- tion results from the first generation of these programs in Colombia, Mexico, and Nicaragua show them to be an effective means for promoting human capital accu- mulation among poor households. In particular, there is clear evidence of program success in increasing school enrollment rates, improving preventive health care, and raising household consumption. These evaluation results have provided policy- makers with rare empirical evidence on program efficiency and effectiveness that has informed administrative reforms, prompted the expansion of programs geo- graphically and to new population groups, and contributed to their continuation despite changes in political regimes. The next generation of evaluations is building on this body of knowledge of condi- tional cash transfer programs by providing evidence on the medium-term impact of programs, the value of new elements, and the impact of new conditional cash transfer programs in Honduras, Jamaica, Turkey, and urban areas of Mexico. These evaluations will confirm or challenge earlier findings, shed light on questions of sustainability and medium-term impacts, and provide policymakers with a better understanding of the impacts of alternative combinations of program inputs and different regional circumstances. These results will be useful in understanding the capacity of condi- tional cash transfer programs to meet the new demands placed on them and ensure that these demands do not interfere with achievement of the program's primary objectives. Laura B. Rawlings and Gloria M. Rubio 51 But even when evaluations of the new generation of conditional cash transfer pro- grams become available, many fundamental questions will remain unanswered-- about the effectiveness, including long-term welfare impacts; synergies between program components; tradeoffs between transfer size and number of beneficiaries; and balance between the short-term transfer objectives and the long-term human develop- ment objectives. There is also a need to assess the effectiveness of these programs both as a permanent institution for addressing chronic poverty and as a temporary instru- ment for addressing vulnerability to shocks, and to explore ways to strengthen the links with suppliers of health and education services to improve access and quality. These long-term questions can be answered only through further evaluation. Improvements in evaluation instruments are also needed. Econometric modeling is being used to simulate the anticipated impacts of program design alternatives, such as transfer size and eligibility criteria. Although not a substitute for impact evaluations, these tools can be very useful, particularly for program design. There is also a need to go beyond impact evaluations of individual programs to improve results-based monitoring and evaluation systems of related programs, as a foundation for effective policy management. Finally, there is a need for impact evaluations to explore the development effectiveness of alternative programs and policies, particu- larly concerning long-term welfare impacts. The benefits of program evaluations go far beyond country boundaries and constitute a global public good. The experience of conditional cash transfer programs to date shows the critical role of evaluations in shedding light on success and failure in the fight against poverty. The evaluations are also contributing to the spread of condi- tional cash transfer programs, which are being replicated around the world. Even so, it should not be assumed from positive evaluation results from a handful of countries that similar successes can be achieved in other countries in different con- texts, especially in areas facing supply constraints in health and education or where the capacity to administer a conditional cash transfer program is limited. Nor do the positive results from one program imply that the program subject to the evaluation is necessarily the best approach to achieving a particular outcome. Ideally, program evaluations would compare alternative interventions for achieving a similar objective to determine the most effective and efficient approach. What evaluations of the impact of conditional cash transfer programs reveal so far is the importance of good program and evaluation design in informing policy decisions and providing evidence for achieving progress in the fight against poverty. Notes Laura Rawlings is the Country Sector Leader for Central America in the Latin America and Caribbean Human Development Department at the World Bank; her e-mail address is lrawlings@worldbank.org. 52 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Gloria Rubio is Deputy Director of Social Programs Evaluation at the Ministry of Social Development (SEDESOL) in Mexico. This paper was written while she was a consultant at the World Bank; her e-mail address is gmrubio@sedesol.gob.mx. The authors are grateful to Orazio Attanasio, Jeanine Braithwaite, Christopher Chamberlain, David Coady, Ariel Fiszbein, Kathy Lindert, John Maluccio, Monica Orozco, Emmanuel Skoufias, Miguel Székely, Andrea Vermehren, and three anonymous reviewers for com- ments received and to Tania Gomez for editorial assistance. 1. In March 2002, Progresa changed its name to Oportunidades and changed some of its objectives and operational features. 2. A review of World Bank projects from 1998 to 2000 analyzing the quality of impact evaluation plans incorporated into the project appraisal process found that only 10 percent of projects had ade- quate plans for a rigorous impact evaluation, though the percentage of projects that included them had doubled over these years (World Bank 2001b). 3. See Kim and others (1999) for a review of Pakistan's primary school female fellowship program; Ravallion and Wodon (2000) for an assessment of Bangladesh's Food for Education program; Kandkher and others (2003) for an evaluation of the Bangladesh Female Stipend program; Yap and others (2001) for a discussion of Brazil's PETIprogram aimed at reducing child labor; and World Bank (2001a) for a review of Brazil's Bolsa Escola education stipend program. 4. This section draws on Rawlings (2004). For a more in-depth description of conditional cash trans- fer programs, see Ilahi and others (2000), Legovini and Regalia (2001), and Morley and Coady (2003). 5. Nicaragua initially stipulated that families would lose their grant if malnourished children did not gain adequate weight, but this requirement was dropped after the first year. 6. For example, to increase its coverage of rural areas, Mexico's Progresa expanded in phases from August 1997 to early 2000. Nicaragua's RPS started with a two-year pilot phase in two departments (Madriz and Matagalpa), whereas in Honduras implementation of PRAF was limited to a subset of municipalities because of funding stringencies. 7. Nicaragua's RPS has completed follow-up measurements after one year and two years of program implementation and plans to conduct a third once demand incentives are eliminated and only the supply intervention remains. Mexico's Progresa collects six rounds of panel data in rural areas every six months. Evaluators of Honduras's PRAFplanned to follow up after one and two years of program implementation. Colombia's FA will have baseline data available for about half of the treatment sample (table 4). 8. Although not strictly part of the evaluation, censuses were conducted in the evaluation areas in some countries. In Mexico, a census collected data to determine household eligibility. In Honduras and Nicaragua, censuses generated a beneficiary registry and a household list from which to draw a repre- sentative sample in treatment and control areas and provided information for simulating inclusion and exclusion errors resulting from a proxy-means test targeting mechanism. 9. As Skoufias (2001) discusses, the use of the Progresa eligibility variable for program evaluation allows the evaluators to estimate the "intention to treat" effect. To the extent that not all eligible house- holds actually receive program benefits, the intention to treat effect underestimates the program mean effect on actual program beneficiaries. 10. For a comprehensive discussion of the education impacts see Schultz (2000a, b, c), Behrman and others (2000), and IFPRI(2002b). References Attanasio, Orazio, and others. 2002. "Evaluación de Impacto del Programa Familias en Acción: Primer Informe de Avance." Institute for Fiscal Studies, London. Attanasio, Orazio, and others. 2003. "Baseline Report on the Evaluation of Familias en Acción." Insti- tute for Fiscal Studies, London. Laura B. Rawlings and Gloria M. Rubio 53 Behrman, Jere, and John Hoddinott. 2000. "An Evaluation of the Impact of Progresa on Pre-School Child Height." International Food Policy Research Institute, Washington, D.C. Behrman, Jere R., and Petra E. Todd. 1999. "A Report on the Sample Sizes Used for the Evaluation of the Education, Health and Nutrition Program (Progresa) of Mexico." International Food Policy Research Institute, Washington, D.C. Behrman, Jere, P. Sengupta, and P. Todd. 2000. "The Impact of Progresa on Achievement Test Scores in the First Year." International Food Policy Research Institute. Washington, DC Coady, David. 2000. "The Application of Social Cost-Benefit Analysis to the Evaluation of Progresa." International Food Policy Research Institute, Washington, D.C. FISE (Fondo de Inversión Social de Emergencia). 2001. "Información General del Programa." Red de Protección Social, ABCno. 1, Managua, Nicaragua Freeman, Howard E., and Peter Rossi. 1993. Evaluation: A Systematic Approach. Newbury Park, Calif.: Sage. Gertler, Paul J. 2000. "Final Report: The Impact of Progresa on Health." International Food Policy Research Institute, Washington, D.C. Grossman, Jean. 1994. "Evaluating Social Policies: Principles and U.S. Experience." World Bank Research Observer 9(2):159­80. Hoddinott, J., E. Skoufias, and R. Washburn. 2000. "The Impact of Progresa on Consumption: A Final Report. September." International Food Policy Research Institute, Washington, D.C. IFPRI(International Food Policy Research Institute). 2000a. "Second Report: Implementation Proposal for thePRAF IDB / Project--Phase II." Washington, D.C. ------. 2000b. "Third Report: Monitoring and Evaluation System PRAF ." Washington, D.C. ------. 2001a. "Informe: Sistema de Evaluación de la Fase Piloto de la Red de Protección Social de Nicaragua." Washington, D.C. ------. 2002. "Informe Final--Sistema de Evaluación de la Fase Piloto de la Red de Protección Social de Nicaragua: Evaluación de Impacto." Washington, D.C. Ilahi, Nadeem, Guilherme Sedlacek, and Emily Gustafsson-Wright. 2000. "Targeted Conditional Transfer Programs in Latin America: An Early Survey." World Bank, Office of the Chief Economist, Latin America and the Caribbean Region, Washington, D.C. Kandkher, Shahidur, Mark Pitt, and Nobuhiko Fuwa. 2003. "Subsidy to Promote Girls' Secondary Education: The Female Stipend Program in Bangladesh." World Bank, Poverty Reduction and Economic Management Group, Washington, D.C. Kim, Jooseop, Harold Alderman, and Peter F. Orazem. 1999. "Can Private School Subsidies Increase Enrollment for the Poor? The Quetta Urban Fellowship Program." World Bank Economic Review 13(3):443­65. Legovini, Arianna, and Ferdinando Regalia. 2001. "Targeted Human Development Programs: Investing in the Next Generation." Inter-American Development Bank, Sustainable Development Department, Poverty and Inequality Advisory Unit, Washington, D.C. Morley, Samuel, and David Coady. 2003. "From Social Assistance to Social Development: A Review of Targeted Education Subsidies in Developing Countries." International Food Policy Research Institute, Washington, D.C. Ravallion, Martin. 2003. "Targeted Transfers in Poor Countries: Revisiting the Trade-offs and Policy Options." World Bank, Development Economics Research Group, Washington, D.C. Ravallion, Martin, and Quentin Wodon. 2000. "Does Child Labor Displace Schooling? Evidence on Behavioral Responses to an Enrollment Subsidy." Economic Journal 110:158­75. 54 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Rawlings, Laura. 2004. "Conditional Cash Transfer Programs: Experimenting with Service Delivery Alternatives." World Bank, Latin America and Caribbean Region, Human Development Department, Washington, D.C. Schultz, T. Paul. 2000a. "Final Report: The Impact of Progresa on School Enrollments." International Food Policy Research Institute, Washington, D.C. ------. 2000b. "Impact of Progresa on School Attendance Rates in the Sampled Population" Inter- national Food Policy Research Institute, Washington, D.C. ------. 2000c. "School Subsidies for the Poor: Evaluating a Mexican Strategy for Reducing Poverty." International Food Policy Research Institute, Washington, D.C. Skoufias, E. 2001. "Progresa and Its Impacts on the Human Capital and Welfare of Households in Rural Mexico: A Synthesis of the Results of an Evaluation by IFPRI." International Food Policy Research Institute, Washington, D.C. World Bank. 2001a. "Brazil: An Assessment of the Bolsa Escola Programs." Human Development Department, Latin America and Caribbean Region, Washington, D.C. ------. 2001b. "Poverty Reduction and the World Bank: Progress in Fiscal Year 2000 and 2001." Washington, D.C. ------. 2001c. "Project Appraisal Document: Social Safety Net Project" Washington, D.C. ------. 2001d. "Project Appraisal Document: Social Risk Mitigation Project for Turkey". 22510-TU. Washington, D.C. ------. 2002. "Conditional Cash Transfer Programs: Operational Experiences Workshop." Washington, D.C. Yap, Yoon-Tien, Guilherme Sedlacek, and Peter Orazem. 2001. "Limiting Child Labor through Behavior- Based Income Transfers: An Experimental Evaluation of the PETI Program in Rural Brazil." World Bank, Washington, D.C. Laura B. Rawlings and Gloria M. Rubio 55 Reassessing Conditional Cash Transfer Programs Jishnu Das, Quy-Toan Do, and Berk Özler During the past decade, the use of conditional cash transfer programs to increase investment in human capital has generated considerable excitement in both research and policy forums. This article surveys the existing literature, which suggests that most conditional cash transfer programs are used for essentially one of two purposes: restoring efficiency when externalities exist or improving equity by targeting resources to poor households. The pro- grams often meet their stated objectives, but in some instances there is tension between the efficiency and equity objectives. The overall impact of a program depends on the gains and losses associated with each objective. The use of conditional cash transfer programs as a means of combating poverty has increased dramatically in the past decade. Programs such as Progresa (now called Oportunidades) in Mexico, Bolsa Escola (now called Bolsa Familia) in Brazil, and the Red de Proteccion Social in Nicaragua aim to balance the goals of current and future poverty reduction by providing cash to finance immediate consumption and foster- ing investment in human capital. Several evaluations show that these programs are technically feasible in that the main stated goals of the programs are actually met in practice and are politically acceptable in that successive governments are willing to continue and even expand program coverage. These results have been a source of encouragement for researchers and policymakers in the development community. Some studies suggest, however, that households would behave very differently if given an equivalent amount of cash with no strings attached: Households would consume less of the conditioned-on good and more of other commodities. In western Kenya, for instance, the incidence of malaria decreased when households were given insecticide-treated bednets (Nahlen and others 2003). But when households were asked what they would do if given an equivalent amount of cash, their priori- ties were different. They would have spent the cash on food and clothing--bednets © The Author 2005. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THEWORLD BANK . All rights reserved. For permissions, please e-mail: journals.permissions@oupjournals.org. doi:10.1093/wbro/lki005 20:57­80 were a distant priority (Alaii and others 2000). Also in western Kenya, a deworming program led to a large increase in school attendance. But surprisingly, even after a year of the program, a small increase in the price of the pills from its initial level of zero resulted in an 80 percent decline in their use (Miguel and Kremer 2003, 2004). The evidence is not limited to Sub-Saharan Africa. For the Bolsa Escola program in Brazil, giving cash conditional on school attendance was critical and successful in increasing school participation; cash given unconditionally would not have had a significant impact (Bourguignon and others 2002; Cardoso and Souza 2003). In all these cases, the conditions induced households to behave differently than they would have had they been given unconditional cash. In fact, the advocates of conditional cash transfer programs point to the ability of the programs to influence behavior as a measure of their success. But how the programs work depends on how households respond. An economic perspective can provide useful information on the efficacy of conditional cash transfer programs by analyzing the impacts on house- hold and individual behavior in relation to program aims. In addition, this perspec- tive can help identify the underlying rationale for inducing behavioral changes, that is, under what circumstances policymakers would like individuals to behave differ- ently from the market-induced outcome. Toward this aim, this article reviews the theoretical and empirical literature on conditional cash transfer programs, which suggests that most programs are used for essentially two purposes (table 1). When an individual's actions do not match soci- etal preferences, conditional cash transfers provide incentives for individuals to alter their behavior. This induced change increases the combined welfare of all individu- als. Thus, a mother who makes schooling decisions for her child may not take into account the long-term benefits of education. In this case, giving cash to parents only if they send their children to school reconciles the (possibly) divergent interests of parents and children. Conditional cash transfer programs have also been used as screening mecha- nisms for targeting resources to poor households. Specifically, when governments are unable to directly observe individual characteristics, conditional cash transfers can induce self-selection so that members of the targeted group participate in the program and others opt out. A classic example is workfare. When cash is given only to people who work on a specific task, say, road building, the rich usually opt out and the poor participate. These two purposes do not form an exhaustive list. Although economists have developed tools to analyze and understand behavioral responses for these two cases, a third justification is that human beings often violate the economist's concept of rationality. As Basu (2003) points out, people are frequently impatient, willing to sacrifice too much to make good things happen too soon, not good at understanding complicated concepts such as compound interest rates, and often lacking in self-control. In addition, they may be less than fully informed.1 In all these cases, conditional 58 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Table 1. Selected Evaluations of Conditional Cash Transfer Programs Study Program Main findings Issues covered Alaii and The western Perceived private benefits from Implementation, others (2003) Kenya insecticide-treated insecticide-treated bednet use is low. physical bednet trial (ITN) Positive externalities imply that externalities insecticide-treated bednets had an effect on the spatial distribution of malaria vectors within 600 m of the control villages and on child mortality, moderate anemia, high-density parasitemia, and hemoglobin levels within 300 m of the control villages. Alderman Food subsidies (Tunisia) Self-targetingthroughchoiceofcommodityis Targeting, and Lindert found to be effective at improving targeting redistribution, (1998) efficiency. The effect is not as great as political mechanisms employed elsewere, such as economy of aid means-tested food stamps in Jamaica and geographically targeted food supplementation in Peru. Behrman and Progresa (Mexico) Nutritional supplements and other child Household Hoddinott health interventions have a positive impact bargaining, (2001) on child growth and probability of being fungibility stunted. Bourguignon Bolsa Escola (Brazil) Ex ante evaluation of the program shows Implementation, and others increases in school enrollment, with larger targeting, (2002) effects for poor households, but no effect on redistribution current poverty levels. Unconditional cash transfers would have no impact on child labor and school enrollment rates. Cardoso and Bolsa Escola (Brazil) Program has no impact on child labor but Targeting, Souza (2003) positive impact on enrollment. fungibility, efficiency Galasso and Plan Jefes y Jefas Program is well targeted, with significant Targeting, Ravallion (Argentina) impact on poverty reduction. Forgone redistribution, (2003) incomes equal roughly one-third of the equity-efficiency cash transfer provided by the plan. tradeoff Galasso and Food for Education Program is mildly propoor. Targeting, Ravallion (Bangladesh) efficiency (forthcoming) Jacoby Nutribun and Milk Poorest households recieve the largest Targeting, (1997) Program (Jamaica) benefits. Benefits per beneficiary child are redistribution, less than 50% of program cost, suggesting efficiency considerable deadweight loss. (Continued) Jishnu Das, Quy-Toan Do, and Berk Özler 59 Table 1. (Continued) Study Program Main findings Issues covered Jacoby School Feeding Program On school days, the caloric intake of Fungibility, (2002) (Philippines) children is higher by 80% of the caloric redistribution value from the school feeding program. Effects are weaker for children in poorer households. Khandker Female Stipend Program Girls' secondary education increases sub- Implementation, and others (Bangladesh) stantially, and boys' education decreases equity-efficiency (2003) in some instances. Benefits from the pro- tradeoff, gram accrue disproportionately to girls fungibility from households with large landholdings. Kremer and Western Kenya Negative social learning effects lead to Learning Miguel Deworming Project free-riding due to the positive externalities externalities (2003) of deworming treatment. Subsidies need to be continued indefinitely. Miguel and Western Kenya Schoolparticipation increased in the control Implementation, Kremer Deworming Project group as result ofa deworming program. physical (2003) There are externalities for students in treat- externalities, ment schools and neighboring schools. cost-effectiveness Miguel and Western Kenya Small increase in price of deworming pills Implementation, Kremer Deworming Project lead to large reduction in take-up rates. cost-effectiveness (2004) Health education and community of program mobilization programs failed. Latrines compared with and boreholes were far less cost-effective alternative than provision of deworming pills. counterfactuals Ravallionand Food for Education Program increases schooling far more Implementation, Wodon (Bangladesh) than it decreases child labor. equity-efficiency (1999) tradeoff, fungibility Schultz Progresa (Mexico) Program has a positive impact on school Implementation, (2001) attainment. For most families in rural equity-efficiency Mexico, the effect is a wealth effect. tradeoff, fungibility Stifel and Vaso de Leche (Peru) Program is well targeted to the poor and Targeting, Alderman has minimal leakage. Program has no fungibility, (2003) impact on the main goal of increasing implementation child height. cash transfers can increase welfare in the society, either by protecting people from their own irrationalities or by providing incentives for them to gather more informa- tion (if forced to send a child to school, parents will find out more about the value of education). 60 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Most of the empirical literature focuses on the efficiency rationale arising from mismatched interests and the equity rationale. The findings are generally positive, in that conditional cash transfer programs often meet their stated efficiency or equity objectives. When used to induce greater investment in human capital, they do lead to increases in schooling and greater use of health resources. Similarly, when used to target resources to the poor, they do ensure that the poor receive more than the rich. However, there is sometimes tension between the efficiency and equity objec- tives. For instance, conditional cash transfer programs may lead to greater school participation, but this may be accompanied by larger transfers to the rich. This tradeoff is not new to policymakers. Several conditional cash transfer pro- grams were implemented with efficiency in mind and with an explicit attempt to ensure that the rich did not receive more than the poor by constructing eligibility criteria that dictated who could receive the cash transfers. The means-tested eligibility criteria ranged from sophisticated measures, such as household income (Bolsa Familia in Brazil), to less demanding correlates of poverty, such as land ownership or employment (Food for Education in Bangladesh). Any such means testing requires additional expenditures and the careful collection of household data. As the popular- ity of these programs grows and the emphasis shifts to quick results on goals set by donors, governments may be less willing to invest in such expensive efforts. The tradeoffs then become starkly apparent. The overall impact of the program will depend on the gains and losses associated with each. The economic analysis of the efficiency and the equity objectives are different, and this article is structured accordingly. It first presents the textbook case of a condi- tional cash transfer, which shows that in the standard economic framework uncon- ditional cash is better than conditional cash. Consequently, whether a conditional cash transfer program is effective depends on how well it addresses market failures arising from mismatched preferences and how well it targets resources to a particu- lar group. The article discusses these issues in turn, with an emphasis on the prob- lems that programs have faced in meeting their stated objectives. It then examines the overall costs and benefits of conditional cash transfer programs, looking at the combined effects of conditionality on efficiency and equity. Conditional Cash Transfers: The Textbook Example and Rationales This article defines any program requiring a specified course of action to receive a benefit (which may be in cash or in-kind) as a conditional cash transfer. This definition includes cash transfersbasedon human capitalinvestments(such asschoolingand health), but it is broad enough to encompass other programs, such as workfare (cash contingent on working in a program) and consumption transfers (in-kind transfers), among others. Jishnu Das, Quy-Toan Do, and Berk Özler 61 Though different in ultimate form, workfare and in-kind transfers are conceptually iden- tical to conditional cash transfers. Workfare is an income transfer conditional on taking the occupation proposed by the program. Similarly, in-kind transfers are cash transfers conditional on the purchase and consumption of the relevant commodity. Not surpris- ingly, the economics of these three types of transfers is similar, and the empirical insights gained from the analysis of one are relevant to the others. Figure 1 shows a conditional cash transfer program from the point of view of the household in a standard economics framework. The household can consume two goods, X (say, education) and Y. The maximum amount that the household can consume prior to the scheme is given by the budget constraint AB. That is, if a household spends all its income on X, it can consume up to the amount B; similarly, if it spends all its income on Y, the maximum it can consume is A. After the pro- gram's implementation, the budget constraint is given by AEDC. Under this new budget constraint, if the household consumes at least X0, it receives an additional income, ED. However, if the household consumes less than X0, it does not receive an Figure 1. Households and Conditional Cash Transfer Programs (Milk) Type I household Y A Type II household D Original budget constraint E Type III household X0 XIII B C X (Education) 62 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) income transfer and remains on its pretransfer budget constraint, AE. X0 represents the condition--to receive the extra income, the household must consume at least this amount of the good. The indifference curves for three types of households are shown as type I (dotted), type II (dashed), and type III (solid). Their behavior is very different. Type I does not participate in the program and remains in the same place on its budget constraint, AE, consuming less than the required amount, X0. Type II moves from its budget constraint, AE, where it consumes less than X0, to consume X0 when it participates in the program. Finally, type III consumes more than X0 before the program is intro- duced and continues to do so after. In this example, conditional cash transfers are worse than distributing an equivalent amount of unconditional cash. If households were given cash without the condition at X0, the budget constraint would be given by the line CD extended upward to meet the y axis. For type I and type II households the conditional cash transfers is strictly worse than an equivalent amount given in cash--these households move to a higher indifference curve when given unconditional cash. For type III households the conditional cash transfer is equivalent to an unconditional cash grant. The logic inherent in a conditional cash transfer program drives the intuition behind these results. By imposing conditions, the policymaker provides incentives for households to take an action that they would not ordinarily take on their own (otherwise why have the condition in the first place?). But if that action is different from what households would have chosen on their own, their resulting welfare must be lower--by distorting the consumption choices of households conditional cash transfer programs reduce welfare compared with unconditional cash grants. The only households for which conditional cash transfers and unconditional grants produce equivalent outcomes are the ones that do not experience any distortion in their consumption decisions, represented by the type III households that were already consuming X0. Implicit in this example is the absence of a market failure, meaning that decisions made by individual households are also optimal for society as a whole. Suppose instead that from a societal point of view, type II households underinvest in educa- tion: when making educational decisions, they do not take into account the effect of their own education on others around them. In this case, a cash transfer conditional on education could lead to a larger increase in social welfare than an unconditional cash grant, even though the welfare of the individual household could be reduced. This is the first rationale for conditional cash transfer programs: they can be used to improve efficiency when there are underlying market failures in the economy. A second rationale for conditional cash transfer programs relates to equity and redistribution. Suppose policymakers are interested in providing cash only to a cer- tain segment of the population, such as poor households. The problems that they face are that poor households are hard to identify and that budgets are small. One Jishnu Das, Quy-Toan Do, and Berk Özler 63 solution is to screen beneficiaries so that poor households participate (and receive the cash) and rich ones opt out. Conditions can achieve this exact result. The idea behind successful screening is that the benefits of the program (the cash received) minus the costs of the condition (the utility loss from being forced to consume a cer- tain amount) will be different for poor households than for the rest of the population. In particular, if the benefits of the cash received exceed the cost of the condition for poor households only, then they will self-select into the program. In figure 1, type II and type III households self-select into the program and type I households opt out. Thus the second rationale for conditional cash transfer programs is their use as screening devices to target participants when individual characteristics cannot be easily observed. These two rationales are very different.2 In the first case, the objective of policy- makers is explicitly to increase consumption of the conditioned-on good, whereas in the second case the condition matters only insofar as it determines who self-selects into the program. Efficiency and Conditional Cash Transfers The example in figure 1 shows that when societal and individual preferences are dif- ferent, conditional cash transfers can help increase overall welfare. This leads to two important questions. First, when do externalities arise and what empirical evidence exists on these market failures, especially in low-income countries? Second, if these market failures are frequent, particularly in certain sectors, what are the problems that conditional cash transfer programs face in achieving their stated objectives? Market Failures and the Use of Conditional Cash Transfers Several examples from the recent literature are useful for examining when external- ities arise.3 Although the studies reviewed here address different concerns, they share a common theme. In each one, an inefficiency arises because of an externality whereby the actions of one individual directly affect the welfare of others and because of the inability of individuals to contract out the externality by appropriately rewarding or punishing others for their actions. Together, these cause a divergence of societal preferences from individual decisions. More efficient outcomes could then be achieved through the use of conditional cash transfers. Direct Externalities. Two types of externalities that have received attention in the recent literature are physical externalities and learning externalities. Physical exter- nalities arise in several contexts. For instance, Miguel and Kremer (2003, 2004) study a program that provided free deworming treatment to rural primary school 64 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) children in Kenya. The deworming program resulted in larger positive health and school participation effects for children in the treatment group than for children in the control group (who received no treatment). Furthermore, the study points out that "there is existing only limited empirical evidence on deworming treatment externalities, but that which exists suggests that school-based deworming may cre- ate substantial externalities" (p. 162). These externalities are confirmed for the pro- gram: The program had positive effects for untreated students in treatment schools and students in neighboring schools as well. The researchers conclude that the externality could be more than 20 times as large as the cost of the program. Starting with the premise that individual households do not take into account the positive benefits for the whole community of their individual consumption of deworming pills, the overall consumption of deworming treatment is likely to be less than the socially optimum amount. Consequently, there is a clear role for policymakers to subsidize deworming treatment to restore efficiency.4 Learning externalities arise frequently in technology adoption scenarios. Once the attributes of a new technology are known to a community, there are significant benefits. Nevertheless, learning about a new technology often involves costly experimentation. Households that do not take into account the benefits that their early experimentation will have on the community's stock of knowledge may be unwilling to invest time and effort, leading to a free-rider problem. These households will prefer to wait and see rather than experiment with the new technology. As in the previous example, individual household decisions lead to underinvestment in learning. There is considerable evidence to suggest that learning externalities play an important role in adopting technology. For example, Foster and Rosenzweig (1995) find that farmers are more likely to adopt a new technology when their neighbors do but are less likely to make the first move, preferring to wait for their neighbors to experiment first.5 Household Bargaining. Market failures that can be remedied through conditional cash transfer programs also arise when decisions are made through a bargaining process within the household.6 For example, recent literature on child labor argues that the inefficiency of child labor and underinvestment in schooling arise due to a mismatch between parents' preferences and children's interests (Baland and Robinson 2000). If children could promise to pay their parents for their own education, educa- tion levels would be higher. But because education decisions for children are made by parents and because children cannot make such promises, parents favor positive short-run outcomes that benefit them relatively more (including increased income from their child's labor) to long-term returns that accrue mostly to children. In empirical studies Kochar (2000a) shows that there are significant differences between the parents' and the child's rate of return to children's education and that parents' returns overwhelmingly determine the amount of education a child receives. Jishnu Das, Quy-Toan Do, and Berk Özler 65 Moreover, education levels in rural households respond strongly to labor market returns in urban areas (Kochar 2000b). Kochar's work suggests that rural families underinvest in education because they are unable to guarantee that their children will continue to reside in the village once they are educated: The inability of children to commit to such action leads to the underprovision of education. In this case, con- ditional cash transfer programs that provide cash only if the child attends school, such as Bolsa Familia, increase efficiency by addressing the difference between parental preferences and children's interests.7 Using Conditional Cash Transfers to Address Market Failures Suppose that a policymaker is convinced of the importance of externalities and therefore that conditional cash transfers are an appropriate solution. Several problems may still arise in the program's ability to meet the stated objective of, say, increasing human capital investments. Two important issues are participation and fungibility. Low Participation. The first requirement for a conditional cash transfer program to achieve its objective is that individuals participate. (This holds true for both programs used to boost efficiency and programs used to target resources.) Thus, programs with stipends for school attendance should lead to higher enrollment and atten- dance rates. Similarly, in workfare designed to target poor households, individuals should be willing to undertake the work required to receive the benefits. With low participation, a program is ineffective. Conceptually, though, the partic- ipation problem is related to the size of the transfer and the cost of the condition. For example, suppose policymakers want to get children into school and start out by offering parents $5 to send their children to school. If participation is low, the amount of cash offered can be increased until the point at which all parents enroll their children. Thus, if participation is low, either the amount of cash offered is too small or the cost of the condition is too large.8 The importance of the participation problem is reflected in the fact that the large evaluation literature emphasizes program uptake. Bolsa Familia in Brazil is a condi- tional cash transfer program with the objective of increasing school enrollment and reducing child labor for children ages 6­15. The program provides transfers to chil- dren from households below a certain income level who attend school at least 85 percent of the time. Bourguignon and others (2002) estimate the ex ante impact of the program and find large effects on enrollment, with a decrease in the share of chil- dren not enrolled (from 5.8 percent to 3.9 percent) and stronger effects among the poor (from 9.1 percent to 4.7 percent). These results are partially confirmed in an ex post evaluation of the same program (Cardoso and Souza 2003). In Bangladesh, two important programs to increase school enrollment, Food for Education and the Female Stipend Program, both resulted in increases in enrollment. 66 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) As with Bolsa Familia, both programs conditioned benefits, food in the first and cash in the second, on school attendance. The effects were large and significant. Ravallion and Wodon (1999) show that under Food for Education an extra 100kg of rice increases the probability of school enrollment by more than 15 percent for both boys and girls. For the Female Stipend Program, Khandker and other (2003) estimate that an additional year of participation in the program leads to a 8 percent increase in girls' enrollment. The Fungibility Problem. The second problem that conditional cash transfer pro- grams face relates to the fungibility of the conditioned-on commodity. Conditions work successfully when individuals are forced to take actions that they would not ordinarily take on their own. But this logic automatically creates an incentive for individuals to try to offset the welfare loss imposed by the condition. The ability of individuals to offset any distortion from the program is the problem of fungibility, which usually arises when there is a close substitute for the conditioned-on com- modity (this could be the commodity itself). In this case, recipients can offset the dis- tortion imposed by the condition by decreasing consumption of the substitute, so that overall amounts are unchanged by the program, even when the condition is satisfied.9 As an example of fungibility, consider a program that seeks to improve learning outcomes by giving parents cash conditional on the purchase of educational materials (such as school supplies). Such a program may report high uptake (that is, all par- ents buy shcool supplies with the cash), but it will have minimal effects on learning outcomes if households cut back on their own funding for school supplies. If parents were already buying school supplies, the program would make school supplies free without increasing the total amount of school supplies available to the child. Follow- ing up on the previous discussion, the behavior of type III households in figure 1 shows how fungibility can be a problem. Because this group was already consuming the necessary amount of the good, the increased consumption of the commodity after the program is introduced is no greater than it would have been under an unconditional income grant. For type III households, the commodity is perfectly fun- gible. In less extreme cases, substitution may take the form of decreasing consump- tion of a close substitute (eating less spinach when given iron tablets), changing patterns of consumption (eating less at home when given food in school), or even reallocating investments within the household (sending fewer boys to schools when girls are given a stipend). How can policymakers evaluate whether fungibility is a problem? The literature discusses two approaches. One approach is to directly estimate the program's impact on close substitutes of the conditioned-on good. Thus in the case of educational materials, the impact of providing supplies in school on the household purchase of supplies would be estimated. The second approach is to examine an outcome that Jishnu Das, Quy-Toan Do, and Berk Özler 67 depends on both the conditioned-on good and the substitutes. Learning outcomes depend on supplies provided both in school and at home. If, as a consequence of pro- viding supplies in school, there is an increase in learning outcomes, there must have been an increase in the overall amount of educational materials available to the child. Both methods have advantages and drawbacks. The first can directly estimate the extent of substitution (the decreased provision of educational materials at home), whereas the second can only indicate whether substitution was one for one. But the first method requires clear knowledge of what the substitutes actually are. The sub- stitutes for educational materials are straightforward, but they may not be for other goods. A mother may substitute for better healthcare in the clinic in a number of ways (washing hands, boiling water, and the like), which may be difficult to list, let alone measure. By concentrating on outcomes, such as child height, that depend on healthcare both in the clinic and at home, the reduced-form impact of the program can be estimated. Jacoby (2002) is one of the few sources of a direct estimate of substitution. Studying a school feeding program in the Philippines, Jacoby assumes that households smooth consumption, so that daily caloric consumption is the same on school and nonschool days. By comparing the caloric consumption of children on school and nonschool days, Jacoby finds that households do not substitute away calories from the program. On school days children's caloric intake is 80 percent higher than on other days--the difference accounted for by the caloric value of the feeding program. Behrman and Hoddinott's (2001) evaluation of Progresa in Mexico and Stifel and Alderman's (2003) analysis of Vaso de Leche in Peru use the second approach. Under Progresa, eligible mothers had to visit clinics regularly and participate in growth monitoring to receive nutritional supplements and cash transfers. Behrman and Hoddinott find that Progresa's preschool child nutrition programs had substantial positive impacts on growth and reduced the probability of stunting for children ages 1­3. Behrman and Hoddinott postulate that the program addresses externalities arising from household bargaining: Progresa directs its resources to mothers based in part on prior evidence from other populations that suggest better child health and nutrition result when women rather than men receive resources. Under Vaso de Leche, the largest social transfer program in Peru, select house- holds receive milk or milk products. Although the program is well targeted to the poor, Stifel and Alderman (2003) find that it fails to achieve its main nutritional objective of improving child height. They argue that the reason for the lack of impact is that half the in-kind transfers are inframarginal (that is, to type III house- holds), meaning that the provision of milk did not have an impact beyond the income transfers it represented. Stifel and Alderman conclude that cash transfers may be a better means of increasing consumption of milk if they are less costly and more efficiently distributed than food transfers. But as van de Walle (1998) 68 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) suggests, in-kind transfers might be more politically feasible than cash transfers. Whether political gains outweigh implementation cost in this instance remains an open question. A natural question is whether fungibility is an issue in all conditional cash transfer programs or only in particular types, such as nutritional interventions. For instance, if a program leads to higher school enrollment, is there an associated fungibility problem? Although there is limited empirical evidence, the answer seems to be yes. In particular, these programs may reduce the amount of education that is given to children at home or affect the composition of education across children. Khandker and others (2003) examine whether providing stipends only for girls in the Female Stipend Program in Bangladesh had an effect on boys' schooling--because the pro- gram reduces the price of girls' education compared with that of boys' education. The evidence is mixed. Using household data, the authors find that the effect on boys' schooling is statistically insignificant. But using school enrollment data they find a 29 percent decline in boys' enrollment in program schools. Equity and Conditional Cash Transfers The use of conditional cash transfers as a screening mechanism has a long history in workfare and food distribution programs. Recent reviews by Ravallion (2003) and van de Walle (1998) document the ways these programs have been evaluated as well as their ability to lift households out of poverty. These reviews show that suc- cess is closely tied to whether the condition simultaneously satisfied two criteria: that the targeted group was willing to participate in the program and that the group not targeted found the condition too expensive compared with the program's rewards. Means Testing Because of the difficulty of meeting these criteria, policymakers adopt targeting pro- grams that completely avoid conditions and directly target cash based on observable household characteristics. This method, known as means testing, collects more data on the identity of individuals and thus avoids the need to devise screening mecha- nisms. Innovative programs with different eligibility criteria have emerged during the past decade. In some programs (such as Progresa in Mexico or Bolsa Familia in Brazil) means tests are based on detailed data collection and identification of house- hold wealth (Bolsa Familia, for instance, provides cash only to households with per capita monthly incomes less than 90 reais). In cases where data collection on house- hold assets may be too expensive or politically infeasible, other proxy indicators of wealth have been used. Jishnu Das, Quy-Toan Do, and Berk Özler 69 Proxy indicators are used in the Food for Education program in Bangladesh and the Vaso de Leche program in Peru. Food for Education provided food to children who attended school at least 85 percent of the time. An attempt was also made to ensure that only poor households received the transfer. In the absence of data on household income or assets, the eligibility criteria were based on easily observable correlates of poverty--whether the household head is a widow, the amount of land owned by the household, and the employment status of the household head. These criteria ensured that poor households received greater transfers--but not by much. Galasso and Ravallion (forthcoming) show that the difference in receipts between rich and poor households was marginal. In the Vaso de Leche program, a committee was in charge of observing the rel- evant characteristics of households to determine eligibility. Despite the potential ambiguities of such a strategy, the program did result in marginally greater transfers to poor households. Using measures of participation and expenditure levels, Stifel and Alderman (2003) show that the Vaso de Leche program resulted in the allocation of more than 60 percent of the value of all transfers to poor households. When means testing is feasible, the problem of targeting resources to a particular subset of households is reduced. The extent to which this becomes less of a problem depends on the sophistication of the data--in cases such as Bolsa Familia, where the means test is based on detailed household information, targeting is more efficient than in, say, the Food for Education program. Screening Where means tests are logistically or politically infeasible, conditional cash transfers can be used as screening mechanisms so that only members of the tar- geted group self-select into the program.10 One example of particular interest is targeting to the poor. In such cases, the requirements for self-selection can be expressed in terms of the income effect of the conditioned-on good, so that tar- geting is successful if the conditioned-on good is inferior.11 Rich households (that consume less of the good to start with) will then find the costs of meeting the condition higher than poor households do and will disproportionately opt out of the program. In figure 1, the condition X0 induces exactly this sort of screening. Type II and type III households (less wealthy) participate in the program, whereas type I households (more wealthy) voluntarily opt out. The good is inferior because consumption decreases with wealth. Examples of conditional cash transfers as screening mechanisms include well-documented workfare programs (van de Walle 1998; Galasso and Ravallion 2003), rationing of food or health subsidies by queuing (Alderman 1987), and packaging of commodities that are unappealing to the rich (Alderman and 70 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Lindert 1998; Jacoby 1997). Nevertheless, the requirements for conditional cash transfers as screening mechanisms may be hard to satisfy. Argentina's Plan Jefes y Jefas was designed to increase employment by providing a subsidy only for individuals who were trying to find employment. Household heads who were initially unemployed could apply for the program, which provided cash if they found employment and undertook 20 hours of activities, such as basic community work, training activities, school attendance, or employment in a private company with a wage subsidy for 6 months (Galasso and Ravallion 2003). Why was the condition of counterpart activities imposed? In Argentina it is hard to verify whether an individual is unemployed, because over half of all employment is in the informal sector. The 20-hour requirement acts as a screening device: Individuals who are already employed are unwilling to sacri- fice their work (or leisure) time for the benefits of the program, whereas those who are currently unemployed face a lower opportunity cost for participating. Galasso and Ravallion (2003) find that the condition was only partially successful at ensur- ing that the already employed did not participate--and even less successful at increasing employment. Only 3 percent of beneficiaries had formal employment to start with, so the program did ensure that those with formal employment did not participate, but 19 percent of beneficiaries were already employed in the informal sector to start with. At the same time, increases in the employment rate were much less than expected due to the program's inability to distinguish between individuals who were inactive (out of the labor force) and those who were unemployed (actively searching for a job but unable to find one). In fact, 38 percent of eventual recipients were inactive, lead- ing Galasso and Ravallion (2003) to conclude that a large share of the participants were women who would not otherwise have been in the labor force. The condition thus failed on two counts: It was not expensive enough to ensure that individuals employed in the informal sector opted out, nor was it expensive enough compared with the opportunity cost of losing 20 hours of housework. In contrast to the previ- ous evaluations, where higher program uptake is always better, this example illus- trates the idea that program uptake can also be too high compared with its stated objectives. Conditional Cash Transfer Programs and the Efficiency-Equity Debate The preceding discussion suggests that conditional cash transfer programs have an impact on both efficiency and equity, regardless of their initial objective. In some cases the program can have a positive impact on both. Particularly when the condi- tion is imposed on an inferior good, the provision of cash could lead to an increase in, Jishnu Das, Quy-Toan Do, and Berk Özler 71 say, human capital while simultaneously ensuring that the rich select out of the pro- gram. In this case, both efficiency and equity would be enhanced. One example of such multiple positive impacts is conditioning cash on attendance in public schools in places where the rich use private schooling (a fairly common practice in South Asia). In this case the transfer would result in greater school enrollments. Further, since the rich do not use public schools, the benefits of the program would be enjoyed (mostly) by the poor--both efficiency and equity objectives would be met. In other instances, though, there can be tension between efficiency and equity. Using conditional cash transfer programs to increase human capital investment could have an adverse impact on equity; similarly, the distortions required for self- selection to work could impose an efficiency cost. Policy-makers tend to be aware of the contradictory roles that conditional cash transfers are sometimes required to ful- fill. Eligibility criteria and means tests are introduced in programs such as Bolsa Familia and Progresa specifically to minimize adverse redistributive impacts. In other cases, the net impact of the tradeoff is less clear. Efficient but Not Equitable? Consider, for instance, the Female Stipend Program in Bangladesh, which was implemented to increase secondary school enrollment of girls. Under this program the government gives stipends to girls who attend at least 85 percent of classes. What differentiates this program from Food for Education is the complete lack of means testing. There are no eligibility criteria besides the attendance requirement. Khandker and others (2003) evaluate the program and find that girls' enrollment rates in secondary education increase substantially. If these rates were inefficiently low to begin with, the program served to decrease the gap between societal prefer- ences and individual actions. But in Bangladesh secondary public schooling for girls is a normal good, with rich households more likely than poor households to enroll their children. For extremely poor households, the opportunity cost of 85 percent attendance in terms of labor lost likely exceeds the benefit of the stipend. Conse- quently, Khandker and colleagues find that the program has adverse distributional impacts: The "currently untargeted stipend disproportionately affects the school enrollments of girls from households with larger land wealth. Targeting towards the land poor may reduce the overall enrollment gains of the program while equalizing enrollment effects across landholding classes" (p. 25). Note that the adverse redistributive impact is not a sufficient reason to discon- tinue the program. One approach to address this problem, as suggested by the authors, is to introduce means tests in addition to increasing the amount of the cash transfer. Indeed, when the condition is imposed on a normal good, means testing becomes a screening tool that is a complement to the conditional cash transfer program, alleviating the adverse redistributive impact of the condition. 72 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Equitable but Not Efficient? Using conditional cash transfers as a screening device for targeting also comes at a cost, which arises from the distortion in consumption and investment choices induced by the condition. An example is food subsidies targeted to the poor, where the quality of the food is low. Because low-quality food is an inferior good (as people's incomes increase, the quality of the food they eat increases), this would presumably lead to greater food subsidies for the poor than for the rich. The Nutribun and Milk program in Jamaica is one such example. Under this program, children in school received a nutritious baked product every day. Jacoby (1997) finds that the program achieved its aims in terms of targeting--lower uptake among the rich implied that the benefits from the program as a share of total consumption fell from 4.8 percent for the poorest decile to 1.2 percent for the richest decile. The effective targeting did not necessarily make for a successful program, how- ever. Jacoby (1997) argues that the deadweight loss of the program was large. By using the notion of equivalent variation--the smallest increment in cash that would have achieved the same objective--Jacoby shows that the benefit per respondent was J$152, compared with the program cost of J$400. Thus more than half of the costs of the program were estimated deadweight losses due to the implied condi- tional nature of the transfer. Whether the benefits of targeting exceeded this loss is an open question. Although Argentina's Plan Jefes y Jefas resulted in greater benefits to the poor, the efficiency cost in terms of forgone incomes from the condition (the 20-hour work requirement) is unclear. Galasso and Ravallion (2003) estimate the average oppor- tunity cost to be around 50 pesos--a third of the amount of cash transferred by the program itself. A possible next step is to trace out the losses in targeting effectiveness from decreasing the work condition against the gains from increasing efficiency. In this case means testing is a substitute for conditional cash transfer programs. The costs of implementing means tests should then be weighed against the efficiency losses necessary to obtain self-selection of participants. Arguably, the optimal redis- tribution tool is a combination of the two. Beyond Efficiency: Political Economy The two rationales discussed in this article, efficiency and equity, form the basis for several conditional cash transfer programs, but they are by no means an exhaustive list. A third rationale relates to the incentives of the institutions that provide the cash for the program, specifically whether they are country governments or donor organizations. Governments that are primarily concerned with targeting resources to poor households can ill afford to ignore the fact that the public may favor work requirements Jishnu Das, Quy-Toan Do, and Berk Özler 73 or school attendance in exchange for public assistance rather than cash payments with no strings attached (van de Walle 1998). On the other hand, some leakage of program benefits to nonpoor households in primarily efficiency-increasing pro- grams may be more appealing to middle-class voters, increasing the total budget available for such programs (Gelbach and Pritchett 2002). These political economy considerations play a vital role in the design of conditional cash transfer programs. Conditional cash transfer programs also serve the direct purpose of imposing donor preferences when tastes over allocations differ (Culyer 1991; Tobin 1970). Even without different preferences, an agency view of policy-making can provide a rationale for why conditional cash transfers are desirable and increasingly popular. Because aid agencies are accountable to donors and ultimately taxpayers, transpar- ent and observable measures of performance need to be implemented to ensure proper monitoring. Conditional cash transfer programs are thus suitable tools for aid agencies to move toward the announced objectives, although the exclusive focus on measurable dimensions of performance may come at the expense of more compre- hensive and efficient development policies, whose impacts are difficult to assess.12 To illustrate this, consider the Millennium Development Goals for education. Target 3 is to "ensure that, by 2015, children everywhere, boys and girls alike, will be able to complete a full course of primary training."13 This objective includes enrollment of all school-age children and universal primary completion. But only the enrollment rate is officially included in the Millennium Development Goal indicators--not the primary completion rate. The first is a clearly identifiable indicator, but the second is fuzzy, with differing definitions and scales across countries. This problem also arises due to an externality--in making schooling decisions, it is unlikely that parents take into account the idea that enrollment might induce additional aid flows from donor countries. Imposing conditions then yields considerable political gains in the pro- duction of verifiable and clearly observable outcomes. The stress on verifiability may come at a cost. The first round of conditional cash transfer programs, including Bolsa Familia and Food for Education, were careful in considering the tradeoffs between equity and efficiency. Later programs, such as the Female Stipend Program in Bangladesh, were not. Because the Millennium Develop- ment Goals for education do not require minimizing the adverse distributional impacts of educational policies, the temptation to implement programs that perform better only on the verifiable dimension may be high. This may lead to conditional cash transfer programs that enhance efficiency but worsen equity or that yield immediate gains but impose long-run costs. This article has compared conditional cash transfers with their unconditional counterparts. Even though the previously discussed rationales may justify the use of conditional cash transfer programs, these programs need to be compared with other policy interventions. For example, Case and Deaton (1999) find that pupil-teacher ratios in South Africa have significant positive impacts on enrollment, educational 74 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Figure 2. Flow Chart Textbook case: Unconditional transfers are better than conditional transfers Why have conditional cash transfers? STEP 1 Efficiency rationale: Redistributive rationale: Improve efficiency by Target transfers to the poor reconciling societal when there is asymmetric preferences with individual information (the identity of the decisions poor is unknown) Why might conditional cash transfers not have the desired impact? Design and implementation: Fungibility: Individuals undermine Very low participation in the the rationale of the program by STEP 2 program because the benefit of changing their consumption or the transfer is less than the cost investment patterns for a close of the condition. This is due substitute of the conditioned-on good. either to poor design (too little Examples include decreasing the cash in the design of the consumption of oranges when given program) or poor vitamin C tablets and cutting back on implementation (corruption, food intake at home when given food leakage) in school Tradeoffs: Is there a tradeoff between the efficiency and the redistributive rationale? That is, do conditional schemes that seek to restore efficiency have adverse redistributive impacts? Alternatively, do redistributive schemes have adverse efficiency impacts? STEP 3 Cost-benefit analysis: How big are the externalities that the conditional cash transfer addressed in the first place? Jishnu Das, Quy-Toan Do, and Berk Özler 75 achievement, and test scores for numeracy. Should South Africa then use main- stream conditional cash transfer programs to improve educational outcomes? Although the programs would arguably have immediate political impacts, the alter- native of reducing class sizes could have long-lasting positive effects that are harder to measure in the short run. These two alternative policy instruments illustrate the tradeoff between short-term and long-term impacts. These results also suggest a quantity-quality tradeoff: Increasing enrollment (through the implementation of conditional cash transfer programs, for example) without supplying additional teachers will be harmful to educational achievements and test scores in the long run. Though an emphasis on demand-side policies (such as conditional cash trans- fers) brings political gains and ultimately future aid flows, it comes at the expense of other policy instruments. These tradeoffs require careful consideration. Conclusion This article examines a wide range of conditional cash transfer programs and tries to understand and interpret various evaluations in the literature. Figure 2 synthesizes the discussion. Two important ideas emerge: · Conditional cash transfer programs can be justified on efficiency grounds to alleviate market failures or on equity grounds to redistribute resources. The choice of the good on which the condition is imposed is crucial. In particular, the extent to which the good is fungible makes a significant difference, depending on the underlying motivation for the program. · The equity and efficiency rationales, while theoretically distinct, exist simulta- neously in several conditional cash transfer programs. The extent to which the efficiency and equity objectives result in a tradeoff can be addressed through the condition requirement, the amount of the cash conditionally transferred, and the refinement and enforcement of eligibility criteria. A review of the literature suggests that researchers have typically addressed these issues separately. However, policymakers need to incorporate several factors in a comprehensive framework to design optimal programs. Notes Jishnu Das is an economist in the Development Research Group at the World Bank; his e-mail address is jdas1@worldbank.org. Quy-Toan Do is an economist in the Development Research Group at the World Bank; his e-mail address is qdo@worldbank.org. Berk Özler is an economist in the Development Research Group at the World Bank; his e-mail address is bozler@worldbank.org. This article draws from Das and others (2004), in which the main arguments made here are discussed in greater detail. 76 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) The authors are grateful to Harold Alderman, Francisco Ferreira, Jed Friedman, Emanuela Galasso, Hanan Jacoby, Michael Kremer, Peter Lanjouw, Martin Ravallion, Carolina Sanchez-Paramo, Adam Wagstaff, and three anonymous referees for comments and helpful discussions. 1. A review of the literature reveals little empirical support for these rationales. This is partly because empirical investigations of departures from rationality, especially with respect to investments in human capital, are hard to implement. In laboratory experiments, behavioral economists have systematically demonstrated that individuals behave very differently from what would be predicted by economic models of a rational human being. Taking these experiments to the field has proved much more difficult and remains an area for future research. Similarly, systematic evidence for informational problems remains sparse. In fact, limited evidence in the case of the deworming experiment suggests that information was not the constraining factor--even after substantial experience with the pills and their effects had been built up, households were unwilling to pay even a tiny sum for their purchase, suggesting that subsidies may have to be continued indefinitely to sustain high take-up of the conditioned-on good (Miguel and Kremer 2003). 2. Another rationale for conditional cash transfer programs is the presence of credit constraints or credit rationing. The theoretical justification in this case is still unclear. Credit-rationed individuals face a tighter static or dynamic budget constraint. As this discussion suggests, credit rationing is optimally alleviated with unconditional income transfers rather than conditional cash transfers because condi- tions introduce an unnecessary distortion. 3. The review presented here is not exhaustive. Some cases of market failures have been omitted. In particular, moral hazard arguments, whereby aid is believed to induce lower effort on the part of recipi- ents, are an important omission (see Ravallion 2003). Targeting aid to poor people might create an incentive for them to remain poor, leading to excess consumption instead of investment. In this case, conditional cash transfers would encourage investment instead of immediate consumption. Another omitted argument (with little empirical evidence) is the case of bounded rationality. If individuals are unable to undertake appropriate investment decisions due to bounded rationality or behavioral prob- lems such as hyperbolic discounting, conditional cash transfers would act as substitutes for individual decisionmaking, with the government behaving essentially as a parental agency (Basu 2003 discusses this issue as it regards child labor). A third rationale relates to the political economy of aid and is addressed in the last section of this article. 4. The authors also state that "it is difficult to draw conclusions about optimal deworming subsidies in the absence of a fully-fledged behavioral and epidemiological model" (p. 207). Although positive externalities were large on average, it is hard to identify marginal externalities because they depend on how many others are being treated. 5. Just as individuals can learn from their neighbors, their preferences can also be influenced by the behavior of others in their community. For example, believing that social norms drive gender discrimination may offer the government scope to positively discriminate in favor of women to induce a communitywide change in beliefs or preferences. Bangladesh's Female Stipend Program can be seen as a conditional cash transfer program with such an objective (see Khandker and others 2003). 6. For discussions of how baragaining in households affects decisions, see Bourguignon and Chiappori (1994) and Basu (1999). 7. Bargaining models also suggest that conditional cash transfers are not the only way to address underinvestment in children. An alternative may be to provide direct income support to members of the household whose preferences are more aligned with the interests of their children. For instance, it is often assumed that mothers allocate resources more consistently with their children's needs than fathers do. This assumption has been empirically tested by Lundberg and others (1997), who examine the change in composition of household expenditures after a policy change in the U.K. child benefit program in the late 1970s. The policy change replaced the universal child benefit, which had consisted primarily of a reduction in the amount of taxes the father had to pay, with a cash payment to the mother. This represented a substantial redistribution of income from husbands to wives (about 8 percent of average male earnings in the United Kingdom by 1980). The authors find a substantial increase in spending on children (and women), relative to men, and conclude that "children do better when their mothers control a larger fraction of family resources" (p. 479). Jishnu Das, Quy-Toan Do, and Berk Özler 77 8. The small size of the transfer can arise either by design (if policymakers are unsure of the right size) or due to poor governance. Few studies have evaluated the governance argument, but studies by Ablo and Reinikka (2000), Reinikka and Svensson (2003), and Das and others (2003) suggest that large differences can exist between executed program budgets and the amount that actually reaches beneficiaries. 9. When households can access a secondhand market where the conditioned-on good is sold, fungibil- ity is a serious issue. Making sure that this problem does not occur is a prerequisite for the conditionality to make any sense. This article assumes, however, that the problem is well known and dealt with to the best of the policymakers' ability. 10. As discussed earlier, self-selection implies that the net benefit of the program is positive for the targeted group and negative for others. This "single-crossing condition" requirement is detailed in Spence (1973) and Rothschild and Stiglitz (1976). 11. The consumption of an inferior good decreases with wealth. By contrast, the consumption of a normal good increases with wealth. 12. For a formal treatment of this problem, see Holmström and Milgrom (1991). 13. See www.development goals.org/education.htm. References Ablo, Emmanuel, and Ritva Reinikka. 2000. "Do Budgets Really Matter: Evidence from Public Spend- ing on Education and Health in Uganda." Policy Research Working Paper 1926. World Bank, Washington, D.C. Alaii, Jane A., H. W. Van Den Borne, S. Patrick Kachur, Halima Mwenesi, John M. Vulule, William A. Hawley, Martin I. Meltzer, Bernard L. Nahlen, and Penelope A. Phillips­Howard. 2003. "Percep- tion of Bed Nets and Malaria Prevention before and after a Randomized Controlled Trial of Permethrin-treated Bed Nets in Western Kenya." American Society of Tropical Medicine and Hygiene 68(4):142­48. Alderman, Harold. 1987. "Allocation of Goods through Non-Price Mechanisms: Evidence on Distribution by Willingness to Wait." Journal of Development Economics 25(1):105­24. Alderman, Harold, and Kathy Lindert. 1998. "The Potential and Limitations of Self-Targeted Food Subsidies." World Bank Research Observer 13(2):213­29. Baland, Jean-Marie, and James A. Robinson. 2000. "Is Child Labour Inefficient?" Journal of Political Economy 108(4):663­79. Basu, Kaushik. 1999. "Child Labor: Cause, Consequence, and Cure, with Remarks on International Labor Standards." Journal of Economic Literature 37(3):1083­119. ------. 2003. "Global Labour Standards versus Freedom of Choice." World Institute for Development Economics Research Annual Lecture, November 10, Helsinki. Behrman, Jere R., and John Hoddinott. 2001. "Program Evaluation with Unobserved Heterogeneity, Selective Implementation, and Imperfectly Targeted Beneficiaries: The Mexican PROGRESA Impact on Child Nutrition." Research Paper 02­006. University of Pennsylvannia, Penn Institute for Economic Research, Philadelphia. Bourguignon, François, and Pierre-André Chiappori. 1994. "The Collective Approach to Household Behavior." In Richard Blundell, Ian Preston, and Ian Walker, eds., The Measure of Household Welfare. Cambridge: Cambridge University Press. Bourguignon, François, Francisco H. G. Ferreira, and Phillippe G. Leite. 2002. "Ex-ante Evaluation of Conditional Cash Transfer Programs: The Case of Bolsa Escola." Policy Research Working Paper 2916. World Bank, Washington, D.C. 78 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Cardoso, Eliana, and André P. Souza. 2003. "The Impact of Cash Transfers on Child Labor and School Attendance in Brazil." Instituto de Estudos do Trabalho e Sociedade, Rio de Janeiro, Brazil. Case, Anne, and Angus Deaton. 1999. "School Inputs and Educational Outcomes in South Africa." Quarterly Journal of Economics 114(3):1047­84. Culyer, A. J. 1991. "The Political Economy of Social Policy." Aldershot, U.K.: Gregg Revivals. Das, Jishnu, Stefan Dercon, James Habyarimana, and Pramila Krishnan. 2003. "Rules vs. Discretion: Public and Private Funding in Zambian Education." World Bank, Washington, D.C. Das, Jishnu, Quy-Toan Do, and Berk özler. 2004. "Conditional Cash Transfers and the Equity-Efficiency Debate." Policy Research Working Paper 3280. World Bank, Washington, D.C. Foster, Andrew D., and Mark R. Rosenzweig. 1995. "Learning by Doing and Learning from Others: Human Capital and Technical Change in Agriculture." Journal of Political Economy 103(6):1176­209. Galasso, Emanuela, and Martin Ravallion. 2003. "Social Protection in a Crisis: Argentina's Plan Jefes y Jefas." Policy Research Working Paper 3165. World Bank, Washington, D.C. ------. Forthcoming. "Decentralized Targeting of an Anti-Poverty Program." Journal of Public Economics. Gelbach, J., and Lant Pritchett. 2002. "Is More for the Poor Less for the Poor? The Politics of Means- Tested Targeting." Topics in Economic Analysis and Policy 2(1):1027. Holmström, Bengt R., and Paul Milgrom. 1991. "Multi-Task Principal-Agent Problems: Incentive Con- tracts, Asset Ownership and Job Design." Journal of Law, Economics and Organization 7:24­52. Jacoby, Hanan G. 1997. "Self-Selection and the Redistributive Impact of In-Kind Transfers." Journal of Human Resources 32(2):233­49. ------. 2002. "Is There an Intrahousehold `Flypaper Effect'? Evidence from a School Feeding Pro- gramme." Economic Journal 112(476):196­221. Khandker, Shahidur R., Mark M. Pitt, and Nobuhiko Fuwa. 2003. "Subsidy to Promote Girls' Secondary Education: The Female Stipend Program in Bangladesh." World Bank, Washington, D.C. Kochar, Anjini. 2000a. "Institutional Deterrents to the Spread of Schooling." Stanford University, Palo Alto, Calif. ------. 2000b. "Migration and Schooling Rates of Return." Stanford University, Palo Alto, Calif. Kremer, Michael, and Edward Miguel. 2003. "The Illusion of Sustainability." Working Paper W10324. National Bureau of Economic Research, Cambridge, Mass. Lundberg, Shelly J., Robert A. Pollack, and Terence J. Wales. 1997. "Do Husbands and Wives Pool Their Resources? Evidence from the United Kingdom Child Benefit." Journal of Human Resources 32(3):463­80. Miguel, Edward, and Michael Kremer. 2003. "Social Networks and Learning about Health in Kenya." University of California at Berkley. ------. 2004. "Worms: Identifying Impacts on Education and Health in the Presence of Treatment Externalities." Econometrica 72(1):159­217. Nahlen, L. Bernhard, John Paul Clark, and David Alnwick. 2003. "Insecticide-Treated Bed-Nets." American Journal of Tropical Hygiene and Medicine 68(4 suppl):1­2. Ravallion, Martin. 2003. "Targeted Transfers in Poor Countries: Revisiting the Tradeoffs and Policy Options." Policy Research Working Paper 3048. World Bank, Washington, D.C. Ravallion, Martin, and Quentin Wodon. 1999. "Does Child Labor Displace Schooling? Evidence on Behavioral Responses to an Enrollment Subsidy." Policy Research Working Paper 2116. World Bank, Washington, D.C. Reinikka, R., and Jacob Svensson. 2003. "The Power of Information: Evidence from a Newspaper Campaign to Reduce Capture." Policy Research Working Paper 3239. World Bank, Washington, D.C. Jishnu Das, Quy-Toan Do, and Berk Özler 79 Rothschild, Michael, and Joseph E. Stiglitz. 1976. "Equilibrium in Competitive Insurance Markets: An Essay in the Economics of Imperfect Information." Quarterly Journal of Economics 90(4):629­49. Schultz, T. Paul. 2001. "School Subsidies for the Poor: Evaluating the Mexican Progresa Poverty Program." Center Discussion Paper 834. Yale University, New Haven, Conn. Spence, A. Michael. 1973. "Job Market Signalling." Quarterly Journal of Economics 87(3):355­74. Stifel, David, and Harold Alderman. 2003. "The `Glass of Milk' Subsidy Program and Malnutrition in Peru." Policy Research Working Paper 3089. World Bank, Washington, D.C. Tobin, James. 1970. "On Limiting the Domain of Inequality." Journal of Law and Economics 13(2):263­77. van de Walle, Dominique. 1998. "Targeting Revisited." World Bank Research Observer 13(2):231­48. 80 The World Bank Research Observer, vol. 20, no. 1 (Spring 2005) Infrastructure Privatization and Regulation: Promises and Perils Ioannis N. Kessides Infrastructure is crucial for generating growth, alleviating poverty, and increasing international competitiveness. For much of the twentieth century and in most countries, the network utilities that delivered infrastructure services--such as electricity, natural gas, telecommunications, railroads, and water supply--were vertically and horizontally integrated state monopolies. But this approach often resulted in extremely weak services, especially in developing and transition economies and especially for poor people. Common problems included low productivity, high costs, bad quality, insufficient revenue, and shortfalls in investment. Over the past two decades many countries have implemented far-reaching institutional reforms--restructuring, privatizing, and establishing new approaches to regulation. This article identifies the challenges involved in this massive policy redirection within the historical, economic, and institutional context of developing and transition economies. It also reviews the outcomes of these policy changes, including their distributional consequences--especially for poor households and other disadvan- taged groups. Drawing on a range of international experiences and empirical studies, it recommends directions for future reforms and research to improve infrastructure performance. For much of the twentieth century and in most countries, network utilities--electricity, natural gas, railroads, water supply, telecommunications--were vertically and hor- izontally integrated state monopolies under ministerial control.1 Infrastructure's enormous economic importance, a desire to protect the public interest in industries supplying essential services, and concerns about private monopoly power led governments to conclude that control over these services could not be entrusted to the motivations and penalties of free markets. Governments also believed that given the large investments involved, public resources were required to increase infra- structure coverage. Accordingly, a single public entity usually controlled every © The Author 2005. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THEWORLD BANK . All rights reserved. For permissions, please e-mail: journals.permissions@oupjournals.org. doi:10.1093/wbro/lki003 20:81­108 aspect of a utility--facilities, operations, and administration--and determined which services to provide to essentially captive customers. The past decade has seen a dramatic change in views about how network utilities should be owned, organized, and regulated (Newbery 2000a). The new model calls for increased reliance on private infrastructure to improve efficiency, promote inno- vation, and enhance services. But after a series of financial crises, corporate scandals, and stock market collapses; the California electricity crisis; and blackouts around the world, clear guidance is needed on what to do about infrastructure--as well as reas- surance about (or qualifications of) earlier, more confident messages. What are the promises and perils of the new model? What principles should guide future efforts to restructure, regulate, an