S P D I S C U S S I O N PA P E R NO. 0542 90506 Brazil’s Bolsa Escola Program: The Role of Local Governance in Decentralized Implementation Alain de Janvry, Frederico Finan, Elisabeth Sadoulet, Donald Nelson, Kathy Lindert, Bénédicte de la Brière and Peter Lanjouw December 2005 Social Safety Nets Primer Series Brazil’s Bolsa Escola Program: The Role of Local Governance in Decentralized Implementation By Alain de Janvry, Frederico Finan, and Elisabeth Sadoulet University of California at Berkeley, Donald Nelson University of Arizona and Kathy Lindert, Bénédicte de la Brière, and Peter Lanjouw The World Bank1 December 2005 1 This paper is not a formal publication of the World Bank. Rather, it presents preliminary results of analysis. Citation and the use of such a paper should take account of its provisional character. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors and should not be attributed in any manner to the World Bank, to its affiliated organizations or to members of its Board of Directors or the countries that they represent. Social Safety Net Primer Series The World Bank Social Safety Nets Primer is intended to provide a practical resource for those engaged in the design and implementation of safety net programs around the world. Readers will find information on good practices for a variety of types of interventions, country contexts, themes and target groups, as well as current thinking of specialists and practitioners on the role of social safety nets in the broader development agenda. Primer papers are designed to reflect a high standard of quality as well as a degree of consensus among the World Bank safety nets team and general practitioners on good practice and policy. Primer topics are initially reviewed by a steering committee composed of both World Bank and outside specialists, and draft papers are subject to peer review for quality control. Yet the format of the series is flexible enough to reflect important developments in the field in a timely fashion. The primer series contributes to the teaching materials covered in the annual Social Safety Nets course offered in Washington DC as well as various other Bank-sponsored courses. The Social Safety Nets Primer and the annual course are jointly supported by the Social Protection unit of the Human Development Network and by the World Bank Institute. The World Bank Institute also offers customized regional courses through Distance Learning on a regular basis. For more information on the primer paper series and papers on other safety nets topics, please contact the Social Protection Advisory Service; telephone (202) 458-5267; fax (202) 614-0471; email: socialprotection@worldbank.org. Copies of related safety nets papers, including the Social Safety Nets Primer series, are available in electronic form at www.worldbank.org/safetynets . The website also contains translated versions of the papers as they become available. An ambitious translation plan is underway (especially for Spanish and French, some in Russian). For more information about WBI courses on social safety nets, please visit the website www.worldbank.org/wbi/socialsafetynets . Papers in the Safety Nets Primer as of November 2005 Theme Author Program Interventions Cash Transfers Tabor, Steve Community-based Health Insurance Tabor, Steve Conditional Cash Transfers Rawlings, Laura Fee Waivers in Health Bitran and Giedion Fee Waivers in Housing Katsura and Romanik Food Related Programs Rogers and Coates Micro Credit and Informal Insurance Sharma and Morduch Mitigating Social Risks Tesliuc, Emil Price and Tax Subsidies Alderman, Harold Public Works Subbarao, Kalanidhi Transferring Cash Benefits Lafaurie and Velasquez Themes Evaluation Blomquist, John Gender Ezemenari, Chaudhury and Owens Institutions de Neubourg, Chris Political Economy Aspects of Targeting Pritchett, Lant Public Attitudes and Political Economy Graham, Carol Safety Nets for Poverty Reduction Ravallion, Martin Targeting Coady, Grosh and Hoddinott Targeting: Lessons from LAC – Overview Lindert et all Targeting in Brazil Lindert and Briere Targeting in Chile (Spanish) Larrañaga, Osvaldo Targeting in Colombia Castañeda, Tarsicio Targeting in Costa Rica (Portuguese) Viquez, Roxana Targeting in Mexico (Portuguese) Orozco and Hubert Targeting in the United States Lindert, Kathy Testing Vietnam’s Public Safety Nets van de Walle, Dominique Country Context Poverty and Aging in Africa Subbarao, Schwartz and Kakwani Transition Economies Fox, Louise Very Low Income Countries Smith and Subbarao Special Vulnerable Group Disability Mitra, Sophie About the Authors Alain de Janvry, Frederico Finan, and Elisabeth Sadoulet* University of California at Berkeley, Donald Nelson** University of Arizona and Kathy Lindert, Bénédicte de la Brière, and Peter Lanjouwo*** The World Bank * Alain de Janvry, Department of Agricultural and Resource Economics, Giannini Hall 207, University of California, Berkeley, California 94720. Tel: 510-642-3348, fax: 510-643-8911, email: alain@are.berkeley.edu Frederico Finan, Department of Agricultural and Resource Economics, Giannini Hall 207, University of California, Berkeley, California 94720. Tel: 510-643-4214, fax: 510-643-8911, email: finan@are.berkeley.edu Elisabeth Sadoulet, Department of Agricultural and Resource Economics, Giannini Hall 207, University of California, Berkeley, California 94720. Tel: 510-642-7225, fax: 510-643-8911, email: sadoulet@are.berkeley.edu ** Donald Nelson, Department of Anthropology, University of Arizona, email: drn@u.arizona.edu *** Kathy Lindert, Senior Economist/Task Manager, Banco Mundial, SCN, Qd. 2, Lt. A, Ed. Corporate Financial Center, Cj. 303/304, Brasília, DF 70712-900. Phone: (55 61) 329-1000, Fax: (55 61) 329-1010, email: Klindert@worldbank.org Bénédicte de la Brière, Economist, The World Bank, email: bdelabriere@worldbank.org Peter Lanjouw, Lead Economist, DECRG, The World Bank, 1818 H Street, Washington D.C. 20433. email: planjouw@worldbank.org Acknowledgement The authors would like to thank Margaret Grosh and anonymous referees for their feedback on earlier versions of this report. We would also like to thank the World Bank’s Research Support Budget, DFID, and UNESCO for generous financial support for this research. Abstract This study analyzes the role of local governance in the implementation of Bolsa Escola, a decentralized conditional cash transfer program for child education in Brazil. It is based on a survey of 260 municipalities in four states of the Northeast. The analysis focuses on program implementation. Results show that there was considerable confusion over the municipality’s role in beneficiary selection and consequently much heterogeneity in implementation across municipalities. Social Control Councils as direct accountability mechanisms were often not in place and poorly informed, weakening their role. However, electoral support for incumbent mayors rewarded larger program coverage, presence of Councils, and low leakages of benefits to the non-poor. JEL codes: H77, I38, I21 Keywords: local government, education, conditional cash transfer Table of Contents Page I. Introduction and Overview………………………………………………………… 1 1.1. Motivation and Context………………………………………………………… 1 1.2. Objectives of the Study………………………………………………………… 2 1.3. Conceptual Framework………………………………………………………… 3 1.4. Main Findings and Conclusions……………………………………………….. 5 1.5. Roadmap………………………………………………………………………. 6 II. Brief Background on the Bolsa Escola and Bolsa Familia Programs…………… 6 III. Survey Design and Implementation………………………………………………... 7 3.1. Sample and Field Work………………………………………………………... 7 3.2. Survey Instruments and Questionnaires……………………………………….. 8 IV. Municipal Survey Results on the Decentralized Implementation ……………… 10 of the Bolsa Escola Program 4.1. Identification and Registration of Potential Beneficiareis…………………….. 10 4.2. Beneficiary Selection………………………………………………………….. 18 4.3. Monitoring and Enforcement of Conditionalities……………………………… 23 4.4. Bolsa Escola’s Social Control Councils: A Short Route to Downward ……… 26 Accountability 4.5. Political Rewards and Re-Elections: Long Route to Downward ……………... 30 Accountability V. Municipal Survey Results on the Decentralized Implementation of the ……… 32 Bolsa Família Program (as of late 2004): A Comparative Analysis 5.1. Identification and Registration of Potential Beneficiaries……………………. 33 5.2 Beneficiary Selection…………………………………………………………. 38 53. Monitoring and Enforcement of Conditionalities …………………………….. 39 VI. Concluding Remarks……………………………………………………………… 41 References ………………………………………………………………………………… 44 Tables: Table 1. Selected socioeconomic characteristics, by states in the Northeast ……… 8 Table 2. List of informants for each section of the questionnaire ………………… 9 Table 3. Identification of potential beneficiaries for the Bolsa Escola program…... 12 (Share of municipalities) Table 4. Correlates of the different methods of beneficiary identification………… 14 for the Bolsa Escola program used by municipalities Table 5. Criteria used to identify beneficiaries for the Bolsa Escola program…….. 16 (Share of municipalities that report using each criterion) Table 6. Additional aspects of beneficiary identification for the Bolsa Escola …… 17 program (Share of municipalities) Table 7. Correlates of the beneficiary selection process for Bolsa Escola………… 20 i Table 8. Criteria used for the selection of beneficiaries for Bolsa Escola ………… 21 program among municipalities that actively selected (Share of municipalities) Table 9. Additional aspects of beneficiary selection in the Bolsa Escola ………… 22 Program (Shere of municipalities) Table 10. Principal conditions imposed on Bolsa Escola recipients……………… 23 (Share of municipalities Table 11. Additional conditions imposed on Bolsa Escola recipients…………….. 24 Table 12. Imposing additional program conditions in the Bolsa Escola program…. 25 Table 13. Monitoring and enforcement of the Bolsa Escola participation ………… 26 requirements (Share of municipalities) Table 14. Existence and Performance of Bolsa Escola Councils…………………... 27 (Share of municipalities and share of council members) Table 15. Bolsa Escola Council correlates…………………………………………. 28 Table 16. Long route to downward accountability: Electoral rewards…………….. 31 Table 17. Bolsa Familia beneficiary identification………………………………… 34 (Share of municipalities) Table 18. Persistence of procedures in the Bolsa Familia program implementation.. 36 Table 19. Criteria used to identify Bolsa Familia beneficiaries……………………. 37 (Share of municipalities) Table 20. Additional criteria used for beneficiary identification for the ………….. 37 Bolsa Familia program Table 21 Additional aspects of Bolsa Familia beneficiary identification………….. 38 (Share of municipalities) Table 22. Bolsa Familia beneficiary selection (Share of municipalities)………….. 39 Table 23. Additional aspects of the conditions imposed by the …………………… 40 Bolsa Familia program Figures: Figure 1. Research framework …………………………………………………….. 3 Figure 2. Achieving accountability in the Bolsa Escola program:..……………….. 4 Short and long routes Figure 3. Distribution of the number of requirements used by municipalities ……. 16 to identify beneficiaries for the Bolsa Escola Program Figure 4. Who selected the Bolsa Escola beneficiaries? …………………………... 18 Figure 5. Number of characteristics that determined selection into ……………….. 22 the Bolsa Escola program (Share of municipalities with active selection) Figure 6. Number of requirements imposed for program participation ……………. 24 Figure 7. Number of secretariats involved in the registration across municipalities.. 33 Appendix: Table A1. Summary statistics of regression covariates……………………………. 45 ii I. Introduction and Overview 1.1. Motivation and Context Incentive-based poverty reduction programs have generated considerable interest in the international development community. One such approach consists in using conditional cash transfers that help reduce short-term poverty through the transfers, while giving households the incentives to invest in the human capital of their children and thereby reduce poverty in the long- run. Central to the success of any social program aimed at poverty reduction is the ability to effectively reach and engage the poor. This has led several countries to experiment with a variety of implementation mechanisms in order to improve the effectiveness of these programs in reducing poverty. And while some evidence exists suggesting that differences in program implementation can have important consequences for program impact, the policy implications of these results remain unclear because the choice of implementation procedures is endogenous.2 Why a service provider adopts a particular approach to targeting and accountability depends not only on the rules and budgets under which it operates, but also on its own socioeconomic and political settings, institutional arrangements, and administrative capacities. Understanding this choice is particularly important when program implementation is delegated to local governments to act as service providers. Identifying what factors influence local governments’ decisions to adopt a particular approach allows us to understand why they may adopt technically sub-optimal implementation strategies from the perspective of program objectives. In addition, understanding these factors heightens our comprehension of the implications of particular implementation strategies for the poor. These insights can help national policy makers better design programs so as to align incentives for service providers with program objectives. Brazil’s experience with the Bolsa Escola Program provides an ideal laboratory to understand how contextual factors affect local governments’ (as service providers) choices of implementation strategies. The Bolsa Escola Program was introduced on a national-scale in 2001, building on earlier municipal-level program initiatives.3 By late 2003, Bolsa Escola had been implemented in almost all of Brazil’s 5,561 municipalities, providing nearly US$500 million in total stipends paid to over 8.6 million children from 5.06 million families.4 In October 2003, Bolsa Escola was merged with three other transfer programs to collectively form the “Bolsa Familia Program,” which is currently being improved and expanded to ultimately reach 11.2 million beneficiary families (or 44 million people). Both the pre-reform Bolsa Escola program and the current Bolsa Familia Program belong to a class of programs called “conditional 2 Coady, Grosh and Haddinott (2003) analyze a database of over 120 antipoverty interventions from 48 countries and find a strong association between different program implementation mechanisms and targeting performance. The authors also find some evidence that a country’s governance, voice, and inequality are positively correlated with program targeting. The authors admittedly have several caveats regarding the comparability of their performance measure and the sample of antipoverty interventions. 3 Unless otherwise indicated, the use of the term Bolsa Escola in the remainder of this paper refers to the federal program. 4 Source: Ministry of Social Development. 1 cash transfers” that provide cash transfers to beneficiary families in exchange for meeting pre- specified conditionalities, usually linked to school attendance and health care utilization. Bolsa Escola (2001-2003) provided mothers of poor households a monthly stipend conditional on their children’s regular school attendance. Many aspects of program implementation for Bolsa Escola were devolved to the municipal governments, including the identification and selection of program beneficiaries, the monitoring and enforcement of conditionalities, and the management of local accountability mechanisms (payment of benefits, however, was made directly from central agencies to the beneficiaries).5 This resulted in considerable variation in the manner in which municipalities chose to implement the program, providing a unique opportunity to explore how differences in institutional settings lead to different choices in targeting, monitoring and enforcement, and accountability instruments and practices. 1.2. Objectives of the Study The objectives of this study are to use this “natural laboratory” created by Bolsa Escola’s decentralized implementation in order to: • Document and analyze municipal variation in implementation of the Bolsa Escola Program (also with some preliminary results for the nascent Bolsa Familia Program) in four key aspects: (a) beneficiary identification; (b) beneficiary selection; (c) conditionality monitoring, verification and enforcement; and (d) implementation of accountability mechanisms. • Analyze how contextual factors, such as socio-economic conditions and indicators of municipal governance, affect the varying implementation strategies adopted at the municipal level. This study is the first of a series of reports that build on field surveys conducted from October to December 2004 in 261 randomly selected municipalities in four states of Northeast Brazil. This report will contribute to our further work, which will analyze how these findings on decentralized implementation and governance affect program outcomes, namely (a) how well Bolsa Escola was targeted to the children of poor households; (b) impact on poverty and inequality; and (c) impact on school attendance and achievement. 5 After Brazil’s move towards decentralization in the 1980’s, municipal governments became particularly important bodies of government. As one of the most decentralized countries in the world, currently transferring 15 percent of its federal resources to local governments, the decision to devolve Bolsa Escola to the municipalities was a natural one. Moreover, the devolution of decision-making power from the central government to the local community has several potential advantages that can lead to better targeting outcomes and improved project performance. Local authorities tend to have more information about the community and can better identify the poor, which should allow for fewer targeting errors. With better information on local conditions and fewer levels of bureaucracy, the local community can deliver goods and administer the program more efficiently than a central government that must rely on monitoring devices. Because local institutions are potentially more accountable to local citizens, this creates further incentives for both better targeting and better program outcomes. 2 1.3. Conceptual Framework. The research framework used in this paper and in our broader research program is outlined in Figure 1. Essentially, we hold that the quality of (decentralized) implementation will affect the degree to which the Bolsa Escola program can improve outcome indicators for recipients. In turn, a variety of context factors in the municipality affect both the quality of implementation and these outcomes, including: (a) political and governance indicators, (b) economic variables, and (c) social variables. The potential impacts of context and implementation factors on program outcomes (particularly education impacts) will be analyzed and presented in future reports. This present paper examines correlations between the context and implementation variables. Figure 1. Research framework CONTEXT IMPLEMENTATION OUTCOMES Political Process for identifying Targeting variables and selecting beneficiaries (leakages, coverage) Monitoring and Economic enforcement of conditionalities Poverty variables Inequality Social Control Councils Social School attendance and variables achievement Risk of drop out The conceptual framework for governance and accountability mechanisms is outlined in Figure 2. There are two routes to social accountability in decentralized service provision (World Bank, 2004). One is the “short route to accountability”, whereby potential and actual program clients can demand accountability from service providers through the direct use of oversight mechanisms. Here, Bolsa Escola services are provided in a decentralized fashion to beneficiaries by a municipal government assisted by a federally mandated Bolsa Escola social council composed of both municipal and civil society representatives. The local provider operates under rules given by the federal Bolsa Escola program. The functions of the local municipal provider are: targeting (beneficiary identification and selection), monitoring and enforcement of conditionalities, and implementation of direct accountability mechanisms through appeals to the social council. Payments are made directly by central agencies to beneficiaries. Two necessary conditions to achieve downward accountability toward potential beneficiaries are transparency (information) and functioning oversight mechanisms through 3 appeals to the social council. For downward accountability of the local provider to obtain, these two complementary conditions must both be met. The other accountability mechanism is the “long route to social accountability” whereby local elected municipal officials derive rewards or punishments from the electorate for their role as service provider. For mayors in their first term in office and aspiring to re-election, successful access to program benefits (number of bolsas per children enrolled), effective program targeting, and availability of a Social Council can be rewarded though the delivery of votes. The perceptions that matter in eliciting votes are not only those of actual and potential program beneficiaries, but also perceptions of the value and implementation of the program by the general local population. In this study, we will analyze how effective each of these two mechanisms is in providing downward accountability to Bolsa Escola potential beneficiaries. Figure 2. Achieving accountability in the Bolsa Escola program: Short and long routes Federal government: Federal Bolsa Escola Program Rules and budgets Decentralized service provider: Municipal government Municipal Bolsa Escola social council Program implementation: Long route to social accountability Targeting: Beneficiary identification and selection Local political retributions Monitoring and enforcement of conditionalities Accountability mechanisms Short route to social accountability Appeals to social council Transparency (information) Bolsa Escola beneficiaries: Demands for downward Potential and actual beneficiaries accountability Perceptions by local population Program outcomes: Poverty reduction and human capital formation Within this framework, we analyze the influence (correlation) of a variety of governance factors on heterogeneity in the quality of decentralized information in addition to specific accountability mechanisms. Some of the aspects of governance on which we focus are mayoral participation in beneficiary identification, and synthetic measures of perceptions of “clientelism” and “patronage.” For the purpose of our analysis, we define clientelism as the exchange between politicians and voters of material goods for votes; and patronage as the distribution of public goods based on political support and not on local need. These definitions correspond to those used by Brusco, Nazareno, and Stokes (2002). The specific questions to characterize clientelism 4 and patronage come from a survey conducted from IBOPE Group (www.ibope.com.br) during the 2000 municipal elections and sponsored by Transparencia International of Brazil (2005). 1.4. Main Findings and Conclusions The general findings of the study are that: (a) there is considerable heterogeneity in implementation quality and strategies by municipality; and (b) contextual factors – including local governance and politics – affect implementation. More specifically, the following six findings stand out: • There was considerable variation across municipalities in the processes used to register potential beneficiaries for both the Bolsa Escola and the Bolsa Familia programs. Cost considerations and political/governance variables were important correlates of the way registration was implemented. • There was widespread confusion concerning the municipality’s role in beneficiary selection for the Bolsa Escola Program. A majority of municipalities understood that such decisions were made not by them but by the Federal Government in Brasilia. Political variables, such as the practice of clientelism and patronage, also affected municipal beneficiary selection decisions. On the other hand, social variables such as literacy (voice) and social councils (accountability) clearly also affected the selection process. Our survey results show that confusion was reduced under Bolsa Familia after the issuance of a regulatory decree that clarified that responsibility for beneficiary selection rests with the federal government (MDS). • There was considerable transparency with respect to the beneficiary identification and selection process, with ample dissemination, public knowledge, and information on the criteria used. • With respect to the monitoring and enforcement of conditionalities under Bolsa Escola, we find that (a) a significant share of municipalities imposed additional conditionalities (beyond the federal requirements) on beneficiaries; (b) there was significant variation in the monitoring and enforcement of conditionalities; and (c) economic and political factors seemed to influence the degree to which these processes were implemented. It is important to note, however, that there is a significant difference between municipalities indicating that they monitor conditionalities and their forwarding of such information to the federal government. In fact, federal statistics from the Ministry of Education show that reporting of conditionalities compliance data by municipalities to the federal government increased substantially under Bolsa Familia (from 19% of schools reporting under Bolsa Escola to 79% under Bolsa Familia in early 2005). • Our municipal survey suggests considerable variation in the existence and effectiveness of social control councils, the instrument designed to insure a short route to downward accountability. Specifically, we find that: (a) although most municipalities operated such councils, about a fifth of them did not establish them at all, despite the federal requirement to do so; (b) in municipalities where social councils did exist, there was a positive impact on the quality of implementation; but (c) even when they did exist, they did not necessarily function properly and their membership seems to have been selected with predominant support for mayors. 5 • Finally, we found encouraging evidence that the longer route to downward accountability via electoral rewards does work. Incumbent mayors gained from the program: their likelihood of re-election increased with the share of school children covered by the program, even though this quota was determined by Brasilia, with no room for influence by them. In spite of this, they were perceived by their local constituency as effective intermediaries in bringing program benefits to the community. Mayors who put into place a social control council also reaped electoral rewards, with a 47 percent higher chance of being re-elected. Finally, we find that perceptions of targeting accuracy do seem to have political dividends, but only if it minimized “errors of inclusion” (leakages to families perceived to be undeserving or non-poor). However, perceived errors of exclusion – poor families that should benefit but do not – did not generate political costs for incumbent mayors. These results give strong evidence that electoral rewards and punishments are an effective instrument for downward social accountability. However, because they take longer to activate (i.e., the length of the political cycle, four years in this case), they are not a substitute for more immediate social accountability through effective social control councils, which could be strengthened. 1.5. Roadmap In this report, we give in Section II background information on the Bolsa Escola and Bolsa Familia programs. In Section III we explain how the municipal survey was designed. We then use the survey data to analyze in Section IV the way the Bolsa Escola program was implemented across municipalities, including beneficiary identification, beneficiary selection, implementation of conditionalities, and accountability mechanisms through performance of the Bolsa Escola social councils. Section V provides a brief analysis of implementation of the Bolsa Familia program, comparatively with implementation of the Bolsa Escola program. Section VI concludes the evaluation of the implementation of the programs. II. Brief Background on the Bolsa Escola and Bolsa Familia Programs Bolsa Escola was a demand-driven education program that provided cash transfers to mothers of poor children throughout Brazil, conditional on their children's continued attendance in school (Abramovay, Andrade, and Waiselfisz, 1998; World Bank, 2001). Initiated in 1995 as municipal programs in Campinas and the outskirts of Brasília, Bolsa Escola became a nationwide federal program in 2001. By the end of 2001, it had been implemented in 98 percent of the 5,561 Brazilian municipalities, providing stipends to over 8.2 million children from 4.8 million families, at a cost of over US$700 million. Having benefited millions of Brazilian school-aged children, the program has served as a source of inspiration for and a point of comparison with similar educational programs throughout the world.6 The targeting of Bolsa Escola cash transfers was implemented in two stages. First, the Federal government decided, based on a determination of need, the number of federally-financed stipends that a municipality could provide to its population (geographic targeting at national 6 For example, similar education programs exist in Argentina, Bangladesh, Chile, Colombia, Ecuador, Honduras, Mexico, Nicaragua, and Pakistan. 6 level). Second, given this number of stipends, the municipality selected which households would receive the program from among qualifying beneficiaries. This devolution of the selection process allowed each municipality to target the program, within the general guidelines, according to its own local objectives and preferences. As a result, the program's impact on schooling and targeting outcomes may vary considerably across municipalities, and expectedly along a number of observable dimensions. This two-stage design thus provides a unique laboratory to analyze how differences in institutional settings affect program implementation: the identification and selection of beneficiaries, the monitoring and enforcement of conditionalities, and the role of social councils in providing a short route to downward accountability. In 2003, Bolsa Escola and three other federal cash transfer programs were unified into a single program called Bolsa Familia. The Bolsa Familia Program has expanded very rapidly – both integrating existing beneficiaries from the pre-reform programs (including Bolsa Escola) and incorporating new beneficiaries. As of October 2005, Bolsa Familia had expanded to reach over 8 million households throughout Brazil, targeting in particular two groups: households with a monthly per capita income of less than R$50 (extreme poor) and households with a monthly per capita income between R$50 and R$100 (moderately poor). These household receive monthly payments ranging from R$15-R$95. The exact amount depends on the household’s income and composition, and is conditional on a set of program requirements. Unlike the Bolsa Escola program which placed requirements on the individual children, the conditionality emphasis of the Bolsa Familia program is at the family level. All relevant family members must comply with a set of key human development requirements that include: (i) children ages 6-15 years old be enrolled and attend at least 85 percent of their classes; (ii) children under the age of seven visit health clinics to have their growth monitored and immunizations updated; and (iii) pregnant women conduct prenatal care. Bolsa Familia also re- centralized beneficiary selection decisions to the federal government, though many other aspects of program implementation remain decentralized, such as registration of potential beneficiaries into the unified registry (the “Cadastro Unico”), monitoring of conditionalities (coordinated by the Ministries of Health and Education), and social controls. III. Survey Design and Implementation 3.1. Sample and Field Work Data collection for our surveys took place between October and December of 2004, in 261 municipalities randomly selected across the states of Ceará, Pernambuco, Paraíba, and Rio Grande do Norte.7 The municipalities of these four states were stratified according to their land inequality, size of public sector, and quota of program beneficiaries; and were randomly sampled from eight strata. The sample was stratified to capture sufficient variation along variables that 7 The choice of these four states was mainly based on our budget constraint and institutional support. Our restriction to Northeast, Brazil stems from the fact that the Northeast represents one of the poorest regions of Brazil and is where we would expect Bolsa Escola to have its most significant impact. It is also a region that has had a long- standing tradition of clientelistic and oligarchic politics and where the local government plays an immensely important role in the economic livelihood of its population. The Northeast is also highly diverse and presents a range of institutional settings in which to analyze the program. 7 may be correlated with governance and vote buying. Our sample is thus representative only for these four states and not necessarily for the Northeast as a whole. To get a sense for how these states differ from the other states of the Northeast, Table 1 presents a set of socioeconomic and demographic characteristics by state. Compared to the other states, the states in our sample are on average slightly better off along several of these welfare measures. Rio Grande do Norte for example has the lowest poverty rate among the states of Northeast, as well as the highest percentage of school enrollment among children ages 7-14. Pernambuco has the highest per capita household income but in terms of income inequality is also among the most unequal state, second only to Ceará. Paraíba is perhaps the exception, as it ranks near the bottom in terms of poverty, infant mortality, and per capita income. Together these states are key participants of the Bolsa Escola program, as at least 43 percent of the total number of households benefited by the program in the Northeast claimed residency in one of these four states. Table 1. Selected socioeconomic characteristics, by states in the Northeast Number of families Poverty Infant mortality Average per benefited by the Rates (per 1000 live births) capita income of Gini Enrollment rates Bolsa Escola State (P0) Age<1 households coefficient (ages 7-14) program 2001 2000 2001 2000 2000 2002 Cear‡ 0.58 41.4 221.4 0.68 94.4 432,736 Para’ba 0.62 51.5 178.4 0.65 93.9 180,918 Pernambuco 0.60 47.3 240.5 0.67 92.1 305,920 Rio Grande do Norte 0.54 43.3 203.8 0.66 94.8 129,710 Alagoas 0.65 49.0 161.8 0.69 89.0 140,183 Bahia 0.59 46.5 216.3 0.67 93.1 674,244 Maranh‹o 0.64 55.4 150.5 0.66 91.6 338,538 Piau’ 0.61 47.3 166.4 0.66 93.7 182,425 Sergipe 0.55 48.5 190.6 0.66 93.3 76,927 Northeast 0.60 NA 209.0 0.60 95.2 2,461,601 Notes to Table 1: These data come from www.ipeadata.gov.br 3.2. Survey Instruments and Questionnaires In each of the 261 municipalities analyzed, two survey instruments were applied: (1) a municipal survey and (2) collection of school records. This current paper presents the main results of the municipal survey – particularly as they pertain to decentralized implementation and governance. The analysis of the school records data is still in progress, and the findings on school outcomes will be presented in a subsequent paper. Given our various research objectives, the municipal survey consisted of 10 parts.8 Table 2 presents who were the main informants for the various sections of the survey. 8 To make clear to whom questions were asked and how the questions were phrased, the tables reproduce in italics the question asked. Footnotes to the tables indicate who the respondents were. 8 Table 2. List of informants for each section of the questionnaire Informants Public administrators Secretary of Member of Coordinator Director of President of human Secretary Coordinator of Bolsa Escola of Bolsa legislative agricultural Parts of the questionnaire resources of finances Bolsa Escola council Familia branch workers union 1. General Characteristics X 2. Administration X 3. Budget X 4. Bolsa Escola X 5. Bolsa Escola Council X 6. Bolsa Familia X 7. Legislative Branch X 8. Legislative Branch - indirect X X 9. Mayor characteristics X X 10. Political variables X X X One module of the survey (Part 10: Political variables) interviewed public administrators to gather information on governance, budgetary procedures, and other municipal characteristics associated with public administration. These questions allow us, for example, to examine measures of “political patronage,” “clientelism,” and other indicators of governance and political influence. Our measure of patronage is based on a question that asks, to three key informants, the proportion of public works that benefit communities for political motives and that were not necessarily undertaken to fulfill an organic need. The measure used in all the regressions is an average of these responses. Our measure of clientelism is based on a question that asks three key informants (see Table 2 for identification of the informants) to rate the level of clientelism in the municipality on a scale of 1 to 7, where clientelism was defined for the informant as the exchange between politicians and voters of material goods for votes. After averaging the responses, we defined three levels of clientelism: low clientelism (less than or equal to 2), medium clientelism (more than 2 but less than 4), and high clientelism (4 and higher). Because of the sensitivity and subjectivity associated with some of these subject matters, these questions were asked multiple times to various segments of civil society. As seen in Table 2, respondents include a Bolsa Escola council member, the director of the legislative branch, and the president of the agricultural worker’s union. Other modules of the survey characterized the implementation of the Bolsa Escola (Parts 4 and 5: Bolsa Escola and Bolsa Escola Council) and Bolsa Familia (Part 6) programs. In these sections, we interviewed the respective program coordinator about how the municipality identified and selected program beneficiaries, and imposed and monitored the program requirements. We also gathered information to assess how transparent the program was in its implementation. For the section on the Bolsa Escola council, a council member (priority given to non-governmental members) was asked several questions about the composition of the council and its level of activity. The remaining parts of the survey interviewed either politicians from the legislative branch of the local government (Parts 7 and 8: Legislative Branch) or key members of the municipality to characterize the major stakeholders in the provision of public 9 goods and government programs (Part 1: General Characteristics; Part 2: Administration; Part 3: Budget). To properly measure the effect of Bolsa Escola on school enrollment and student achievement, we collected children’s school records for approximately 500 eligible children in each municipality during the period of 1999-2003. To gather these records, two schools were randomly drawn proportional to the number of Bolsa Escola recipients (data which were obtained from the payments records of the Ministry of Education) within each selected municipality. Information on the grades, enrollment, and approval for each child in the school were collected. These results will be analyzed and presented in a future report. In sum, four field teams, one in each state, assembled a unique database comprised of municipal information on 261 municipalities, and comprehensive school records for over 130,500 eligible children spanning the years 1999-2003. IV. Municipal Survey Results on the Decentralized Implementation of the Bolsa Escola Program As mentioned above, Bolsa Escola was a federal program whose implementation was devolved to the municipality.9 Given this decentralized design, one can expect considerable variation in the manner in which municipalities chose to implement the program. In this section, we document this variation, focusing on four key aspects of program implementation that could potentially have an important impact on program outcomes: (1) beneficiary identification by the municipality; (2) beneficiary selection by the municipality or Brasilia; (3) monitoring and enforcement of the conditionality; and (4) social controls over program implementation. 4.1. Identification and Registration of Potential Beneficiaries Data collection and registration of potential beneficiaries is a crucial step in any transfer program, in particular since the pool of potential families interviewed and registered greatly influences the eventual selection of beneficiaries. This section presents the degree to which municipalities vary in the approaches used to identify potential beneficiaries of the Bolsa Escola Program in several aspects. We find a fair degree of uniformity in some aspects, such as the questionnaire used to collect household data, and the ways municipalities notified households about the program. This is not surprising since the Federal Government gave provided the standard questionnaire for municipal use. We find more heterogeneity in the decentralized implementation of the following processes: (a) the actors responsible for identifying potential beneficiaries; (b) the location of where registration took place; (c) the criteria used for identifying potential beneficiaries (including whether or not municipalities adopted geographic criteria for prioritizing areas within 9 For the purpose of this report, Bolsa Escola refers to the Bolsa Escola Federal program initiated in 2001 under President Fernando Henrique Cardoso. Bolsa Escola-type programs exist at the state and municipal levels in some areas. It is, however, much less common for the Northeast. In our sample there were only 3 municipalities that participated in a state or municipal program. 10 their jurisdictions as well as apparent criteria used to identify specific households to be registered); and (d) the degree to which municipalities made attempts to verify the information collected. Again, this heterogeneity is not surprising, since municipalities were given significant leeway in carrying out these aspects of the program (with few federal guidelines). We also examine the possible influence of some of the context factors (economic, political/governance, and social) on the variation observed in some of these aspects, such as the location of registration activities. Questionnaires Used for Data Collection and Registration. We find that most municipalities (92%) used a standard questionnaire designed specifically for Bolsa Escola for the collection of household information. This questionnaire was established by the Federal Government and gathers basic information on household demographics and per capita incomes. This Bolsa Escola questionnaire was replaced in 2001 by the federal questionnaire Cadastro Único, which became a unified Federal Registry across multiple programs (Bolsa Escola, Bolsa Alimentação, Auxilio Gas and the Cartão Alimentação) and is currently used by the integrated conditional cash transfer, the Bolsa Familia Program. Promoting Awareness about the BE Program and Registration Process. We likewise find a fairly high degree of homogeneity in terms of provision of public information about the program and its registration process. Table 3 presents basic summary statistics to characterize how municipalities notified households about the program (and how they identified and registered their potential beneficiaries, discussed below). Almost all municipalities (94%) use schools to notify individuals about the program. Schools, however, were not the only source of information about the program, as it was also advertised on the radio (66 percent) and in public announcements (53 percent). On average, municipalities used at least three (standard deviation of 1.14) different channels to notify citizens about the program. Actors Responsible for Registration. Public administrators and school teachers were mostly responsible for identifying the beneficiary population. Our municipal survey results suggest that 70 percent of the municipalities used school teachers or school administrators and 82 percent used public administrators to register potential beneficiaries into the registry used for selecting BE beneficiaries, while 55 percent of the municipalities used both. Consequently, most of the potential beneficiaries were interviewed either at the schools (85 percent) or at the mayor’s office (55 percent). Health agents, who are active members in a community and visit households frequently to provide information on preventative health measures, participated in the registration in 32 percent of the municipalities, while members of the civil society, such as nongovernmental organizations, municipal councils, and volunteers were used in only 11 percent of the municipalities interviewed. Location of Interviews. Most of the registration process took place in public locations, with schools again providing the natural setting. Only 28 percent of all municipalities in the sample registered individuals at their home, and among these municipalities, the median percentage of households interviewed was only 20 percent. In fact only 4 municipalities performed the entire registration process using home visits. 11 Table 3. Identification of potential beneficiaries for the Bolsa Escola program (Share of municipalities) Number of Standard observations Mean deviation Who registered the potential beneficiaries? School teachers or administrators 260 0.700 0.459 Health agents 261 0.318 0.467 Public administrators 261 0.820 0.385 Members of civil society 261 0.107 0.310 Where were potential beneficiaries registered? Escolas 256 0.848 0.360 Post de saude 256 0.086 0.281 Prefeitura 256 0.547 0.499 Comunidades 256 0.367 0.483 Casas 259 0.278 0.452 Percent of households registered at home 72 36.417 33.565 How were household notified about the registration Radio 261 0.655 0.476 Television 261 0.061 0.240 Newspapers 261 0.180 0.385 Community leaders 261 0.609 0.489 Schools 261 0.935 0.247 Public announcement 261 0.529 0.500 Did the municipality prioritize geographically in the registration? 261 0.383 0.487 Among the municipalities that prioritized, what were the criteria used? Poor neighborhoods 100 0.620 0.488 Greater number of schools 99 0.293 0.457 Ease of access to target group 99 0.424 0.497 Distance from municipal head 100 0.130 0.338 Notes to Table 3: Questions in italics are from the questionnaire. Geographic prioritization refers to whether the municipality prioritized some areas of the municipality to identify beneficiaries. The respondent for these questions was the Bolsa Escola coordinator. Use of Geographic Targeting Within Municipalities. Moreover, only approximately 38 percent of the municipalities used some form of geographical targeting to decide in which areas to begin the registration. Among those municipalities that did prioritize specific areas, 62 percent targeted the poorest neighborhoods. Other considerations in geographical prioritization included the number of schools in an area, ease of access to the target group, and the rural nature of a community (measured by the distance to the municipal head). Factors Associated with Location of Registration and Geographic Targeting: Room for Political Manipulation? The decision regarding the location of registration activities could influence which groups of eligible households are eventually selected into the program and can therefore be a critical choice of the program’s implementation. The choice location of registration could also suggest political intentions, for example if potential beneficiaries are registered at the mayor’s office – which could promote political linkages or a sense of favoritism with potential interactions with the mayor and his/her staff that could result in political retribution. One could expect lower clientelistic associations if registration were conducted 12 through home visits or at other public locations. To explore this hypothesis, Table 4 investigates to what extent specific mayor and municipal characteristics are associated with each of the different approaches that municipalities used in deciding where and how to register potential beneficiaries. Column (1) reports the estimates for the probability that a municipality registers its eligible households at the mayor’s office, and column (2) reports the estimate of an OLS regression where the dependent variable is the percentage of households that were registered at their home.10 The dependent variable in column (3) is an indicator for whether the municipality prioritized some areas in the municipality in the registration process (geographic targeting). Each regression controls for the same set of mayor characteristics, municipal characteristics, program characteristics at the municipal level, and political characteristics of the municipality. Summary statistics of the covariates in all the regression presented in this report are given in Appendix Table A1. The results of column (1) suggest that a municipality’s ability to effectively reach eligible families may be an important deterrent in a municipality decision to register beneficiaries at the mayor’s office. For example, the size of the municipality (measured both in terms of its number of districts and population density), the extent of its rural sector, and the number of families that can be benefited in the program (Bolsa Escola quota as a share of all children enrolled in primary and secondary school) are all negatively associated with registration in the mayor’s office. In addition to these efficiency considerations, there is also evidence that municipalities that register beneficiaries at the mayor’s office may do so to garner political support. Municipalities with higher levels of patronage are more likely to register beneficiaries in the mayor’s office. And municipalities with higher levels of clientelism also tend to prefer this approach, although these estimates are measured with less precision (at only 80 percent confidence). Similarly, the share of functioning oversight councils and the existence of a judiciary district – institutions that in theory limit executive power and reflect greater local democratic practices – are also negatively correlated with registering eligible households at the mayor’s office. Column (2) suggests that cost efficiency may have been an important consideration in limiting a municipality’s decision to perform home visits. Municipalities that are more rural and with greater population density are less likely to pursue this type of approach to beneficiary identification. Municipalities with more catholic churches, which often play an important role in reaching and identifying poor households, are more likely to perform home visits. Assuming that the radio is an effective medium to notify eligible households about the program (65 percent of the municipalities did use the radio for this purpose), the number of radio stations may reduce the need for home visits, thus explaining the negative association between the number of radio stations and home visits. By contrast to registration at the mayor’s office, we do not see any evidence that political patronage or clientelism influenced the decision to use home visits. 10 Estimating a Tobit model to account for left censoring yields marginal effects that are quite similar. 13 Table 4. Correlates of the different methods of beneficiary identification for the Bolsa Escola program used by municipalities (1) (2) (3) Registration at the mayor's office Home visits Geographic Targeting Mayor characteristics Education 0.042 0.137 0.016 [0.019]* [0.745] [0.019] Gender (male=1) -0.041 -11.113 -0.005 [0.139] [7.493] [0.120] Second-term 0.032 -4.457 0.09 [0.078] [3.247] [0.074] Political experience -0.091 -1.358 -0.011 [0.037]* [1.229] [0.034] Member of an elite family 0.086 -8.386 -0.06 [0.101] [4.396]+ [0.099] Municipal Characteristics Population density (Persons/km) -0.034 -0.753 -0.017 [0.019]+ [0.315]* [0.009]+ Number districts -0.053 0.57 -0.008 [0.021]* [0.626] [0.011] Share of rural households -0.532 -26.937 -0.317 [0.267]* [11.317]* [0.246] Share of literate population 0.501 23.52 -0.345 [0.728] [31.089] [0.737] Log per capita income -0.188 -4.645 0.41 [0.232] [9.895] [0.217]+ Gini -0.091 10.462 0.116 [0.783] [29.869] [0.750] Number of radio stations -0.021 -3.762 -0.077 [0.039] [1.184]** [0.034]* Number of catholic churches -0.004 0.413 0.007 [0.006] [0.239]+ [0.004]+ Proportion of councils that function -0.738 -9.597 0.145 [0.313]* [9.511] [0.244] Judiciary district -0.154 -2.574 -0.018 [0.093]+ [4.268] [0.088] Program characteristics Bolsa Escola quota -0.061 0.161 0.217 [0.018]** [0.441] [0.412] Received training 0.138 4.178 -0.011 [0.077]+ [3.087] [0.075] Bolsa Escola Council Exists -0.104 -1.269 0.216 [0.096] [4.443] [0.095]* Political Characteristics Patronage 0.004 -0.032 0.005 [0.002]+ [0.068] [0.002]* Medium clientelism 0.121 2.367 0.079 [0.096] [4.116] [0.097] High clientelism 0.137 -5.233 0.104 [0.108] [3.773] [0.109] Observations 248 250 252 R-squared 0.18 Notes to Table 4: Marginal effects from a Probit model are reported in columns 1 and 3. Column 2 reports coefficient estimates from an OLS regressions. Robust standard errors in brackets; significantly different from zero at (+) 90%, (*) 95%, (**) 99% confidence. In addition to the variables displayed, each regression controls for the mayor’s age, the share of the population that is employed by the local government, the number of newspapers in the municipalities. We define elite family as a family that has had a long political tradition in the municipality; patronage is defined as the average share of public goods provided to communities for political reasons and not need. Our measure of the degree of clientelism in a municipality is based on a scale of 1 to 7. Medium clientelism corresponds to a value of above 2 but below 4, high clientelism corresponds to a value of above a 4. Summary statistics for the covariates are presented in table A1. 14 Column (3) presents some of the mayor and community characteristics that correlate to a municipality’s decision to prioritize some geographical areas of the municipality in the registration process. We find that municipalities that have a lower population density and have a higher average per capita income are more likely to target geographically. This would make sense if the poor are harder to identify in wealthier municipalities and if there is greater concern for effective poverty reduction in the wealthier communities. Another motivation to target geographically, however, can also be political, and we find evidence consistent with this interpretation. Municipalities with higher levels of patronage are positively associated with geographical targeting, suggesting that the program may have been used to reward certain communities for political support. The effects of clientelism are again positive but imprecisely measured. Criteria Used to Identify Potential Beneficiaries. The federal government specified three criteria for eligibility to the program. First, the child must come from a household that earns not more than R$90 per capita per month. Second, the child must be enrolled in primary or lower secondary school.11 And finally, the child must be between the ages of 6 to 15. Despite these federal eligibility criteria, only 85 percent of the municipalities used all three criteria to identify the potential beneficiary population. Moreover, 73 percent of the municipalities used other criteria in addition to the three federal ones, suggesting that these municipalities screened eligible (by the federal standards) households prior to the actual selection process (see Figure 3).12 Panel A of Table 5 reports some of the additional factors used to identify potential program beneficiaries among the municipalities that respected the federal criteria. In addition to the federal requirements, several municipalities took into consideration the number of children in the household (79 percent), and the household’s living conditions as evidenced by the dwelling characteristics (54 percent). Some municipalities (33 percent) gave preference to whether the child was enrolled at a municipal school, which might reflect some political considerations.13 Panel B of Table 5 presents the same factors for the 15 percent of municipalities that did not abide by all 3 federal eligibility criteria. As can be seen, the child enrollment conditionality (58 percent of the municipalities) and the child’s age (20 percent of the municipalities) were the least respected of the federal criteria. For these municipalities, the household’s per capita income was the most important determinant of whether or not the household was consider eligible. 11 This corresponds to grades 1-8 (ensino fundamental). 12 It is not clear why a municipality would perform a pre-selection. One reason might be to reduce the cost of registering potential beneficiaries by shrinking the eligibility pool. Alternatively, a municipality may have been confused about the distinction between identification and selection. Equally confusing to us is why some municipalities did not follow the federal guidelines in the identification. Unfortunately our data do not provide much insight into these questions. 13 Enrollment at municipal schools determines the amount of resources that a municipality receives for education from FUNDEF (Fund for the Implementation and Development of Basic Education). To boost enrollment in these schools, mayors have a strong incentive to target the program to children enrolled in the municipal schools. In some of the case studies we conducted, there were incidences of mayors offering free bicycles to induce children to enroll in municipal schools. 15 Figure 3. Distribution of the number of requirements used by municipalities to identify beneficiaries for the Bolsa Escola program 90 83 80 70 60 60 50 45 38 40 30 20 13 11 10 10 1 0 1 2 3 4 5 6 7 8 Table 5. Criteria used to identify beneficiaries for the Bolsa Escola program (Share of municipalities that report using each criterion) Number of Standard First Second observations Mean deviations priority priority Panel A: Municipalities that respected the Federal requirements Federal eligibility criteria Per capita income 221 1.000 0.000 86.92 9.23 Enrolled 221 1.000 0.000 5.38 41.54 Age of the children 221 1.000 0.000 2.31 10.00 Other criteria used Number of children in the household 221 0.787 0.410 1.54 33.08 Living conditions 221 0.538 0.500 1.54 1.54 Enrolled in a municipal school 221 0.326 0.470 0.08 3.08 Placed weights on these items 220 0.591 0.493 Panel B: Municipalities that did not respect the Federal requirements Federal eligibility criteria Per capita income 40 0.925 0.267 75.00 25.00 Enrolled 40 0.575 0.501 16.67 50.00 Age of the children 40 0.200 0.405 0.00 8.33 Other criteria used Number of children in the household 40 0.225 0.423 0.00 0.00 Living conditions 40 0.125 0.335 0.00 8.33 Enrolled in a municipal school 40 0.175 0.385 0.00 0.00 Placed weights on these items 40 0.300 0.464 Notes to Table 5: The respondent for these questions was the Bolsa Escola coordinator. 16 Household Awareness of Registration Criteria. In 91 percent of the municipalities, households were aware of the criteria used to determine eligibility (see Table 6). In 67 percent of the municipalities, town meetings were held to diffuse information on the qualification criteria, and in 50 percent of the municipalities the criteria were announced over the radio and in 41% over billboards. Only 19 percent of municipalities informed the households at the time of the interview and, despite the importance schools have had in the registration process, only 12 percent of the municipalities used them to convey information about the criteria. Overall, these results suggest that the process followed to identify program beneficiaries was fairly transparent and quite effective. Verification of Household Information. Table 6 also reports considerable variation in the types of information verification mechanisms used by municipalities. Even though the federal government did not require a municipality to verify self-declared information, 65 percent of the municipalities did perform some type of verification. Among those that did, 45 percent asked for proof of income, while 59 percent of those that did any verification did so through a home visit, which functioned like an informal proxy-means test. The majority of the municipalities (88%) that verified information did so by consulting members of the community. Given that the program was decentralized at the level of the municipality, where the median population is only 13,522 persons, this method of information verification may be fairly reliable. Table 6. Additional aspects of beneficiary identification for the Bolsa Escola program (Share of municipalities) Number of Standard observations Mean deviation Households knew the criteria used to identify beneficiaries 261 0.912 0.284 If so, how did the public know about the criteria? Town meetings 238 0.672 0.470 Newspaper 238 0.155 0.363 Radio 238 0.496 0.501 Internet 238 0.013 0.112 Billboards 238 0.408 0.492 Television 238 0.109 0.313 Schools 238 0.122 0.328 Public Announcements 238 0.067 0.251 At the interview 238 0.185 0.389 Municipality verified self-declared information 260 0.646 0.479 If so, how did the municipality verify the information? Asked for proof of income 168 0.452 0.499 Conducted home visits 168 0.589 0.493 Consulted members of the community 168 0.881 0.325 Notes to Table 6: The respondent for these questions was the Bolsa Escola coordinator. 17 4.2. Beneficiary Selection The Federal government, along with establishing the criteria for program eligibility, specified for each municipality quotas indicating the number of households that could participate in the Bolsa Escola program. These quotas, which were based on the estimated number of households in the municipality that would meet the three federal eligibility requirements based on census/household survey estimates, were often insufficient to meet the municipal demand. In our sample more than 97 percent of the municipalities had qualified children who were rationed out of the program. For these municipalities, an estimated 49 percent of eligible household were left out of the program. Among the pool of eligible households, it was thus the municipality’s responsibility to select the program beneficiaries and to thus ration these benefits. Confusion About Institutional Roles for Beneficiary Selection under Bolsa Escola. Under the Bolsa Escola Program, municipalities were responsible for selecting beneficiaries within these quotas and according to the federal criteria. Yet, despite this discretion, one of the more striking results is the extent of confusion about who was officially responsible for selecting the beneficiaries (see Figure 4). Among the municipalities sampled, 63 percent responded that the Federal Government in Brasilia selected who received the program, while 30 percent claimed that the municipal Bolsa Escola council had selected the beneficiaries.14 For the 63 percent of municipalities that did not actively select its beneficiaries, it may have been the case that the order in which they sent the qualifying households to the Federal Government in Brasilia determined program inclusion. For these municipalities, 37 percent did in fact prioritize the sending of the files to Brasilia, implying perhaps an inadvertent selection process. This does, however, suggest that at least 40 percent of the municipalities in the sample did not knowingly select the beneficiary population. Figure 4. Who selected the Bolsa Escola beneficiaries? 7% 40% 63% 30% 23% Council Mayor's office Brasilia -Random Brasilia - Prioritized 14 Confusion about the selection process was also a common observation in the case studies. Program coordinators and school teachers often remarked that they could not understand how Brasilia had decided upon the list of beneficiaries when so many more deserving children were excluded from the program. 18 Factors Associated with Understanding of Institutional Roles. Column (1) of Table 7 reports an OLS regression of an indicator for whether the municipality thought “Brasilia” (the Federal Government) selected the beneficiaries (i.e., misunderstood the selection process) on the same mayor, municipal, program, and political characteristics used in the previous specifications. We find that more educated municipalities, as measured by the share of the population that is literate, are much less likely to have misunderstood the selection process. Somewhat surprising is the fact the characteristics of the mayor, and most notably his education level, have little predictive power. Mayors in their second consecutive term are less likely to have misunderstood the program, but we can only reject that this estimate is different from zero with 83 percent confidence. Whether the municipality received training about the program also did not have any effect. One factor that clearly influences understanding of the selection process is the existence of the Bolsa Escola social controls “Councils.” Specifically, our analysis suggests that misunderstanding regarding institutional responsibilities for the selection process is strongly correlated with not having established a Bolsa Escola council, suggesting that municipalities that misunderstood the selection process also failed in other important aspects of implementation. Lack of civil society participation through a Bolsa Escola council suggests greater likelihood for misinterpretation of, or greater scope for discretion with, program rules. Factors Associated with Municipal Selection of Beneficiaries. Within the sub-group of those municipalities that did acknowledge actively selecting beneficiaries (95 out of 252 municipalities), Column (2) then explores the decision to have the mayor select the beneficiaries versus the Bolsa Escola council. The results suggest that higher income inequality (measured by the Gini Coefficient) is strongly associated with selection by the mayor, whereas having more radio stations and a judiciary district is negatively correlated with mayor selection of beneficiaries. There is also evidence that a mayor’s involvement in beneficiary selection is based on political gain. Municipalities with high levels of clientelism are 27 percentage points more likely to have the mayor select the beneficiaries (significant at 90 percent confidence). A similar finding on the role of clientelism is also reported in column (3) for municipalities where the mayor approved the list of beneficiaries.15 15 Roughly 12 percent of the municipalities had the mayor approve the list of beneficiaries (see Table 9). 19 Table 7. Correlates of the beneficiary selection process for Bolsa Escola (1) (2) (3) (4) Mayor Mayor Misunderstood participated approved list Used weights in the selection beneficiary of the selection process selection beneficiaries criteria Mayor characteristics Education 0.014 -0.002 -0.074 0.015 [0.016] [0.022] [0.025]** [0.026] Gender (male=1) 0.151 -0.327 0.29 0.035 [0.125] [0.211] [0.130]* [0.147] Second-term -0.095 -0.049 0.116 0.188 [0.069] [0.105] [0.098] [0.094]* Political experience 0.004 0.014 -0.003 -0.029 [0.030] [0.059] [0.047] [0.045] Member of an elite family -0.108 -0.081 -0.141 -0.039 [0.082] [0.114] [0.112] [0.138] Municipal Characteristics Population density (Persons/km) -0.015 -0.014 -0.002 0.02 [0.005]** [0.011] [0.012] [0.009]* Number districts 0.012 -0.005 -0.034 -0.008 [0.012] [0.014] [0.013]* [0.012] Share of rural households -0.253 -0.792 -0.133 0.562 [0.198] [0.341]* [0.292] [0.292]+ Share of literate population -1.371 -0.357 -1.499 -0.548 [0.664]* [0.827] [0.827]+ [0.850] Log per capita income 0.18 0.116 0.077 0.083 [0.224] [0.291] [0.261] [0.267] Gini 0.399 1.657 1.35 -0.534 [0.689] [0.780]* [0.736]+ [1.048] Number of radio stations 0.014 -0.153 0 0.076 [0.028] [0.060]* [0.029] [0.038]* Number of catholic churches 0.003 0.017 -0.005 -0.005 [0.003] [0.007]* [0.006] [0.005] Proportion of councils that function -0.053 0.214 -0.449 -0.154 [0.248] [0.339] [0.364] [0.322] Judiciary district -0.154 -0.229 0.131 0.041 [0.086]+ [0.131]+ [0.111] [0.125] Program characteristics Bolsa Escola quota 0 0.006 0.016 0.003 [0.006] [0.011] [0.010] [0.021] Received training 0.064 0.121 0.071 0.12 [0.071] [0.109] [0.100] [0.096] Bolsa Escola Council Exists -0.206 -0.159 -0.232 -0.022 [0.096]* [0.201] [0.157] [0.154] Political Characteristics Patronage -0.004 -0.001 -0.001 0 [0.002]* [0.003] [0.003] [0.003] Medium clientelism 0.046 0.062 0.109 0.018 [0.088] [0.103] [0.107] [0.123] High clientelism 0.092 0.273 0.213 0.021 [0.107] [0.152]+ [0.122]+ [0.152] Observations 252 95 96 152 R-squared 0.2 0.41 0.41 0.17 Notes to Table 7: Coefficient estimates from an OLS regressions are reported. Robust standard errors in brackets; significantly different from zero at (+) 90%, (*) 95%, (**) 99% confidence. In addition to the variables displayed, each regression controls for the mayor’s age, the share of the population that is employed by the local government, the number of newspapers in the municipalities, and state intercepts. We define elite family as a family that has had a long political tradition in the municipality; patronage is defined as the average share of public goods provided to communities for political reasons and not need. Our measure of the degree of clientelism in a municipality is based on a scale of 1 to 7. Medium clientelism corresponds to a value of above 2 but below 4, high clientelism corresponds to a value of above a 4. Summary statistics for the covariates are presented in table A1. 20 Criteria Applied in Beneficiary Selection. Among the municipalities that did acknowledge actively selecting beneficiaries, household per capita income (99% of cases) was the most frequent criteria used in making such decisions (see Table 8).16 The age of the child (82%) and the number of children in the household (84%) were also important, the first as required by the Federal Government in Brasilia and the second likely as a correlate of poverty. Other characteristics of the family that entered into the decision included the condition of the house (56%) and whether the mother was single (48%), two other correlates of poverty. Interestingly, 35 percent of the municipalities that did target the program also prioritized children, who in their opinion were at risk of dropping out of school, indicating concern with the potential educational gains of the CCT. Figure 5 plots the number of factors that entered into the selection process for each municipality. Twelve percent of the municipalities only used one item to select beneficiary families, while over 85 percent consider at least 3 factors. Table 8. Criteria used for the selection of beneficiaries for Bolsa Escola program among municipalities that actively selected (Share of municipalities) Number of Standard First Second observations Mean deviations priority priority Per capita income 155 0.987 0.113 89.80 7.00 Number of children 155 0.839 0.369 4.08 59.00 Single mothers 155 0.477 0.501 1.02 2.00 Age of child 155 0.819 0.386 2.04 12.00 Condition of house 155 0.561 0.498 0.00 3.00 Enrolled in other programs 155 0.361 0.482 0.00 3.00 Child at risk of dropping out 155 0.348 0.478 0.00 0.00 Placed weights on these items 155 0.624 0.486 Notes to Table 8: The respondent for these questions was the Bolsa Escola coordinator. Sixty-two percent (152) of the municipalities that selected beneficiaries used explicit weights to prioritize among the various criteria. As seen in column (4) of Table 7, second-term mayors are much more likely to place weights on the criteria, as well as municipalities that are more rural and have more radio stations. Political motives, however, do not appear to have been a factor in this decision (measures of clientelism and patronage were not significant in the regression). Transparency and Public Information Regarding Beneficiary Selection Criteria. Table 9 reports some aspects of transparency in the selection process. Among the municipalities that actively selected their beneficiaries, the population was informed about the selection criteria in 92 percent of the municipalities. Moreover, in 85 percent of the municipalities the list of families participating in the program was made publicly available, with billboards (76%) being 16 Note that for beneficiary selection the question asked whether or not the municipality among the qualified households further prioritized households according to their income level. In contrast, per capita income for beneficiary identification simply referred to whether or not the household qualified for having a per capita income of no more than R$90. And similarly, for the age of the child. 21 the most common form of disclosure. These results are consistent with the results presented in the beneficiary identification, and support the conclusion that the program was implemented in a fairly transparent manner. Figure 5. Number of characteristics that determined selection into the Bolsa Escola program (Share of municipalities with active selection) 35 30 30 28 27 25 23 21 20 15 12 10 10 4 5 0 1 2 3 4 5 6 7 8 Table 9. Additional aspects of beneficiary selection in the Bolsa Escola program (Shere of municipalities) Number of Standard observations Mean deviations Who approved the list of beneficiaries Mayor 260 0.119 0.325 Council 261 0.444 0.498 Community 261 0.080 0.273 Brasilia 260 0.592 0.492 Did the public know the criteria used for selection? 156 0.923 0.267 Was the list of beneficiaries released publicly? 261 0.854 0.353 If so, how was the list released publicly? Meetings 223 0.193 0.395 Newspapers 223 0.283 0.451 Radio 223 0.395 0.490 Internet 223 0.022 0.148 Billboards 223 0.762 0.427 Notes to Table 9: The respondent for these questions was the Bolsa Escola coordinator. 22 4.3. Monitoring and Enforcement of Conditionalities According to Federal rules, participants to the Bolsa Escola program receive monthly payments conditional on attending school at least 85 percent of the time. There are several steps involved in monitoring and enforcement of conditionalities, each with its own institutional arrangements under the Bolsa Escola Program: (a) monitoring of compliance at the school level (responsibility of municipalities to compile reports of school directors); (b) forwarding of information by municipalities to the Ministry of Education (MEC) and the Caixa Economica Federal (Caixa); and (c) linking of non-compliance information to payments for eventual penalties (MEC, Caixa). As discussed below, most municipalities report monitoring of conditionalities. In practice, however, data from the Ministry of Education suggest that the actual forwarding of information on conditionalities compliance to the central level (MEC, Caixa) was rather weak: at its highest ever, only MEC received such information for a maximum of 19 percent of schools reporting. Consequently, the imposition of penalties for non-compliance was even more rare. Imposition of Additional Conditionalities. As Table 10 reports, virtually all municipalities (99 percent) in our municipal survey reported imposing the school-attendance conditionality, which was essentially the only federally-imposed behavioral requirement of the program.17 In addition to this requirement, however, some municipalities reported imposing other conditions upon parents. Approximately 33 percent of them required that parents either attend school meetings (30 percent) or maintain and clean the school (3 percent). Four percent of the municipalities made other demands such as to provide receipts for how the money was spent, to have vaccination cards up to date, or to require parents to attend school and learn how to read (see Table 11). Overall 28 percent of the municipalities imposed at least 2 requirements for program participation (see Figure 6). Table 10. Principal conditions imposed on Bolsa Escola recipients (Share of municipalities) Obs Mean Std. Dev School attendance 261 0.992 0.087 Help clean or maintain the school 261 0.027 0.162 Attend meetings 261 0.299 0.459 Other conditionalities imposed on the parents 261 0.042 0.201 Notes to Table 10: The respondent for these questions was the Bolsa Escola coordinator. 17 These questions were responded by program coordinators. It should be noted that these responses do not imply that monitoring data were necessarily sent to the Federal Government in Brasilia. In fact, information from the Ministry of Social Development suggests that at most only 19% of municipalities nationwide ever sent in such information (only 13% by the end of the program). 23 Table 11. Additional conditions imposed on Bolsa Escola recipients Number of cases Apply the money responsibly 3 Learn how to read and write/attend school 2 Participate in school activities 6 Continue to live in the municipalities 6 Children progress academically 1 Keep the children vaccinated 1 Notes to Table 11: The respondent for these questions was the Bolsa Escola coordinator. Figure 6. Number of requirements imposed for program participation 80 68.2 70 Percentage of Municipalities 60 50 40 27.59 30 20 10 4.21 0 1 2 3 This decision to impose additional conditionalities is mostly associated with the relative size of the municipality’s quota as a share of the number of children enrolled in primary and secondary school and the proportion of municipal councils that function (see Table 12). The negative relation between a smaller municipal Bolsa Escola quota and a larger number of additional conditionalities is consistent with expectations about the allocation of scarce goods (rationing). Interpretation of the negative relation between the proportion of councils that function and imposition of additional conditionalities is less clear. Lame-duck mayors are also less likely to impose additional program requirements, but this estimate is only measured at an 87 percent confidence level. Less explicit conditionalities means more space for political maneuvering. This is supported by the result in Ferraz and Finan (2005) who show that second term mayors are more prone to engage in corrupt practices than first term mayors that have to be concerned with re-election. 24 Table 12. Imposing additional program conditions in the Bolsa Escola program Imposed other conditionalities Mayor characteristics Education 0.005 [0.016] Gender (male=1) -0.07 [0.116] Second-term -0.105 [0.068] Political experience -0.014 [0.028] Member of an elite family -0.106 [0.085] Municipal Characteristics Population density (Persons/km) 0 [0.005] Number districts 0.003 [0.012] Share of rural households 0.337 [0.234] Share of literate population -0.916 [0.642] Log per capita income 0.279 [0.219] Gini -0.457 [0.722] Number of radio stations 0.003 [0.025] Number of catholic churches -0.002 [0.004] Proportion of councils that function -0.424 [0.236]+ Judiciary district 0.063 [0.087] Program characteristics Bolsa Escola quota -0.013 [0.006]* Received training 0.008 [0.068] Bolsa Escola Council Exists 0.082 [0.075] Political Characteristics Patronage 0 [0.002] Medium clientelism -0.001 [0.089] High clientelism 0.047 [0.107] Observations 252 R-squared 0.14 Notes to Table 12: Coefficient estimates from an OLS regressions are reported. Robust standard errors in brackets; significantly different from zero at (+) 90%, (*) 95%, (**) 99% confidence. In addition to the variables displayed, the regression controls for the mayor’s age, the share of the population that is employed by the local government, the number of newspapers in the municipalities, and state intercepts. We define elite family as a family that has had a long political tradition in the municipality; patronage is defined as the average share of public goods provided to communities for political reasons and not need. Our measure of the degree of clientelism in a municipality is based on a scale of 1 to 7. Medium clientelism corresponds to a value of above 2 but below 4, high clientelism corresponds to a value of above a 4. Summary statistics for the covariates are presented in table A1. 25 Public Awareness of Conditionalities. As seen in Table 13, among the municipalities sampled, 90 percent responded that all the beneficiaries were notified about the conditions of the program. Town hall meetings (77 percent) and schools (89 percent) were the more commonly used sources of information, with over 98 percent of the municipalities notifying recipients of their responsibilities either at the school or in town hall meetings. Somewhat surprising is that only 9 percent of the municipalities notified the families about these requirements during the course of the interview. Table 13. Monitoring and enforcement of the Bolsa Escola participation requirements (Share of municipalities) Obs Mean Std. Dev Did all of the beneficiaries know about the conditionalities? 261 0.900 0.300 If so, how did they all know? Town meetings 235 0.774 0.419 Home 235 0.234 0.424 School 235 0.889 0.314 Radio 235 0.068 0.252 At the time of the interview 235 0.089 0.286 Did you monitor attendance? 261 0.950 0.218 Notes to Table 13: The respondent for these questions was the Bolsa Escola coordinator. 4.4. Bolsa Escola’s Social Control Councils: A Short Route to Downward Accountability To participate in the Bolsa Escola program, a municipality was required to create a social council (Conselho de Controle Social) designed to approve the list of selected families, verify the school attendance of beneficiary children, and address complaints about the program. Moreover, in further stipulating that at least 50 percent of the council be comprised of non-governmental members, the council was to provide program oversight and voice for the various segments of civil society. Mayors, however, were given full discretion in selecting the council members. Our municipal survey suggests the following conclusions with respect to these Social Controls councils as a “short route” to downward accountability: (a) although most municipalities operated such councils, about a fifth of them did not establish them despite the federal requirement to do so; (b) in municipalities where social councils did exist, there was a positive impact on the quality of program implementation; but, (c) even when they did exist, they did not necessarily function properly and their membership seems to have been selected with predominant support for the mayors. Establishment of Social Controls Councils: Incomplete. One of the more striking results presented in Table 14 is that a council existed in only 81 percent of the municipalities, despite being a requirement of the program. Among those that did establish a council, some were simply integrated into a pre-existing council (such as the education oversight council).18 18 Instead of creating an entirely new council, municipalities had the option of assigning an existing council to perform these various functions of the program. In our sample, among the municipalities where a Bolsa Escola council existed (81 percent), 40 percent of these municipalities incorporated the Bolsa Escola council into an existing one. 26 Table 14. Existence and Performance of Bolsa Escola Councils (Share of municipalities and share of council members) Number of Standard observations Mean deviation Council exists 261 0.812 0.391 Council functions 212 0.675 0.470 Council deliberative 212 0.571 0.496 Percentage of members that particiapate at meetings 209 0.828 0.240 Meets at least once a month 261 0.383 0.487 Has the right to remove a child from the program 210 0.600 0.491 Has the right to include a child into the program 207 0.599 0.491 Council monitors the attendence of the children 212 0.736 0.442 Council maintains an updated list of beneficiaries 208 0.543 0.499 Member of the legislative branch participates in the program 261 0.100 0.300 Percentage of councils that: Supports the mayor 213 0.730 0.261 Relative of the mayor 213 0.066 0.108 Beneficiary of the program 213 0.090 0.174 Public employee 213 0.272 0.192 Member of local legislature 213 0.077 0.097 Members of large farmer unions 213 0.004 0.021 Agricultural unions 213 0.053 0.079 Member of teacher Association 213 0.112 0.133 Member of parent Association 213 0.079 0.112 Health agents 213 0.023 0.056 Representatives of the catholic church 213 0.103 0.138 At least one relative of the mayor is on the council 213 0.343 0.476 At least one program recipient is on the council 213 0.324 0.469 At least one non-parent program recipient is on the council 213 0.225 0.419 Notes to Table 14: The respondent for these questions was a member of the Bolsa Escola council, when it existed. Factors Associated with Existence of Councils. Columns (2) and (3) of Table 15 report correlates of whether or not a Bolsa Escola council existed in the municipality. Column (2) estimates the OLS regression for the entire sample of 252 municipalities, whereas the estimation sample used in column (3) excludes municipalities that simply incorporated the Bolsa Escola council into a preexisting one, reducing the sample to 171 municipalities. Column (2) shows that a mayor’s political experience and the share of functioning councils are both negatively correlated with the existence of the council. A mayor’s political experience, measured by the number of times he has held a political office, suggests lack of political competition, and hence less pressure to engage into downward accountability. The share of the population that is literate is positively correlated with an existing council, indicating greater ability to demand social accountability. Bolsa Escola councils are also less likely to exist in municipalities with medium to high levels of clientelism: The probability that a council exists is 17 percent lower in municipalities with high levels of clientelism compared to municipalities with low levels. The results in column (3) are broadly similar. They suggest that use of the program for clientelistic purposes is done at the cost of a loss in accountability. 27 Table 15. Bolsa Escola Council correlates (1) (2) (3) Legislator is a Council exist beneficiary Entire sample Restricted sample Mayor characteristics Education 0.014 0.02 0.04 [0.018] [0.013] [0.017]* Gender (male=1) 0 0.147 0.153 [0.088] [0.110] [0.148] Second-term 0.009 0.04 0.041 [0.054] [0.051] [0.067] Political experience 0.06 -0.051 -0.06 [0.032]+ [0.026]+ [0.032]+ Member of an elite family -0.16 0.042 0.041 [0.109] [0.072] [0.099] Municipal Characteristics Population density (Persons/km) -0.02 0.004 0.006 [0.028] [0.006] [0.019] Number districts -0.008 0.005 0.011 [0.008] [0.006] [0.008] Share of rural households 0.408 0.193 0.388 [0.222]+ [0.165] [0.252] Share of literate population -0.678 0.969 1.226 [0.553] [0.527]+ [0.797] Log per capita income 0.4 -0.068 -0.103 [0.183]* [0.168] [0.248] Gini 0.529 0.025 -0.297 [1.069] [0.582] [0.778] Number of radio stations -0.046 0.009 0.023 [0.035] [0.024] [0.034] Number of catholic churches -0.005 0.004 0.005 [0.003] [0.004] [0.006] Proportion of councils that function 0.023 -0.298 -0.449 [0.208] [0.134]* [0.193]* Judiciary district -0.122 0.057 0.095 [0.091] [0.070] [0.098] Program characteristics Bolsa Escola quota 0.014 -0.007 -0.004 [0.005]** [0.005] [0.005] Received training 0.127 0.014 0.005 [0.058]* [0.054] [0.074] Political Characteristics Patronage -0.001 0 0 [0.002] [0.002] [0.002] Medium clientelism -0.053 -0.103 -0.154 [0.092] [0.067] [0.089]+ High clientelism 0.027 -0.144 -0.177 [0.115] [0.075]+ [0.098]+ Number of legislators 0.031 [0.015]* Share of legislator that opposes the mayor 0.349 [0.193]+ Share of secretaries related to a legislator -0.259 [0.142]+ Number of jobs a legislator can appoint -0.045 [0.024]+ Observations 243 252 171 R-squared 0.16 0.24 0.3 Notes to Table 15: Coefficient estimates from an OLS regressions are reported. Robust standard errors in brackets; significantly different from zero at (+) 90%, (*) 95%, (**) 99% confidence. In addition to the variables displayed, each regression controls for the mayor’s age, the share of the population that is employed by the local government, the number of newspapers in the municipalities, and state intercepts. We define elite family as a family that has had a long political tradition in the municipality; patronage is defined as the average share of public goods provided to communities for political reasons and not need. Our measure of the degree of clientelism in a municipality is based on a scale of 1 to 7. Medium clientelism corresponds to a value of above 2 but below 4, high clientelism corresponds to a value of above a 4. Summary statistics for the covariates are presented in table A1. 28 Benefits of Existence of Social Controls Councils. As reported in the sections above, the existence of a BE social controls council seems to have been associated with higher quality program implementation by municipalities. Specifically, municipalities with social controls councils were far more likely to have a clear understanding (less confusion) about institutional responsibilities for beneficiary selection (see Section 4.2 above). Reliance on mayor selection of beneficiaries – without involvement of BE social controls councils in the selection process – also appears to be correlated with higher income inequality and higher perceptions of clientelism. Uneven Functioning of Social Controls Councils. Despite these apparent benefits, our municipal survey suggests that there is ample room for strengthening the functioning of these councils. When council members were asked if their council functioned properly, only 68 percent of those municipalities where a council existed claimed that they did. The fact that councils met at least once of month in only 38 percent of the municipalities perhaps best describes the lack of functionality. Another stark observation is that only 54 percent of the councils maintained an updated list of the beneficiaries. As such, CCT programs should go beyond simply requiring that municipalities establish social controls councils and provide additional guidance and support for their functioning. Composition of Social Controls Councils: Mayoral Support. The composition of the council also demonstrates the limited ability of these councils to serve as proper watchdogs or to truly represent the interests of civil society. Among the 213 municipalities with a council, on average 73 percent of the council members were supporters of the mayor and 7 percent were actual relatives of the mayor.19 In fact, having a relative of the mayor serve on the council appeared in 34 percent of the municipalities. Some Instances of Political Manipulation of Bolsa Escola? Table 14 also reports that, on average, 27 percent of the council members were actual beneficiaries of the program.20 While it was common to have a representative of the beneficiaries participate in the council, when we exclude these types of representatives, 22 percent of municipalities still had a program recipient. Perhaps more astonishing is that in 10 percent of the municipalities, a member of the legislative branch was a recipient of the program.21 Although it is technically possible for a local politician to be eligible for the program, at an average salary of R$1,400 per month, the politician would have to be the sole earner in a household of over 14 members to meet federal qualification rules. A more plausible explanation is that the mayors used the Bolsa Escola program in exchange for support in the legislative branch. 19 The statistic that on average 73 percent of the council supports the mayor does not necessarily reflect the fact that the mayor appoints the council. It is more likely the result of the requirement that only 50 percent of the council be composed of nongovernmental members. 20 Although this number appears large, one would need to know what a random draw would predict in order to properly assess this magnitude. 21 Several newspapers have reported incidences of fraud associated with the Bolsa Escola program. For example, Folha de São Paulo reported based on an audit done by the CGU (Federal Comptroller’s Office) that in the municipality of Cachoeira do Piriá (PA) , the president of the mayor’s cabinet, a principal of the a primary school, and a member of the local legislative branch were all found participating in the program. (Folha de São Paulo 9-14- 2003). 29 Column (1) of Table 15 explores this possibility, reporting the results of a regression that estimates the probability that a member of the legislative branch benefited from the program. The specification is similar to those presented in the previous tables, except that it includes variables to measure the mayor’s level of support in the legislature. The results show that the higher the share of legislators that oppose the mayor, the more likely it is (35 percent greater chance) that a legislator will participate in the program. Receiving the program is also a substitute to other types of political exchanges.22 Legislators are less likely to participate in the program in municipalities where the share of secretaries related to a legislator is higher, and where legislators have control over more public appointments. Legislators are also more likely to participate in the program in municipalities that are more rural, where the municipality’s quota is higher, and where the mayor has more political experience. These indicators clearly support an interpretation where the allocation of these bolsas is used as an explicit element of clientelism and political rents. 4.5. Political Rewards and Re-Elections: Long Route to Downward Accountability An important conclusion of this report is that Bolsa Escola Councils are important as a potential mechanism to promote a short route to program accountability in the short term – but that in practice, these mechanisms did not function to their full potential (as discussed above). In this section, we explore the extent to which an alternative yet longer route to accountability functioned via electoral rewards and punishments. In this regard our survey yields several interesting findings: (a) the size of Bolsa Escola program quotas yielded important political dividends to incumbent mayors; (b) mayors that created Bolsa Escola Councils were far more likely to be re-elected; (c) targeting accuracy pays off for politicians if it is perceived as minimizing “errors of inclusion;” but (d) perceived errors of exclusion – poor families that should benefit but do not – did not generate political costs for incumbent mayors. Short-Route Accountability Rewarded. Column (1) of Table 16 reports an OLS regression of an indicator for whether the mayor was reelected in the 2004 municipal elections on an indicator for whether the Bolsa Escola council existed and the municipality’s quota of program beneficiaries, while also controlling for the same mayor, municipal, and political characteristics presented in previous specifications.23 From this straightforward regression, we see that elections did serve as an important mechanism to achieve accountability in the delivery of this social program. Compared to the mayors of municipalities where a social council was not put into place, mayors who created a Bolsa Escola council were 26 percentage points more likely to get reelected. At a baseline of 55 percent (re-election average), this estimate indicates a 47 percent increase in reelection rates. 22 In every municipality, the mayor has a number of public positions, called cargos de confiança, which he can allocate to whomever he chooses. In talking to several people, it is widely recognized that these jobs are clear instruments of patronage and even nepotism, which is remarkably high in Brazil. For example, a sub-secretary of education of a particular municipality said that he was originally secretary of education until the mayor appointed the cousin of a vereador secretary in exchange for a thousand votes. In order to keep the former secretary of education, since he was quite competent in his work, the mayor created the new position of sub-secretary. 23 The sample consists of only those mayors that were in their first term and hence eligible for reelection. Brazil limits mayors to two consecutive terms. 30 Table 16. Long route to downward accountability: Electoral rewards Dependent variable: Mayor was reelected in 2004 (1) (2) (3) Bolsa Escola council exists 0.264 0.262 0.206 [0.133]+ [0.128]* [0.147] Public denouncement for Type II (inclusion) error -0.263 -0.25 [0.111]* [0.121]* Public denouncement for politics -0.003 -0.053 [0.131] [0.142] Public denouncement for Type I (exclusion) error 0.011 0.031 [0.119] [0.122] Registered beneficiaries in mayor's office 0.034 [0.131] Registered beneficiaris using home visits 0 [0.003] Registered beneficiaries with geographic priorization 0.132 [0.120] Misunderstood selection process 0 [0.115] Quota 0.016 0.013 0.011 [0.006]** [0.006]* [0.006]+ Mayor characteristics Y Y Y Municipal Characteristics Y Y Y Political Characteristics Y Y Y Observations 108 108 105 R-squared 0.38 0.43 0.45 Notes to Table 16: Coefficient estimates from an OLS regressions are reported. Robust standard errors in brackets; significantly different from zero at (+) 90%, (*) 95%, (**) 99% confidence. Municipal characteristics include population density, number of districts, share of rural households, share of literate population, log per capita income, Gini coefficient, number of radio stations, number of catholic churches, proportion of councils that function, judiciary district, number of newspapers, share of population employed by the local government, state intercepts; mayor characteristics include education, gender, second-term, political experience, age, member of an elite family; political characteristics include patronage, medium clientelism, high clientelsim. Quota is expressed as a share of the total number of children enrolled in primary or lower secondary school. Political Dividends of the Bolsa Escola Program. The estimation results reported in column 1 also suggest that the program provided political benefits to the incumbent mayor. The greater the number of children that the municipality could benefit with the program (expressed as a share of the total number of children enrolled in primary or lower secondary school) the more likely the incumbent mayor would gain reelection. This positive association appears despite the fact that the quota was exogenously determined and Bolsa Escola is a federal program. The devolution of program implementation allowed local mayors to reap political rewards when they were perceived as effective intermediaries for potential beneficiaries in the municipality. Political Economy of “Targeting” Errors. Column 2 presents a similar specification to column 1, but adds additional covariates to distinguish municipalities that had received different types of public complaints about the program. Not only do the results from column 1 remain robust to these controls, but we also find that errors of inclusion are associated with significant political costs. The perceived inclusion of non-poor households decreases the probability of the incumbent’s reelection by 26 percentage points, which in magnitude is similar for not 31 implementing a social council. On the other hand, perceived errors of exclusion and complaints of mistargeting based on political motives made no difference at the polls. The final specification, presented in column 3, includes as controls various aspects of beneficiary identification and selection in addition to the other covariates presented in column 2. The results from this regression show that a municipality’s decision to target geographically, or to register beneficiaries at the mayor’s office, or to use home visits did not affect election outcomes of incumbent mayors. The other effects, discussed in column 2, continue to be robust although the effects of the existence of a social council are slightly attenuated and measured only with 85 percent confidence. In summary, the results presented in Table 16 suggest the following two points. First, incumbent mayors gained politically from the introduction of the program. This gain is likely the result of segments of the population being better off with the transfers, in addition to the mayor’s ability to target the program politically. The fact that several aspects of the implementation strategies adopted by the municipality were correlated with levels of clientelism and patronage does lend support to the latter effect. Second, the longer route of political accountability through electoral rewards and punishments does seem to be effective. Mayors, who did not target the program properly or failed to provide civil society a forum for voice and appeal, experienced a significant degree of political cost. While this result is in some respects reassuring, a reliance on political accountability is clearly only a second best option. More immediate mechanisms of social accountability, such as effective Social Councils, are also necessary to optimize delivery of social benefits delivered by local service providers. This is all the more important when social programs are ephemeral creatures that might not extend beyond a political cycle. V. Municipal Survey Results on the Decentralized Implementation of the Bolsa Família Program (as of late 2004): A Comparative Analysis In 2003, Bolsa Escola and three other federal cash transfer programs were unified into a single program called Bolsa Familia. With this merger, the Bolsa Familia program instituted a number of important reforms, including, inter alia: (a) merging conditionalities (education, health and nutrition) for greater synergies; (b) shifting the assistance unit from a focus on the individual (e.g., the specific child within a household) to the focus on the entire household (all relevant family members must comply with conditionalities, not just some); (c) increasing the unit transfer benefits; and (d) reforming and clarifying many of the institutional arrangements surrounding the program. Examples of the latter, which are particularly relevant for this study, include shifting responsibility for beneficiary selection to the federal government (Ministry of Social Development). Nonetheless, many features of Bolsa Familia remain decentralized, such as data collection and registration of potential beneficiaries into the unified registry (Cadastro Unico), monitoring of conditionalities, and social controls. In this section, we document the decentralized implementation of the Bolsa Familia program using information collected in our municipal survey (as of late 2004), and contrast some of these approaches with those of the Bolsa Escola program. It is important to note, however, that the Bolsa Familia program was still 32 very young at the time our survey was conducted, and many reforms to implementation have been instituted since the survey was carried out (particularly in 2005). 5.1. Identification and Registration of Potential Beneficiaries Even under Bolsa Familia, municipalities retain responsibility for identifying and registering potential beneficiaries into the unified registry (Cadastro Unico). This section reports several aspects of decentralized implementation of this process, contrasting them with our findings for Bolsa Escola. Actors Involved in Carrying Out Registration. Our municipal survey reveals an important shift in the actors involved in registering potential beneficiaries under Bolsa Familia. Whereas teachers and school administrators played a prominent role in identifying and registering potential beneficiaries under Bolsa Escola, school-based staff are less likely to be involved in registration for the Bolsa Familia program. Municipalities in our survey report that teacher participation in the registration process occured in only 20 percent of the municipalities (see Table 17), compared to 70 percent for the Bolsa Escola program. Health agents (54 percent of municipalities) and contracted individuals (26 percent) have compensated for this decline in teacher involvement. This is not surprising given the shift in focus under the Bolsa Familia program, which integrates education, health and nutrition conditionalities and maintains the whole family as the assistance unit: community health agents maintain an on-going relationship with the entire family (often conducting home visits for health care services). In addition, while responsibility for the Bolsa Familia program has shifted from education secretariats to “social action” (or social assistance/development) secretariats (76 percent of municipalities), there are considerably more secretariats involved in this program at the municipal level. As Figure 7 shows there are on average 0.2 more secretariats involved in the Bolsa Familia program than in the Bolsa Escola program (difference is significant at 99 percent confidence). Figure 7. Number of secretariats involved in the registration across municipalities 200 180 160 140 120 100 80 60 40 20 0 1 2 3 4 5 Bolsa Familia Bolsa Escola 33 Table 17. Bolsa Familia beneficiary identification (Share of municipalities) Number of Standard observations Mean deviation Which secretariat was responsible for registering the beneficiaries Education 260 0.404 0.492 Health 260 0.373 0.485 Social Action 260 0.758 0.429 Culture 260 0.023 0.150 Agriculture 260 0.096 0.295 Who registers the potential beneficiaries Health agents 259 0.541 0.499 Public administration 259 0.776 0.418 Contracted individuals 259 0.255 0.437 Teachers 259 0.197 0.398 Portal da Alvorada 259 0.066 0.248 Comite of Zero Hunger 259 0.031 0.173 How were individuals informed about the registration Radio 259 0.649 0.478 Public announcement 259 0.371 0.484 Television 259 0.054 0.227 Newspapers 259 0.147 0.355 Community leaders 259 0.649 0.478 Schools 259 0.622 0.486 Health agents 259 0.301 0.460 Did the registration take place in a public place 260 0.900 0.301 Schools 233 0.399 0.491 Health posts 233 0.197 0.399 Public administration 233 0.764 0.426 Communities 233 0.442 0.498 Geographic targeting 259 0.471 0.500 Poor neighborhoods 122 0.680 0.468 Greater number of schools 122 0.230 0.422 Ease of access to target group 122 0.434 0.498 Distance from municipal head 122 0.107 0.310 Did not register households at home 257 0.385 0.488 Percentage of household registered at home >0 158 56.259 33.628 Notes to Table 17: The respondent for these questions was the Bolsa Familia coordinator. 34 Public Knowledge About the Program. There are few differences between Bolsa Familia and Bolsa Escola in terms of how the population learned about the program. The main difference is that in only 62 percent of the municipalities’ schools inform potential beneficiaries of Bolsa Familia compared to almost 94 percent for Bolsa Escola, but this is again consistent with amplification of the program’s focus. Health agents again play an important role in promoting the program (30 percent of municipalities). Location of Registration Activities. Registration activities under Bolsa Familia are more likely to occur in the public administration buildings (76 percent) and health centers (20 percent) than under Bolsa Escola. While 84 percent of the municipalities had used schools to registered Bolsa Escola recipient, only 40 percent of the municipalities use schools to register eligible families into the Bolsa Familia program. There is also a significant increase in the percentage of households that are registered at home under Bolsa Familia. Sixty-two percent of the municipalities registered some proportion of the households at their home, compared to only 28 percent conducting home visits for the Bolsa Escola program. Again, this likely reflects the shifted emphasis to the family (rather than the individual) under the Bolsa Familia program. Use of Geographic Targeting for Prioritizing Registration. The percentage of municipalities that used geographic targeting also increased slightly under Bolsa Familia (a 9 percentage points increase which is significant at a 95 percent confidence level). In addition, municipalities are more likely to verify the information households provide for the Bolsa Familia program than they were for the Bolsa Escola program (71 percent in Table 20 compared to 65 percent in Table 6 for Bolsa Escola). Persistence of Municipal Approaches From BE to BFP. Even though municipalities have taken some different approaches towards the implementation of the Bolsa Escola and Bolsa Familia programs, there is considerable persistence in their procedures. Table 18 reports separate regressions for whether the municipality performed home visits, targeted geographically, and verified household information. In addition to the controls presented in the previous regression tables (mayor, municipal, and political characteristics), each specification includes, among others, a variable that can be thought of as a lagged dependent variable. For example, in column (1) we regress the percentage of households registered at home for the Bolsa Familia program on the percentage of households registered at home for the Bolsa Escola program in addition to a set of other controls. We find that municipalities that used these implementation methods for Bolsa Escola are much more likely to use them for Bolsa Familia. For example, if a municipality performed home visits for the Bolsa Escola program, it is 62 percent more likely to perform home visits for the Bolsa Familia program. Table 18 also demonstrates that program training has little influence on any of these decisions. The exception is the decision to verify the household’s information which is positively correlated with the training a municipality received for the Bolsa Escola program and not the Bolsa Familia program (see column 3). However, public denouncements of the Bolsa Escola program for large Type I and II targeting errors do not seem to encourage municipalities to perform more home visits or verify a household’s information. 35 Table 18. Persistence of procedures in the Bolsa Familia program implementation (1) (2) (3) Geographical Home visits prioritization Verification Home visits in Bolsa Escola 0.388 -0.001 [0.113]** [0.001] Geographic targeting Bolsa Escola 0.317 [0.071]** Verification Bolsa Escola -0.069 -0.032 0.117 [5.422] [0.072] [0.070]+ Training Bolsa Familia -0.199 0.049 -0.035 [8.676] [0.098] [0.087] Training Bolsa Escola 1.169 0.037 0 [5.347] [0.000] [0.071]* Public denouncement for Type I error -0.188 -0.071 -0.044 [5.683] [0.073] [0.065] Public denouncement for politics 3.22 -0.039 0.072 [7.030] [0.091] [0.089] Public denouncement for Type II error 7.272 0.012 -0.025 [6.810] [0.085] [0.083] Number of persons involved in the Bolsa Familia registration 0.107 0.001 -0.001 [0.081] [0.001] [0.001] Mayor characteristics Y Y Y Municipal Characteristics Y Y Y Political Characteristics Y Y Y Observations 240 246 243 R-squared 0.23 0.23 0.19 Notes to Table 18: Coefficient estimates from an OLS regressions are reported. Robust standard errors in brackets; significantly different from zero at (+) 90%, (*) 95%, (**) 99% confidence. Municipal characteristics include population density, number of districts, share of rural households, share of literate population, log per capita income, Gini coefficient, number of radio stations, number of catholic churches, proportion of councils that function, judiciary district, number of newspapers, share of population employed by the local government, state intercepts; mayor characteristics include education, gender, second-term, political experience, age, member of an elite family; political characteristics include patronage, medium clientelism, high clientelsim. Public denouncement of type I error correspond to complaints during the Bolsa Escola program about individuals receiving the program that should have. Public denouncement of type II error corresponds to complaints during the Bolsa Escola program about individuals that are eligible about the program but were left out. Table 19 reports some of the criteria reportedly used in registering potential beneficiaries according to program coordinators. Per capita income is again clearly the most important determining factor for whether or not a household is registered (98 percent of municipalities). Other important characteristics include family size (75 percent), age of the children (68 percent), and living conditions (60 percent). Interestingly, 8 percent of the municipalities also reported any health deficiency among household members as an important consideration. Table 20 provides some of the additional criteria used for beneficiary identification that were given by the program coordinators. Municipalities in our sample report prioritizing households with unemployed members, with several elderly members, and children enrolled in school. Municipalities also report discriminating against families that have several pensioners. In 89 percent of the municipalities, individuals were reportedly made aware of these criteria (see Table 21). 36 Table 19. Criteria used to identify Bolsa Familia beneficiaries (Share of municipalities) Number of Standard observations Mean deviations First priority Second priority Per capita income 259 0.977 0.151 92.11 5.92 Family size 259 0.753 0.432 2.63 58.55 Age of the children 259 0.676 0.469 0.00 17.76 Living conditions 259 0.595 0.492 1.32 10.53 Has a health deficiency 259 0.081 0.273 0.00 0.00 Placed weights on these items 256 0.594 0.492 Notes to Table 19: The respondent for these questions was the Bolsa Familia coordinator. Table 20. Additional criteria used for beneficiary identification for the Bolsa Familia program Number of cases There are no retired members in the family 6 Families with pregnant women 1 All families in the municipality 2 Unemployed 4 People with documentation 4 Number of elderly in the household 6 Single mothers 3 Number of children 2 Those who were in other programs 6 Illiterate people 1 Matriculated children 7 Married households 1 Households that live in rural areas 1 Households that vaccinate their children on time 1 Is not a public employee 1 Notes to Table 20: The respondent for these questions was the Bolsa Familia coordinator. 37 Table 21 Additional aspects of Bolsa Familia beneficiary identification (Share of municipalities) Number of Standard observations Mean deviation Individuals knew of the criteria 261 0.889 0.315 Meetings 232 0.608 0.489 Newspaper 232 0.099 0.299 Radio 232 0.483 0.501 Internet 232 0.004 0.066 Television 232 0.099 0.299 Health agents 232 0.138 0.345 At the interview 232 0.207 0.406 Schools 232 0.061 0.240 Public announcements 232 0.057 0.233 Verified information of the register 258 0.709 0.455 Proof of income 183 0.383 0.487 Home visits 183 0.738 0.441 Spoke to member of the community 183 0.869 0.338 Health agents 183 0.071 0.258 Received training for the Cadastro Unico 259 0.846 0.362 How days of training 213 2.559 2.623 Number of months that the registration has taken 257 22.825 13.599 Number of people working on the registrations 256 19.020 29.368 Hired a firm or individuals to help with the registration 260 0.242 0.429 Notes to Table 21: The respondent for these questions was the Bolsa Familia coordinator. 5.2. Beneficiary Selection An important policy shift under Bolsa Familia was to explicitly shift responsibility for selecting beneficiaries to the federal government (“Brasilia”) under the auspices of the Ministry of Social Development (MDS). This was explicitly clarified in the regulatory decree for the program, which was issued on September 17, 2004. Interestingly, the influence of this policy clarification is evident in the results of our municipal survey, which was carried out in the months following this declaration. Specifically, a much larger share (88 percent) of municipalities reported that the Federal Government in “Brasilia” is responsible for selecting beneficiaries, with only 12 percent suggesting that municipal administrators and/or the local council are responsible for such selection (see Table 22). As such, there was already less “confusion” regarding responsibility for selecting beneficiaries under Bolsa Familia – even in the months just following the decree – than there was under Bolsa Escola. Nonetheless, even with these clarifications, about 40 percent of municipalities reported not knowing what their program quota was at that time (which is not surprising given that the quota policy was still being clarified at that time). 38 Table 22. Bolsa Familia beneficiary selection (Share of municipalities) Number of Standard observations Mean deviation Knew quota for Bolsa Familia 261 0.609 0.489 How many families can participate in Bolsa Familia 159 2030.088 4505.812 How families are enrolled in the program 235 2082.196 4216.933 Who selected the beneficiaries? Brasilia 260 0.877 0.329 Public administration 260 0.038 0.193 Council 260 0.085 0.279 Were the questionnaires sent in a particular order? 224 0.384 0.487 Knew the criteria for selection 114 0.904 0.297 Meetings 103 0.650 0.479 Newspapers 103 0.184 0.390 Radio 103 0.495 0.502 Internet 103 0.000 0.000 Billboard 103 0.340 0.476 Television 103 0.107 0.310 Health agent 103 0.155 0.364 Public announcement 103 0.078 0.269 At the interview 103 0.126 0.334 Approved the list of beneficiaries Public administration 260 0.054 0.226 Council 260 0.308 0.462 Community 260 0.035 0.183 Brasilia 260 0.735 0.442 The list was divulged to the public 260 0.842 0.365 Meetings 219 0.196 0.398 Newspaper 219 0.283 0.452 Radio 219 0.438 0.497 Internet 219 0.041 0.199 Billboard 219 0.767 0.424 Notes to Table 22: The respondent for these questions was the Bolsa Familia coordinator. 5.3. Monitoring and Enforcement of Conditionalities Another important policy change under Bolsa Familia was the unification of education, health and nutrition conditionalities from the four pre-reform programs (mainly from Bolsa Escola and Bolsa Alimentação). With this integration, Bolsa Familia imposes three requirements on program participants. First, every member of the household between 6-15 years old must maintain at least 85 percent daily school attendance (which implicitly requires enrollment). 39 Second, children under the age of seven must seek health care with growth monitoring and vaccinations up to date. Third, pregnant women are required to seek prenatal care. Our municipal survey suggests that just 9 percent of the municipalities sampled imposed additional conditionalities (see Table 23), such as requiring that the parents continue their education or provide receipts for items purchased with the money (this latter is likely reminiscent of one of the pre-reform programs, the Cartão Alimentação under Fome Zero, which temporarily required proof of food purchases). Table 23. Additional aspects of the conditions imposed by the Bolsa Familia program Number of Standard observations Mean deviation Impose other conditional than those required by the program 260 0.085 0.279 Advised everyone about the conditionalities 261 0.713 0.453 Advised about the conditionalities Meetings 236 0.602 0.491 Home visits 236 0.419 0.495 Health agents 236 0.076 0.266 Time of interview 236 0.106 0.308 School 236 0.064 0.244 Radio 236 0.233 0.424 Always accompany the conditionalities 261 0.613 0.488 Reasons for not always monitoring the conditionalities Too much work 101 0.347 0.478 The families depend on the program 100 0.270 0.446 Not necessary 100 0.250 0.435 Federal government doesn't care 100 0.650 0.479 Who monitors the conditionalities Secretary of education 231 0.494 0.501 Secretary of health 230 0.452 0.499 Secretary of social action 230 0.687 0.465 Secretary of culture 230 0.052 0.223 Bolsa Familia Comite 230 0.113 0.317 What happens if the family does not comply Does not receive the transfer 225 0.342 0.476 Does not receive a component 225 0.244 0.431 Cut from the program 225 0.320 0.468 Receives a visit from the administration 225 0.418 0.494 Nothing if they can justify it 225 0.613 0.488 Public denouncement of the program Did not receive the payment 261 0.843 0.365 Mistargeting Type 1 261 0.713 0.453 Political targeting 261 0.230 0.422 Mistargeting Type 2 261 0.854 0.353 Notes to Table 23: The respondent for these questions was the Bolsa Familia coordinator. At the time the survey was carried out (October – December 2004), MDS was just clarifying the policy regarding the monitoring and verification of conditionalities. In fact, with the transition of the integration of the four programs, formal reporting of conditionalities 40 compliance to the central government agencies (Ministries of Education and Health) had been temporarily suspended in the early months of 2004. By October 2004, following the issuance of the regulatory decree for Bolsa Familia in September, the Ministry of Education reinitiated the collection of information from municipalities regarding compliance with education conditionalities for Bolsa recipients. The first reporting period under Bolsa Familia covered the period from October-December 2004, the same months as our survey. During that first reporting period, just 70% percent of municipalities reported their attendance records for Bolsa beneficiaries.24 This corresponds closely to the share of municipalities in our sample that reported that they monitor conditionalities regularly (61 percent, see Table 23). This is lower than the share reported in our sample for Bolsa Escola. However, it is important to note that there is a significant difference between monitoring conditionalities at the local level and transmitting this information to the Ministry of Education for incorporation into decisions regarding penalties under the program (such decisions are taken by MDS after MEC compiles the information). In fact, the highest share of schools ever to report compliance information to the federal government (via municipalities) was 19% under Bolsa Escola – as compared with 79% of schools reporting information (95% of municipalities) for the reporting period of February-April 2005 under Bolsa Familia.25 Thus it seems that monitoring of conditionalities did recover after an initial temporary lull under Bolsa Familia, and is now higher than previously under Bolsa Escola (which may have had a large share of municipalities monitoring the conditionalities, but this information was not passed on to MEC for more than 19% of schools). According to our municipal survey, the municipalities that do not always monitor beneficiary compliance with conditionalities cite several reasons for not doing so such as that it is too much work (35 percent), or that the family depends on the program (27 percent), or that they lack infrastructure for monitoring (not shown in table). Sixty-five percent of these municipalities also cited the Federal government’s indifference for why they do not always monitor the conditionalities of the program – which likely reflects the temporary suspension of enforcement by the Federal Government in early 2004 (later reversed in October 2004). VI. Concluding Remarks In 2001, the Brazilian government initiated Bolsa Escola as a nationwide education program designed to improve the schooling of children from poor households. The program provided cash transfers to mothers of poor children, conditional on their children’s continued attendance in school. An important feature of this program, and of the current Bolsa Familia program, is that its implementation was devolved to the municipalities, thus offering an ideal case study to explore differences in the targeting, monitoring and enforcement, and accountability strategies of conditional cash transfer programs when the service provider is a local government. In this report, we document significant variation in how municipalities have implemented the Bolsa Escola program in 261 municipalities randomly selected across four states of the Northeast. Despite considerable variation across municipalities, several consistent patterns emerged from the data. 24 Source: Ministry of Education. 25 Source: Ministry of Education. 41 The general findings of the study are that: (a) there is considerable heterogeneity in implementation quality and strategies by municipality; and (b) contextual factors – including local governance and politics – affect implementation. More specifically, the following six findings stand out: • There was considerable variation across municipalities in the processes used to register potential beneficiaries for both the Bolsa Escola and the Bolsa Familia programs. Cost considerations and political/governance variables were important correlates of the way registration was implemented. • There was widespread confusion concerning the municipality’s role in beneficiary selection for the Bolsa Escola Program. A majority of municipalities understood that such decisions were made not by them but by the Federal Government in Brasilia. Political variables, such as the practice of clientelism and patronage, also affected municipal beneficiary selection decisions. On the other hand, social variables such as literacy (voice) and social councils (accountability) clearly also affected the selection process. Our survey results show that confusion was reduced under Bolsa Familia after the issuance of a regulatory decree that clarified that responsibility for beneficiary selection rests with the federal government (MDS). • There was considerable transparency with respect to the beneficiary identification and selection process, with ample dissemination, public knowledge, and information on the criteria used. • With respect to the monitoring and enforcement of conditionalities under Bolsa Escola, we find that (a) a significant share of municipalities imposed additional conditionalities (beyond the federal requirements) on beneficiaries; (b) there was significant variation in the monitoring and enforcement of conditionalities; and (c) economic and political factors seemed to influence the degree to which these processes were implemented. It is important to note, however, that there is a significant difference between municipalities indicating that they monitor conditionalities and their forwarding of such information to the federal government. In fact, federal statistics from the Ministry of Education show that reporting of conditionalities compliance data by municipalities to the federal government increased substantially under Bolsa Familia (from 19% of schools reporting under Bolsa Escola to 79% under Bolsa Familia in early 2005). • Our municipal survey suggests considerable variation in the existence and effectiveness of social control councils, the instrument designed to insure a short route to downward accountability. Specifically, we find that: (a) although most municipalities operated such councils, about a fifth of them did not establish them at all, despite the federal requirement to do so; (b) in municipalities where social councils did exist, there was a positive impact on the quality of implementation; but (c) even when they did exist, they did not necessarily function properly and their membership seems to have been selected with predominant support for mayors. • Finally, we found encouraging evidence that the longer route to downward accountability via electoral rewards does work. Incumbent mayors gained from the program: their likelihood of re-election increased with the share of school children covered by the program, even though this quota was determined by Brasilia, with no room for influence by them. In spite of this, they were perceived by their local constituency as effective 42 intermediaries in bringing program benefits to the community. Mayors who put into place a social control council also reaped electoral rewards, with a 47 percent higher chance of being re-elected. Finally, we find that perceptions of targeting accuracy do seem to have political dividends, but only if it minimized “errors of inclusion” (leakages to families perceived to be undeserving or non-poor). However, perceived errors of exclusion – poor families that should benefit but do not – did not generate political costs for incumbent mayors. These results give strong evidence that electoral rewards and punishments are an effective instrument for downward social accountability. However, because they take longer to activate (i.e., the length of the political cycle, four years in this case), they are not a substitute for more immediate social accountability through effective social control councils, which could be strengthened. 43 References Abramovay, Miriam, Carla Andrade, and Julio Jacobo Waiselfisz. 1998. Bolsa Escola: melhoria educacional e redução da pobreza. UNESCO. Brusco, Valeria, Marcelo Nazareno, and Susan C. Stokes. (2002). “Clientelism and Democracy: Evidence from Argentina.” Conference on Political Parties and Legislative Organization in Parliamentary and Presidential Regimes. Yale University. Coady, David, Margaret Grosh, and John Haddinott. 2003. “The Targeting of Transfers in Developing Countries: Review of Experiences and Lessons.” Washington D.C.: International Food Policy Research Institute, Working Paper. de la Brière, Bénédicte, and Kathy Lindert. 2003. “Brazil’s Cadastro Único”. Country Case Report prepared for the World Bank and DIFID. Transparencia International of Brazil (2005). www.transparencia.org.br World Bank. 2001. “Brazil: An Assessment of the Bolsa Escola Program.” The World Bank: Workshop on Conditional Cash Transfer Programs (CCTs): Operational Experiences. World Development Report. 2004. Making Service Work for Poor People. Washington D.C.: The World Bank. 44 Table A1. Summary statistics of regression covariates Number of Standard observations Mean deviation Mayor characteristics Age 260 48.323 9.815 Education 261 6.326 2.104 Gender (male=1) 261 0.916 0.278 Second-term 260 0.577 0.495 Political experience 261 2.515 1.176 Member of an elite family 261 0.816 0.388 Municipal Characteristics Population density (Persons/km) 261 1.195 4.860 Number districts 260 3.054 3.355 Share of rural households 261 0.459 0.195 Share of literate population 261 0.671 0.070 Log per capita income 261 4.203 0.253 Gini 261 0.519 0.057 Share of population employed by public sector 261 0.042 0.021 Number of newspapers 260 0.385 2.017 Number of radio stations 260 1.165 1.433 Number of catholic churches 260 7.473 11.356 Proportion of councils that function 260 0.919 0.156 Judiciary district 260 0.588 0.493 Program characteristics Bolsa Escola quota 259 0.516 2.691 Received training 257 0.665 0.473 Political Characteristics Patronage 261 13.798 16.187 Medium clientelism 261 0.544 0.499 High clientelism 261 0.284 0.452 Number of legislators 260 11.562 3.688 Share of legislator that opposes the mayor 260 0.354 0.181 Share of secretaries related to a legislator 251 0.098 0.183 Number of jobs a legislator can appoint 261 0.540 1.538 Notes to Table A1: The variables measuring political experience, member of an elite family, received training, and all the political characteristics are taken from the survey. All other variables are from secondary data. 45 This study analyzes the role of local governance in the implementation of Bolsa Escola, a decentralized conditional cash transfer program for child education in Brazil. It is based on a survey of 260 municipalities in four states of the Northeast. The analysis focuses on program implementation. Results show that there was considerable confusion over the municipality’s role in beneficiary selection and consequently much heterogeneity in implementation across municipalities. Social Control Councils as direct accountability mechanisms were often not in place and poorly informed, weakening their role. However, electoral support for incumbent mayors rewarded larger program coverage, presence of Councils, and low leakages of benefits to the non-poor. HUMAN DEVELOPMENT NETWORK About this series... The World Bank Social Safety Nets Primer is intended to provide a practical resource for those engaged in the design and implementation of safety net programs around the world. Readers will find information on good practices for a variety of types of interventions, country contexts, themes and target groups, as well as current thinking of specialists and practitioners on the role of social safety nets in the broader development agenda. Primer papers are designed to reflect a high standard of quality as well as a degree of consensus among the World Bank safety nets team and general practitioners on good practice and policy. Primer topics are initially reviewed by a steering committee composed of both World Bank and outside specialists, and draft papers are subject to peer review for quality control. Yet the format of the series is flexible enough to reflect important developments in the field in a timely fashion. The primer series contributes to the teaching materials covered in the annual Social Safety Nets course offered in Washington DC as well as various other Bank-sponsored courses. The Social Safety Nets Primer and the annual course are jointly supported by the Social Protection unit of the Human Development Network and by the World Bank Institute. The World Bank Institute also offers customized regional courses through Distance Learning on a regular basis. For more information on the primer paper series and papers on other safety nets topics, please contact the Social Protection Advisory Service; telephone (202) 458-5267; fax (202) 614-0471; email: socialprotection@worldbank.org. Copies of related safety nets papers, including the Social Safety Nets Primer series, are available in electronic form at www.worldbank.org/safetynets. The website also contains translated versions of the papers as they become available. An ambitious translation plan is underway (especially for Spanish and French, some in Russian). For more information about WBI courses on social safety nets, please visit the website www.worldbank.org/wbi/socialsafetynets.