Report No. 24413-BR Brazil Musnicipal Education Resources, Incentives, and Results (In Two Volumes) Volume II: Research Report December 20, 2002 Brazil Country Management Unit Latin America and the Caribbean Region Document of the World Bank E. CuCrTency Equivaiiexts Currency Unit: Dominican Peso (DR$) Exchange Rate US$1.00 = DR$16.35 (as of March 22, 2000) 2. FfrCRs Year January I - Dec 31 O Acrornyms sad AbbrevWations CB Central Bank CEA State Sugar Council C DE National Electricity Corporation FEyD Fundacion Economia y Desarollo FTZ Free Trade Zones GDP Gross Domestic Product INESPRE Price Stabilization Institute LAC Latin America and the Caribbean ONAPLAN National Planning Office ONAPRE National Budget Office SPI Banking Sector Superintendency WTO World Trade Organization Vice President David de Ferranti Country Director Orsalia Kalantzopoulos Sector Director Guillermo Perry Task Manager John Panzer Acknowledgements This report was prepared by a team consisting of Suhas D. Parandekar (Task Manager, LCSHE), Emanuela di Gropello (Human Development Economist, LCSHE), and William Dillinger (Lead Public Sector Management Specialist, ECSPE). The report was written under the general supervision of Vinod Thomas (Director, LCC5C - Brazil Country Management Unit), Joachim von Amsberg (Lead Economist, LCC5C) and Indernit Gill (Lead Economist, LCSHD). The peer reviewers were Shanta Devarajan (Chief Economist, HDNVP), Elizabeth King (Lead Economist, DECRG) and Peter Moock (Lead Economist, EASHD). Claudia Willemsens Kastrup and Maria Paula Savanti at HQ provided research assistance, Veronica Yolanda Jarrin was the administrative assistant, and back-up was provided by Fabiana Rezende Imperatriz from the RM Brasilia. The team is grateful to many persons in the government and research community in Brazil and from the Bank's Human Development team for Brazil. Valuable advice, guidance, and data were provided by the following individuals:- National Fund for Educational Development (FNDE): M6nica Messenberg Guimaraes, Mauricio Antonio do Amarol Carvalho, Vinicius de Lara ; Secretariat of Fundamental Education (SEF): Ulysses Cidade Semeghini and Valentina Pereira Buainain; Institute of Applied Economic Research (IPEA): Jorge Abrahao de Castro and Jose Aparecido; National Union of Municipal Secretaries of Education (UNDIME): Francisco Potiguara; National Institute of Educational Research (INEP): Joao Batista Gomes Neto, Cai los Eduardo Moreno Sampaio and Liliane Brant; Foundation for Administrative Development (FUNDAP): Vera Cabral Costa, Monica Maia Bonel Maluf, Fernando Ortega, and Eduardo Fagnani; Fundaao Joaquim Nabuco: Osmil Galindo, Ivone Aquino de Medeiros and Paulo Ferraz Guimaraes; National Association of Educational Policy and Administration (ANPAE): Romualdo Portela de Oliviera; Faculty of Economics, University of Sao Paulo: Prof. Jose Alfonso Mazzon; Faculty of Economics, State University of Campinas: Rinaldo Fonseca; UNDP Brasilia Office: Maristela Marques. Thanks are also due to the numerous people in State and Municipal Secretariats of Education who provided their time for the team. In the Bank, other than those cited above, advice and support was provided by members of the Brazil Team and the Regional HD Team - Alberto Rodriguez, Andrea Guedes, Dorte Vemer, Madalena dos Santos (HD Sector Leader), Marito Garcia (Sector Manager), Robin Horn and Ricardo Silveira. This report has been prepared in consultation with the Ministry of Education (MOE) of the Government of Brazil (GOB) and has benefited from the comments of the MOE. The views expressed in this report are solely those of the World Bank. Abbreviations and Acronyms ANPAE National Association of Educational Policy and Administration (Associa,do Nacional de Polftica e Administra,co da Educa,cao) BNDES National Bank for Economic and Social Development(Banco Nacional de Desenvolvimento Economico e Social) FIPE Foundation of,the Economic Research Institute, University of Sao Paulo (Funda9do Instituto de Pesquisas Economicas, Universidade de Sdo Paulo) FNDE National Education Development Fund (Fundo Nacional de Desenvolvimento da Educa ,do) FPE State Participation Fund (Fundo de Participa,cdo dos Estados) FPEX Fund for the Compensation of Exports (Fundo de Exporta,do) FPM Municipal Participation Fund (Fundo de Participa,do dos Municipios) FUNDAP Foundation for Administrative Development (Funda,do do Desenvolvimento Administrativo) FUNDEF Fund for the Development of Primary Education and Valuing of Teachers (Fundo de Manuten,do e Desenvolvimento do Ensino Fundamental e de Valoriza,ao do Magisterio) FUNDESCOLA Fund for Strengthening of Schools (Fundo de Fortalecimento da Escola) IBGE Brazilian Institute of Geography and Statistics (Instituto Brasileiro de Geografia e Estatistica) ICMS State Tax on the Circulation of Goods and the Rendering of Interstate and Intermunicipal Transportation and Communication Services (Imposto sobre opera,coes relativas a Circula,ido de Mercadorias e sobre presta,ces de servi,os de transporte interestadual e intermunicipal e de comunica,do) INEP National Institute for Educational Studies and Research (Instituto Nacional de Estudos e Pesquisas Educacionais) IOF Federal Tax on Financial Operations (Imposto sobre Opera,5es Financeiras) IPEA Institute of Applied Economic Research (Instituto de Pesquisa Econ8mica Aplicada) IPI Federal Tax on Industrialized Products (Imposto sobre Produtos Industrializados) IPI-Ex Federal Tax on Industrialized Products relative to Exports (Imposto sobre Produtos Industrializados, proporcional as exporta,ces) IPTU Municipal Tax on Urban Land and Territorial Property (Imposto sobre a Propriedade Predial e Territorial Urbana) IPVA State Tax on Ownership of Motor Vehicles (Imposto sobre a Propriedade de Vehiculos Automotores) IR Federal Tax on Income and Profits of any nature (Imposto de Renda) IRRF Withholding Income Tax (Imposto de Renda Retido na Fonte) ISS Municipal Tax on Services of any Nature (Imposto sobre Servi,os de qualquer natureza) ITBI Municipal Tax on Real Estate Transfers (Imposto sobre a Transmissao de Bens Imoveis e de Direitos a eles Relativos) ITR Federal Tax on Rural Territorial Property (Imposto sobre a Propriedade iv Territorial Rural) LC 87 Complementary Law 87 (Lei Complementar #87/86 - trata dos recursos relativos a desonera,do de exporta,ces de produtos primdrios) LDB The National Law of Education (Lei de Diretrizes e Bases da Educafdo Nacional) LRF Fiscal Responsability Law (Lei de Responsabilidade Fiscal) MDE Development & Maintenace of Educational Systems (Manutengdo e Desenvolvimento do Ensino) MEC Ministry of Education (Ministerio da Educa,cao) PDDE Money to Schools Program (Programa Dinheiro Direto na Escola) PMAT Program to Modernize Municipal Revenue Administration (Programa de Modemiza,cao da Administra,do Fiscal e da Gestdo dos Setores Sociais Bdsicos) PNAFEM Program to Modernize Local Revenue Administration and Social Sectors (Programa de Apoio e Gestao Administrativa e Fiscal dos Municipios Brasileiros) SE Education Payroll Tax (Saldrio Educa,do) SEADE State System Foundation for Data Analysis (Funda,co Sistema Estadual de Andlise de Dados) SIAFI Integraded Financial Management System (Sistema Integrado de Administragdo Financeira do Governo Federal) SMU Sector Management Unit (Unidade de Gestdo Setorial) STN Secretariat of the National Treasury (Secretaria do Tesouro Nacional) UNDIME Union of Municipal Directors of Education (Unido dos Dirigentes Municipais de Educagco do Estado do Rio de Janeiro) v Table of Contents Part I - Brazil Municipal Education: Resources, Incentives and Results. Acknowlegments .............. ............... iii Abbreviations and Acronyms .iv Preface .vii Section 1: Introduction. 1 A. The Policy Context .I B. Structure of the Education System. 3 C. Financing of Education. 5 D. Scope of the Study .11 E. Data Sources, Methodology and Audience for the Study .14 Section 2: Additional Resources to Municipalities for Education .16 A. Evolution of Educational Expenditures 1995-2000 .16 B. Redistribution of Resources from State Govs.to Municipal Govs .26 C. Redistribution of Resources among Municipal Govemments .29 D. Fungibility of Municipal Resources and FUNDEF .32 E. FIJNDEF and impact on Pre-School Education .35 F. Further Issues regarding Additionality .37 Section 3: Management of Resources: Composition of Expenditures .40 A. Overview .40 B. Capital Expenditures .41 C. Recurrent Expenditures .50 D. Transportation Expenditures .59 Section 4: Management of Resources: Teacher Remuneration and Career Pogress .62 A. Teacher Remuneration .62 B. Number of Teachers .65 C. Expenditure on Training of Teachers .66 D. Career Progress of Teachers .67 Section 5: Impact of Resources: Educational Results .70 A. Introduction and Variable Description .70 B. Econometric Results: Estimation and Specification Issues .73 C. Main Results: Impact of Expenditures .79 D. Main Results: Composition of Expenditure .83 E. Main Results: Degree of Municipalization .86 vi Part 11 - Appendix Appendix for Section I ......................................................... 90 Table 1.1 Distribution of Federal Government Education Expenditure, 2001..... 90 Table 1.2: Distribution of State Government Education Expenditure, 2000 ........ 90 Appendix for Section 2 ............................................................. 91 Table 2.1: Evolution of Education Expenditures ....................................... 91 Table 2.2: Distribution of Education Expenditures by Level of Government ...... 91 Table 2.3: Representativeness of the STN Sample .................................... 91 Source of Table 2.4: (A) Regression of Log Change in Total Revenues ................ 92 Source of Table 2.4: (B) Regression of Log Change in Own Revenues ................. 93 Source of Table 2.4: (C) Regression of Log Change in Transfer Revenues ............ 94 Source of Table 2.4: (D) Regression of Log Change in Total Expenditures ............ 95 Source of Table 4.4: (E) Regression of Log Change in Education Expenditures ...... 96 Table 2.14: Simulation of Change in Per Student Floor for FUNDEF (Year 2000) 97 Appendix for Section 3 ......................................................... 98 Table 3.1: Primary Education Expenditures,1999 .98 Table 3.2.a: Municipal Primary Education Expenditures per Region, 1999 .99 Table 3.2.b: State Primary Education Expenditures per Region, 1999 .100 Table 3.3: Primary education Expenditures per Municipal Size, 1999 .101 Table 3.4: FUNDEF Education Expenditures, 1999 .102 Table 3.5: FUNDEF Expenditures per Region, 1999 .103 Appendix for Section 5 ........................................ 104 Table 5.1: Variable description and data source .104 Table 5.2: Some summary statistics .108 Table 5.3: OLS Estimates- Dependent Variables: AGEGRADIS14; DROPOUT14; PASSRATE4 .109 Table 5.4: OLS Estimates- Dependent Variables: AGEGRADIS58; DROPOUT58; PASSRATE58 .110 Table 5.5: OLS Estimates- Dependent Variables: AGEGRADIS 14; DROPOUT14; PASSRATE14 .111 Table 5.6: OLS Estimates- Dependent Variables: AGEGRADIS58; DROPOUT58; PASSRATE58 .112 Table 5.7: OLS Estimates- Dependent Variables: AGEGRADIS14;DROPOUT14 PASSRATE14 (all municipalities excluding North/Northeast) 113 Table 5.8: OLS Estimates- Dependent Variables: AGEGRADIS14; DROPOUT14; PASSRATE14 (Northeast municipalities) .114 Table 5.9: OLS Estimates- Dependent variable: MENROLSH14 .114 Table 5.10: OLS Estimates- Dependent variable: MENROLSH 58 .114 vii Preface This document is the second volume of a two-volume report that deals with the use of public resources to provide educational services in Brazilian municipalities. This volume is the research report and the companion volume is a policy report which presents the key findings and the related policy conclusions. This volume has 5 sections, in addition to a set of Annexes. Section 1 provides an introduction of the policy context and outlines the objectives and methodology of the study. Section 2 deals with the first major theme of the study - the extent to which municipalities in Brazil benefited from efforts designed to provide additional resources for education. The period from 1996 to 2002 has seen substantial redirection in resources from State Governments to Municipal Governments, and from prosperous municipalities to relatively less prosperous ones. This section describes and analyzes the patterns of the changes in the availability of resources. Section 3 discusses and analyzes the composition of expenditures and deals with issues surrounding the municipal management of resources. Section 4 deals with the specific issues around teachers related to municipal management of resources. Finally, Section 5 presents a detailed econometric analysis of the educational impact of the additional resources and the management shift of resources from States to Municipalities. The educational impact is measured in terms of changes in indicators of internal efficiency of municipal education systems. The policy recommendations that follow from the research report are presented in Volume I. viii Section 1: Introduction A. The Policy Context 1.1 The main feature that characterized primary education in Brazil in the beginning of the 1990s was the inequality of educational provision, reflecting and reinforcing the high level of inequality in living standards. Brazil was committed to the goal of providing a quality education for all, but there were still a number of children in the age group of 7-14 years, who were not attending school. Especially in the North East and Northern regions of Brazil, even the children who did attend school received an education of uncertain quality, and the system was plagued by high rates of absenteeism, repetition and dropout. Financing of education in Brazil was a combination of historic trends and negotiated political agreements, with scant attention paid to the educational needs of the population. In a number of Brazilian states, municipalities with sizeable student enrollments had insignificant revenue streams, at the same time as states with huge student populations had failed in attempts to devolve the responsibility for primary education to the municipal level. The educational financing system was not creating accountability and was not providing incentives in order to resolve the problems of access, quality and equity. 1.2 It was clear that mere tinkering with policy alternatives would not lead to an effective and long-lasting solution to the problems confronting education in Brazil. Brazil was in need of a systematic and ambitious reform program and the federal government instituted just such a wide-ranging reform in the second half of the 1990s. The Lei de Diretrizes e Bases da EducaVao Nacional (LDB), approved in 1996, clearly laid out the roles and responsibilities of various levels of government to provide quality education. The federal government was assigned the lead role in national policy formulation and guaranteeing equity and quality assurance. Accompanying the new LDB, standards were set in place for the school curriculum and for teacher qualification. The National Institute for Educational Research and Studies (INEP) was made responsible for the creation and production of educational statistics and student assessment. State and municipal govemments were to cooperate in the provision of primary education (Grades 1-8). Municipal govemments were further assigned the priority for pre-school education and state governments with secondary education.' 1.3 The reforms were accompanied by a marked increase in the amount of educational expenditures. Our estimates indicate that educational expenditures in Brazil increased from 4.2% of GDP in 1995 to 5.6% of GDP in 2000, an increase of R$ 20 billion in real terms. The reforms also established a new legal and management l The topic of secondary education was the subject of a joint Word Bank-IDB report called "Secondary Education in Brazil: Time to Move Forward", March 30, 2000, Report No.19409-BR. Higher education is discussed in detail in a recently produced Bank study "Brazil: Higher Education Sector Study", June 30, 2000, Report No. 19392-BR. Brazil's education reforms were also the subject of another piece of sector work titled "Brazil: Teachers Development and Incentives: A Strategic Framework", February 2001, Report No. 20408-BR. framework for the use of public resources aimed at improvement in the delivery of the quantity and quality of educational services. A substantial portion of public resources for sub-national levels began to be distributed on the basis of student enrollment, which stimulated enrollments at the municipal level, with particularly strong impacts in the North East. Teacher remuneration was increased significantly under the new financing regime, and measures were instituted to enhance civic participation in the use of public resources. The combination of increase in resources and incentive mechanisms regarding the use of the resources have led to an improvement in the quality of educational inputs and in the quantity of educational services. It is yet too early to expect impact of the reform on educational quality as measured by test scores, but the evidence suggests that educational quality has improved as measured by indicators of internal efficiency such as repetition rates and percentage of over-age children. 1.4 At the time of heightened skepticism about the effectiveness of additional public expenditures in social areas such as education, the case of Brazil provides an illuminating example of public policy that works.2 In the particular case of municipal education in Brazil, there is compelling evidence of how ineffective governance can be transformed, though much remains on the agenda to extend good performance to all municipalities. The ongoing change from policy based on political interests often at odds with good governance, to effective public service delivery of municipal education forms the core subject matter of this report. The provision of resources, the incentives in place regarding the use of the resources, and the results obtained from additional resources and new incentive structures form the basic framework of this report. However, before we turn to an in-depth explanation of each of these three issues, it is useful to outline the basic structure of the educational system in Brazil and the nature of financing of education for Brazilian municipalities. 2 A thoughtful analysis of the process of effective policy design and implementation in the Northeastern state of Ceara is provided in the book appropriately titled "Good Government in the Tropics" by Judith Tendler, JHU Press, Baltimore, 1997. 2 B. Structure of the Education System 1.5 The education system in Brazil up to the level of secondary education is best explained by looking at Table 1.1 below:- Table 1.i: Structure of theBrazilianhEducatiohal System l' Age Grade Term Used in Brazil English Translation 0-3 Creche Creche 4-6 Pr6-escola Pre-School 7 1 Ensino Fundamental Prmary or Basic 8 2 Sene 1-4 Education 9 3 10 4 11 5 Ensino Fundamental 12 6 Sene 5-8 13 7 14 8 _ _ _ _ _ _ _ _ _ 15 9 Ensino Medio Secondary Education 16 10 17 11 1.6 Under the Brazilian constitution, municipalities are autonomous sub-national entities, i.e., municipalities are not hierarchically subservient to the states to which they belong. The LDB defines priorities for each level of government - municipalities have the priority for pre-school education and for primary education. It is only after a municipality can ensure that adequate service is available for these two levels that the municipality can provide other levels of education. The LDB also provided a clear delineation within the pre-school level, between creches for children 0-3 years old, and pre-school education for children from ages of 4 to 6 years. Creches have traditionally been mainly from the private sector, with a small public presence in the municipality, under secretariats of Social Assistance rather than Education. Since 1996, creches have been progressively transferred to be part of the educational system. Pre-school education is tied to primary education, and is most often provided in the same educational establishment as primary education. The LDB assigned state governments with the priority for primary and secondary education. As primary education is a shared responsibility between the state and the municipality, there is ground for both co-operation and for conflict, an issue that is discussed at length in this report. Higher education is the responsibility of the federal government, though some state governments also provide higher education services. 1.7 Table 1.2 indicates the distribution of enrollment across the levels of government, and also includes private provision, which is particularly important at the creche and pre-school levels. The table shows that in accordance with the assignment of responsibilities across the federative levels of government, public enrollment in pre- school education is mainly municipal, and that state governments are the principal public providers for secondary education. With states accounting for 97% of public secondary enrollment, and municipalities accounting for 93% of public enrollment 3 prior to primary education, there is not much scope for inter-state variation. However, there is high variance within the level of primary education, with municipalities predominating in the North East and in the state of Rio de Janeiro, and states predominating on other regions. Municipal governments also command a greater share in the enrollment of students in Grade 1-4 as compared to Grades 5-8. Table 1.2: Enrollment Across Levels of Governmenit in 2001 Millions of Students Municipal State Private Total Early Education Creches 0.7 0.0 0.4 1.1 Pre-School 3.3 0.3 1.2 4.8 Literacy Classes 0.4 0.0 0.2 0.7 Basic Education _ Grades 1-4 12.5 5.6 1.7 19.8 Grades 5-8 4.7 9.4 1.5 15.6 Secondary Education Grades 9 -11 0.2 7.0 1.1 8.3 Adult Education _ _ All Levels 1.4 2.0 0.4 3.8 Source: INEP / School Census 2001 4 C. Financing of Education 1.8 The framework for education financing in Brazil is determined by the Federal Constitution of 1988, with important amendments in 1996. Article 212 of the Constitution of Brazil, established in the year 1988, specifies that the federal government is obliged to spend a minimum of 18% of tax revenues (not including social security contributions) on educational expenditures, and states and municipalities are obliged to spend a minimum of 25% of their tax and transfer revenues on education. Tax collection is distributed across the three levels of government in Brazil, with non-discretionary, formula based transfers to determine the allocation of transfers to the sub-national entities.3 The main tax revenues of the federal government are the Income Tax (IR) and the Tax on Industrial Products (IPI). The federal government is the residual claimant of these revenue streams after transfers to state governments (FPE), and to municipal governments (FPM). The federal transfers have strong re-distributive elements, with the bulk of the FPE going to states in the underdeveloped regions of the country, and with the FPM favoring small municipalities away from the state capitals. The main source of state revenues is the ICMS or Value Added Tax on Goods and Services. States, in turn share part of the ICMS revenues with municipalities. Municipal own revenues come mainly from a Property Tax (IPTU) and from a tax on services (ISS). Another important public resource for education in Brazil is an earmarked 2.5% payroll tax called the Saldrio-Educaqdo (SE). One-third of the collection from SE accrues to the National Education Development Fund (FNDE), which funds the federal government activities such as the School Feeding and Textbook distribution programs. Two-thirds of the SE collection accrues to state governments, some of which share the proceeds with municipalities. 1.9 This study is dedicated to an analysis of municipal education expenditures, but it is important to place municipal spending in the context of spending by other levels of government. Education expenditures in Brazil are predominantly sub-national. The federal government spends the biggest portion (more than 60%) of its educational budget on higher education, but the federal government funds important programs towards improving equity and quality for other levels of education. Appendix Table 1.1 shows the composition of federal govemment expenditures in 2001. The federal government spent approximately R$ 1.6 billion on programs to improve the quality of primary education, and a further R$1.9 billion on equity related programs. The magnitude of approximately R$ 3.5 billion of federal investments in pre-school and primary education pales when compared to the over R$ 30 billion spent by state and 3 The structure of Brazilian fiscal federalism is the subject of a huge literature, including sector work done by the World Bank. An overview of trends regarding tax and expenditure assignment across levels of government is provided in "Fiscal Decentralization and Sub-national Fiscal Autonomy in Brazil: Some facts of the Nineties", by M6nica Mora and Ricardo Varsano, IPEA Working Paper, Rio de Janeiro, December 2001. This and a range of other papers are available in the Data Bank prepared by BNDES. This information source has been specifically set up to provide statistics, details about the legislation as well as news and academic research and policy papers. The website is hrtp://federativo.bndes.gov.br. 5 municipal governments on primary education alone. However, the role of the federal government goes beyond the effect of its own expenditures, because of the leverage afforded to it by the constitutional responsibility to frame national educational policy and to guarantee equity and quality. 1.10 The federal government spends substantial resources on equity enhancing programs in municipalities, as well as on programs that seek to improve the quality of inputs for fundamental education. Amongst the various federal programs, the biggest is the school-feeding program, for which the federal government spent upwards of R$1 billion in 2001. Another critical federal program is the provision of textbooks. Schools are provided textbooks directly by the federal government - schools have a choice of options regarding the particular textbooks that they would like to use for their students, and the federal government mails the textbooks to the schools on the basis of the enrollment in each school.5 The federal government spends approximately R$0.5 billion on a cash transfer program to enhance demand for schooling by poor families, and another R$0.5 billion as complementary financing to the sub-national fund for primary education. In addition to these flagship programs, the federal government spends resources on a range of other programs that provide resources directly to schools and parents or to sub-national governments. 1.11 In accordance with the mandate to the federal government to inform and guide policy regarding all levels of education, it incurred an expenditure of R$ 152 million on evaluation, statistics and related activities. This amount does not include various grants provided to federal universities and other institutes of higher learning for research activities in the field of education policy. The federal expenditures on educational research contain an important lesson regarding the role of the center in a federal framework. The relative expenditures on statistics and research may be small in relative terms, but the magnitude belies the critical importance of these expenditures. The success in the use of objective methods for distribution of public resources begins with accurate and reliable educational statistics. The possibility of accessing educational statistics in Brazil today is the result of a persistent policy choice to invest in these activities. Moving beyond the availability of data to the analytical use of data to improve the efficiency and efficacy of public education expenditures is an important item that remains on the agenda. 1.12 State government expenditures are important to our analysis because state governments share resources for primary education with municipalities. The disposable revenue for state governments consists of own revenues and transfers from the federal government, less transfers made by state governments to municipalities. State governments in Brazil are important collectors of tax revenue, 4 The resources are made available to municipalities and state governments on the basis of the number of students, at the rate of 13 centavos per student, for the 200 school days in a year, or R$26 per student per 5'ear. In the year 2000, a total of approximately 130 million textbooks were distributed for the approximately 35 million fundamental education students enrolled in Brazil. 6 accounting for 28% of the total tax collection in Brazil. The main source of state tax revenue is the ICMS, which accounts for 84% of the tax collection of state governments. State governments also receive transfers from the federal government (FPE and FPEX), at the same time as state governments transfer a quarter of their ICMS revenues to municipalities. State and municipal governments are the direct providers of education for pre-school education, primary education, and secondary education. Part of the state and municipal resources for education consists of the reallocation of state and municipal revenues for FIJNDEF, with an additional allocation for some states from the federal government. FUNDEF is described in greater detail later in this introductory section. Appendix Table 1.2 shows the destination of state government expenditures for education. 1.13 Municipal Revenues consist of own revenues and transfers from the federal and state governments. On average, own-source fiscal revenues account for 35% of total municipal revenues, with transfers accounting for 65%, though the situation varies with municipal size. The own sources of revenue for municipalities are taxes on personal and professional services (ISS), the urban property tax (IPTU), and the tax on real estate transfers (ITBI). The main transfer of general revenues from the federal government is based on a formula funding mechanism that distributes 22.5% of the collections from the income tax and the industrial products tax (FPM or the Municipal Participation Fund) to the municipalities. The main transfer from the state government is derived from the ICMS (Tax on Goods and Services) collection of the state. Smaller municipalities are favored in the FPM formula, and the ICMS distribution is based on origin of revenue,, that favors medium and large municipalities. Part of the ICMS revenue of state governments is also directed to municipalities in the form of FUNDEF. Table :1.3: eDistiribution ~ot Munl c ^i > ve emt:Eduatl6ov~ ExP6nditurbe, 000. -' /Rrolect ~~~~~~~E'xp--6ndit__ui9,r ___________________________~ ~ ~ ~~~~~~ec"~ R e&tages___~ Item 'of Expend itur'eroet ^. rRals -i. A. Early Education 4,042 18.1% Creches 643 2.9% Pre-School 3,031 13.6% Literacy Classes 367 1.6: B. Fundamental Education 16,631 74.5°h Grades 1-4 12,087 54.1% Grades 5-8 4,545 20.3% D. Secondary Education 307 1.4% E. Adult Education 1,354 6.1% TOTAL 22,334 100.0% Source: Own Calculations from INEP and STN data 1.14 The scenario regarding educational expenditures in Brazilian municipalities is determined greatly by the reforms to the Constitution of 1988. The LDB approved in 1996 established clear guidelines about what items could and could not be considered as educational expenditures (see Box). These guidelines are enforced at various stages of the budget formulation and budget execution process for the 7 federal government as well as sub-national governments. While the list has been instrumental in removing the worst forms of wasteful expenditures, one important item has not been resolved - the issue of pension payments for retired teachers. While the LDB legislation does not specify payments of pensions as eligible educational expenditures, neither does it specifically prohibit such expenditures. The use of educational expenditures for pension payments reduces resources for the classroom, a problem that is particularly acute in large municipalities. However, FUNDEF resources cannot be used for the payment of pensions, an interesting example of how federally imposed restrictions on local expenditures may prove to be beneficial to municipal students. Box 1.1: Definition of Educational Expenditures ELIGIBLE EXPENDITURES I. Remuneration of Teachers and other Educational Personnel II. Construction and Maintenance of School Physical Plant Ill. Goods and Services for Education IV. Statistics and Research for improving Education V. Activities essential for functioning of Education Systems VI. Scholarships for Students in Public or Private Schools VII. Amortization of debt incurred for Education VmI. Didactic Materials and School Transport NON-ELIGIBLE EXPENDITURES L Research not directly linked to improve Education II. Assistance to Sport or Cultural Institutions III. Training of Public Sector Employees IV. Social Assistance Programs for Health or Nutrition V. Construction Projects other than School Physical Plant VI. Personnel not employed directly for Education Source: Articles 70 and 71 of the LDB, as quoted by Castro, 2001 1.15 By far the most important education financing reforms introduced in Brazil were the laws that established FUNDEF (Emenda Constitucional 14/96 and Lei 9424/96). The FUNDEF financing mechanism addresses the divergence between resource needs and resource availability for sub-national governments. FUNDEF assigns a part of the 25% of resources available for education specifically for the purpose of primary education. The FUNDEF mechanisms works as follows:- FUNDEF collects resources from state and municipal governments in a single fund dedicated exclusively to primary education. Each of the 26 states in Brazil has its own FUNDEF - the FUNDEF for each state consists of 15% of tax collection from specific tax and transfer sources(The FPM and FPE, ICMS, the IPI-Ex and LC 87/96). The money collected by the FUNDEF for each state in this manner is then divided by the number of primary education students enrolled in that state in the previous academic year, whether in state or municipal systems. This unit collection is then multiplied by the number of students enrolled in each sub-national primary education system, and distributed to the state government and municipalities Every sub-national level of government makes a contribution to FUNDEF, but the 8 resources get back to the entity depending on the number of students enrolled under its respective primary education system. The federal government has also established a national floor of per student educational spending - if the FUNDEF amount for a particular state is lower than the floor, the federal government "tops up" the Fund with resources drawn from its general revenues. 1.16 FUNDEF is not exclusively a transfer from other levels of government - the municipality recovers its own tax revenues from the fund to the extent that the municipality has students, and needs those revenues. It is important to note that FUNDEF does not include the entire 25% of the four revenue sources tapped by FUNDEF, that municipalities are obligated to spend on education - each municipality also needs to spend at least 10% of the same sources on education. The municipality also is required by law to spend 25% of the revenue sources that are not tapped by FUNDEF. The Federal Constitution also mandates that a minimum of 60% of educational expenditures must be directed towards primary education, i.e., 15% of overall tax and transfer revenues. Figure 1.1 below provides a diagrammatic explanation of the flow of revenues to a municipality towards educational expenditures. The figure only shows the contribution that the municipality makes to FUNDEF. Of course, the municipality also gets back resources from FUNDEF, depending on the number of enrolled students. Figure 1.1: Explanation of Minimum Municipal Resources for Education r~ ~ ~ ~~~ 0 for Pe-Sho or Primar _LC 87 15 % to FUNDEF Tax and Transfer _ IPTU 15% for Revenues ISS Fundamental ITBI Education ITR 10%o for Pre-School or Primary IOF Education 1.17 A key characteristic of FUNDEF is the timely availability of resources to municipalities. Money accrues to the FUNDEF account for the state/municipality established with the Banco do Brasil. The periodicity of account replenishment depends on the particular revenue source, but accrual varies from a minimum of 10 days to a maximum of 30 days. There are no intermediaries involved in the 9 distribution of FUNDEF resources, and the flow of FUNDEF funds appears to be free of bottlenecks. 60% of FUNDEF resources have to be spent by the state/municipality on the payment of teachers and others directly involved in the provision of educational services. The other 40% can be spent on a variety of expenditure heads within the overall area of primary education. Even though the law establishing FUNDEF was passed in 1996, the implementation of the law did not take place until 1998. At this time, FUNDEF has been established for 4 years, and the year 2002 is the fifth year of implementation. The legislation stipulated that FUNDEF would be a program of specified duration (10 years), and would thus terminate in the year 2007. The law also established that in the first few years of the program (ending December, 2001), states and municipalities that employed teachers with very low levels of qualifications could spend FUNDEF resources on the providing of training of those teachers as part of the 60% to be devoted towards the payment of frontline educational workers. 1.18 The financing arrangement of FUNDEF has been a revolutionary way for state governments and municipal governments to cooperate with one another by sharing resources in an equitable manner. However, this introductory section on financing would be incomplete without mentioning that there is currently no mechanism at work that leverages the financial co-operation into the broader need for sub-national entities to work together to enhance quality of service delivery. In fact, quite to the contrary, state governments are prone to make the claim that as FUNDEF brings so much of resources to municipalities, they do not need to do anything more. For instance, this argument is used regarding the sharing of resources from the Salario Educacional (SE). States receive 2/3rds of the revenue collection from SE, the same source of which the federal government receives the other 1/3rd to fund textbook distribution and school lunches for the entire student population. Yet, few states share the resources with municipalities, using the argument that municipalities now have FUNDEF resources. FUNDEF represents a drastic advance from the earlier situation when municipalities were completely dependent on discretionary transfers from higher levels of government, but much remains on the policy agenda. 10 D. Scope of the Study 1.19 This study originated from a CMU/SMU discussion about the lending program for education in Brazil, in the context of the CMU's dialogue with the Brazilian Ministry of Finance. The Bank has investment programs with the Federal Ministry of Education and with State Secretariats of Education. As a significant proportion of educational services are provided by municipalities, the Bank would like to see enhanced co-operation between state and municipal governments, so that the education investment needs of municipalities are addressed, particularly in municipalities with high poverty rates. The discussion about the lending program led to the assertion from the side of the state govemnments that municipalities in Brazil are now flush with FUNDEF funds, and therefore are not in need of additional investment. The question of whether or not it is true that municipalities are flooded with money raises another important question in need of an objective answer - the question of the incentives and institutional capacity of municipalities to allocate the expenditures wisely. To the extent that municipalities now receive additional funds and spend them in accordance with educational priorities, this would ultimately be reflected in the final goal of providing quality education in an equitable manner to all Brazilians. These motivating factors lead the study to ask the following questions:- 1.20 Ouestion 1: To what extent did FUNDEF actually lead to additional resources? The policy implication of this question is to determine the magnitude of the additionality of resources and whether the additionality of the resources from FUNDEF was related to educational needs and capacities. As resources are fungible, we also seek to examine whether FUNDEF merely substituted for other resources, thus leaving overall resource availability the same as before, or whether there was some substitution of resources but not enough to wipe out all the gains for expenditures on education. Equity is always an overriding policy concern in Brazil, and we attempt to gauge the extent to which the redistribution entailed by FUNDEF has improved the equity of municipal educational resources - whether the simplicity of the FUNDEF formula leads to undesirable distortions which might be removed by introducing compensatory elements or whether the formula is working just fine as it stands. 1.21 Question 2: How do municipalities deploy the general educational resources available to them beyond FUNDEF? FUNDEF only accounts for 15% of a specified set of revenue sources for a municipality - this still leaves a minimum of 10% of those revenue sources plus the entire 25% of other revenue sources untapped by FUNDEF that the municipalities are obliged to spend on education. Municipalities also receive significant amount of in-kind transfers from the federal government (School Lunch, Transportation, Textbooks, to name but a few). At the same time, the criticism is held against FUNDEF that because of its singular focus on primary 6 A highly successful example of such programs is the FUNDESCOLA initiative, currently in its third round of financing from the Bank, with a heavy emphasis on institutional development for local governments and focusing on school improvement. education, municipalities have sacrificed the provision of other levels of education - particularly pre-school education that is the responsibility only of municipal govemments. An important policy question is whether the FUNDEF mechanism should be extended to cover other levels of education or whether it is primary education that needs greater protection as FUNDEF resources are not adequate. A related policy implication is regarding the extension of FUNDEF to other revenue sources - if FUNDEF resources are well-spent, but other resources are spent in a wasteful manner, there may be a reason to extend the FUNDEF mechanism to cover additional sources of municipal revenue. 1.22 Question 3: How do municipalities spend the money that they get from FUNDEF and from other sources? In many ways, this is the central question of the study, as it is linked with each of the three stages of analysis regarding the amount of resources, the management of resources and the production of educational outcomes. Did municipalities spend the money mainly on increasing the salaries of teachers regardless of the initial level of salaries, did municipalities establish graded salary schemes and invest in professional development of teachers, what about changes in capital expenditures and spending to improve the quality and quantity of didactic materials or compensatory programs for poor children? Is the composition of expenditures from FUNDEF different from the composition of expenditures from other municipal sources? What processes and structures are in place to determine allocations across different expenditure heads? How are resources distributed across schools within the municipality? Is there a flow of technical assistance that accompanies the resources that are redistributed from states to municipalities? Are the incentives for state and municipal govemments aligned towards enhancing the quality of educational services? 1.23 Ouestion 4: Is "municipalization" of education associated with greater community participation and higher standards of accountability? A significant portion of enrollment was already in municipal systems and the reforms have led to an increase in municipal proportion of enrollment. Does the benefit of being closer to the point of service delivery actually benefit the citizenry by improving service delivery? The law that established FUNDEF lays down norms regarding the establishments of councils to oversee the use of FUNDEF resources - the councils should have representation from school teachers and parents as well as civil society. However, there is great debate in Brazil about the efficacy of the mechanisms of social control - at one extreme, some commentators allege that the mechanisms of social control exist only on paper - once a council is constituted, it never meets again and the mayor effectively has a carte blanche to do whatever he or she likes with the money. Even the training is of little use, because the council members have two-year terms, and they are out of the door by the time that they have really begun to understand their responsibilities. On the other hand, there is also some evidence of the effective role of the councils in reducing graft and ensure proper use of resources. The federal government has also attempted to stimulate decentralization to the school level. Are municipalities in a condition to capitalize on these efforts ? 12 Are there incentive mechanisms in place for municipalities to be more accountable to the population they serve? 1.24 Question 5: What is the efficiency of educational expenditures incurred by municipalities? Different municipalities achieve different levels of educational output with the same level of inputs. These differences in efficiency are caused due to differences in institutional mechanisms and in the incentives to agents responsible for service delivery, as well as differences in costs and socio-economic conditions of students. What is the definition of the frontier of efficiency for Brazilian municipalities? How far away are municipalities from this frontier? What policy conclusions can be reached to enhance the performance of municipalities that make inefficient use of public resources. 1.25 Ouestion 6: What has been the educational impact of the recent reforms? Two levels of educational impact have been associated with the recent reforms- the first is the direct impact in terms of enrollment and teacher characteristics; the second is the impact that would be expected from additional resources and from the changed management of resources. The enrollment impact has been well documented, and we know that teacher salaries increased but we wish to know if increases in teacher salaries as well as other improvements on the input side of the equation had any educational impact. Did enrolled children actually stay enrolled and progress through the grades in a timely manner? Have the notoriously high repetition rates in Brazil being reduced and can such reduction be associated to the provision of additional resources and changed incentives for accountability? What policy conclusion can be derived from empirically established patterns regarding the results that have been obtained in association with the recent reforms? 13 E. Data Sources, Methodology, and Audience for the Study 1.26 The study uses a combination of quantitative and qualitative methodologies, drawing from primary as well as secondary data sources. The primary data was collected from a sample of 40 municipalities drawn from 5 states from different regions in Brazil, and constituted of structured in-depth interviews with municipal and state secretaries of education and their staff and representatives of municipal community associations.7 We also had more informal discussions with others such as mayors, representatives of UNDIME, and representatives of teachers' unions as well as academic researchers in Brazil who have been working in the field of municipal education financing. At the Federal level, we interviewed staff at the Federal Ministry of Education's FUNDEF unit as well as at FNDE, which provides resources for federal programs directed towards municipalities. Primary data collection was carried out with the help of accomplished Brazilian researchers at FUNDAP (Sao Paulo) and the Funda,cao Joaqim Nabuco (Recife). 1.27 The secondary data sources, available in general for all of Brazil, included: a) Data from the School Census conducted for the years 1996 and 2000; two years selected because they mark periods before and after the reforms were established, (INEP); b) Data from the Population Census of Brazil for 2000; (IBGE); c) Data regarding Municipal GDP, Geographic Area, and Human Development Index, (IBGE and IPEA); d) Data regarding expenditure composition and institutional arrangements from a Survey of 300 Brazilian municipal and state systems of education that represent 80% of the enrollment in Brazil; (FIPE); (e) Data about FUNDEF revenue collection and funds transferred to municipalities, (SEADE) and (f) Data about Federal, State and Municipal Government revenues and expenditures for various years (STN). The study team ended up using literally tens of gigabytes of data, as the subject of inquiry was vast, extending from more than 5,500 municipalities ultimately to some 200,000 schools, our final unit of analysis. 1.28 The methodology used in this report is essentially dictated by the need to base policy recommendations on empirical facts and opinions within Brazil. The approach has been to generate policy issues from the qualitative segment of the data collection effort - the discussions with the stakeholders, to raise the main questions of interest. We then try to identify the empirical data that would help make a determination about the policy recommendation. In spite of the great deal of data collected for the study, it was not possible to get the exact data for every policy issue, and we are limited to making plausible conjectures. This approach leads to some variance in the strength of the empirical foundation of the recommendations made by this report. For instance, we had high quality data about various indicators 7 The states chosen were Paraiba and Pernambuco in the North-East, Rio de Janeiro and Sao Paulo in the South East, and Parana in the South. The purposive selection of states was based on the need to get a regional spread at the same time as we wished to look at states where the Bank has an ongoing policy dialogue. Municipalities within a state were chosen with a view to generate a diversity of relevant information, as characterized by municipal population, degree of municipalization of education and whether the municipality was a gainer or loser to FUNDEF. 14 of internal efficiency to compare the universe of state and municipal schools, and we are able to use rigorous econometric techniques to judge our propositions. In the case of overall trends regarding municipal revenues we are able to use the database compiled by STN, but this database is not universal, and our analytical technique is limited to drawing simple co-relations and OLS regression. At the other extreme regarding the strength of the empirical foundation, we are aware that the political affiliation of the mayor as compared to the political affiliation of the state Governor may be an important factor influencing municipal outcomes, but we did not collect data about party affiliations, and our conclusions are based on anecdotal evidence. 1.29 This study is intended for a Brazilian audience of those who make policy (government officials and legislators) and those who influence policy (academic researchers and the media). We hope to discuss the findings and policy conclusions of this report in workshops and seminars within Brazil. The contribution of this study is targeted beyond the mere discussion of the findings. For instance, we hope that some of the methodological approaches used in the study, such as the analysis of efficiency of municipal expenditures and the use of 'positive deviant' municipalities to help bring about a change in other municipalities is actually adopted and practiced by municipalities in Brazil, possibly through the Bank's technical assistance that is associated with education sector investment loans. 1.30 The Bank audience is a secondary one for this report. The report includes descriptions of the Brazilian context that would already be familiar to most Brazilians, but not to the Bank audience. However, such information is required for readers within the Bank to be able to comment on the appropriateness of the policy recommendations to be discussed with Brazilians. Also, the Brazilian case provides a fascinating real example of the power of public policy to influence the lives of people, especially the poor who are most dependent on social services such as Public Education. With measures such as Fiscal Responsibility Law (LRF) to improve local governance, and the creation of community councils to oversee the use of public educational expenditures, just to mention two examples, there is much material for other countries to learn from the Brazilian example. 15 Section 2: Additional Resources to Municipalities for Education A. Evolution of Educational Expenditures 1995-2000 2.1 Public expenditures in education in Brazil have grown steadily over the past few years, rising from 4.2% of GDP in 1995 to grow to 5.6% of GDP in 2000, the latest year for which reliable estimates are available.8 At the same time, the weight of the different levels of government have changed, with municipal governments now accounting for nearly 38% of expenditures, compared to 27% of expenditures in 1995. The gain for municipal governments at the aggregate level for the entire country has come mainly from a reduction in the share of the federal government, whereas state governments as a consolidated group have declined slightly in their overall contribution to education expenditures. Municipalities as a group spent nearly R$ 24 billion on education in the year 2000, nearly twice of what they were spending, in real terms, in the year 1995 (Appendix Table 2.2). In other terms, of the R$20 billion increase in educational expenditures in Brazil between 1995 and 2000, R$ 12 billion accrued from municipalities, R$ 7.5 billion from state governments, and federal education expenditures increased marginally by R$ 0.5 billion. 2.2 We turn to an analysis of the sources of gains in municipal resources for education. Gains would be possible from a combination of three sources. The gains could have come from a) an increase in overall revenues for municipalities, with the percentage of resources for education remaining the same; or b) the gains could arise in the context of overall municipal resources remaining the same, with an increase in the percentage of resources devoted to education; and c) particular municipalities would have gained because of a redistribution of resources towards these municipalities. The analysis shows that even as there was an overall increase in municipal resources as shown in Appendix Table 2.2, the main phenomenon that characterizes municipal resource availability in the period since 1996 is the increase in resources to those municipalities that were earlier the most starved of resources. 2.3 An analysis of resources for municipalities is based on a dataset compiled by the Secretaria de Tesouro Nacional (STN) of the Federal Ministry of Finance. The dataset has a common series of information for 2983 municipalities, which together represent about 60% of the population of Brazil. The sample is fairly representative of all the 5,500 municipalities in Brazil in terms of size, but it under-represents the number of municipalities in the North and North East, as seen by the comparative table presented in the appendix.9 The STN data indicates that in constant prices, the amount of municipal resources for the sample increased by about 35%, from R$45 billion to R$ 60 billion (Table 2.3). Own revenues remained stable over the time 8 Appendix Table 2.1 indicates the evolution in overall education expenditures, representing a consolidation of all levels of education and all levels of government. 9 Appendix Table 2.3 also shows the distribution of population across the 5,500 Brazilian municipalities - it is useful to note that the 4,011 very small municipalities with population less than 20,000 represent 73% of the number of municipalities, but contain only 20% of the population of Brazil. 16 period, increasing from R$10.6 billion to R$11.5 billion. Recurrent transfers (from federal and state governments), on the other hand, increased from R$27 billion to R$ 38 billion, an increase of about 42%. Education expenditures for the sample increased from R$10.6 billion to R$ 15.5 billion, with the bulk of the increase coming from the increase in revenue receipts rather than a marked increase in the percentage spending on education. 2.4 In order to understand which kind of municipalities were affected by changes in each of the parameters being considered here (total revenues, own revenues, transfers, total expenditures, and education expenditures), we carry out a regression analysis using the STN database. The dependent variable in the regression analysis was the logarithm of the change between 1996 and 2000 in each of the revenue or expenditure parameters being considered. The regressors are the respective levels in 1996, regional dummies (with the North East as the excluded dummy, so that the coefficients can be interpreted as comparisons to the North East), the population and the GDP of the municipality in 1996. The objective of the regression analysis is to determine a) the extent to which there was a tendency for convergence in the variables, as would be shown with a negative coefficient on the level variable (higher gains associated with lower 1996 levels); and b) whether the coefficients on the regional dummies show benefits for the poorer regions in Brazil. The population and GDP per capita variables are intended as controlling variables. Logarithmic transformations are used because of the large range in the dependent variable as well as the regressors, as well as to aid interpretation of the coefficients. Coefficients that are statistically significantly different from 0 are marked with an asterisk. Full results including standard errors of the coefficients and other statistics are presented in the Appendix. Table 2 -:9Re resslonbf.Changesln Revenues and Expenditur (1996.to'2000)-on BasebVa lues in 1996,9 Log (A Total Log (A Own Log (A Log (A Total Log (A Edu Revenues) Revenues) Transfers) Expenditures) Expenditures) Intercept -1.37' -3.26' -9.07- -0.51 -8.1 9^ Log(Total Revenues) 0.81 Log (Own Revenues) 1.11^ Log (Transfers) 0.003 Log (Total Expenditures) 0.95^ Log (Edu Expenditures) 0.09 Dummy NORTH -0.06 0.64 -0.17 0.12 -0.06 Dummy SOUTH-EAST -0.06 -0.05 -0.07 0.28* 0.16 Dummy SOUTH -0.09* 0.01 -0.17 -0.57* -0.24 Dummy CENTER-WEST 0 04 -0.07 0.01 .0.20* 0.30 Log (Population in 1996) 0.16' 0 32^ 072^ 0 33^ 0.54^ Log (Municipal GOP in 1996) 0.03 -0.04 0.16* -0.20^ 0.16^ Source: STN Database 2.5 In the first column of Table 2.4, the regression of the change in total revenues (1996 - 2000) on the base value for 1996 shows a statistically significant coefficient value of 0.81 - since the variables are measured as logarithms, the result implies that if the 1996 value for a municipality was 1% higher than another municipality, the 2000 value was higher on the average by 0.81%. The second column shows that the corresponding relation considering own revenues was similar, with a coefficient 17 value of 1.11. In the third column, looking at transfer revenues, we find that the correlation was statistically insignificantly different from zero. What story do these three coefficients tell us? They tell us the story that rich municipalities got richer and even though poor municipalities received additional revenues in the 5-year period, there was not a convergence between the rich and the poor in regard to total municipal revenue. However, when we consider transfer revenues, this story was not repeated, i.e., higher levels of transfers in 2000 were not associated with a high level of transfers in 1996. Even though we do not see a compensatory trend, at least we can say that the changes in transfers were not regressive. The results hint that inequality of municipal revenue continues as before in Brazil, but that something might be going on in particular portions of municipal budgets.'0 2.6 An even more interesting story in Table 2.4 comes from looking at a) the negative coefficients on the regional dummies in the total expenditure regression; and b) the difference between the results for total expenditures as compared to education expenditures. The coefficients on the dummy variables in the total expenditure regression indicate that municipalities in the North East spent proportionally more in 2000 than in 1996 as compared to municipalities in the prosperous South and South East regions. In the same regression of total expenditures, the negative coefficient on the 1996 municipal GDP indicates that poorer municipalities (in terms of GDP) gained more in terms of expenditure increases. 2.7 The zero-valued coefficient on educational expenditures as compared with the positive value on total expenditures indicates that there was a shift in the pattern of educational resources that was different from other areas of expenditures - municipalities that spent low amounts on educational expenditures did not continue to spend low amounts, at least not in relative terms. These results are important in arriving at the policy conclusions presented at the end of this section -note that we are looking at all education expenditures, and not just at FUNDEF expenditures, and that the FUNDEF formula does not include any explicit compensatory criteria. 2.8 The preceding analysis led up to some patterns regarding increases in total municipal education expenditures. However, total education expenditures could have increased together with an unequal rate of increase in the enrollments of students, leading to a different pattern in terms of per student expenses. We now examine the difference in per student municipal expenditures using the data for 1996 and 2000. Chart 2.1 presents the change in education expenditure per student, treating students across the different levels as the same unit. The chart indicates an increase of per student education expenditures by about 40% in the period from 10 Note that our analysis begins at 1996, at which time poor municipalities had already benefited from compensatory elements in the FPM transfers established in the Constitution of 1988. The changes in transfers that we are capturing in the regression pertains to the period since 1996, which saw the initiation of policies aimed to transfer revenues to poor municipalities towards improving social services such as health and education by moving from discretionary and politicized agreements towards transfers based on rules. 18 1996 to 2000. The increase was higher in the North and North East, and lower in the more prosperous South and South East regions. Chart 2.1: Per Student Increase in Municipal Education Expenditures 1,800 1,600 1.400 1.200 1,000 800 600 400 200- E199%6 452 1 465 t34s t1,468 1,183 859 10200 779 591 1.651 1,786 1,306 1.205 Constant 2001 Reals 2.9 Even as the regional pattern of educational resources per students continues to show high variation, there is a substantial narrowing of the distribution of municipal per student education expenditure when we consider all of Brazil. We extend the analysis of mean expenditure per student to the issue of the distribution of the expenditure per student across municipalities. Table 2.5 shows the changing shape of the distribution from 1996 to 2000. The table shows substantial reductions above the median value, and positive increments below the median value. A rough measure of inequality, comparing the 95th percentile to the 5th percentile, declines from 22 to 8, and the corresponding number for the 75th and 25t'h percentile declines from 4 to 2. These figures indicate a substantial movement towards improved equity in municipal education expenditures. Table 2.5.: Changing Dlstribution of ,Educational Expenditures per Student7 - . 1996 2000 Maximum 19,043 13,569 95 Percentile 7,258 4148 75 Percentile 2628 2093 Median 1325 1400 25 Percentile 714 921 5 Percentile 326 500 Minimum 122 197 Ratio 95/05 22 8 Ratio 75 /25 4 2 Source: STN 19 2.10 Just how such a large decrease in inequality came about requires a more detailed comparison between the situation in 1996 and the situation in 2000. The reforms introduced in 1996 were required because of the existence of a wide divergence between sub-national resources for education and the requirements in terms of enrolled students. Table 2.6 shows the number of students enrolled in municipal systems in 1996, and the percentage of public enrollment represented by those numbers for each state. The 26 states in Brazil are arranged in the table according to the geographical region. Table 2.6: Munici al Enrollment and Share of Total Enrollinent in 1996 Pre-School Basic Secondary Muni Muni Muni _ OOs Share OOOs Share OOOs Share Brazil 2,49 76% 10,921 o30 312 7% North 134 49% 926 36% 5 2% Rond6nia 12 45% 89 34% 1 3% Acre 4 31%1 37 32% 0 3% Amazonas 16 57% 176 35% 1 1% Roraima 2 19% 2 4% 0 0% Para 83 54% 505 39% 3 2% Amapa 2 14% 15 15% 0 0% Tocantins 15 55.40% 100 32% 1 2% Northeast 777 75% 4,948 54Y% 164 18%Y' Maranhio 134 79% 791 65% 24 25.30% Piaui 56 63% 296 55% 2 5% Ceara 141 85% 808 61% 22 20% R. G. do Norte 45 70% 231 46% 7 11% Paraiba 49 77% 290 50% 3 6% Pernambuco 84 82% 751 51% 43 21% Alagoas 39 79% 306 65% 7 26% Sergipe 43 73% 159 45% 7 22% Bahia 187 67% 1,318 50% 48 18% Southeast 1,240 89% 2,803 24%0- 127 6% Minas Gerais 256 75% 845 25% 61 13% Espirito Santo 44 61% 135 25% 9 9% Rio de Janeiro 110 73% 1,097 64% 23 8% Sio Paulo 829 100% 727 13% 34 2% South 257 70% 1,627 400/o 8 1% Parana 104 93% 762 46% 0 0% Santa Catarina 101 71% 255 29% 3 2% R. G. do Sul 62 49% 609 39% 6 2% Centre-West 71 38% 617 34% 8 3% M. G. do Sul 21 64% 153 40% 3 6% Mato Grosso 19 53% 157 34% 0 1% Goias 30 49% 307 32% 4 3% Source: INEP/School Census 1996 2.11 All sub-national entities were obligated to spend a minimum of 25% of their tax and transfer revenues on education, and even as some states and municipalities exceeded the constitutional minimum, there were wide disparities in resource availability. At 20 the same time, there was a large variation in the enrollment of students across different state and municipal education systems. The two disparities did not match, especially for municipal governments in the North East, as the very municipalities that had a high enrollment of students were the ones with low levels of resources, in spite of compensatory transfers. 2.12 Table 2.6 shows that the municipal share for pre school enrollment was consistently high across all regions except for the North. Primary education enrollment was primarily municipal in the North East and in the state of Rio de Janeiro in the South East. Municipal enrollment in primary education was very low in the South East in the three other states in the region. The share of municipal enrollments at the secondary level was consistently low across all the regions, except for some of the states in the North East, where it was relatively higher, accounting for 18% of public Secondary enrollments. The differing degrees of municipal enrollment reflected the effects of historical attempts to municipalize primary education in the two decades prior to the beginning of the 1990s, efforts that had been more successful in the North East than other regions. The table does not explicitly show the state enrollments at the respective level, but the state enrollments can be inferred as the complement of the municipal enrollments. States had generally low levels of enrollment at the pre-school level and relatively high levels of enrollments in regions other than the North East and the state of Rio de Janeiro. 2.13 The data on enrollments in municipal and state systems provides a view of the resource needs for education in 1996. If resource availability had matched the variation in enrollment, there would not have been the need of a sweeping reform of educational financing. In fact, the pattern of resource availability did not match resource needs, as shown by the patterns observed in Table 2.7. The first column of data in the table shows the minimum amount of resources available for education to each state government - 25% of tax and transfer resources. The second column shows the actually reported education expenditures of state governments. The third column presents the availability of minimum resources to states in per-student terms. The top entry in the column shows that the average resource availability for all of Brazil for students enrolled in state systems was R$650 per student (1996 reals). This amount is used to create an "Index of Minimum Revenue" to compare resource availability, across states and regions. The same base value of R$650 is used to compute an index value of reported educational spending, shown in the last column of Table 2.7. The index value of 111 for all of Brazil for actual education expenditures indicates that in per student terms, the spending exceeded the constitutionally established minimum by 11%. States in the North and North East have lower values for either of the indices. The state of Rio de Janeiro is a notable outlier, with a value of 202 for the index on minimum revenues, and a value of 264 for the index on spending. We will discuss the state of Rio de Janeiro again when we look at allocations after the financing reforms. 21 Table 2.7: State resource availability in:1996 _ - 25% of Tax Actual Minimum Index of Actual Index of and Transfer Spending in Revenue per Minimum Spending Actual Revenues 1996 Student Revenue per Student Spending ( s000 Reals) (OOOs Reals) Brazil 14,904,396 16,489,596 650 100 719 111 North 1 ,261 ,081 1 278,134 577 89 584 90 Rondonia 125,146 123,126 578 89 569 88 Acre 96,371 97,715 969 149 983 151 Amazonas 427,570 300,546 1,030 158 724 111 Roraima 66,253 68,702 884 136 916 141 Para 301,223 425,650 303 47 428 66 Amapa 110,843 122,698 936 144 1,036 159 Tocantins 133,676 139,697 498 77 520 80 Northeast 2,905,842 2,684,127 568 87 525 81 Maranhio 266,984 323,154 504 78 610 94. Piauf 184,783 192,850 591 91 617 95 Ceara 426,508 292,589 667 103 457 70 R. G. do Norte 192,449 190,076 563 . 87 556 85 Paraiba 252,097 237,397 700 108 659 101 Pemambuco 467,406 335,700 523 80 376 58 Alagoas 163,420 154,868 853 131 808 124 Sergipe 156,652 162,263 670 103 694 107 Bahia 795,543 795,231 494 76 493 76 Southeast 7,760,089 8,809,611 715 110 812 125 Minas Gerais 1,384,784 1,860,223 457 70 614 94 Espfrito Santo 369,848 240,649 720 111 469 72 Rio de Janeiro 1,197,147 1,559,871 1,316 202 1,715 264 Sio Paulo 4,808,310 5,148,868 752 116 805 124 South 2,204,590 2,871,065 671 103 874 134 Parana 726,086 1,104,520 594 91 904 139 Santa Catarina 503,058 495,611 636 98 626 96 R. G. do Sul 975,446 1,270,934 766 118 998 153 Centre-West 772,793 846,659 515 79 565 87 M. G. do Sul 177,050 230,186 585 90 761 117 Mato Grosso 237,719 231,242 631 97 614 94 Goiis 358,024 385,232 436 . -67 469 72. Source: INEP/STN 2.14 The analysis of resource availability for municipalities in 1996 is shown in Table 2.8. In order to facilitate comparison with resource availability for state Governments, we continue using the same base of R$ 650 as a base value for construction of the indices. This method enables us to see that for Brazil as a whole, Municipalities had resources that amounted to approximately 93% of the resources for State Governments. However, the striking figures in Table 2.8 are the ones that relate to the municipalities in the North and North East regions. The state of Para in the North had a minimum index value of only 33, and the state of Maranhao a value of 20. At the other extreme, the municipalities in the state of Sao Paulo were constitutionally enjoined to spend resources that accounted for an index value of 279. These wide disparities indicate how public spending in education in 1996 was working to increase rather than ameliorate the inequality that has defined much of Brazilian economic development. 22 Table 2.8: LbJunicil ial resource availabilitf in 1996 25% of Tax Actual Minimum Index of Actual Index of and Transfer Spending in Revenue per Minimum Spending Actual Revenues 1996 Student Revenue per Student Spending (OOOs Reals) (000s Reals) _ . Brazil -7804,967 8,165,407 606 , 93. 634 98 North 341,074 '' 349,833 320 5491 328 -*'-.5 Rond6nia 29,319 19,269 288 44 189 29 Acre 20,113 30,605 487 75 741 114 Amazonas 89,725 89,059 464 .71 460 71 Roraima 11,552 7,060 2,957 455 1,807 278 Para 128,662 131,793 218 : .:. 33 223 34 Amapa 13,771 5,092 776 119 287 . 44 Tocantins 47,932 66,955 413 '64 577 89 Northeast' .1,575,051 1,986,474 267 41 337 '52 Maranhio 120,577 159,680 127 . 20 168 26 Piaui 67,213 83,181 190 . 29 235 36 Ceara 274,037 348,965 282 43 359 55 R. G. do Norte 109,032 103,162 386 59 365 . 56 Paraiba 116,983 146,002 342 - 53 427 66 Pernambuco 270,043 317,798 308 47 362 - 56 Alagoas 94,989 121,545 270 . .42 346 ',53 Sergipe 68,007 89,227 326 .50 427 66 Bahia 454,168 616,914 293 45 397 : 61 Southeast . 4,249,149 3,745,749 1,108 1.70 977 - 150 Minas Gerais 986,166 882,678 850 . .131 760 .117 Espirito Santo 122,889 235,267 657 '101 1,258 194 Rio de Janeiro 859,208 424,140 698 107 345 553 Sao Paulo 2,280,887 2,203,665 1,815 -279 1,754 . 270 South , ... 1,156,293 1 ,486,105 830 128 1,066 . 164 Parana 560,193 742,552 647 100 858 .132 Santa Catarina - - . R. G. do Sul 596,100 743,552 1,129 ,. 174 1,408 217 'Centre-West 483,399 597,247 695 107 859 132 M. G. do Sul 118,317 154,885 666 103 872 134 Mato Grosso 128,646 174,913 728 112 990 i52 Goias 236,437 267,449 693 107 784 121 Source: INEP/STN 2.15 The story regarding municipal and state finances in 1996 is brought together in graphical form in Chart 2.2a. In addition to the observations regarding the North East and the two states of Rio de Janeiro and Sao Paulo, two further interesting elements can be observed from Chart 2.2a. Firstly, comparing the solid bars (state index values) with the dashed bars (municipal index values), the variation is observed to be much greater for the municipal index values. Numerically, the standard deviation for the municipal index (72) is more than twice as high as the standard deviation for the state index (34). Secondly, the index value for the state government is higher than the index value for the municipalities in the state for a majority of the cases. 23 Chart 2.2a: Resource Availability 1996:States and Municipalities Compared 300 250 200 150 100 50 Z 0 0 O N 2.16 The policy solution to the divergence between resource needs and resource availability was the creation of FUNDEF, a funding mechanism that ensured that resources were linked to the enrollment of students. As municipalities and states were enjoined by the Constitution to cooperate in the provision of primary education, it was feasible to create the common pool of resources between the state government and the municipalities, and then redistribute the resources according to the enrollment of students for primary education. A number of concerns of the formula funding mechanism of FUNDEF were expressed when the legislation was initiated in 1997. These concerns included:- a) the possible harmful effect on other levels of education that did not benefit from the formula funding; b) the fungibility of municipal resources, so that municipalities that received additional resources from FUNDEF reduced education spending from their non-FUNDEF resources; and c) the federal government has attempted to set the floor low so that it does not have to include too many states in the list of states that receive federal complements to FUNDEF; d) the obligation to spend a minimum of 60% of FUNDEF resources on teacher salaries set up rigidities that may have been contrary to local preferences, leading to loss of efficiency. This section of the report presents a brief discussion of the first three of the issues mentioned above. The issue of composition of expenditures is considered in detail in a later section. However, before we examine the issues regarding FUINDEF, it is useful to look at the situation regarding all, resources for municipal education in 2000, and then examine the empirical data regarding the redistribution of resources under FUNDEF from states to municipalities and from richer to poorer municipalities. 24 2.17 Table 2.9 presents the resource situation of municipalities in 2000. The table points to a significant improvement in resource availability to municipalities. Using the value of R$873 per state student as the base, we see that there was a parity between states and municipalities, with the overall index of minimum revenue moving from 93 in 1996 to 104 in 2000. The index for actual spending was 129. These two values compare with respective values of 93 and 98 for 1996, indicating the shift in resource availability in the period from 1996 to 2000. ablz 2.9: Munrci al resoucce availalblZity In 2000 25% of Tax Actual Minimum Index of Actual Index of and Transfer Spending in Revenue per Minimum Spending Actual Revenues 1996 Student Revenue per Student Spending ( s000 Reals) (000s Reals) Brazil 16,129,870 19 935,032 - 08 104 1,122 129 Novth 633,393 741,303 620 71 725 83 Rond6nia 99,461 114,645 696 ; 80 803 92 Acre 35,488 49,321 730 84 1,015 116 Amazonas 164,031 203,179 613 : 70 759 87 Roraima 27,248 29,731 2,494 286 2,722 .312 Para 205,907 234,730 481 55 548 63. Amapi 21,728 22,004 903 103 915 105. Tocantins 79,529 87,693 795 91 877 100 Kce5s 2,993,093 3,967,845 415 48 551 63 Maranhio 252,766 385,731 283 32 432 50 Piaui 139,765 186,280 428 .. 49 570 . 65 Ceari 464,097 596,234 362 42 465 53 R. G. do Norte 199,060 242,501 624 71 760 87. Paraiba 236,423 355,672 485 56 729 83 Pernambuco 455,476 508,519 489 56 545 62 Alagoas 214,416 272,852 403 46 513 59 Sergipe 122,275 169,677 505 58 701 80 Bahia 908,814 1,250,380 415 48 571 65 Southeast 8,935,140 10,796,461 1 ,377 158 1,663 191 Minas Gerais 1,802,596 2,440,059 993 114 1,345 154 Espfrito Santo 339,203 473,351 1,050 120 1,465 168 Rio de Janeiro 1,790,397 1,834,544 1,163 133 1,191 136 Sao Paulo 5,002,945 6,048,507 1,778 204 2,150 246 South 2,806,948 3,490,635 1,236 142 1,537 176 Parani 1,057,583 1,242,238 1,120 128 1,316 151 Santa Catarina 600,012 795,103 1,166 134 1,545 177 R. G. do Sul 1,149,352 1,453,294 1,415 162 1,789 205 Centr-West 761,297 938,788 . 986 113 1,216 139 M. G. doSul 207,424 261,971 881 101 1,112 127 Mato Grosso 264,473 378,740 960 110 1,375 157 Goids 289,401 298,077 1,108 127 1,141 131 Source: INEP / STN 2.18 Table 2.9 also shows improvement in the index values for municipalities in the North and North East. Resources for the municipalities are still low compared to the national average, but the difference with richer municipalities has narrowed down considerably. We turn now to an analysis of how FUNDEF was instrumental in bringing about these changes. 25 B. Redistribution of Resources away from State Governments towards Municipal Governments 2.19 FUNDEF implies a redistribution of resources away from state governments towards municipal governments. Overall, in the year 2001, 24 of the 26 state governments in Brazil together made a net contribution of R$ 2.3 billion to FUNDEF, resources that were redirected towards municipal governments. The distribution of the "losses" incurred by state governments is skewed, with a few states accounting for the bulk of the redistribution. The state government of Rio de Janeiro alone accounts for nearly a fourth of the state to municipal transfers. The 9 North-Eastern States together account for nearly a half of the redistribution from FUNDEF, as shown in the following graph (North-Eastern states marked in red dashed bars):- Chart 2.2b: Net Gain of State Govemments from FUNDEF in 2001 100.00 o 0.0 ._ 00.00 - Ul~~~~ 0 3~~~~~~~~ *i-400 00- .500.00 Soutrce: MEC-FUNDEF, Juv 2001 -400.00 2.20 While an analysis of absolute amounts redistributed from state to municipal governments is important to understand the context, it is of further interest to examine the weights of the FUNDEF transfers relative to the spending floors of the state governments on primary education. For Brazil as a whole, the sum of R$ 2.3 billion redirected from states to municipalities accounts for at most about a fourth of the state government spending on primary education. However, the incidence of this redistributive effect is even more concentrated for a few state governments, as shown in Figure 2.1 below:- 26 Figure 21 State L to FUNDEF as Peroentage of State FUNDEF Spedling GL_SPEND | to 10% RRi 0.0 Ism So% ore ,0 50 % Source MEC-FUNDEF Julv2001 2.21 The primary reason for resources to be allocated away from states towards municipalities is because FUNDEF monies follow the student. As FUNDEF pools together the state and municipal resources, the redistribution of these resources takes place according to student enrollment in primary education. A key question here is whether the simple formula funding mechanism of FUNDEF is adequate when one considers equity of expenditures. Is the magnitude of transfers away from state governments related to the inequity between states and municipalities prior to transfers? We answer this question by examining the correlation between the magnitude of transfers with the size of the state-municipality gap prior to the transfers. In other words, were the states that transferred relatively more resources to their municipalities also the states whose spending was more disparate from their municipalities prior to the reforms? The variable used to measure the magnitude of the state-municipality gap are:- a) the difference in per student spending between State and Municipal spending prior to transfers (the variable is named DIFFSMF); and b) - the average salary of teachers in 1997 in the respective systems, as reported in INEP's Census of Teachers (the variable is named DiFFSAL). 2.22 Table 2.6 represents the results of a regression equation with data from the 26 states in Brazil. The dependent variable is the net gain (receipts from FUNDEF less contribution to FUNDEF) per student as the dependent variable. Two alternative specifications regarding municipal-state resource gaps are used as regressors - Specification I uses DIFFSMF and Specification II uses DIFFSAL. A common set of control variables includes a) Population in 2000, b) Per Capita Income of the State, c) Human Development Index and d) Percentage of Enrollment in State System. The table shows that there indeed was a direct relationship between the net transfers of resources away from states (expressed in per student terms) and the state-municipality resource gap, especially when proxied by the difference in average teacher salaries. The state of Rio de Janeiro was an outlier, as in addition to having very high transfers, the state was one of the exceptions where average 27 municipal teacher salaries were higher in 1997 as compared to average state teacher salaries. Running the regression with Rio purposively excluded accentuates the relationship, as shown in the last two rows of Table 2.6. Table 2.10: Regression of Net Gain per Student on Municipal State Resource Ga p , ._ " _ _ , , - _ , _ _ _ _ _ _ _ ' _ ' _ Specification I Specification II Variable Coeff S.E Coeff S.E % State Enrollment 7.236 0.637 9.856 1.597 GDP per Capita -0.118 0.032 -0.141 0.035 Population (in millions) 11.581 3.752 13.134 3.907 HD Index 16.222 5.28 18.145 5.711 DIFFSMF -0.109 0.056 DIFFSAL -0.123 0.146 DIFFSMF* -0.083 0.041 DIFFSAL* -0.253 0.096 Source: MEC/FUNDEF/IBGE/INEP 2.23 Chart 2.3 depicts a scatter plot between net gain per student enrolled in a state system of education, and the salary difference between municipal and state teachers. The chart shows a bi-variate slope of -0.5, nearly twice the regression coefficient shown on the last line of Table 2.10, indicating that part of the correlation is explained by the other variables included in the regression equation. Chart 2.3: Correlation between Net Gain per Student and State-Municipal Teacher Salary Difflerence in 1997 Qca99 100.00 8W000 -2 f° > * ~~~~100 20g 300o 400 500 9 0~~~~~~~ 40o 00 000.00 ~ ~ ~ Salary DlNterence In Reals per month 28 C. Redistribution of Resources among Municipal Governments 2.24 In addition to the R$2.3 billion that was redirected away from 24 of the 26 states towards the municipalities in each of the states, FUNDEF also induces a redistribution of resources among the municipalities themselves. Of the 5,560 municipalities in Brazil, we have data from a subset of 5,386 municipalities. Data from 2001 indicates that of the 5,386 municipalities, 2,033 were net 'losers' to FUNDEF, i.e., they contributed more to FUNDEF than they got back. The total net contribution of the 'losers' was approximately R$ 0.8 billion. 3,342 municipalities were net 'gainers', and the sum total of the net gain was approximately R$ 3.6 billion."I The distribution of gains and losses is heavily skewed, even more so than the distribution of net losses of the state governments. Chart 2.4: Net Gain of Municipal Govemments from FUNDEF in 2001 1.80 160 C i 1.40 1.20 - o 1.00 CDF of Uniform I; 0 80 Distribution 080 CDFof Net 040 ~~~~~Gain E 0 20 0 00 Ordering of Municipalites 2.25 Chart 2.4 shows the cumulative distribution function (CDF) of net gains of municipalities. The figure also includes the CDF of a hypothetical uniform distribution with an identical range for purposes of comparison. The skewness of the distribution is apparent by looking at the 20% mark. The empirical distribution shows that 346 municipalities accounted for 20% of the contribution made to the overall redistribution of municipal resources. If the net contribution had been declining uniformly from the highest net loser (the municipality of Guarulhos in the state of Sdo Paulo; net contribution R$ 42 million) through zero and on to the The R$ 0.8 billion contributed to FUNDEF by the 'loser' municipalities, plus the R$2.3 billion contributed by State Governments and the balance of R$0.5 billion contributed by the Federal Government. 29 positive side of net gains to the other end of the range (the municipality of Rio de Janeiro, in the state of Rio de Janeiro; net gain R$ 292 million), the 20% mark would have been crossed at municipality number 2610 rather than number 346. The data underlying Chart 2.4 also shows that the first 8 'loser' municipalities accounted for R$152 million of the losses (about 5% of the total redistribution) and the last 8 'gainer' municipalities accounted for R$618 million (more than 20% of the same total). TABLE 2.11: Distribution of Net Gains from FUNDEF amo g Municipalities in 2001 Number of Muncipalities Amount of Net Gain (R$) Distribution Across States Distribution Across States STATE Total Gainers Losers Total Gain Loss Acre 0.4% 0.7% 0.0% 0.8% 0.6% 0.0% Amazonas 1.2% 1.6% 0.4% 2.2% 1.8% 0.2% Amapa 0.3% 0.4% 0.2% 0.3% 0.2% 0.0% Para 2.7% 4.1% 0.2% 8.6% 7.2% 2.1% Rondonia 1.0% 1.4% 0.2% 1.1% 0.8% 0.0% Roraima 0.3% 0.2% 0.4% -0.2% 0.0% 1.1% Tocantins 2.6% 2.6% 2.5% 0.8% 0.7% 0.5% NORTH 8.3% 11.0% 3.9% 13.5% 11.4% 4.0% Alagoas 1.9% 2.9% 0.1 % 3.5% 2.9% 0.7% Bahia 7.7% 11.9% 0.7% 15.3% 12.6% 3.0% Ceara 3.4% 5.5% 0.0% 9.0% 7.1% 0.0% Maranhao 4.0% 6.4% 0.1% 9.5% 7.6% 0.7% Paraiba 4.1% 5.1% 2.6% 2.7% 2.2% 0.6% Pernambuco 3.4% 5.2% 0.4% 6.1% 4.9% 0.3% Piaui 4.1% 6.2% 0.6% 3.2% 2.5% 0.1% Rio Grande do Norte 3.1% 4.1% 1.5% 1.9% 1.6% 0.4% Sergipe 1.4% 1.9% 0.6% 2.0% 1.7% 0.5% NORTH-EAST 33.1% 49.2% 6.7% 53.2% 43.0% 6.2% Goias 2.3% 0.8% 4.8% 0.0% 0.6% 2.5% Mato Grosso do Sul 1.4% 1.0% 2.2% 1.0% 1.1% 1.4% Mato Grosso 2.3% 1.9% 3.0% 1.4% 1.3% 1.2% CENTRE-WEST 6.1% 3.7%/6 10.0% 2.4% 3.0% 5.1 % Espirito Santo 1.4% 1.6% 1.2% 1.7% 1.4% 0.4% Minas Gerais 15.8% 10.7% 24.3% 3.2% 4.5% 9.0% Rio de Janeiro 1.7% 1.8% 1.5% 19.0% 15.2% 1.7% Sao Paulo 12.0% 7.1% 19.7% -1.5% 11.3% 57.6% SOUTH-EAST 30.9% 21.2% 46.7% 22.4% 32.4% 68.7% Parand 7.4% 6.5% 9.0% 3.5% 3.5% 3.5% Rio Grande do Sul 8.7%/6 5.5% 13.8% 4.0% 4.8% 7.7% Santa Catarina 5.4% 2.8% 9.8% 1.0% 1.9% 4.8% SOUTH 21.5% 14.8% 32.6% 8.6% 10.2% 15.9% BRAZIL 100.0% 100.0% 100.0% 100.0% 100.0% 100.0% Source: MEC-FUNDEF 2.26 The examination of the univariate distribution of the net gains from FUNDEF accruing to municipalities indicates, as did the analysis of state net gains, that FUNDEF entails a massive redistribution of resources. We now seek to characterize the 'losers' and 'gainers' among the municipalities. The first step is to look 30 separately at the 'losers' and 'gainers' in the list of 5,386 municipalities. Table 2.11 provides an overview of gainers and losers by state and by region. 2.27 As shown in Table 2.11, the gainers were predominantly the municipalities in the North East (both in terms of numbers and municipalities and the amount of net gain), and the losers were the municipalities in the South and South East. Indeed the municipalities in the rich states of the South and South East together account for 85% of the losses incurred by municipal governments. The municipalities of the North-East account for 53% of the gains made by municipalities, though they represent only 33% of the number of municipalities. Much is made of the fact that FUNDEF as a policy instrument is restricted to redistributive forces within state boundaries, whereas the biggest inequalities in Brazil are across states. While the reasoning behind such an argument is valid, such is the nature of sub-national revenue streams in Brazil and the incidence of enrollment in municipal systems of fundamental education that the relatively poor North East municipalities end up being the primary gainers. It is extremely interesting to note the two factors together, that a) State governments are net losers, contributing an overall R$2.3 billion to municipalities through FUJNDEF and b) 'Losing' municipalities are much smaller in number and the total of municipal losses (R$ 0.8 billion) is only a fraction of the total of municipal gains of R$ 3.6 billion. 31 D. Fungibility of Municipal Resources and FUNDEF 2.28 It is well known that the receipt of a grant from a higher level of government for a specific purpose such as primary education does not necessarily lead to an increase in resources for that purpose because of the fungibility of resources. The sub- national government that receives the grant, say for primary education, may reduce its expenditure assigned for primary education, leading to a possibly much lower net effect on resource availability for that specific purpose. If compensatory financing is one of the purposes of the grant, such an objective may be undermined if local preferences do not coincide with nationally imposed ones, leading to a reduced impact of the federally designed intervention. In the particular case of FUNDEF resources for Brazilian municipalities, there are a number of factors that would influence such an eventuality. First of all, FUNDEF is not in the nature of a grant from a higher level of government - the municipality gets its own monies back from FUNDEF, as well as additional monies, if the municipality has enough enrolled students. Hence, only the part of FUNDEF resources that are in excess of the municipality's own contribution can be considered as additional resources. Secondly, the constitutionally established floor of 25% of revenues and transfers places a limitation on the extent to which expenditures can be cut back. Therefore, the question of substitutability arises only for those municipalities that were spending in excess of the floor of 25%. In fact, for these municipalities, it is possible that there was the need to maintain or even increase expenditures over and above the extra resources from FUNDEF. Finally, municipalities are also enjoined to provide services for pre-school education, which does not benefit from FUNDEF. As municipalities had increased responsibility for pre-school after the LDB assigned the specific responsibility to municipalities, there was less flexibility for the municipality to reduce expenditures. It would not be feasible for municipalities to reduce teacher salaries for one group of teachers (pre-school) at the same time as they increased the salaries for another group of teachers (primary education). The slack in expenditures would only come from lowering the number of additional teachers hired or from lower expenditures in non-salary expenditures. We explore the composition of expenditures in a later section of this report. At present, we turn to an empirical determination of the effect of FUNDEF resources on municipalities that were net gainers. 2.29 Data limitations restrict the nature of possible analysis regarding fungibility of resources. This is because we have to pull together data from a combination of sources from municipalities that do not follow the same standard of accounting practices. We use an extract of the STN database of 1500 municipalities that has uniform reporting of the composition of revenue sources. The database is sufficiently large to reduce the possibility of bias, but the analysis may not accurately represent the situation for all Brazilian municipalities, a shortcoming that can only be overcome with a larger version of the STN database. We choose the two years of 1996 and 1999 to perform the analysis, comparing revenue and expenditure assignments across the two years. We first compute the percentage of revenues spent on education in 1996, before the creation of FUNDEF. We then 32 compute the net gain from FUNDEF as a percentage of revenues, and then examine whether reported expenditures after FUNDEF was more or less than the difference created by the additional resources from FUNDEF. A few numerical examples would illustrate the computation. The municipality of Quatis, a small municipality of population 10,000 in the state of Rio de Janeiro spent 29.42% of its revenue (from tax and transfers) on education in 1996. With FUNDEF, the municipality accrued gained the equivalent to 5.33% of its revenues. If there were a one-for-one increase in educational expenditures due to FUNDEF, the municipality would have spent 29.42+5.33=34.75% in 1999. The reported expenditure for Quatis in 1999 was 35.88%, i.e., 1.13% more than the pure gains from FUNDEF. As an example in the other direction, the municipality of Granito, of population 6,000 in the North East state of Pernambuco, spent 31.51% of revenues on education in 1996. FUlNDEF gains were 15.1 1 % of revenues, but education expenditures went up only by 11.96%, thus the effect was about 3 percentage points lower than a one-for-one increase. The data from the STN sample of 1500 used for this analysis indicates a substantial variation in the extent to which the effects from FUNDEF varied away from a unitary effect. The average for the entire sample was that gains were at 15% of the revenues in 1996, but that the increase in educational expenditures was 2.7% lower than this figure. However, there is a wide variation in this statistic, which we term as deviation from a unitary effect. 2.30 Table 2.12 presents the means from the STN sample of 1500 municipalities used to analyze the patterns of co-variation that explain the variation in the deviation from a unitary effect. Table 2.12: Variables Explaining Deviation from Unitary' Effect for FUNDEF, -, 1 l Standard Variable Description Mean Deviation SPEND96 Education Spending as % of Revenue 1996 34.75 6.58 SPEND99 Education Spending as % of Revenue 1999 47.07 7.57 NETGAINS Net Gainsfrom FUNDEFas% of Rev 1999 15.04 7.54 DEVIATION Deviation from One for One Effect of NETGAINS -2.72 9.92 STUMIL6 Thousands of Municipal Basic Education Students '96 2.22 4.94 MUNSH96 Municipality Share of Basic Education Enrollment 96 38.68 20.65 MUNSHOO Municipality Share of Basic Education Enrollment '00 52.36 18.72 OTHSH96 Share of Other Levels of Edu in Municipal System 96 22.36 18.23 OTHSHOO Share of Other Levels of Edu in Municipal System 00 18.29 10.20 REVPCAP Total Revenue of Municipality per Capita 279.05 145.72 OWNREV Own Tax Revenue as % of Total Municipal Revenue 4.31 5.93 DUM_N Dummy for Region North 0.031 0.17 DUM_CW Dummy for Region Centre-West 0.089 0.29 DUM_S Dummy for Region South 0.279 0.45 DUM_SE Dummy for Region South-East 0.356 0.48 Source: INEP/STN 2.31 The table shows the overall increase in percentage spending on education from 34.75% to 47.07 %. The municipal share of primary education enrollments increased from 39% to 52%, and the share of other levels of education dropped from 22% to 18%, primarily due to reduction in municipal secondary enrollments. 33 Dummy variables were constructed with the North East as the excluded group. The means for the dummy variables indicate a very low representation for municipalities in the North, but sufficient representation in other regions. 2.32 Table 2.13 shows the results of a regression with the deviation from unitary effect as a dependent variable. Positive coefficient values on the regression coefficients in Table 2.13 can be interpreted either as leading to a higher positive deviation or a lower negative deviation. Table 2.13.:Ftegression ResuIts explaining deviation from fltaryf ff ectfor FUNDEF- , P - value for Zero Variable Coefficient Standard Error Null Intercept 31.57 1.26 <0.001 Enrollment _ STUMIL96 -0.18 0.04 <0.001 MUNSH96 -2.86 1.39 0.04 MUNSHOO 8.12 1.41 <0.001 OTSH96 -1.47 1.32 0.26 OTHSHOO 1.55 2.18 0.48 Financial REVPCAP -0.002 0.001 0.09 OWNREV -9.63 3.13 0.002 NETGAINS -0.56 0.03 <0.001 SPEND96 -0.85 0.03 <0.001 Geography DUM_N -3.07 0.97 0.002 DUM CW -0.21 0.77 0.79 DUM_S 2.93 0.56 <0.001 DUM_SE 3.74 0.57 <0.001 Dependent Variable: DEVIATION from one-for-one impact of FUNDEF resources Sample Size: 1508, R-squared 0.62; F-value of regression: 184.48 Source: STN Database; INEP for Enrollments 2.33 The table shows three groups of explanatory variables. The first group constitutes of variable related to enrollment. The negative sign on the variable STUMIL96 indicates a scale effect - the larger the municipal system, the more likely that the municipal system was approaching the bounds on resource availability, and the more difficult to sustain higher expenditures after FUNDEF. The signs of the coefficient on the enrollment shares tend to confirm this hypothesis. The negative sign on the 1996 variables shows the difficulty to increase expenditures with already high enrollments; the positive coefficients on the 2000 variables indicate the dynamic introduced by FUNDEF. After controlling for the 1996 enrollments, municipalities that were more able to increase enrollments after FUNDEF were better able to secure additional resources for education from other sources as well. Amongst the financial variables, the level of net gains has an expected negative sign - the higher the net gains, the lower the chance of there being a one-for-one effect on resource availability. The negative coefficient on the variable SPEND96 mirrors 34 the finding regarding enrollments in 1996. Revenue per capita of the municipality and the percentage of own revenue have a negative impact. This is an interesting finding from the policy perspective - the FUNDEF reforms were designed to help poorer municipalities, such as those in the North-East, that are heavily dependent on transfer revenues. The results of Table 2.13 indicate that such municipalities were more likely to benefit from additional resources beyond merely the amount of net gains that they received from FUNDEF. Finally, the coefficients on the regional dummies are statistically insignificant for the Center-West, and statistically significantly positive for the South and South East. E. FUNDEF and impact on Pre-School Education 2.34 As FUNDEF resources are meant only for primary education, there is a concern that other levels of education have suffered because of the lack of a similar financing mechanism. The issue of financing for secondary education (Grades 9 to 11) is outside the scope of this report, as this level of education is assigned to state governments rather than to municipalities. Furthermore, the joint World Bank -IDB study on secondary education (Secondary Education in Brazil: Time to Move Forward; Report No. 19409-BR) provides a detailed analysis of the financing of Secondary Education and the policy options. The case of pre-school education is different, because pre-school education is a responsibility of municipalities. We examine the hypothesis that the gains for primary education due to FUNDEF came at the expense of pre-school education. The first concern was that as states progressively reduced their involvement in pre-school education, municipalities were not in a position to take up the responsibility to provide pre-school services. There are two perspectives to examine the issue of pre-school education - the perspective of resource availability and that of actual enrollments. 2.35 FUNDEF taps 15% of four specific revenue streams - the FPM or FPE, ICMS, IPI- exp, and LC 87/96. Of these four stream, the FPM/FPE and ICMS account for over 90% of resource availability. This leaves a minimum of 10% from these sources for municipalities to invest in pre-school education. Municipalities are also obligated to spend 15% of other revenue components on primary education, though these resources do not enter the FUNDEF pool for the entire State. The 10% remaining from these other components is also available to be spent for pre-school education. The FIPE 2001 survey used for this analysis determines whether municipalities reduced or increased outlays for pre-school education since 1997, the last year before the introduction of FUNDEF. Of the 5,319 municipalities to which the FIPE sample can be extrapolated for this question, 7 % reduced their expenditure on pre- school education in the period from 1997 to 2000, 55% increased their expenditure, and 36% had expenditures on pre-school education unchanged. The average decrease amongst the 7% of municipalities that that decreased expenditures was 66%, and the average increase amongst the 55% that increased expenditures was 83%. The percentage of increase was even across the group of municipalities that were net gainers or net contributors to FUNDEF. 35 2.36 One of the major effects of FUNDEF was the increase in salaries of primary education teachers in municipalities where the salaries were comparatively low. Part of the effect on increases in salaries could be explained by the injunction on municipalities to spend a minimum of 60% of FUNDEF receipts on salaries. Together with stricter standards regarding teacher qualifications, the policy motivation to irtcrease salaries had been to make teaching a more attractive profession and thus improve the pool of qualified teachers. Even as there was a general increase in expenditures on pre-school education, such an increase may have been associated only with an increase in coverage. If pre-school teacher salaries did not keep pace with salaries for primary education, there might be an adverse selection of less productive teachers for pre-school. The avallable data indicates that pre-school teacher salaries did not suffer such an adverse impact, as shown in Table 2.14. Table 2.14: Reported lncregses in Pre-School. eacher Salaries, 1997-2000 All Level of Increase Municipalities Net Gainers Net Losers Upto 10% 14%k 7% 29% 1 0%to 20% 17%1/ 17% 18% 20% to 30% 12% 7% 23% 30% to 50% 23% 24% 20% 50% to 70% 8% 9% 7% 70% to 90% 6% 8% 0%N 90% to 120% 6% 9% 0% 120% to 150% 6% 8% 0% 150% to 200% 4% 5% 0%/ More than 200% 4% 5% 2% Estimated Average Increase 55% 68% 28% Number of Municipalities 4656 3199 1457 Source: FIPE Database, 2001 2.37 The FIPE survey does not provide equivalent data regarding salary increases for primary education teachers, but it does determine the relation between salary increases for the two levels of education. 68% of the municipalities that reported increases for salaries for pre-school teachers also reported that the salary increases were the same as for primary education teachers. 23% reported salary increases of a smaller magnitude, and 9% reported salary increases of a larger magnitude. It is important to note that salary increases for pre-school teachers have not come at the expense of increase in coverage. INEP estimates that the coverage for 0-3 years (Creche) has gone up from 247,000 in 1998 to 665,000 in 2001, and for 4-6 years (pre-school) has gone up from 3.5 million to 4.2 million students in the same period. 36 F. Further Issues regarding Additionality 2.38 Researchers in Brazil have debated the issue of the level of the floor on per student expenditures, currently at R$ 418 per student for Grades 1-4 and R$429 for Grades 5-8. The argument is centered mainly on legal grounds of articles in the legislation that established FUNDEF. It is held that the legislation called for the computation of a single value of per student resources from the combined FUNDEF baskets of all states in Brazil, divided by the total number of primary education students in all of Brazil. The resulting unit value was to be used as the national floor for FUNDEF, so that states for which the FUNDEF basket resulted in a lower amount than this national norm would receive complementary transfers from the Federal Government. Our analysis of the School Census data of 2000 and FUNDEF revenues indicates that this would have had a nearly five fold impact on Federal Government outlays, from about R$630 million to over R$3,000 million (Appendix Table 2.14). Currently, the Federal Government's top-up only gets to 6 out of the 9 North East states and the state of Para in the North. Increasing the floor would lead to all the North East states needing complementary transfers from the Federal Government, 3 more states from the North becoming eligible for transfers, and 3 states from other regions becoming eligible as well. 75% of the increases would accrue to the North East states currently receiving a Federal complement, and the two states of Pernambuco and Minas Gerais would account for nearly 20% of the additional federal transfers. 2.39 Two related questions have been raised regarding the role of the federal government regarding FUNDEF. The Constitution also enjoined the federal government to carry out a detailed analysis of the costs of providing primary education of minimum quality and relate it to the revenues currently available from FUNDEF. The logic was that the FUNDEF per student amount is based only on the aspect of revenue collection. However, providing a minimum quality of primary education may cost more or less than the available resources. The federal government did attempt to carry out such a study of costs, though the technical issues proved to be too difficult to surmount, and the results of the study were never released publicly. It is alleged by some in the policy community that the results were not released because the study showed that the costs per student are indeed higher than the FUNDEF amount per student in a number of the states. The federal government has the additional responsibility of periodically carrying out detailed evaluation of the impact of the FUNDEF reforms. The federal government has indeed carried out such evaluations, and published two reports regarding FUNDEF12. The data sets used in the evaluations and the methodological details for these evaluations have never been publicly released. The Bank's study team is the first team outside of the federal 12 Balango do Primeiro Ano do FUNDEF, MEC, Brasilia, 1999; and Balango do FUNDEF 1998-2000, MEC, Brasilia, 2001. 37 government that has been permitted access to the richly detailed database used in these evaluations, under strict conditions of maintaining confidentiality.'3 2.40 Without contesting the legal issues regarding the floor of expenditures, from an analytical point of view, we have seen that the redistribution caused by FUNDEF has led to significant increases in availability of resources to municipalities, particularly the poorer municipalities in the North East. For instance, the comparison of Table 2.8 and Table 2.9 has shown patterns such as the doubling of minimum per student municipal resources in the state of Maranhao from R$127 to R$283 (including all resources and all levels of education). The mismatch between revenues and needs was so great within the states in Brazil, especially between state and municipal governments, that there has been a significant shift towards equity, just relying on transfers within states. As we will see later in the report, the evidence is mixed regarding the effectiveness of the additional expenditures and the absorptive capacity of the municipalities to spend the resources prudently. The analysis of Appendix Table 2.14 shows that the biggest beneficiaries of raising the floor would be the same municipalities that have already received an infusion of resources. The increases in enrollment have led to a situation of nearly universal enrollment, so the incentive effect of FUNDEF to increase enrollments has clearly already been played out. It is difficult to justify a call for yet more additional resources without tying those additional resources to new policy mechanisms regarding efficiency and quality. 2.41 Another reason to be careful about raising the floor of FUNDEF is provided in an insightful analysis reported in "Descentralizqdo da Educaqdo Fundamental: avaliaqdo de resultados do FUNDEF', by Marcos Mendes, Working Paper, Economics Consultant to the Senate, Brasilia, 2001. Mendes carries out an analysis comparing cases of fraud and possible misuse of FUNDEF resources. Using the "natural experiment" offered by the presence of North East states that received or did not receive federal complementary transfers, Mendes reaches an extremely interesting conclusion. When the FUNDEF basket is restricted to shared resources of the state and municipalities, there is a greater vigilance regarding the distribution and application of FUNDEF resources. This is because FUNDEF is a zero-sum game, and the greater appropriation of resources by one municipality means correspondingly fewer resources for the other contributors in the state. On the other hand, as regards federal transfers, the gain of one municipality does not come at the expense of the others, as the transfers are a windfall gain, and Mendes find evidence of an increased tendency towards misuse of resources. This line of enquiry needs to be investigated in further research. 13 The legal agreement establishes that the Bank will not release the data to outside investigators. While legalities surrounding the data collection effort may justify the inability of the government to release the past data, we strongly recommend that future data collection and dissemination regarding public accounts and other matters of eminent public interest should be carried out in open collaboration with Brazilian academic researchers. 38 2.42 One place where there is certain evidence in favor of increasing additional resources is the issue of local revenue mobilization. Associated with the Fiscal Responsibility Law (LRF), the Federal Ministry of Finance and the National Development Bank (BNDES) have initiated programs to help strengthen tax and expenditure administration in municipalities.'4 Though it is difficult to quantify the revenue gap, it has been suggested that administrative bottlenecks and the lack of institutional capacity rather than tax delinquency constrains municipal tax effort (see "Brazil: An Evolving Federation", by Jose Roberto R. Afonso and Luiz de Mello, Working Paper presented at the IMF/FAD Seminar on Decentralization, Washington, D.C., November, 2000). The authors cite studies showing that the gap between the potential and effective revenue collection in Brazilian municipalities is in the neighborhood of 20 percent. Our review of municipal educational financing mechanisms has shown that municipalities receive significant amount of revenues from transfers, and that there does not exist a mechanism such as the requirement for matching own contributions from municipalities. Introducing such measures would complement the efforts to improve tax collection in municipalities, with consequent implications on educational expenditures. 14 These include the Program to Modernize Municipal Revenue Administration (PNAFEM) of the Ministry of Finance and the Program to Modernize Local Revenue Administration and Social Sectors (PMAT) of the BNDES. Studies have shown effectiveness in such interventions to raise municipal revenues. A highly successful example is the case of a program for the municipality of Rio de Janeiro, with the eighth largest sub-national budget in Brazil. 39 Section 3: Management of Resources: Composition of Expenditures A. Overview 3.1 Section 2 of this report has provided wide-ranging and conclusive evidence regarding the fact that municipalities in Brazil indeed did receive substantial additional resources for education. A key concern that has motivated this study is to examine whether municipalities applied those resources towards productive ends. Before turning to answer this question, it is useful to point out some data related issues. The diversity of municipalities in Brazil is only compounded when it comes to the diversity of accounting practices and the public reporting of those accounts. It is only with the LRF that municipalities have received clear-cut guidelines regarding the reporting of accounts, but the ready availability of consolidated public accounts at the level of expenditures within a sector such as education is still a few years away. We rely on the FIPE database for 2000 that collected detailed data on expenditures on primary education and FUNDEF expenditures for the Year 1999, in a carefully chosen sample of 267 municipalities that is representative for all of Brazil. Even though more recent data is not available, 1999 represents two years since the implementation of FUNDEF, and would be sufficient to draw out relevant policy lessons. A ready policy conclusion already emerges from the difficulty of accessing the data - the imperative for municipalities to follow uniform accounting practices and for the federal Ministry of Education to actively collaborate with the federal Ministry of Finance and BNDES in their efforts to modernize municipal budgeting practices. 15 3.2 Appendix Tables for this section provide a detailed overview of the composition of expenditures. Appendix Table 3.1 details a Brazil wide comparison between state and municipal primary education expenditures. Appendix Tables 3.2a and 3.2b provide a break up by region and Appendix Table 3.3 provides a comparison by municipal size. Finally, as one of the issues under investigation is the constraining nature of the restrictions placed on FUNDEF regarding the composition of expenditures, Appendix Table 3.4 provides details regarding the composition of municipal resources only from FUNDEF, and Appendix Table 3.5 provides a break- up by region. We avoid a potentially tedious commentary in the text on the data presented in the Appendix, though the tables serve as useful tools of reference. We instead focus in the main text on key policy issues regarding the composition of expenditures. We attempt to answer the complex questions surrounding the issue of the quality of municipal expenditures by looking at various items of expenditure, beginning with capital expenditures. 15 It bears reiterating that the data described and analyzed in this section represents the first time in Brazil that municipal education expenditures have been presented at this level of detail. The lack of previous work on this issue is due to the complications facing most researchers in getting access to data. Brazil's very modern and progressive public policy regarding the transparency of public expenditure data at the level of the federal government and a number of state governments needs to be extended to municipal levels of government. 40 B. Capital Expenditures 3.3 Capital expenditures accounted for 8.84% of all primary education expenditures for municipalities in 1999. Capital expenditures increased from 6.87% of all primary education expenditures in 1997, the year before the implementation of FUNDEF. Capital expenditure for all levels of education combined was 7.39% in 1997, but the corresponding figure for 1999 is not available. The limitation of our analysis to primary education is not such an important issue for pre-school education or adult education, since both facilities and often teachers are mostly shared across these levels. However, it is an issue regarding education in creches for children 0-4 years of age. We do know from our visits to municipalities that there has been a significant amount of construction of creches in a number of municipalities, especially the more prosperous ones, but unfortunately we lack hard quantitative data to further investigate this issue. 3.4 The issue of capital expenditures is important from the policy perspective for a number of reasons. Prior to FUNDEF, there was a tendency for municipalities to overspend on construction projects, because such expenditures were politically favorable to mayors (Mendes, 2001). At the aggregate level, between 1996 and 2000, the number of schools in Brazil has stayed more or less constant, signifying the impact of reforms since that time. Apart from FUNDEF, the Fiscal Responsibility Law (LRF) may also have put a brake on wasteful capital expenditures. The LRF enjoins stringent restrictions on expenditures in the final term of the administration of mayors, who changed with municipal elections in 2000, and are elected for 4-year terms.'6 The LRF also put in place the need for medium terrn planning for municipal capital expenditures, at the same time as it restricts municipality expenditures on Personnel. The LRF also initiated rewards and penalties for improved budget and expenditure management, with a stress on transparency and civic involvement, and federal investment in training projects to improve municipal resource management. In the period up to the initiation of the reforms, the administrators stood to gain politically for the standard pork barreling purposes. In the current policy environment, there is a greater emphasis on good governance and provision of services that leads to prudent management in areas such as the capital budget. 3.5 As capital expenditures are lumpy investments, it is difficult to draw conclusive lessons from one or two years of data. However, the presence of cross-sectional variation in expenditures allows some inferences to be made. We seek to determine whether the incentives introduced by policy reforms had an impact on allocation decisions by municipalities. An indicator of the availability of physical resources is 16 A detailed exposition of the LRF and its implications regarding municipalities is provided in "Lei de Responsibilidade Fiscal: Principais Aspectos Concernentes aos Municipios", by Weder de Oliviera (Camara dos Deputados, Brasilia, 2000). The penalties of the LRF are not limited to indemnification of municipalities regarding access to federal transfers or access to credit. The limits extend to criminal liabilities of individuals occupying public posts. In our interviews with mayors and municipal secretaries, quite a few mentioned the LRF as a motivator for instituting multi-year capital budgets. 41 the class size as measured by the number of students. Information is not available regarding the actual floor area of classrooms, but a larger class size can be interpreted as a tendency towards more crowding of classrooms. Even though physical class size may vary from one school to another, or even within the same school, it seems plausible to assume a random distribution of such classes across the student population, so that larger enrollment in classes on the average translates to more crowded classrooms. If the policy reforms created incentives for better service delivery, we should expect to see greater expenditures for classroom construction in municipalities that faced higher class-sizes because of historically determined availability of classroom space. Conversely, municipalities that had relatively smaller class sizes would invest relatively smaller amounts in increasing classroom space if the municipalities were accountable to make expenditure decisions according to educational needs. 3.6 Another measure of classroom size is also related to the capital expenditure needs of the municipality. The population distribution of students is typically uneven across a municipality, and good municipal planning and resource management can be determined by examining the variation in class-size across different schools in the same municipality. The assumption here is that efficient resource managers would be able to reduce imbalances in class-size. As with the mean class size, it is not a perfect measure of good governance, but a higher standard deviation would indicate poorer planning and resource management by the municipal secretariat. In fact, quite apart from efficiency considerations, the presence of wide disparities in class size is also a poor signal of the concern of the municipality regarding equity, as class size would tend to be associated with school quality. Of course there is more to school quality than class size, but the presence of crowded schools in poor neighborhoods and relatively more spacious schools for children of the better off definitely does not improve the quality of schooling experience of the poor. 3.7 Given that the same student population is shared by municipal and state systems alike in the same municipality, it is also useful to compare the class sizes between the two systems in a given municipality. An issue of policy interest is the question of co-operation between the state and municipality in the provision of educational services. FUNDEF creates incentives for each of the systems to vie for student enrollments, but if states are reluctant to handover schools to municipalities, and there is a lack of co-operation between the two providers, this would manifest itself in greater imbalances in class sizes across the two systems. Municipalities with available resources would go in for construction of more classrooms and possibly even schools at the same time as state schools function with half-empty classrooms, as municipalities would not wish up to give up students to state systems. In the absence of time to carry out a locality wise analysis, we restrict ourselves to compare mean class sizes between the two systems within the municipality. 3.8 A simple correlation exercise was carried out between measures of Capital expenditures and measures of the mean and standard deviation of class sizes. A positive correlation would indicate that municipalities are taking allocation 42 decisions based on technical criteria of the needs of the educational system rather than being motivated by political concerns. The panel below represents the findings for all municipalities as a group, and separately for gainers and losers from FUNDEF. Table 3.1: Correlationi between Cpital Exoenditures andLMuiiicipal Cli~ss Size'(All Muriicipalities) Mean I Correlaon Coefficient Absolute Level of K-spending K-Spending as % of Total Spending _ Basic Education FUNDEF Basic Education FUNDEF Class Size in 1996 Mean Grades 1-4 26.69 0.18 Mean Grades 5-8 31.05 0.16 . Std. Deviation Gr 1-4 5.79 0.13 0.17 Std. Deviation Gr 5-8 6.17 ** * * Class Size in 2000 ,Mean Grades 1-4 27.101 * 0.12 lMean Grades 5-8 31.971. Table 3.2:' Correlation between ,apital 6 Expediture's7'a'nd' MutnicipaI Clasi SSke(Gainaners) . Mean Correiation Coefficient Absolute Level of K-spending K-Spending as % of Total Spending Basic Education FUNDEF Basic Education FUNDEF Class Size in 1996 Mean Grades 1-4 27.56 0.18 Mean Grades 5-8 31.70 Std. Deviation Gr 1-4 5.96 _ * Std. Deviation Gr 5-8 6.36 Class Size in 2000 X Mean Grades 1-4 27.66 * Mean Grades 5-8 32.361 Table 3.3: Correlation'between Capital Expen'ditures anidMunicipal Clsise.,Size (Losers) Mean Correlation Coefficient Absolute Level ot K-spending K-Spending as , of Total Spending Basic Education FUNDEF Basic Education FUNDEF Class Size in 1996 Mean Grades 1-4 23.57 0.45 0.46 Mean Grades 5-8 28.51 0.38 0.42 * Std. Deviation Gr 1-4 5.13 * 0.30 0 46 Std. Deviation Gr 5-8 5.18 * * * Class Size in 2000 Mean Grades 1-4 25.38 0.48 0.55 0.30 0.28 Mean Grades 5-8 30.34 0.46 0.60 0.27 0.33 Note: Only Statistically Signiticant Coefficients (at minimum 5% level) shown ^ denotes the correlation was statistically insignificantly different from zero Source: FIPE Database 2000 3.9 The panel of Tables 3.1 to 3.3 shows mixed evidence, but it generally favors the notion that municipal capital expenditures are related to system needs. The empty cells are ones where the correlation coefficient is not shown, as it was not statistically significant at the 5% level. None of the various pairs of correlation coefficients had a negative value, lending credence to the hypothesis that municipalities took responsible allocation decisions. The tables show that the correlation was stronger amongst municipalities that were 'losers' to FUNDEF, or those that contributed more to FUNDEF as compared to what they received from 43 FUNDEF. This marked effect shows the incentive effect of FUNDEF. The losing municipalities would presumably wish to become gaining municipalities by enrolling more students. However, enrolling students requires that places be made available for them, and the more crowded the existing infrastructure, the greater the need to add classroom space. It is also to useful to note from Table 3.3 that the effects are similar whether we consider all the resources available for primary education or only the resources obtained from FUNDEF. It would appear that the FIJNDEF restriction to spend a minimum of 60% of the funds on the salaries of teachers does not create strong distortions in expenditure allocations. 3.10 Comparing the deviation in mean classroom size between state and municipal systems by municipality, we find evidence of convergence between the two. The mean classroom size for municipalities has usually been smaller than that of states, but the difference has diminished between 1996 and 2000 as shown in Table 3.4 below. The table shows that the pace of convergence varies across the regions and cycle of primary education (Grades 1 to 4 or Grades 5 to 8). The tendency for municipalities to have larger class sizes is related to the increase in pace of municipalization, as municipalities move beyond the responsibility for smaller rural schools to bigger schools in urban areas. The table does not show the number of schools, but the period between 1996 and 2000 saw a slight decrease in the number of municipal schools from 131,400 to 129,600, at the same time as state schools fell from 42,000 to 33,000, pointing to a process of consolidation of schools. Since every school represents variable costs in addition to the payment of personnel, these changes point to a more favorable management of public resources. TabIe'63.4: Converi,nce in Class'Sizebetween State and Municipal Systems Difference Municipal over State Class Size Grades 1_4 Grades 5-8 1996 2000 1996 2000 All Brazil -4.10 -1.93 -4.35 -3.35 North -1.55 -1_.86 -2.74_ -4.48 North-East -5.41 -1.90 -5.02 -3.88 Centre-West -4.31 -0.48 -3.94 -2.73 South -3.21 -2.45 -4.41 -2.14 South-East -4.28 -2.28 -4.20 -3.06 Source: INEP School Census, 1996 and 2000 3.11 Capital expenditures are incurred for purposes other than classroom or school construction. In particular, the quality of educational services is related to the provision of adequate equipment. If municipalities indeed face greater accountability for service delivery, there ought to be evidence of pressure on them to provide adequate equipment for schools. The INEP school census for 2000 recorded information about the availability of certain kinds of equipment for pedagogical support. Information is available about 5 kinds of inputs - school libraries, school laboratories, computers, television and video equipment. About 8% of municipal schools and 48% of state schools have a computer. Municipal schools in the South and South East are much better equipped in terms of computers. Some 44 municipal secretaries in the state of Sao Paulo speak with pride about the computers that their children have access to - one even told us of virtual reality interactive devices that his municipality was acquiring for the students "estao comprando uns equipamentos para a escola de ultima geraCdo (E um tipo de computador, mas o aluno manipula um objeto concreto que aparece na tela....)907. 3.12 The INEP data does not provide us with details regarding the make of the computer or the use made of it, but it is clear that there are extreme regional disparities. The reforms of the LDB and FUNDEF have probably eliminated wasteful expenditures in the wealthier municipalities of Sao Paulo, so we did not get to hear examples such as the school band being sent to Europe to meet the constitutional earmarking requirement regarding educational expenditures. However, quite apart from FUNDEF, the more prosperous municipalities have enough resources of their own to invest heavily in expenditures allowed as education expenditures. As North East municipalities are more dependent on FUNDEF, and FUNDEF allowances per student are two or three times less than those in the South East, North East municipalities remain very distant in provision of equipment. Given Brazil's federal structure and unequal regional development, it seems unlikely that there would ever be a time when all public schools in Brazil offer education of equal quality. However, computers do not make a school, and we will see later in the report that there has been quite a degree of convergence across regions regarding other inputs, such as teachers and pedagogical support personnel for schools. 3.13 Next to computers comes the provision of Television and Video, which is at a higher level than for computers. The availability of TV equipment is owed to the federally sponsored TV Escola program that provides an audio-visual kit to every Brazilian public school with more than 100 students, though municipalities clearly have made their own investment. Unfortunately, the situation regarding laboratories and libraries is not so good. Only 1.5% of municipal primary education schools report having laboratories, and even that average number comes from a much higher 5% in the South East (only 0.09% of schools in the North-East had a laboratory). It is difficult to imagine how students can learn about science without even a basic laboratory with a stock of materials to help children conduct simple experiments. It does not seem that there are pedagogical reasons why schools do not need to have laboratories - if this had been the case, we would not have seen 37% of state schools in the South (even higher in some states) equipped with a laboratory. The picture is only slightly better in the case of libraries, with only 11% of municipal schools possessing libraries (4% in the North East, 28% in the South-East). Table 3.5 shows the percentage of schools with the different kinds of equipment, comparing across regions and state and municipal systems. 17 "(we are buying) school equipment of the most upto date technology - it is a type of computer, but the student manipulates a physical object that appears on the screen" 45 Table 3.5: Equipment for Schools Providing Basic Education | |Percentage of all Schools that Possess Equipment T Municipal Computer TV Video Library Laboratory North 1.92% 10.70% 9.76% 4.46% 0.09% North East 1.46% 16.28% 14.91% 3.89% 0.16% Centre West 14.65% 37.11% 35.89% 13.59% 1.29% South East 26.46% 50.78% 49.33% 28.36% 5.57% South 25.17% 47.51% 45.80% 29.62% 4.92% All Brazil 8.20% 24.55% 23.21% 10.65% 1.49% State Computer TV Video Library Laboratory North 17.06% 55.49% 52.78% 32.94% 2.39% North East 22.52% 77.24% 73.90% 33.56% 3.66% Centre West 54.24% 96.12% 94.41% 52.91% 8.70% South East 62.53%° 84.60% 83.28% 61.61% 23.98% South 69.60%j 91.09% 89.83%k 76.63% 36.77% All Brazil 47.63%1 81.25% 79.24%° 53.11% 17.37% Source: INEP School Census 2000 3.14 The analysis of capital expenditures using quantitative averages could arise simply out of a pattern established by only a portion of well-performing municipalities. Indeed, even the quantitative evidence is mixed, generally positive regarding infrastructure, and somewhat negative regarding equipment. Our qualitative interviews in 5 selected states indicated problems about capital expenditures on the part of municipal secretaries in the North East states of Paraiba and Pemambuco as well as in the state of Rio de Janeiro, though it was not mentioned in our discussions in Parana and Sao Paulo. There continue to exist many schools, mostly, but not exclusively in rural areas, that are in terrible conditions. They are characterized by poor lighting, lack of sanitation, and absence of protection from the elements such as rain and flooding.'8 The availability of financing has created conditions that make it possible for municipalities to be able to take action to redress such problems regarding infrastructure, but the resources need to be backed up by the adoption of a set of minimum operational standards to ensure that all students can depend at least on a basic minimum regarding physical infrastructure. The variation in the performance of municipalities is illustrated below with an interesting story from field work done for this study by study team members. 3.15 One of the municipalities in our sample was the small North East municipality of Sao Jose dos Cordeiros, (Population 4,136) in the interior of the state of Paraiba. The municipality had 921 students enrolled in primary education: 496 of whom were in the state system, and 425 in the municipal system. There do not exist any school principals in the municipal system as the schools are run directly by the municipal secretariat, a fact that seems intriguing, until we find out that there were 18 The study "Brazil: A Call to Action: Combating School Failure in the North East of Brazil" has described similar conditions dating to 1996. While much has changed since that time, it is clear that such conditions persist even to today in some schools, in spite of all the reforms that have been put into effect. 46 30 municipal schools (all rural) in the municipal system, with an average of 15 students per school, and some of the "schools" enrolling only 5 or 6 students.,9 The state system students are all in one big urban school of grades 1 through 11, that enrolls 496 students, in the municipal headquarters. The municipal secretary of education did not know how much of resources the municipality got from FUNDEF, leave alone how much was allotted to capital expenditures, and was unable to discuss policy options regarding possible school construction or consolidation. All the resources for education were managed out of the mayor's office, and though there did exist a FUNDEF council as mandated by the law, complete with a representative of parents and other representatives, they did not have much of an answer either, because the council had never held a meeting since it was constituted. 3.16 Switching to the municipality of Sao Gon,alo (Population 891,119) on the periphery of the city of Rio de Janeiro, we found a rather different story. The municipality had a municipal enrollment of 41,930 students and a state enrollment of 66,670 students. The municipality did not have the issue of very small schools, but there was still a problem regarding school construction. The lack of school spaces in municipal systems had implied that there had been about 8,000 students who had sought enrollment in a municipal schools, but had to be turned away to state schools, where they enrolled, taking about R$ 6 million of FUNDEF resources with them. State schools in the municipality of Sao Goncalo have substantial over- capacity in terms of both teachers and school spaces, so the State probably did not have to make any additional investments to attend to these 8,000 students. Data is not available to determine how much of the R$ 6 million was allocated by the State secretariat to the municipality of Sao Gonqalo. 3.17 When we talked to the municipal secretary of education in Sao Gon,alo, he mentioned a number of policy options regarding school construction - the municipality could look for an arrangement to take charge of state schools, a lot of which were literally empty; the municipality could look for an innovative arrangement with private schools, and the municipality could construct schools of its own at a total cost over 4 years of some R$46 million. Politically, these options could probably be arranged in descending order of difficulty. In spite of the excellent relations between the municipality and the state government, the most difficult option would likely be the first one, the transfer of state schools, such as the 30 schools in the municipality with a capacity for 2,000 students each, constructed some years back in a state program of model schools designed by the internationally renowned Brazilian architect, Oscar Niemeyer. Whichever of the options or combination of options were to be eventually chosen, the municipality of Sao Gon,alo was lucky to count with a strong leadership and technical capacity to manage its fiscal responsibilities without compromising the delivery of school places to its students. 19 The absence of school principals was also reported in "Call to Action", which showed evidence that the absence of principals or head teachers is linked to teacher absentism and poor academic results. 47 3.18 There is a danger in jumping to conclusions from anecdotal accounts such as the one presented above. There are quite a few schools in the state of Rio de Janeiro that come close to the description in Paraiba. In fact, there exist extremely small schools without amenities even in urban areas in some municipalities. We also visited municipalities similar in scale to Sao Jose dos Cordeiros, such as the municipality of Mataraca (Population 5,500) in Paraiba, where effective solutions had been found, including the establishment of a transportation system to ferry children to consolidated schools where they were assured of a basic quality of infrastructure. These variations are in themselves a policy lesson regarding 'positive deviants' that we will discuss again in this report. 3.19 These stories, and the accompanying quantitative analysis of class-size provide an insight into the role played by incentives, and the institutional capacity needed to capitalize on those incentives. The incentive created by FUNDEF to enroll students does lead municipalities such as Sao Goncalo to look for ways to enroll more students. This is a desirable policy outcome, because there is reason to believe that municipalization in general leads to better public service delivery. In the absence of such an incentive, with clear economic stakes, it seems difficult to believe that Sao Gon,alo would have sought pro-actively to increase enrollments. Yet, such an incentive is not enough to ensure outcomes. In the case of Sao Gon,alo, a dynamic political leadership may find creative ways to enroll new students. Yet, institutionally, such a municipality in general would find it tough to reach an agreement with the corresponding state government. The current policy regime does not include policy incentives for state governments to cooperate with municipal governments. In the case of the forsaken schools of Sao Jose dos Cordeiros, we find further evidence regarding incentives. Clearly, there was an incentive for the municipality to enroll additional students, which is why the municipality did not close down some of those schools with 5 or 7 students, though whether the children would benefit from that enrollment is a moot point. Incentives may be critical, but we find in general that the incentives were not complete. 3.20 Even if we assume that the incentives could be set right by finding policy mechanisms designed to make different levels of government to collaborate with one another, there still remains the matter of institutional capacity. It takes a certain amount of technical expertise to make a multi-year plan of investments in physical infrastructure, which is only one of the responsibilities of a municipal education system. However, it would be difficult for many municipalities except perhaps the largest ones such as the State Capitals, to be able to put together a team of qualified professionals to perform the activities needed to run a school system effectively. Our research leads us to believe that it would be naive to imagine that some kind of training program for municipal secretaries of education would be sufficient to overcome the constraints of institutional capacity. The problems here are rather similar to the problems facing decentralization efforts in the health sector, and reported in ongoing Bank sector work. For example, there are considerable economies of scale in the provision of complex-care by hospital facilities - every municipality has patients, but there cannot be a hospital in every municipality. This 48 simple fact leads to institutional arrangements for municipalities to come together in micro-regions to provide hospital care. No such collaboration as a matter of general policy is seen in the education sector, though there are examples of successful programs in some states, as well as the federally managed Fundescola program 3.21 The presence of a well functioning system in Mataraca, the other municipality in Parai'ba, can be related as well to institutional features. One marked difference between the two municipalities with opposing indicators of service delivery was the independence of the municipal secretariat of education, and the strong role played by the community at large. The establishment of a functioning municipal education secretariat which relies on community participation requires that the secretariat be given the autonomy to control its own resources. FUNDEF has changed the scenario regarding the relative power of educational secretaries, but the phenomenon of efficient secretariats, accountable to the education community that they serve has not been universal. We also find that legal mandates are not enough to lead to the creation of adequate institutional arrangements. FUNDEF includes a legal mandate for community councils to oversee the use of funds, but there are many cases where the ultimate objective is easily subverted by the creation of councils that practically exist only on paper because they do not hold meetings. The policy imperative in Brazil is a) to build on the initial steps of the FUNDEF reforms and amplify the enrollment incentives to include educational quality and efficiency; and b) to realize that incentives may be necessary, but they are not sufficient to lead to an automatic generation of institutional capacity to take advantage of those incentives. 49 C. Recurrent Expenditures 3.22 Three important policy relevant questions can be raised regarding recurrent municipal expenditures on education -a) the issue of personnel expenditures on administration as compared to expenditures on teachers in the classroom; b) expenditures on training of teachers; and c) non-personnel recurrent expenditures. In addition to these three general issues, two further issues are of importance in the Brazilian context - d) the issue of transport expenses, important when some schools stop at Grade 5 and children need access to another school for higher grades of primary education as well as for secondary education, an issue that is complicated by the fact that most of those higher grades are provided by state schools rather than municipal schools; and e) the issue of pension payments to retired teachers (termed inativos, the Portuguese plural word for "inactive") from the general budget, permissible for general revenues but not allowed for FUNDEF resources. 3.23 Excessive administrative expenses are a ready signal of poor governance - it is politically profitable to hire a lot of administrative personnel, especially those who can be employed at the will of the mayor, whether in the municipal secretariat or at the school level, but every dollar for administrative expenses implies a dollar less to pay for more teachers or more qualified teachers. At the same time, research on educational policy suggests the need for support mechanisms to encourage effective school administration and adequate teaching and learning practices - too high the proportion allocated for the teachers' wage bill would indicate a neglect of these important systemic requirements. Clearly, there is an optimal level of administrative expenses, though the optimal level would vary considerably from one system to another. In the face of the sheer size of some of the state education systems, it is possible that the increased focus on municipalities associated with the recent reforms may have led to reduction in some of the most wasteful expenditures on a complex bureaucratic apparatus. The question of a training budget is important because even if higher teacher salaries were able to attract the best possible teachers, their effectiveness would be curtailed with lack of training. Finally, the issue of non-personnel recurrent expenditures is important for similar reasons regarding governance. Unlike the case of capital expenditures or expenditures on the teacher wage bill, non-personnel recurrent expenditures would tend to have a more direct impact on educational quality in the classroom rather than on the political influence of the stewards of the system. Of course, if the political incentives facing these stewards were aligned to the needs of the educational system, such a trade-off would not need to be made. As with other issues treated in this report, the determination of the incentives faced by the agents in the system is an empirical matter. 3.24 The issue of training of teachers forms a part of the package of policy incentives regarding teachers, a topic which deserves closer attention and is developed in another section of this report. This subsection deals with the issue of administrative expenses, non-personnel recurrent expenses and the payment of inativos. The issue of transportation expenses is further elaborated in a consequent sub-section. 50 3.25 Personnel expenses formed 55% of all municipal expenditures for primary education, and they contributed 75% of all FUNDEF expenditures. FUNDEF puts the restriction that 60% of resources be spent on personnel engaged actively in the delivery of education, which thus excludes the inativos, but permits the use of funds for paying administrative and support personnel. The fact that a much higher percentage of FUNDEF resources are spent on personnel expenditures suggests that the 60% limitation is not a binding constraint to municipalities. This is an important finding considering the worry about taking away the flexibility of local decision- making at the time the legislation was introduced (for instance, see Draibe, 1998). Whereas the legislation introduced the proviso regarding 60% of expenditures on salaries to ensure that the additional funds from FUJNDEF would in fact be used by municipalities to hire more teachers and support personnel or -to pay them higher salaries, there had been a danger that some municipalities might be forced to make allocation choices not congruent with their preferences. However, the data show that municipalities were able to use other available resources for non-personnel expenditures. Table 3.6 provides the proportions of personnel expenditures by region, separately for all Primary education resources considered together, and separately considering only the FUNDEF resources. The table also provides a break-up by municipalities that were net gainers or losers to FUNDEF. Tabie _306 P*ersonnel Expen'ditures-aAs% of Total'Experidituresi(1-999) All Basic Education Funds FUNDEF Resources Gainers Losers Gainers Losers Municipal North 59.40% 45.11% 75.56% 66.27% North East 49.60% 39.67% 64.82% 79.94% Centre West 37.41% 36.49% 69.98% 92.18% South East 48.25% 36.20% 71.84% 81.77% South 57.20% 52.16% 86.79% 81.81% All Brazil 51.30% 40.87% 72.09% 81.99% State North 68.44% 81.47% North East 57.81% 82.93% Centre West 66.63% 76.49% South East 57.71% 77.79% ,South 67.15% 80.84% _AIl Brazil 62.57% r 80.84% Source: FIPE 2000 3.26 Table 3.6 shows that States spent a much higher 63% on personnel expenditures as compared to municipalities that spent 48% on the average of all primary education resources. The table does not show the corresponding figures for 1997, but the figures for 1997 do not indicate any major shifts in resource allocation. It is interesting to note that municipalities that were net losers to FUNDEF allocated lower percentages to personnel expenditures out of general resources, but allocated higher percentages out of FUNDEF resources. There do not exist major variations by region, with some exceptions. In particular, the gainers amongst North East municipalities spent 65% of their FUNDEF resources, much closer to the floor of 51 60% as compared to gainers in the South, who spent nearly 87% on personnel expenditures. This finding confirms the earlier result that schools in the South East are relatively well-equipped so that they can afford to devote more resources to non- personnel expenditures. 3.27 Turning to personnel expenditure on administrative staff as a percentage of personnel expenditure, we find patterns that mirror the findings regarding personnel expenditures as a percentage of total expenditures. There is a slightly higher percentage of administrative expenditure out of general funds as compared to FUNDEF resources, and there are not any marked regional patterns. On the average, municipalities spent 30% out of general primary education resources on personnel other than teachers in the classroom, and 17% out of FUNDEF resources for the same purpose. Table 3.7 shows the disaggregated figures. Municipalities that gained from FUNDEF tended to spend less on administrative resources as compared to losing municipalities. The pattern is similar for general resources as well as those from FUNDEF, though the difference between gainers and losers is smaller for FUNDEF resources. Both Table 3.6 and Table 3.7 indicate that the FUNDEF restrictions do not appear to have caused major distortions in decisions regarding the allocation of expenditures. Table 3.7: Administrative Expenditures as % of Personnel Expenditures (1999) 7All Basic Education Funds FUNDEF Resources Gainers Losers Gainers Losers Municipal North 32.06% 40.23%k 26.54% 39.30% North East 25.63% 39.25%/ 22.20% 29.85% Centre West 37.23% 25.37% 28.00% 8.56% South East 33.04% 31.74% 15.07% 13.74% South 24.02% 44.48% 7.68% 11.93% All Brazil 27.88% 35.22% 18.38% 14.66% S ta te_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ North 24.91% 24.91% North East 25.22% 20.02%/C Centre West 19.39% 12.55% South East 18.17% 8.10% South 25.20%1 18.39% Ajl Brazil 23.45%1 ° 18.51%° Source: FIPE 2000 3.28 The question of whether administrative expenses are too high or too low is difficult to answer in absolute terms because the need for administrative expenditures depends on the context facing each municipality. One way to calibrate the percentage of administrative expenditures is to use comparisons across the municipalities, and relate the expenditures to desirable outcomes of the educational system. One of the key indicators of the effectiveness and internal efficiency of the educational system is the pass rate of students. Even though it is a somewhat imperfect measure of educational quality as high pass rates may be a result of lax learning standards rather than better pedagogical and administrative management, it 52 is a useful measure because of its ready availability at every level of the system. The data for Brazilian municipalities shows a high level of correlation between the level of administrative expenses per student and the pass rate, though the relation is non-linear, as shown in Chart 3.1. Chart 3.1: Municipal Administrative Expenses per Student and Student Pass Rate 100. * *. ; + 90 , ' . 60 **S a * C +~~~~~C * C C 80 40 - 0 100 200 300 400 500 600 700 800 900 Admin. Exp. per Student (RS per year) 3.29 Chart 3.1 shows that higher administrative expenses are associated with better student approval rates, with the concavity showing that the association peters out at higher levels of per student expenditure. The chart shows that municipalities must be doing something right with the resources allocated to administrative expenses, though it does not necessarily show a causal relationship. More prosperous municipalities may be able to afford a higher level of administrative expenditure, and at the same time, benefit from students from a more favorable family background, and thus show higher pass rates. Also, expenses on other items of allocation would be correlated with administrative expenses, thus the municipalities administrative expenses may be only be a part of the reason for better performance. In fact, Chart 3.1 does not provide great insight into the quality of municipal administrative expenses, as the simple bivariate relationship does not provide information about the process used by the municipality to translate administrative expenses into results such as better pass rates. Before exploring patterns of causality, it would be useful at this stage to provide a comparison of municipal administrative expenses with State government administrative expenses. 3.30 Enrollment at the municipal level for primary education increased from 33% in 1996 to 49% in 2001, but States are still heavily involved in the provision of education at this level. The 15 million students in State systems are distributed 53 across 26 states (excluding the federal district), while the 12 million municipal students are enrolled across more than 5000 municipal education systems. State systems are huge as compared to municipal by many orders of magnitude. Even though there do exist some large municipal systems, such as the ones for the cities of Rio de Janeiro and Sao Paulo, State systems are big in size and dispersed over much wider geographic regions than even the biggest municipal systems. It is of empirical interest to determine whether the size of state systems is a positive factor or a hindrance to better service delivery. Brazilian policy makers over the decades seem to have taken the latter view, witness the many attempts towards municipalization, of which FUNDEF is the most recent and more successful example. 3.31 State governments do not maintain records about the level of expenditures in the state system across the municipalities in the state. Though a number of the states have intermediary regional offices to help manage systems that enroll millions of students, and regional expenditure statistics may be available for some states, the general situation is one where such information at the municipal level is not compiled. We make the assumption that state expenditures are evenly distributed over the state. Based on this assumption, Chart 3.2 shows the relation between administrative expenses and student pass rates at the municipal level but in state systems of education. Unlike in the case of municipalities, the correlation between the two variables is zero. Chart 3.2: State Administrative Expenses per Student and Student Pass Rate 110 100 40 L 1-0g - . 70 : 40 30 - . * 20 - o 50 100 150 200 250 300 350 Admin. Exp. per Student (R$ per year) 54 3.32 As stated before, the correlation between per student municipal administrative expenditures and higher student pass rates is not an indication of causality, and further analysis is required in order to uncover patterns of causality. The approach used here is to use the empirical data from the municipalities themselves to determine the efficiency in the use of resources. Pass rates varied from a low of 40% to a high of 98%. At the same time, there was a wide variation in the administrative expenditure per student - 25% of the municipalities spent less than R$68 on administrative expenses; at the other end of the distribution, 25% of the municipalities spent amounts in excess of R$215 per student. Amongst the group of municipalities, clearly there would be some municipalities that were able to obtain high pass rates for their student with low expenditure allocations - these municipalities define the frontier of efficiency. We use the technique of Data Envelopment Analysis (DEA) to estimate the efficiency frontier, and then determine efficiency scores for the municipalities that were below the frontier, with the municipalities on the frontier indexed at an efficiency score of 100.20 3.33 We use the pass rate of municipal schools in the municipality as a measure of the output of the educational system. We consider three inputs into the production of this output - the amounts of administrative expenses, expenses on teachers, and capital expenditures per student. As external factors that wodld influence the need for these expenditures, we consider the per capita income of the municipality and the geographical area of the municipality. It is important to note that the DEA technique is completely different from the regression approach - a regression model applied to this case would seek to find the average conditional relationship between the inputs and the output. Outlying municipalities, such as highly efficient ones that are able to extract much better results for the same level of inputs, or highly inefficient municipalities would tend to dilute the regression results, leading to statistically insignificant correlations. DEA, on the other hand, relies precisely on the outlying 'super-performing' municipalities to define the shape of the level of efficiency that is possible. The reliance of the DEA technique on these super- performers is both a strength and a weakness. On the one hand, the method makes the common sense assumption that if a particular municipality is able to get superior results than another, the fault for the poor performing municipality lies in its own processes of decision making and policy implementation. This is really a powerful 20 DEA is a technique with a long history of use in the world of business, where managers need to identify the efficiency of different units. The technique has also been used intensively to determine efficiency of hospitals and schools as well as the efficiency of public expenditures. It is outside the scope of this report to provide a detailed description of the technique. An excellent theoretical development with empirical examples is available in "The Measurement of Productive Efficiency: Techniques and Applications" by Harold 0. Fried, C.A. Knox Lovell and Shelton Schmidt (Eds.), Oxford, New York, 1993. An application in the context of municipalities in Brazil is a paper that compared DEA with alternative frontier estimation techniques such as FDH (Free Disposable Hull). The difference between DEA and FDH lies in assumptions regarding the shape of the Production Possibilities frontier -FDH assumes that if two municipalities are on the production frontier, hypothetical municipalities using a convex combination of inputs used by the two municipalities also constitute a technically feasible input vector. The DEA technique used in this paper only considers real municipalities as constituting the production frontier. Further research should explore the robustness of the findings presented here to the choice of technique of frontier estimation. 55 assumption, because it does not rely on any externally imposed criteria of efficiency. As a drawback of the method, it imposes stringent conditions on all municipalities in the group, as the presence of a small number of highly efficient municipalities form the benchmark for the entire group - this results in some municipalities receiving very low efficiency scores. One way to increase the discriminating power of the DEA technique is to identify groups of municipalities in clusters that are as homogenous as is possible to identify with available data. In the current application, we identify such groups by the size of the municipal population, under the premise that municipalities of similar size face similar constraints not measured by the variables considered as inputs into the production of pass rates for students. 3.34 Municipalities from the FIPE 2000 dataset used for the analysis were grouped into 5 population categories - very small municipalities (pop. < 20,000); small (pop. 20,000 -50,000); medium (pop. 50,000 - 100,000); large (pop. 100,000 to 500,000), and very large (pop. > 500,000). Each of these groups is considered in turn. We begin by looking at the group of very small municipalities. Chart 3.3: DEA Efficiency Scores and Administrative Expenses per Student 120 KihScr,LowExpecture 100 . * .~~~~~~ *o 80 60~~~~~~ a .... ...... - ... 0- ..* 20 - * * * 0 i~~~ O00 5. I . 40 50 10 150 200 250 30 0 400 450 Admin. Exp. pe Student (RS per year) 3.35 The chart shows a mixed performance regarding efficiency scores and admninistrative expenses per student. Of the 59 very small municipalities that were used in the analysis (they formed roughly 25% of the 250 municipalities that form the FIPE sample with data available on all the parameters), a sizeable number are in the upper left hand quadrant of Chart 3.3, and fewer are in the bottom right hand quadrant. The average efficiency score for the group was 61%, and the average administrative expense per student was R$150. In the group of municipalities that 56 were below the average on expenditures, and above the average on scores, there were 14 municipalities, of which 10 were gainers from FUNDEF and 4 were losers. In the group below the average on scores and above the average on expenditures, there were 5 gainers and 10 losers. This is an interesting finding as it points to the ability of municipalities who gained more resources from FUNDEF to apply such resources to good use. Table 3.8 below shows the full range of variables used in the DEA Analysis for the group of very large municipalities . Note that a number of the municipalities with 100% efficiency scores are municipalities in the North East. Table 3.8: Comrutation of DEA Efficiency Scores (Very La Municipa lities) Teacher Capital Admin Expenses Expenses Expenses Pass Rate Income per per per per of Efficiency Municipali State Area (Km2) capita student student student Students Score CONTAGEM Minas Gerais 195 3,946 341 82 147 80.1 100 FORTALEZA Cearg 312 3J728 292 68 63 75.7 100 SAO GONCALO Rio de Janeiro 251 2,901 377 61 83 74.1 100 TERESINA Piauf 1,673 3 285 285 .15 122 70.4 100 NATAL R. G. do Norte 169 5,239 358 12 108 69.1 100 JABOATAO DOS G PemrnabuCo 25 2,124 184 18 131 68.5 100 JOAO PESSOA Paralba 210 4,219 353 3 1241 62.4 100 SALVADOR Bahia 325 5,681 183 . 32 135 67.5 99.3 MANAUS Amazonas 11,410 5,866 244 180 136 75.2 91.43 CURITIBA Parana 430 9,071 387 21 766 92.4 88.08 BELEM ParA 1,065 6,306 615 39 132 78.9 79.57 CAMPO GRANDE M. G. do Sul 8,096 6,504 357 44 160 79.2 76.74 UBERLANDIA Minas Gerais 4,103 4,907 389 29 176 79.2 76.26 MCEIO AlaSos 511 3,494 367 28 148 65.7 71.35 GOIANIA Goids 741 5,932 407 40 186 77.7 68.1 CAMPINAS S&o Paulo 796 7,432 495 20 593 81.8 68.01 RIO DE JANEIRO Rio de Janeiro 1,261 8,688 488 100 154 90.3 65.63 SAO PAULO SSo Paulo 1,525 10,341 813 59 259 91.5 51.6 BELO HORIZONTE Minas Gerais 331 10,449 732 45 314 93 50.34 PORTO ALEGRE R. G. do Sul 496 10,800 718 84 381 79.5 35.07 3.36 The analysis of efficiency scores is a very useful technique when used appropriately and judiciously. The patterns of variation in efficiency have policy implications that are discussed further in the concluding part of this section. For the immediate purpose of understanding the issue of recurrent expenditures, the analysis shows that municipalities did not just use the additional resources that they have received in the past few years merely to increase teacher salaries at the expense of everything else. Municipalities invested in the payment of administrative and support personnel as well as teachers. Furthermore, there is evidence that municipalities have used the resources in an efficient manner, though there remains substantial ground for improvement. We turn now to the other items on the list of recurrent expenditures, the question of non-personnel recurrent expenditures, and the payment of inativos. 3.37 Non-Personnel recurrent expenditures formed 28% of municipal expenditures on primary education in 1999. The biggest contribution to non-personnel expenditures was for transportation expenses (9%), followed by expenses on didactic materials and miscellaneous consumables (8%) and the school-feeding program (6%). Repair and maintenance of schools accounted for 4%. Maintenance and Repair expenses averaged R$ 8,300 for all of Brazil, but their was great regional disparity, with per 57 school expenses accounting for R$ 2,200 in the North East, and as high as R$ 19,000 per school in the South East. These figures do not include the in-kind transfer of textbooks from the federal government directly to schools, which was equivalent to 2% of the total primary education expenditure of municipalities. The figures for all levels of education combined and for primary education in 1997 are slightly higher, suggesting that the reforms did not lead to major shifts in resource allocation away from non-personnel expenditures. The reductions between 1997 and 1999 are accounted for mainly by reductions in outlays on didactic materials and other consumables, which were responses to the federal textbook distribution program initiated in 1998. 3.38 In the definition of permissible expenditures regarding education in the LDB, the law is silent on the question of payment of pensions of retired teachers (inativos)out of the educational budget. As such payments do not benefit students in the classroom, high outlays on this item of expenditure would pose a grave risk for educational quality. FUNDEF resources cannot be used to pay inativos, but there is no such restriction for general educational resources. Fortunately, the data on municipal expenditures indicates that the payment of inativos is not as serious a problem as it is for state governments, with the exception of large municipalities, for whom it is an area of policy concern. As there were about 1000 municipalities that were created since the early nineties, and yet other older municipalities did not have educational systems of their own, approximately 15% of Brazilian municipalities do not yet have any retired teachers at all. 3.39 Amongst the rest of the municipalities, there are about 21% of municipalities that do pay pensions out of the educational budget, with two-thirds of these municipalities belonging to the South and South East regions. The majority of retired teachers form part of a pension scheme for retired public workers, though the data does not provide details of the kind of pension scheme. About a fifth of pensions come out of the municipalities' general budget that does not go towards the 25% floor on educational expenditures. Amongst the municipalities that do pay inativos out of the general budget, the expenditures on inativos accounts for about 5% of the overall budget, comparable to the amount spent on repair and maintenance of schools. Again, the magnitude is higher in the South and South East, with municipalities in the North East paying only about 3% of expenditures on inativos.. 3.40 Inativos are a significant problem for very large municipalities (with population in excess of 500,000). As compared to the 5% overall allocation to inativos out of the education budget, very large municipalities spent 12% of their educational budget on inativos. Data gathered as part of preparation of a Bank loan for municipal pension reform shows that the 26 state capitals ran a combined pension deficit of R$2.1 billion that accounted for 80% of an aggregate municipal pension deficit of R$2.6 billion in 1999. Together with 23 other municipalities of population larger than 400,000, the group accounted for about 96% of total pension deficits. Some of the pension deficits are very large, such as the R$ 565 million pension deficit for the city of Rio de Janeiro and R$ 1.1 billion deficit for the city of Sao Paulo. 58 C. Transportation Expenditures 3.41 The question of transportation expenses deserves special attention because transporting students can be a cost-effective way to provide a higher coverage to students, especially in municipalities with dispersed population or a high percentage of population living in rural areas. Also, in the case where the municipal school stops at the fourth grade and for providing access to secondary school, municipalities need to have a sound policy regarding transportation. The issue of transportation becomes sticky because it often involves municipalities sharing resources with other municipalities and with the state government. A policy concern is that students would lose out on educational opportunities because municipalities would be wary of providing transportation for state government students as the economic cost of transportation now also includes losing a student and FUNDEF resources with the student to another educational system. 3.42 The federal government introduced a scheme to help municipalities with transportation costs by providing resources to acquire a vehicle, though maintenance and running costs are the responsibility of the municipality. The scheme benefits mainly small municipalities, as there is a limit of R$ 50,000 to each municipality to acquire vehicles - about 860 municipalities availed of this offer in the year 2000. An interesting way for private businesses to contribute to public education is the fact that private transportation companies offer subsidies to students going to school - we do not have data about how much of these subsidies by private operators are in turn sponsored by the municipal or state government. In general school transportation provided by the education system is linked to the rural population - there was a positive correlation of 0.3 between the percentage of rural population and the percentage of municipal expenditure on transportation. 3.43 It would appear that students who need to avail of transportation benefits have not lost out because of any adverse impact of the lack of co-operation between different levels of government. There was an increase of about 80% in the number of students transported by officially provided transportation. The figure includes students of all levels of education, as the break up by level of education is not available. However, there is a break-up available by state and municipal students transported by municipal transportation systems, and the figures show equivalent increases for both types of students, as seen in Table 3.9 below. Table 3.9: Trend 199i to 2001 realrding Trns ortation of Students by Munfci0ility |________ Municipal - State | Total 1997 2001 % Gain 1997 2001 % Gain 1997 2001 0/%Gain North 64,816 260,582 302% 21,653 110,015 408% 86,469 370,597 329% North East 852,278 2,131,097 150% 394,716 810,137 105% 1,246,994 2,941,234 136% Centre West 106,517 250,312 135% 50,250 142,426 183% 156,767 392,738 151% South East 1,243,845 1,761,040 42% 449,300 795,307 77% 1,693,145 2,556,347 51% South 536,093 759,143 42%° 607,609 853,700 41% 1,143,702 1,612,843 41% All Brazil 2,803,549 5,162,174 84%k 1,523,528E2,711,585 78%1 4,327,077 7,873,759 82% Source: FIPE 2001 59 3.44 The reasons why the fears regarding lack of co-operation between levels of government did not materialize is to do with the power of economic incentives in influencing decision making of the providers of educational services. FUNDEF places a great incentive for municipalities to enroll students - in the poorer municipalities, where there was still in 1997 a significant proportion of children not enrolled in school because of problems of access, the ratio of costs and benefits to the municipal administration of providing transportation works out heavily in favor of providing transportation services. If schools already had vacant places, and there were not a need to build more classroom space or to hire more teachers, every additional student would bring in more than R$ 300 to the system, while the transportation outlay would only be a fraction of that expense. 3.45 It is true that there would be a potential problem due to free-riding students enrolled in the state system, but the empirical increase in the provision of transportation services points to a plausible explanation. The state students transported by the municipality would have on the balance been students for whom the municipality was not in a direct competition, such as secondary school students. True, there was a potential externality due to which there would be tendency for under-provision, as the additional state students transported by the municipality would not bring additional revenue to municipalities. 3.46 State governments seem to have been made aware of the need to compensate municipalities, which is why there was an increase in the availability of state resources for municipal transportation - such payments accounted for 6% of municipal revenues for transportation in 1997, and they increased to 11% in 2001. State governments would save on the fixed cost of elaborate own transportation with low capacity utilization, and municipalities would be able to cover the gap in marginal benefits, thus reducing or removing the under-provision. It is interesting to note that state students accounted for 34% of students transported by the municipalities, but that state payments accounted only for 11% of revenues. 3.47 Transportation seems to be another area where the simple incentives introduced by FUNDEF have had meaningful impact. The area is worthy of further research, as the lessons on state-municipal cooperation may be extended to other fields where such co-operation is needed but is lacking. For instance, one of the exceptions in service provision regarding transportation is that nearly two-thirds (more in the case of the North-East) of transportation services were provided by contracted private providers, rather than by systems run by a municipal agency. There is a great scope for flexibility with such provision, and clearly municipal educational administrations, especially small ones in rural areas that most need transportation are not the best equipped to provide such services directly. Furthermore, private providers can clearly enjoy benefits of scope, by providing transportation services to other users outside the educational system at the time when buses are not required to transport school children. At the same time, private providers can enjoy scale economies by signing agreements with more than one municipality or with the state 60 government that would further lower the cost of provision.21 Not many municipal secretaries of education in the North East would have doctoral degree in Economics from the University of Sao Paulo, but it appears that they did not need to have such a certificate to think clearly in economic terms! 21 The use of private providers is not restricted to transportation services, but it also extends to services such as the provision of school lunch. 61 Section 4: Management of Resources: Teacher Remuneration and Career Progress A. Teacher Remuneration 4.1 Prior to the reforms, teachers in the North East received salaries that were a third of the salaries offered to teachers in the South East. Even though part of the differences could be explained due to differences in the cost-of-living, teachers in the North East were underpaid relative to other professions, and teaching was not an attractive profession except for the few teachers motivated by non-pecuniary concerns. The difference between state and municipal salaries for teachers also reflected a difference in resource availability. Table 4.1 shows the patterns of differences in teacher salaries for primary education across the regions in Brazil. Table 4.1 Comparing M unicip- State Teacher Remuneration: 1997 | rimary | Primary | . Qualification Incomplete jComplete -Secondary Tertiary Overall ._________ All Brazil Municipal 213 247 486 1.079 620 State 333 345 679 965 834 % Difference 56% 40%i 40% -11% 35% North Municipal 302 314 410 821 447 State 330 306 447 780 519 % Difference 9% -3% 9% -5%1 16% _________ ____ _ North East Municipal 178 166 289 621 309 State 350 255 442 5221 473 % Difference 97% 54% 53% -17%[ 53% Centre West Municipal 3671 378 493 Z50 548 State 4901 364 593 9247 763 % Difference 34% -4% 20% 39% South East Municipal 1 803 1268 1052 State 899 1125 1074 % Difference . 12% -11% 2% South Municipal 669 955 756 State 558 811 741 % Difference -17% -15% -2° Source: Balanco FUNDEF 1998-2000; MEC 4.2 Table 4.1 shows that state systems paid their teachers an average 35% higher than municipalities, though there is substantial regional variation. States in the North East paid a salary that was 53% higher than municipalities. States in the South and South East show a different pattern. The last column in the table shows weighted averages, with weights being the number of teachers at the different qualification levels. The case of the South East is dominated by the case of the state of Sao 62 Paulo, that in general paid higher salaries for teachers. However, municipalities that were prosperous enough to hire tertiary educated teachers at higher salaries results in an average relationship for the region that shows parity in salaries. Municipalities made use of the additional resources to increase teacher salaries, though as we have seen in the section of this report on composition of expenditures, other municipality outlays increased at the same time. Table 4.2 shows the situation regarding teacher salaries after the reforms. Table 4.2: Comparing Municipal - State Teacher Remuneration: 2000 rPrimary lPrimary I Qualification | Incomplete |Complete Secondary jTertiary jOverall All Brazil Municipal 319 391 662 1,299 826 State 501 583 788 1,266 1044 % Difference 57% 49% 19% -3% 26% North Municipal 351 518 561 985 593 State 484 463 640 968 716 % Difference 38% -11%° 14% -2% 21% North East Municipal 1 2951 3241 5041 8241 526 State 669 451 5981 7221 649 % Difference 127%1 39% 19% -12%1 23% Centre West Municipal | 4571 5271 606 1002 711 State r 5481 5641 593 1186 951 % Difference | 20% 7% -2% 18% 34% South East Munitipal .1005 1531 1291 State 996 1554 1335 % Difference t 1%i 2% 3% South Municipal l 858J 1168 955 State 644 954 867 % Difference | -25% -18% -9% Source: Balanco FUNDEF 1998-2000; MEC 4.3 The table shows a narrowing in state-municipal salary differences after the reforms had increased resource availability to municipalities. Municipal salaries still lag behind states, but the difference has narrowed down considerably for certain regions and certain categories of teachers. The category of secondary educated teachers in the North East is an important case in point. Such teachers account for a great majority of municipal teachers, and the salary gap for this group fell from a difference of 53% to 19%. This is an interesting part of the story about how the recent reforms altered incentives. Municipalities could not hope to attract better quality teachers if candidates had the option to chose a much higher paying teaching job. Better salaries could form a powerful incentive to improve the pool of potential applicants to new teaching positions, and help to retain those that were already employed. Interestingly, the salary gaps between the two systems in the North East 63 did not narrow down as much for the group of teachers with primary education completed (for Brazil as a whole the gap shows an increase). This last fact shows the attempt by municipalities to upgrade the qualification level of teachers, as it made more sense to focus resources on raising salaries of better educated teachers if improving the general level of qualification was the underlying motive behind the salary increases. 4.4 Municipal Secretaries believe in drawing power of higher salaries, as indicated in an interesting question in the FIPE 2001 survey that asked their opinion about the primary incentives for someone to seek a job as a teacher. Table 4.3 provides a summary of the responses:- Table 4.3: Oplinion Regarding Reason for Entering Teaching Career Order of Priority of Reasons Rank 1 Rank 2 Rank 3 Rank 4 Rank 5 Rank 6 Total N Salary 43%/o 27% 18% 8% 1% 3%k 100% 1 69 Stability 37% 34% 18% 8% 3% 1% 100% 195 Idealism 5% 18% 24% 28% 17% 8% 100% 128 Vocation 12% 19% 29% 26% 11% 3% 100% 152 Family Tradition 2% 10%° 18%° 20%° 31% 20% 100% 91 No Other Options 25% 22%1 23% 14% 10% 6% 100% 187 Source: FIPE 2001 Survey 4.5 Salary and Stability are the primary reason mentioned by municipal secretaries as motivations to enter the teaching profession. The ranks were orders of priority for each of the reasons, which is why the rows total up to 100%. The last column shows the number of secretaries who thought that the reason was significant at all, out of a sample of 220 secretaries who were asked the question (as the N differs, it is not meaningful to make direct quantitative comparisons across a single column of percentages). Amongst the non-pecuniary reasons, the important one is the belief that some people join the teaching profession because they may have a vocation for it. The disturbing number is the high figure for the reason that people have limited career options other than teaching. The data in Table 4.3 does tend to vindicate the thinking that economic incentives are very important in the field of education just as in any other area of the labor market. The table shows that economic reasons such as stability and salary, and the presence of career alternatives are considered to be the prime motivations for teachers. At the same time, reasons such as vocation or idealism are not to be discounted away, as they may be important secondary motivators for the bulk of the teachers, and be the primary motivation for a core group of dedicated teachers. These somewhat subtle patterns in the opinions are important because they reflect a considered thinking on the part of municipal secretaries of education, who presumably have taken decisions on the basis of such opinions. 4.6 The data regarding teacher remuneration certainly runs counter to the hypothesis that poor institutional capacity led municipalities with additional resources merely to increase teacher salaries as they did not have any other way to use the additional resources. Teacher salaries did increase substantially, but they did not increase at 64 the expense of other items of expenditure, and they did not leap drastically, as the data only indicate a convergence towards salaries paid by state systems in the same states. At the same time as teacher salaries increased, there were substantial increases in the number of teachers, reflecting in part the increase in enrollments that ensued due to the reforms, as well as an attempt to improve school quality. B. Number of Teachers 4.7 The number of municipal teachers for primary education increased from approximately 600,000 teachers in 1997 to about 750,000 teachers in 2000, an 22 increase of approximately 24%. The increase in the number of teachers has come about because of the efforts to increase enrollment for primary education since 1997 - municipal enrollment increased from 12.4 million students to 16.3 million students in the same period - this increase of 34% in the number of students implies an increase in the municipal students-teacher ratio from 20.5 to 22.2 students per teacher. The number of state teachers declined at the same time from 709,000 to 690,000, with state students declining from 18.1 million to 15.8 mnillion. Even with the increase in the municipal students-teacher ratio to 22.2, this ratio is less than the 22.9 ratio of state students to teachers. 4.8 The change in number of teachers was accompanied by a change in the distribution of teachers by level of educational qualification, as shown in Table 4.4 below (the Appendix provides details by region). The table shows the progressive increase in the level of teacher qualifications. The changes were even more pronounced in the North East where one out every 5 primary school teachers themselves had only a primary qualification. Even in the year 2000, out of the 39,500 such teachers, 29,200 were from the North East. As municipalities have made sustained attempts to reduce the number of the lay teachers (termed leigos in Portuguese), these numbers would have been reduced even further. Table_4.4: Change inumber of Municipal Teachers: 1997.- 2000 l__ _ _ 1 _ _ _ 1 1 IPost Qualification jPrimary |Secondary IGraduate iGraduate |Total All Brazil 1997 74,422 346,313 166,390 20,468 607,593. 2000 39,530 456,082 208,689 49,121 753,422. % Across'97 12, _ 2 57%/o 27Y% 30j 100% % Across '00 5% 61% 28% 7Yc 100% % Change 97 to 00 _ -47% 32% 25% 140% 24% Source: Balanco FUNDEF 1998-2000; MEC 4.9 One of the key interventions of the LDB regarding teachers had been the strong emphasis on the replacement of lay teachers with qualified teachers (lay teachers are those who teach at pre-school or in the first four grades of primary education 22 Combined with the increase in salaries, the result for all municipalities in Brazil was an increase in the wage bill of teachers of 42% (the wage bill for all of the personnel in primary education went up by only 29%, suggesting that wages of non-teachers did not increase as much as the wages of teachers). 65 without having a secondary education degree, or teachers who teach from Grades 5 to 8 without a tertiary education degree.). As an incentive towards reducing the number of leigos, the FUNDEF law also allowed the 60% of resources for salaries of education professionals to be used for training programs to convert lay teachers to regular teachers up to the end of the year 2001. According to the Bank's recent study on teachers (Brazil: Teachers Development and Incentives: A Strategic Framework; Report No. 20408-BR, February 2001), the legislation further requires that all new primary education teachers need to have a college education by the year 2007. The same study reports a simulation of the fiscal sustainability of such a measure, and suggests difficulties in financing unless sub-national governments could take some cost cutting measures. These measures could include a reduction in the number of teachers. The study also indicated the possibility of cost savings from measures to improve internal efficiency, the establishment of minimum retirement age for teachers and parallel measures to de-link pensions from current salaries. 4.10 As in the case of teacher salaries, municipalities appear to have responded judiciously to the incentives to increase the number of teachers on the payroll. The number of teachers has increased, but this increase has been commensurate with the needs of increased enrollment. There is now a higher level of parity between the students-teacher ratio for state and municipal systems, a development that reflects the increased weight to Grade 5 to 8 students in municipal systems as compared to the period before the reforms. Municipalities have also reduced the number of lay teachers in the system, as they were required by law, at the same time as the additional resources made it possible to comply with the law. C. Expenditure on Training of Teachers 4.11 Municipalities spent approximately 1.29% of total resources for primary education on teacher training, roughly spent equally on training of lay teachers as well as general training provided to all teachers. Table 4.5 shows the regional distribution of allocation on teacher training. Table 4.5: Outlays on Teacher Training as % of Basic Education Spending)- Lay Teachers Non-Lay Teachers All Training All Brazil 0.66% 0.62% 1.29% North 0.93% 0.17% 1.11% North East 1.49% 1.19% 2.68% Centre West 0.35% 0.86% 1.21% South East 0.00% 0.39% 0.39% South 0.16% 0.25% 0.42% Source: FIPE 2000 4.12 It is difficult to comment on the adequacy of these training expenditures or on the quality of the training provided due to the lack of data covering such aspects. There is anecdotal evidence to suggest that there is substantial variation in the efficacy of training expenditures, with some municipalities being extremely innovative and 66 others providing training of limited value. The innovations include a federally sponsored distance education program for teachers, as well as demand driven institutional arrangements such as the provision of scholarships to teachers to enable them to attend university courses. The Bank's study on teachers provides a detailed and fairly comprehensive description of the policy measures initiated regarding the training of teachers, as well as a battery of policy recommendations drawing from international best practice as well as successful experiences within Brazil. D. Career Progress of Teachers 4.13 The reforms associated with the LDB and FUNDEF go beyond providing incentives to pay teachers adequately and for training of teachers. The Law 9424/96 that established FUNDEF also included an imperative for municipalities and state governments to promulgate new teacher career plans (Plano de Carreira e Remunera!do do Magist&rio Publico Municipal). Conventionally, teachers moved according to a loosely defined pay scale, depending on the years of experience and educational qualifications. However, the career plan did not provide incentives linked to professional development, and generally did not include systems of performance evaluation and feedback. The difference in pay levels across education levels and years of experience was typically the result of an accretion of isolated decisions over the years, without any technical considerations. In fact, a number of municipalities simply did not have a special teacher's career plan at all, relying simply on the norms regarding all municipal workers. As an evidence of the lack of foresight regarding teachers' careers, up until 1997, there was not even a minimum retirement age for teachers. Teachers could retire with 20 years of service at a benefit level of 70% of salary, and their benefits were directly linked to any increases received by employed teachers. A number of teachers held various part time positions across schools and even education systems, adding to the lack of control. There was clearly a need to clean up the way that the most important asset for the education system was being managed. 4.14 The LDB and associated reforms introduced a minimum retirement age for teachers (48 for females and 53 for males) and eliminated the possibility of early retirement. Parallel measures under the Fiscal Responsibility Law (LRF) have introduced stricter rules regarding new hires of public sector employees, made possible the termination of employment, and streamlined the flow of pension resources. FUNDEF resources were accompanied by the proviso that municipalities would prepare new Planos de Carreira for teachers within a period of six months. These new plans were to include elements thought to incorporate the learning from successful experiences around the world. The federal Ministry of Education prepared comprehensive training packages including customized state of the art software to help municipalities prepare the plans. Amongst the elements mandated in the new Planos de Carreira were a) the career plan would include all educational professionals, not just classroom teachers; b) recruitment of teachers would be based on open public competition; c) the plan would include a points based system of horizontal and vertical differentiation in remuneration; d) the plan would 67 incorporate the requirements regarding upgrading of teacher qualifications; e) incentives would be provided in the plan for qualifications only from accredited institutions and would include a periodic assessment of teacher competencies; defined to include teaching and learning practices used in the classroom as well as enhanced co-operation of the parents and the community f) bonus payments and other benefits would be explicitly defined on the basis of educational objectives.23 4.15 Unfortunately, if the above description seems to be "too good to be true", it was. The article of the LDB regarding the Plano de Carreira was never implemented. The Supreme Court had ruled the article of the law to be unconstitutional, and various legislative alternatives that were advanced in subsequent years never reached successful passage in the legislature. The federal ministry retained the part regarding the Plano de Carreira in its initiatives to aid the implementation of the reform measures, but such efforts lacked the powerful incentive provided by legislation. More than a third of the municipalities to this date do not have career plans, and the quality of those that do have such plans is unknown. Of those that do have new career plans, these are said to include measures of performance, and to include incentives for teacher professional development, but the weights attached to such measures is not known. 4.16 Political economy considerations explain the lack of progress on teacher career plans, as established teachers who generally have more power in teachers unions have vested interests in maintaining the status quo.24 The recruitment stage offers a higher probability of successful intervention, as demonstrated by the experience of the province of Ceara that instituted state-wide competitions to jointly recruit teachers. One successful, though short-lived experience to provide performance incentives was a program introduced by the state of Rio de Janeiro, which could serve as a model for municipalities all over Brazil. Teachers are resistant to merit pay schemes linking salaries to student assessments because factors beyond the control of teachers, such as the socio-economic background of students would influence their test scores. Merit pay schemes further involve problems such as adverse incentives regarding co-operation between teachers and moral hazard issues surrounding test preparation and application. Rather than abandoning altogether the idea of rewarding better performance, the state of Rio de Janeiro introduced a program called "Nova Escola" that provides performance incentives to all teachers in the school to collectively enhance their performance. Instead of focusing merely on student test scores, the incentives were linked to a combination of performance 23 Training material prepared for municipal secretaries of education by the federal Ministry of Education's FUNDESCOLA unit includes "Progressdo na Carreira do Magisterio e Avaliagdo de Desempenho" by Mariza Abreu and Sonia Balzano (in "PRASEM III: Guia de Consulta" by Maristela Marques and Monica Giagio (Organizers); FUNDESCOLA/MEC, Brasilia, 2001. The paper provides an excellent description of the requirements of the law regarding teachers' career plans. 24 The Bank has recently commissioned a consultancy to make an institutional assessment regarding teachers unions in the state of Rio de Janeiro as part of project preparation activities. This study would hopefully identify the stumbling blocks to the revamping of teacher career plans. Without such institutional analysis, there does not seem to be much value in a policy recommendation to improve the career plans. 68 on three dimensions - school administration, internal efficiency indicators and student assessments. The group of measures on school administration includes factors such as the state of the physical plant and the integration of the school community. Internal efficiency measures include the pass rate of students and the age-grade distortion rate, long a bane of public school systems in Brazil. Student assessment was not limited to tests of Portuguese and Mathematics, but included areas such as economic and cultural awareness as well as civic responsibilities. Unfortunately for the students and parents, the program has been abandoned after two years of implementation as a beleaguered state administration attempted to resolve a teachers' strike that lasted for many months. 69 Section 5: Impact of Resources: Educational Results A. Introduction and Variable Description 5.1 An econometric analysis was carried out to measure the impact of the recent financing reforms on educational results, decomposed into three main effects (additional resources, expenditure composition and induced municipalization). Three main types of outcomes are examined: the dropout rate, the age-for-grade distortion rate and the passing rate , sub-divided by sub-cycle of basic education (1 to 4 grades and 5 to 8 grades). They are all regressed on the same set of determinants, which include a set of variables related to municipal expenditures and a set of socio-economic variables as controls. 5.2 Since we have panel data, more specifically, information for both 1996 (before the FUNDEF reform) and 2000 (after the FUNDEF reform), we are able to estimate both a cross-section (CS) and a first-difference (FD) model, aggregated at the municipal level, exploiting the change in space and time of the indicators. The well- known advantage of panel data is that such data makes it possible to discriminate among alternative interpretations and draw more reliable conclusions on the causal nature of the relationship between variables. In particular, FD models26, by construction, remove time-invariant unobservables from the error term, getting rid of possible heterogeneity biases. A limitation of the analysis is that since the FUNDEF reform is a nation-wide reform implemented in all municipalities, and across both the State and municipal sub-sectors, at the same time, we could not establish a "control group" which would allow us to control also for time-variant unobservables through a difference-in-difference type of analysis. The same nature of the reform also suggests, however, that no selection bias in participation in the reform is to be feared making the absence of a control group less of a problem. 5.3 As highlighted in the report, a clear first effect of FUNDEF is on the amount of resources available for municipalities and states. To capture this effect, we used an indicator of municipal and state expenditure in basic education per student (respectively named MUJNEXPEN and STATEXPEND). Over the time-period under analysis, municipal expenditure would capture the redistribution of resources from the states brought about by FUNDEF, but would also be related to the evolution of other possible sources of revenues and/or municipal choices on the allocation of resources to education -including possible substitution effects induced by the new FUNDEF resources- and, finally, to movements in student enrollment. State expenditure would also capture any possible resource transfers between the municipal and state sub-systems brought about by FUNDEF, the evolution of other state sources of revenues and/or choices on the allocation of resources to education 25 All these indicators are calculated annually on the basis of the National School Census undertaken by INEP (Ministry of Education). 26 An equivalent altemative to the FD, would have been to apply OLS to deviations from individual means (within-group estimator), however, having only two times periods, a PD seems the most indicate methodology. 70 and, to a large extent, the movements in student enrollment, in particular the decreased enrollment in the state sub-system. 5.4 Assessing the impact of the additional expenditure per student in both the municipal and the state systems is not sufficient to assess thoroughly the impact of the FUNDEF reform. This reform not only had an impact, directly or indirectly, on the amount of spending for education, but, through the allocation rule that it attaches to the resources (60% of them have to be spent on active teachers), the accountability mechanisms which accompany them (such as the creation of the social control councils at each level of government27) and the way they are channeled (they flow directly to specific bank accounts, without passing through the municipal governements), also on the composition of expenditure across alternative 28 uses 5.5 This implies that the marginal impact of FUNDEF resources on outcomes might be somewhat different from the one of non-FUNDEF resources and we attempt to capture this by measuring the share of FUNDEF resources in municipal and State expenditure (respectively called MFUNDSHARE and SFUNSHARE). The size of this share is not necessarily correlated with the additional resources (in an extreme case, we could have, for instance, no additional resources because FUNDEF resources crowded-out municipal own resources devoted to education, but a high share of FUNDEF resources)29. 5.6 Finally, a third effect of the FUNDEF reform discussed at length in the report is its impact on the municipalization of basic education induced by the link between resources and students. We measured the extent of municipalization by the relative share of enrollment in the municipal sub-sector in the 1 to 4 and 5 to 8 sub-cycles within each municipality (respectively called MENROLSH14 and MENROLSH58). A high share would imply a high level of municipalization in the sub-cycle, and vice-versa. 5.7 Table A.5. 1 (see Appendix) provides a synthesis of these and the other variables included in the econometric estimations, indicating how these variables have been measured, units of measure and data sources. On measurement, we should point out that while all outcomes and the municipal enrol]ment variable are measured 27 With the explicit purpose of monitoring the decision, transfer and utilization of FUNDEF resources. 28 To illustrate this point, according to the "Balance do pnmeiro ano do FUNDEF' (MEC/FUNDEF, 1999), 75% of the municipalities had increased the acquisition of teaching materials and 58% had adopted measures for increasing teacher training in the first year of activity of the Fund. An important consequence of the reform was also the impressive reduction of teachers who do not meet the minimum requirements to teach (i.e. with only complete primary education), either through dismissals or through adequate training (FIPE, 2000). Additionally, it is also shown that from December 1997 to June 2000, the average salary of public school teachers increased by 30% (increase that was particularly noticeable in municipal schools) (FIPE, 2000). 29 As shown in Section 2, Part (D) of this report the substitution effect was slightly positive. The correlation coefficient between the FUNDEF share in municipal expenditure and the change in municipal expenditures is zero. 71 separately for the 1 to 4 and 5 to 8 basic education sub-cycles, the expenditure variables were only available for the basic education cycle as a whole. It is also worth pointing out that State expenditure per municipality was constructed assuming the same unit cost across the municipalities of a same state and the calculation of the state FUNDEF share per municipality was based on the same simplification. This leads to statistical peculiarities which should be taken into account when analyzing the results. Table A.5.2 provides some summary statistics on all the variables included in the regressions. 72 B. Econometric Results: Estimation and Specification Issues 5.8 As mentioned above, both a cross-section (CS) and a first-difference (FD) are estimated. To discriminate among alternative interpretations and draw more reliable conclusions on the causal nature of the relationship between variables, panel data are essential. A possible model of educational performance using panel data is the first-difference (FD) model, which attempts to explain the change in educational performance over a period of time by the change in the inputs over that same period of time, getting rid of time-invariant unobservables that can lead to biased coefficients of the included variables (this would happen when the same unobservable affects both the outcome and one or more regressors producing correlation with the error term and simultaneity bias)30. Box 5.1 below summarizes this argument making use of equations. Box 5 1: Cross-section and first-difference equations The cross-section (CS) estimations are of the following type: EO,=a0 +PfXfi+XAXCX++u, (1) where: EO, = Educational Outcome of the ith municipality Xr, = Matrix of variables related to FUNDEF of the ith municipality X,, = Matrix of socio-economic control variables of the ith municipality ui = random disturbance term distributed nornally Rewriting equation (1) at two points in time and adding a fixed unobservable in the error term, we get: EO,t = a0 + fXfit + XXfit + Gj + ujt (2) EO.t, = a., + BfXfiI + X Xci, l + TlGi + u,- (3) And substracting (3) from (2) we get: EOjt-EO,.t_ = (or.-co.*) + lKf(Xnt - Xfit-,) + X,(X&t-Xcit-t) + u,t - u,t1 (4) which is the FD equations estimated (with no fixed unobservable). 30 See Deaton (1995) for a useful discussion of genuine simultaneity bias and omitted heterogeneity with possible remedies to it. 73 5.9 If this is an improvement, it should, however, be kept in mind that this methodology relies on some strong assumptions and introduces some new issues (in comparison with the CS) which should be highlighted31. The model relies on two main assumptions. Firstly, we have to assume that the heterogeneity takes the form of additive fixed effects. Secondly, first-differencing requires one to assume the time- stationarity of all the coefficients across the two periods of time considered32. Additionally, two main issues are typically raised by first-difference equations. Firstly, when the included regressors are positively correlated over time, differencing them will reduce variation, producing a lack of precision in the estimates. In other words, in cases where the variable changes little, it will usually be very difficult to get precise estimates and to be able to discriminate among alternative hypotheses. Secondly, in the presence of white noise measurement error in the explanatory variables, differencing will not only reduce the variability of the signal (variability in the true x's), but it will also inflate the ratio of noise to signal in the regressors. This combination of loss of precision and increased attenuation bias might induce some coefficients which were significant in the other models to turn small and insignificant even in the absence of heterogeneity bias. In our case, and in spite of the relatively short time-span of the analysis, the lack of precision should be attenuated by the fact that the time-period analyzed was a period of many changes, with, consequently, significant changes recorded in most of the variables. Given the above-mentioned limitations, care will in any case need to be taken when interpreting the FD coefficients and the results will always need to be compared with the CS ones. 5.10 The CS and FD equations are estimated pooling the State and municipal sub- systems' outcomes together. We initially estimated unrestricted models where the coefficients of all the independent variables were allowed to be different across State and municipal outcomes (which is equivalent to estimating two separate equations for State and municipal outcomes)3, with the exception of the coefficient on the municipal enrollment share which we want to be common across these two types of outcomes to measure the impact of management on overall outcomes. After testing for the equality of coefficients on the same variable across State and municipal outcomes, we ended up with the specifications shown in Tables 5.1 and 5.2, reported below, where the coefficients on expenditure, expenditure composition and fiscal capacity were maintained separate across the two types of outcomes34. To then correct for the possible downward bias of the standard deviations and 31 An informal treatment of FD models, with and without lagged dependent variables, with advantages and disadvantages, is provided in Deaton (1995) and a formal application of a FD model with lagged dependent variable applied to wage equations is provided by Newy, Holtz-Eakin and Rosen (1988). Finally, Pitt, Rosensweig and Gibbons (1993) provide a very useful application of a FD case applied to the assessment of education and health programs. 32 More general models, with no need for this assumption, are presented in Boardman and Murname (1979) and in Newey, Holtz-Eakin and Rosen (1988). Among the problems of these models, however, is the fact that they are hampered by the collinearity among the included variables. 33 See Greene (1993), Chapter 7 ("Tests of Structural Change"). 34 Since, according to the F-tests that were carried out, the coefficients on these variables appear to be significantly different across State and municipal outcomes. 74 consequent over-estimation of t-ratios produced by regressing variables with a lower level of aggregation on variables with a higher level of aggregation35 -such as what is produced by the regression of the separate categories, State and municipal outcomes, per municipality, on municipal or State wide variables such as the municipal enrollment share and municipal or State PIB p/c -, we used corrected standard errors36. 5.11 Beyond the "reduced form" model described by equations (1) and (4) of Box 5.1 and shown in Tables 5.1 and 5.2, we also estimated (in Tables A.5.3 and A.5.4) a similar set of equations replacing the FUNDEF related variables with other inputs more directly related to outcomes (precisely, class size, proportion of graduate teachers and class hours37) which, through auxiliary regressions, we then relate to the expenditure and municipalization variables. A significant relation between these variables and the selected inputs in a context where these inputs are in turn related to the age-for-grade distortion, drop-out and passing rate would cast light on possible pathways through which the FUNDEF related variables work. An additional specification, that we present in Tables A.5.5 and A.5.6, estimates the impact of the FUNDEF related variables together with the three selected inputs on outcomes. A comparison between this specification and the "reduced form" model can offer some further insight on the pathways through which the FUNDEF related variables work. If these variables work to a significant extent through class size, the proportion of graduate teachers and class hours the magnitude and level of significance of their impact will change considerably when introducing the three inputs in the regressions, otherwise, their impact will remain generally insensitive to the inclusion of the inputs. The robustness of our results across the "reduced form" and "input augmented" specifications will allow us to conclude that the FUJNDEF related variables work through channels other than the three identified inputs. 5.12 Finally, we decided to stick to estimations run separately on the two sub-cycles of basic education (1 to 4 and 5 to 8), in view of the difficult interpretation of the results obtained on the overall cycle. In particular, we want to avoid "spurious" variations in average educational outcomes associated with the relative weight of enrollment between the two sub-cycles38. This problem is solved running the 35 See the Moulton critique (1990). 36 Specifically the White's heteroskedastic consistent standard errors. 37 Unfortunately, we only have information on class size, proportion of graduate teachers and class hours per sub-cycle of basic education in the municipal and State sub-systems, limiting somewhat the usefulness of our analysis. 38 Consider for instance that, given a quite homogeneous high share of municipal enrollment in the grades I to 4, the main cause of variation of the average municipal enrollment share included in the regressions lies in the different municipal share in the grades 5 to 8 (which fluctuates more than the enrollment in grades 1 to 4 across municipalities, in both levels and first-difference, as can be seen from Table 2). Since this grade range is normally associated with relatively higher drop-out rates and age-for-grade distortion and lower passing rates than the grade range I to 4, we might end up finding that the level of municipalization is negatively related to the municipal sub-system's outcomes and non-significantly related to overall outcomes (if we assume that a higher municipalization in the grades 5 to 8 will, conversely, have a positive impact on State outcomes) for "spurious" reasons (reflecting a mere "structural" issue, i.e. the difference in outcomes between the grades 1 to 4 and 5 to 8). 75 regressions separately on the two sub-cycles. As indicated by the R2, the regressions are generally stronger in the 1 to 4 than 5 to 8 grades. 5.13 Concluding with a word on the control variables, it was decided to control for states' unobservables in the cross-sections including one dummy per State. We also attempted to control for possible interventions outside FUNDEF, in both the cross- section and first-difference regressions, including a variable capturing (in a proxied way since we do not have population by age in 1996) the enrollment rate in pre- basic education. Finally, in the first-differences, in the absence of information on municipal income at two points in time, we attempted to control for the changing income including the change in state income over the 1996/2000 time-period. 76 Tabie 5. CiLS estiinates- Dependent variable: AGEGRAjDIS4; DROPOUT14; PASSRATE14 (reduced model) Dependent variable: Dependent variable: Dependent variable: AGEGRADIS14 (a) DROPOUT14 (a) j PASSRATE14 CS (OLS FD (OLS- CS (OLS FD (OLS- | CS (OLS FD (OLS- robust Robust robust Robust robust Robust estimates- estimates) estimates- estimates) estimates Estimates) with State with State with State fixed-effects) fixed fixed effects) effects) Variables Coefficients Coefficients Coeff. Coeff. Coeff. Coeff. (t-ratios) (t-ratios) (t-ratios) (t-ratios) (t-ratios) (t-ratios) Expenditure STATEXPEND -0.0008 -0.001 -0.004 -0.004 0.004 -0.00 (-0.95) (-2.16)** (-5.09)*** (-7.07)*** (4.75)*** (-0.91) MUNEXPEN -0.0002 0.00 -0.0001 -0.0002 0.0003 0.0003 (-3.07)*** (0.20) (-2.53)** (-3.04)*** (2.58)** (2.52)** Composition of expenditure SFUNDSHARE 0.029 -0.063 -0.043 -0.015 0.044 0.035 (1.90)* (-8.35)*** (-3.69)*** (-2.47)** (2.80)*** (4.40)*** MFUNDSHARE 0.095 -0.098 0.005 -0.079 -0.032 0.088 (10.29)*** (-11.97)*** (078) (-10.20)*** (-3.20)*** (9 34)*** Municip aization MENROLSH14 -0.010 -0.030 -0.010 -0.0037 0.011 0.017 (-1.72)* (4.43)*** (-2.21)** (-0.70) (1.72)* (2.37)** Control variables SFISCAP -0.021 0.038 0.076 0.048 -0.046 0.035 (-1.11) (1.93)* (497)*** (2.24)** (-2 40)** (1.55) MFISCAP -0.093 -0.05 -0.034 -0.05 0.069 0.07 (-4.74)*** (-1.20) (-1.93)* (-1.50) (2.93)*** (1.53) POPULATION 0.00 -0.00 -0.00 -0.00 0.00 0.00 (0.20) (-2.41)** (-0.21) (-0.19) (0.81) (0.88) URBPOP -0.068 0.012 0.0065 (-11.20)*** (2.65)*** (1.03) PEBPC -0.0001 0 003 -0.00 0.002 0.00 -0.001 (-284)*** (8.73)*** (-3.11)*** (5.88)*** (2.78)*** (-3.55)*** PREBASENR -0.035 0.016 -0.013 -0.0080 0.022 0.018 (-2.65)*** (0.67) (-1.00) (-0.58) (1.47) (1.00) R2 0.77 0.043 0.43 0 030 0.48 0.031 Nof observations 7663 6944 7838 6936 7838 6913 77 Table 5.2: OLS estiintes- Dependent variable: AGEGRADIS58; DROPOUT58; PASSRATE58 (reduced,model) '___-______ Dependent variable: Dependent variable: Dependent variable: AGEGRADIS58 (a) DROPOUT58 (a) PASSRATE58 CS (OLS FD (OLS- CS (OLS FD (OLS- CS (OLS FD (OLS- robust Robust robust Robust robust Robust estimates- estimates) estimates- estimates) estimates Estimates) with State with State with State fixed-effects) fixed fixed effects) effects) Coefficients Coefficients Coeff. Coeff. Coeff. Coeff. Variables (t-ratios) (t-ratios) (t-ratios) (t-ratios) (t-ratios) (t-ratios) Expenditure STATEXPEND -0.001 0.002 -0.003 0.001 0.002 -0.00 (-1.02) (2.96)*** (-3.41)*** (1.53) (2.40)** (-0.89) MUNEXPEN -0.002 0.0007 0.00 -0.0005 0.001 0.001 (-4.74)*** (2.37)** (0.14) (-1.48) (2.89)*** (2.01)** Composition of expenditure SFUNDSHARE -0.027 -0.12 -0.016 -0.073 0.023 0.062 (-1.33) (-12.85)*** (-0.95) (-7.83)*** (1.19) (5 51)*** MFUNDSHARE 0.067 -0.023 0.036 -0.059 -0.028 0.026 (4.44)*** (-2.03)** (3.26)*** (-5.35)*** (-2.25)** (1.94)* Municipalization MENROLSH58 -0.023 0.0021 0.009 0.0067 -0.024 -0.031 (-3.48)*** (0.19) (1.55) (0.56) (-3.67)*** (-2.25)** Control variables SFISCAP -0.030 -0.084 0.10 0.013 -0.079 -0.042 (-1.41) (-3.91)*** (5.67)*** (0.66) (3.61)**.* (-1.65)* MFISCAP -0.083 -0.02 0.035 -0.07 -0.065 -0.07 (-2.67)*** (-026) (I I .9) (- 1.12) (- 1.92)* (4097) POPULATION 0.00 -0.00 -0.00 -0.00 0.00 0.00 (1.38) (-1.88)* (-1.24) (-1.86)* (0.44) (2.01)** URBPOP -0.054 0.032 -0.067 (-8.20)*** (6.20)*** (-10.66)*** PIBPC -0.00 -0.004 0.00 -0.0009 -0.00 0.0004 (-0.84) (411.32)*** (-1.24) (-2.51)** (-0.03) (1.08) PREBASENR -0.048 -0.023 -0.032 -0.035 0.037 0.051 (-2.52)** (-0.50) (-3.03)*** (-2.15)** (2.39)** (2.46)** R2 0.76 0.088 0.33 0.020 0.32 0.016 N of observadons 6875 5888 7038 5870 7038 5854 *Significant at 10%; ** Significant at 5%; *** Significant at 1%; (a) A negative sign indicates a positive contribution of the independent variables to the outcome. 78 C. Main Results: Impact of Expenditures 5.14 Expenditure is expected to have a positive impact on outcomes through its relation with the availability and quality of the material (infrastructure, teaching materials) and human resources. Its impact has been analyzed at length in the literature, showing, in general, surprisingly little impact on outcomes39. This could be because expenditure allocation is ultimately more important than expenditure per-se which can be easily spent inefficiently. It is particularly useful to analyze this impact in the context of the FUNDEF reform which, as mentioned above, led to a net increase of resources for the municipal governments. 5.15 Our results on municipal expenditure40 show that this variable has a positive impact on educational outcomes in 9 out of the 12 CS and FD reduced models estimated. It is also positively significant in 8 out of 12 of these regressions. However, the size of its coefficient is also very small, practically economically insignificant, in most of the CS and FD equations (with the exception of the passing rate equations in the 5 to 8 grades, where it appears that an increase of 100 Reales per student would lead to an increase of 0.1% points in the passing rate). The results are particularly consistent in the 1 to 4 grades, where expenditure, both in level and in change terms, is positively and significantly related to outcomes in 5 out of the 6 reduced models, and in the passing rate regression where expenditure is positively and significantly related to the outcome in all 4 equations. 5.16 This indicates that, albeit generally weak (confirming the findings of the literature), spending undertaken by the municipal governments has a positive impact on outcomes. This is also generally true over the time-period under analysis, as indicated by the FD results41, implying that resources matter independently of their correlation with municipal wealth or other fixed effects which might have an influence in the CS and that the additional resources are being spent productively by the municipalities, in particular in the grades 1 to 4. 5.17 What are the pathways through which municipal expenditure influences outcomes ? Tables 5.3 and 5.4, below, illustrate auxiliary regressions between municipal expenditure and three inputs (class-size, proportion of graduate teachers and class hours), which have just the purpose of showing the association between these 39 Analyzing the impact of expenditure per pupil on educational achievement, Hanushek (1986), for instance, finds the expenditure per pupil variable to be significantly positively related to outcomes in only 13 of the 65 studies surveyed (with 16 cases significant or insignificantly negatively related) and Fuller and Clarke (1994) find a significant positive impact of expenditure on outcomes in 5 out of the 11 cases surveyed. 40 Very close to the ones that we would get regressing only the municipal schools' outcomes on expenditure and the other included variables (as was done in a previous version of this section) since the coefficients on municipal and State expenditure have been kept separate, as the ones on expenditure composition and fiscal capacity. 41 With the exception of the age-for-grade distortion outcome where, in the grades 5 to 8, both municipal and State expenditure appear to be negatively and significantly related to the outcome. 79 variables (not causality, since we expect causality to go the other way round). Both the levels and the changes are illustrated. 5.18 The positive and significant relation of municipal expenditure with the proportion of graduate teachers, class hours and a smaller class size in level terms in both the sub- cycles I to 4 and 5 to 8, all positively related to the outcomes (see Tables A.5.3 and A.5.4), indicates that municipal expenditure works at least partly through these variables in the CS. Its little sensitivity to the inclusion of the three inputs in the CS regressions of Tables A.5.5 and A.5.6 also indicates, however, that expenditure works mostly through other pathways. In change terms, municipal expenditure is strongly correlated with a smaller class size. This, combined with its positive association with class hours in the 5 to 8, might explain part of the positive impact of municipal expenditure on the passing rate at this level since both a smaller class size and class hours are positively and significantly related to the outcome in that FD regression. However, this leaves unexplained the negative impact of expenditure in the age-for-distortion in the grades 5 to 8 (also positively associated with a smaller class size and class hours), suggesting that expenditure also works through other pathways and that a deeper analysis of how the net additional resources have been spent would be necessary to understand fully the impact of these resources on the different outcomes in the 1996/2000 time-period. This is also confirmed by the results of the FD on the I to 4 grades: municipal expenditure is more consistently spent well at this level through a pathway which is not the one reflected by the three analyzed inputs (not significantly related to outcomes in Tables A.5.3 and A.5.4).The expertise and capacity acquired by the municipalities in managing the 1 to 4 grades schools across the years (they have been traditionally much more present in this sub-cycle) can provide an explanation for the higher consistency at this level. Table 5.3: OLS estimnates- Dependent variable: MUNEXPEN (grades 1 to 4) - ' CS FD variables coefficients coefficients (t-ratios) (t-ratios) CLASSIZE14 -78 -24.7 (-20)*** (-9)*** GRADTEAC14 15.6 -1.9 (14)*** (-1.7)* CLASSHOUR14 767 -62.5 (12)*** (-1.5) R2: 0.16 0.03 *Significant at 10%; ** Significant at 5%; *** Significant at 1% 80 Table 5.4: OLS estimates- Dependent.variable: MUNEXPEN (grades 5 to 8) CS FD variables coefficients coefficients (t-ratios) (t-ratios) CLASSIZE58 -47 -9.8 (-28)*** (-4.5)*** GRADTEAC58 6.9 -0.9 (18)*** (-1.5) CLASSHOUR58 117 53 (6)*** (3)*** R2: 0.36 0.03 *Sigmficant at 10%; ** Significant at 5%; * Significant at 1% 5.19 Compared to municipal expenditure, state expenditure appears to be less consistently positively contributing to the outcomes (the variable has a positive and significant impact on educational outcomes in only half of the 12 regressions42). However, in the cases where it is positively and significantly related to the outcomes, its impact appears to be stronger, as indicated by a higher coefficient: this is particularly clear in the grades I to 4 where an increase of 100 Reales in expenditure per student would lead to a decrease of between 0.39% points (CS) and 0.47% points (FD) of the drop-out rate and an increase of 0.44% points (CS) in the passing rate. 5.20 The following points can be made in interpretation of this mixed scenario. Firstly, over the 1996/2000 time-period, increases in expenditure per student in the State system occurred to a large extent through the decrease in enrollment, produced by the municipalization process43, in a context where some of the fixed costs could not be easily reduced. A number of state schools, in particular in the 5 to 8 grade range, which were already under-utilized, were for instance further deserted following the municipalization process in the grades 5 to 8, leading to an increase in unit costs without any significant (and possibly even negative) impact on outcomes. This would explain why the impact of state expenditure is particularly poor in the grades 5 to 8 and over the time-period covered by the FUNDEF reform. This also tends to be confirmed by the results of Table A.5.5 which shows the 1 to 4 grades FD run mostly on the South/Southeast municipalities. It is in the Southeast municipalities that state expenditure per student increased the most and that occurred in parallel with the particularly strong municipalization process in the 1 to 4 grades that 42 As for the municipal case, these results are very close to the ones that we would get regressing only the State schools' outcomes on expenditure and the other included variables since the coefficients on municipal and State expenditure have been kept separate, as the ones on expenditure composition and fiscal capacity. 4 And, in the case of some States, like Sao Paulo, also through a redistribution of resources from municipalities promoted by FUNDEF. 81 occurred in this region4, meaning that most of the increase was associated with now under-utilized schools. The non-significant or negative impact of state expenditure in the regressions of Table A.5.5 supports the "fixed cost" theory. 5.21 Secondly, there is evidence that, when they had the choice, states transferred the weakest schools to the municipalities in the grades 1 to 4 (i.e. of a positive selection bias for the states). That could well explain, together with some statistical peculiarity due to the measurement of state expenditure (as explained in part A), why the state expenditure coefficient is generally higher than the municipal one in the grades I to 4 in both the CS and the FD. 5.22 As in the case of municipal expenditure, auxiliary regressions of state expenditure per student on class size, the proportion of graduate teachers and class hours show that state expenditure works at least partly through these variables in the CS. Again, however, the little sensitivity of state expenditure to the inclusion of the three inputs in the CS shows that this expenditure works mostly through other pathways. The results in change terms show that, in the grades 1 to 4, state expenditure was significantly associated with a higher proportion of graduate teachers and class hours over the time-period under analysis. This could also point to the positive selection bias highlighted above: graduate teachers might have increased following the transfer of the weakest schools (with the weakest teachers) to the municipalities. In the grades 5 to 8, the weak relation between the three inputs, all significant in explaining outcomes at this level, and state expenditure confirm that the evolution of expenditure over the 1996/2000 time period was dictated by other factors, such as fixed costs. Table 5.5: OLS estimates- Dependent variable:' STATEXPEND ( rades1 to 4) CS FD variables coefficients coefficients (t-ratios) (t-ratios) CLASSIZE14 -19 3.7 (-18)*** (4)*** GRADTEAC14 2.8 1.7 (12)*** (7)*** CLASSHOUR14 323 41 (19)*** (3.5)*** R2: 0.19 0.02 *Significant at 10%; ** Significant at 5%; *** Significant at 1% 44 The share of municipal enrollment in public enrollment increased by 41% in the Southeast in the 1996/2000 time-period, against an average increase of 14% over the same time-period. 82 Table 5.6:'OLS estimnates- Deependent variable: STATEXPEND ( rades 1 to 4) CS FD variables coefficients coefficients (t-ratios) (t-ratios) CLASSIZE58 -15 -1.4 (-17)*** (-1.6) GRADTEAC58 2.8 0.2 (17)*** (1) CLASSHOUR58 464 20 (26)*** (1.8)* R2: 0.28 0.002 *Significant at 10%; ** Significant at 5%; * Significant at 1% D. Main Results: Composition of Expenditure 5.23 Our hypothesis that FUNDEF resources might have a marginal impact on outcomes different from non-FUNDEF resources due to the allocation rules linked to these resources, the way they are channeled and the accompanying accountability mechanisms brings us to assume that the impact on outcomes of the FUNDEF resource share, which we use as a proxy for the composition of expenditure, should be significant. It is expected to be positive if the decentralization of resources directly to the education departments, the strengthened accountability mechanisms and the allocation of the resources on active teachers more that counterbalance the slight decrease in local autonomy which follows from the compulsory allocation rules and which might lead to less efficient outcomes. 5.24 Our results show that the FUJNDEF resource share in municipal resources is significantly negatively related to the three educational outcomes in all but one of the CS but positively and significantly related to all three outcomes in the FD equations. Why is this ? It is likely that the negative sign in the CS be due to the fact that FUNDEF expenditures represent a higher share of total expenditure in poorer municipalities (as indicated by an higher than average FUNDEF share in the North and North East of the country) with traditionally lower levels of expenditure per student. The negative impact would then simply be explained by the association with unobserved municipal poverty or other characteristic associated with high FUNDEF proportions and low educational outcomes. In the FD, controlling for municipal poverty and other municipal fixed-effects, it turns out that the municipalities where FUNDEF expenditures represent a higher share of total expenditure (i.e. where, from an initial null share in 1996, the FUNDEF share increased the most) have performed better in all three outcomes over the 1996/2000 time period. As concerns the quantitative significance of this relationship, in the 83 grades 1 to 4, a 10% points increase in the FUNDEF share would, in a quite uniform way, bring to an improvement of between 0.8 and 1% points in all three outcomes. In the grades 5 to 8, a 10% points increase in this share would bring to an improvement of between 0.2 and 0.6% points in outcomes. Therefore, the impact is significant and stronger in the grades 1 to 4 than 5 to 8. 5.25 This result needs to be taken with some caution due to the limitation of FD analyses, among which the fact that they do not control for time-variant unobservables. It could well be that other programs, including special interventions, have taken place in the education sectors at the same time that FUNDEF and that they can contribute to explain educational achievement over that same period. We have tried to control for this introducing the pre-basic education enrollment rate variable which, having both the characteristics of not being directly influenced by the FUNDEF reform (focused on primary education) and sensitive to other targeted or non-targeted interventions which happened in Brazil over the time-period under analysis, can capture the impact of these other interventions. As a way of controlling for possible targeted interventions undertaken in the North and the North East of the country (the poorest regions) under the time-period under analysis and which could explain why the poorest municipalities (which also have the highest FUNDEF shares) have performed better, we have also run again the I to 4 FD regressions, where the impact of the FUNDEF share is stronger, over all the municipalities excluding the ones in the North and North East. The results, presented in Table A.5.7, are consistent with the ones obtained on all the country's municipalities, even if the coefficients are lower, indicating that the positive and significant impact of the FUNDEF share is confirmed when "controlling" for the effect of targeted interventions in the North/North East. 5.26 The results on the FUNDEF share for state expenditure show a variable positively related to outcomes in the CS (even if insignificantly so in the grades 5 to 8) and very similar to the ones of municipal expenditure in the FD (i.e. the FUNDEF share appears to be consistently positive and significant in the explanation of outcomes). Compared to municipal expenditure, coefficients are, however, lower in the 1 to 4 grades (an increase of 10% points of the FUNDEF share would bring to an improvement of between 0.15% and 0.6% points in the outcomes) and higher in the 5 to 8 (the same increase would bring to an improvement of between 0.6 and 1.2% points in the outcomes). An explanation for the higher consistency of results between the CS and the FD results lies in the fact that there is no clear correlation between the FUNDEF share and poverty in this case. The difference in coefficient size across sub-cycles shows that the expenditure composition promoted by FUNDEF benefits more the grades 5 to 8, where expenditure, as seen above, is more inefficient, than the grades 1 to 4. 5.27 Finally, we analyze possible relations between the FUNDEF share and the three selected inputs too see if this cast some light on the pathways through which this resource composition affected outcomes. We show below the results for the FUNDEF share in municipal expenditure. In level terms, the results confirm that the 84 municipalities with the highest FUNDEF shares are the least well endowed explaining the negative relation with outcomes in the CS (the results for the FUNDEF share in state expenditure did not show a similar close relation). The analysis in change terms suggests that the FUNDEF share works mostly through variables that have not been identified here (teacher training, proportion of teachers with secondary education, teacher salary, teaching materials, proportion of administrative expenditure as % of personnel expenditure, etc). Similar results hold in the state case. The little sensitivity of the FD results to the incorporation of class size, proportion of graduate teachers and class hours also indicates that the FUNDEF share (in both municipal and state expenditure) works through other channels. Table 5.7: OLS estiiates- ependent v'ariable: MWUNDSHARE (grades 1 to 4> CS FD coefficients coefficients (t-ratios) (t-ratios) CLASSIZE14 1.6 -0.04 (29)*** (-0.6) GRADTEAC14 -0.4 -0.4 (-24)*** (-11)*** CLASSHOUR14 -12 2 (-13)*** (1.7)* R2: 0.3 0.04 *Significant at 10%; ** Significant at 5%; * Significant at 1% Tabile 5.8 OLS-est iiites"-fependent variable: MFUNDSHARE(grA st8) CS FD coefficients coefficients (t-ratios) (t-ratios) CLASSIZE58 1.3 0.2 (16)*** (2.4)*** GRADTEAC58 -0.2 0.04 (-23)*** (1.3) CLASSHOUR58 -4 0.2 (-6)*** (0.3) R2: 0.29 0.006 *Significant at 10%; ** Significant at 5%; * Significant at 1% 5.28 In conclusion, as things stand, we find that the allocation pattern of the resources, as measured by the share of FUNDEF expenditures, has a significant higher impact on outcomes than the additional resources per-se, which, overall, have a positive but 85 weak effect on outcomes. Let's turn now to the assessment of the municipalization process that was promoted by FUNDEF. E: Main Results: Degree of municipalization 5.29 We measure the extent of municipalization by the relative share of enrollment in the municipal sub-sector within each municipality. A growing amount of literature addresses the issue of the effect of decentralization on the social efficiency, technical efficiency and quality of delivery and argues that this effect should be positive. The basic assumption of all this literature is that sub-national units have better access than the central government to information on local preferences, needs and conditions which will make them take decisions which will be more responsive to these local aspects. This will increase both the social efficiency of delivery- through a better fit with local preferences- and the technical efficiency- as well as the quality of it- through the innovative and creative approaches adopted to fulfill needs and characteristics and the higher level of external accountability produced by the closer link between providers and users. If this holds, we would expect that the transfer of responsibility in the delivery of primary education from the states (closer to the central government) to the municipalities in Brazil makes it possible to deliver a service of higher quality through a better match with local needs and characteristics and the increased accountability of the service providers to the local community. In both cases, the positive impact on the quality of education would be enhanced by high levels of participation in the decision-making process of the users (teachers, parents, students), as this would enhance external accountability (through local monitoring and control) and the fit with local needs and characteristics (through the direct expression of users' needs)45. 5.30 Given this, in assessing the impact of municipalization in the Brazilian case, we should keep in mind that, under the FUNDEF reform, municipalization basically occurred under three main different forms. In some cases, the municipal sub-sector took over schools from the state sub-sector, in some others, municipal schools were created where they did not even exist or extended (which also occurred releasing students from the state sub-sector) and, finally, in others, municipal enrollment increased due to extension of the coverage of primary education (i.e. taking students who were not enrolled previously, not even in the state system). This, together with possible selection biases in the transfer of schools and students, might influence the outcome of the recent municipalization process. The FD model will make it possible to disentangle the impact of the recent change in municipalization from the level of 4sA whole strand of literature specific to education argues that a decentralized and participatory decision- making process at the sub-national, and, above all, school level, has good potential for improving student performance (taken as a proxy of educational quality) through mechanisms which are basically the above mentioned ones. Recent studies on the impact of decentralized management on the quality of education include the studies of King and Ozler (1998) on school autonomy reform in Nicaragua, Jimenez and Sawada (1999) on El Salvador's EDUCO's schools, Filmer and Eskeland (2002) on autonomy and participation in Argentinian schools, Paes de Barros and Mendonca (1998) on the determinants of educational achievement in Brazil, and Jimenez and Paqueo (1996) on the impact of local financial decentralization on public schools in the Philippines. 86 municipalization on outcomes, relating the recent changes in the municipal enrollment shares induced by FUNDEF to changes in outcomes. 5.31 The regressions run on the I to 4 grades show that the degree of municipalization in the same four grades is positively related to outcomes in all the CS and FD equations and significantly so in all regressions but one. For two out of the three outcomes (age-for-grade distortion and passing rate), they also show that the change in municipalization is more significantly related to outcomes than the level. This indicates that once we remove fixed-effects through the first difference procedure the degree of municipalization becomes more significant, but also that the municipalization process induced by FUNDEF in the grades 1 to 4 was particularly effective. 5.31 In quantitative terms, the coefficients of MENROLSH14 in the age-for-grade and passing rate FD indicate that an increase of 10% points in the municipalization ratio would bring to a decrease of 0.30% points in the age-for-grade distortion and to an increase of 0.17% points in the passing rate. This is a small impact which however shows that, after controlling for the increased expenditure in both the municipal and state sub-systems and the expenditure composition induced by FUNDEF, the stronger the increase in the municipalization of the grades 1 to 4 over the 1996-2000 time period, the better municipalities performed overall. 5.32 What does this mean? This means that, beyond any expenditure impact, the higher local accountability and better informed choices of municipal governments relative to the state governments are being translated into better schools and outcomes. This impact can also be related to the high level of expertise acquired by municipalities over the years in the 1 to 4 grades (where they have been traditionally much more present than in the 5 to 8 grades) and their active role in policy making in these same grades which put them in a favorable position for handling successfully the school and student transfer from the states (in spite of a likely adverse selection bias). This, together with the strengthening of the local accountability mechanisms and of the municipal education departments, empowered by the direct transfer of resources under FUNDEF, can probably also explain why the latest municipalization process in the grades 1 to 4 has been particularly effective. 5.33 Some care needs to be taken in interpreting these results, since, while the relative municipal enrollment share is aimed at measuring a management impact, it could also, capture mere transfer effect in a context where transferred state schools were higher performers than municipal schools. We do not have enough information to quantify exactly the extent and pattern of transfers of schools between the state and the municipal sub-sectors to be able to assess the impact that this would have on our results. In any case, the fact that the municipalization variable has an even stronger impact in the Northeast46, region where municipalization mostly occurred by 46 As shown in Table A.5.8 which reports the size and level of significance of the municipalization coefficient within the exact same FD specification used in Tables I and 2 but only run on the Northeast municipalities. 87 extending existing schools and not through school transfers, indicates that the management impact is prevalent. 5.34 A brief analysis of the relation between MENROLSH14 and the three selected inputs (reported in Table A.5.9) shows that while the level of municipalization is negatively associated with the proportion of graduate teachers and class hours, which are positively and significantly related to outcomes in the I to 4 grades CS (see Tables A.5.3 and A.5.4), the change in municipalization is, in contrast, positively associated with these two variables, which, however, are non significant in explaining outcomes in the I to 4 grades FD. In both cases, this suggests that municipalization works through other variables on outcomes (and this is confirmed by the little sensitivity of this result to the inclusion of the three inputs in the regressions). This is not surprising as we would expect municipal management to be working mostly through "non-quantitative" types of inputs such as the applied pedagogical practices and other characteristics of the teaching-learning process which should be highly related to outcomes. 5.35 The regressions on the 5 to 8 grades show a much less clear cut situation on the impact of municipalization in these grades on the selected outcomes. A first consideration is that the impact of MENROLSH58 is very heterogeneous across the three outcomes in the CS (positive significant in the age-for-grade distortion, negative insignificant in the drop-outs and negative significant in the passing rate). A second one is that it is always negative, even if significantly so only in the passing rate regression, in the FD. 5.36 In conclusion, in contrast to the 1 to 4 grades, there is no clear evidence, and maybe even some slight proof of the contrary, that municipalization in the 5 to 8 grades, both in terms of current levels and change over the time period analyzed, is having a positive impact on outcomes. A first reason for that is that municipal governments have always been traditionally less involved in these grades than in the 1 to 4 ones, they have developed less expertise in managing these grades and have been and still are less involved in policy making. Consequently, they have had less capacity of influencing outcomes in these grades47. Another reason is that several 5 to 8 grades in state schools are offered in big 5 to 11 grades schools which are very well equipped and prone to get good results. 5.37 As MENROLSH14, MENROLSH58 is negatively related to the proportion of graduate teachers and class hours in level terms (see Table A.5.10), both positively related to outcomes in the 5 to 8 CS. This cannot quite explain the changing impact of MENROLSH58 in the CS. In change terms, this relation does not clearly hold anymore and others seem to be the factors that explain the impact of the variable on outcomes in the 5 to 8 FD (as confirmed again by the regressions including the three 47 Capacity which is also decreased by the high level of heterogeneity in municipal provision in the 5 to 8 grades across all the country's municipalities which makes it difficult for the municipalities to organize themselves as a group. 88 inputs). The increased class size that accompanies the municipalization process48 can however offer some explanation for the poor results of municipalization in these grades in the FD. 48 And that could be a diect squence of the increasingly over-crowded municipal schools, which more than compensates the reducition in size of State classes (which could easily be merged after students left with no reduction in class size). 89 Appendix for Section 1 Table 1.1: Distribution of Federal Government Education Expenditure, 2001 Expenditure Item of Expenditure /Project (Million Reals) Percentages A. Higher Education 7,132.7 61.3% B. Equity Enhancing Programs 1,862.3 16.0% School Feeding 1,036 8.9% Cash Transfer to Poor Children 501 4.3%o Adult and Special Education 310 2.7% School Health 15 0.1% C. Fundamental Education 1,635.9 14.1% Contribution to FUNDEF 476 4.1% Textbook Distribution 470 4.0%/o Cash to Schools 302 2.6% Other Programs 388 3.3%o D. Secondary Education 510.9 4.4%/i E. Policy: Evaluation, Research and Communication 151.8 1.3% F. Miscellaneous Programs 180.5 1.6%/ G. General Administration 163.2 1.4%9 TOTAL 11,637.3 100.0%° Source: SIAFI/STN Tablel.2: Distribution of State' Government Education Expenditure, 2000 Expenditure: Item of Expenditure /Project [Milliin R nliQ Percentages A. Early Education 237 0.9%/ 7 Creches 12 0.0% Pre-School 215 0.8% Literacy Classes 9 0.0% B. Fundamental Education 15,436 58.2%/ Grades 1-4 5,767 21.8% Grades 5-8 9,669 36.5% D. Secondary Education 8,726 32.9%k E. Adult Education 2,101 7.90/% TOTAL 26,500 100.0%k Source: Own Calculations from INEP and STN data 90 Appendix for Section 2 Table 2.1: Evolutionof Education:Expendtures (in bilions of Reals) _ 1995 1997 1998 1999 2000 (a) Education Spending (Nominal) 27.19 40.06 48.88 52.56 59.45 (b) Education Spending (Real) 43.97 50.97 59.31 60.98 63.85 (c) GDP Nominal 646.2 870.7 914.2 963.9 1,086.7 (d) GDP Real (2001 as Base Year) 1,045.0 1,107.9 1,109.3 1,118.3 1,167.1 Education Spending as % of GDP 4.2% 4.6% 5.3%1 5.5% 5.5% Source: 1995 (IPEA); 1997-2000 (FIPE) Table 2.2: Dlatribution of Education Expenditures by Level:of-Goviernment (Blillion of 2001, Rels) _________ _ I 1995 - .-1 1997 1998 1999 2000 Federal 10.96 '249 1085 21.3% 11 98 20.2% 12.74 20.9% 1141 17.9% State 20,99 47J% 24 65 48.4% 27 48 463% 26 61 436/a 28.46 44 6% Municipal ~12,02 273%fo 15 47 30.4% 19 85 33.5%/ 216E3 35 5% 23 98 37 6% Total 4397 o100% 50 97 100% 59 31 100% 60.98 100%x 63.85 100% Source: (1995) IPEA; 1997-2000 for Sub-National Spending (FPIPE); 1997-2000 for Federal Spending (PRODASEN) Table 2,3: Repre3 ntativeness of theSTN Sa ple_. . - - | Universe STN Sample Number Pereentaae % Population Number Percentage % Population Municipal Size < 20,000 4011 73% 20% 2153 72% 17%o 20-50,000 967 180% 17% 494 17% 15% 50-100,000 302 5% 13% 179 6% 12% 100-500,000 190 3% 23% 138 5% 22% > 500,000 30 1% 27% 19 1% 29% Region North 448 8% 8% 49 2% 2% North East 1787 32% 29% 584 20% 21% Centre West 445 8% 6% 221 7% 4% South 1159 21% 15% 939 31% 21% South East 1661 30% 43% 1190 40% 53% Total 5500 100% 100% 2983 100% 100% 91 Source of Table 2.4: (A) Regression of Log Change in Total Revenues The REG Procedure Model: MODELI Dependent Variable: LY1 Analysis of Variance Sum of Mean Source DF Squares Square F Value Pr > F Model 7 572.54009 81.79144 1980.31 <.0001 Error 517 21.35333 0.04130 Corrected Total 524 593.89342 Root MSE 0.20323 R-Square 0.9640 Dependent Mean 3.41154 Adj R-Sq 0.9636 Coeff Var 5.95714 Parameter Estimates Parameter Standard Variable Label DF Estimate Error t Value Pr > Itl Intercept Intercept 1 -1.36811 0.28713 -4.76 <.0001 LX1 1 0 81291 0.02450 33.18 <.0001 POPTOTAL Pop Total 2000 1 0.16245 0.02205 7.37 <.0001 PIB96 1 0.02982 0.01655 1.80 0.0722 DUM_N 1 -0.06134 0.10733 -0.57 0.5679 DUM_SE 1 -0.05799 0.03898 -1.49 0.1374 DUn_S 1 -0.09134 0.03987 -2.29 0 0224 DUM_CW 1 0.04176 0.05493 0.76 0.4475 92 Source of Table 2.4: (B) Regression of Log Change in Own Revenues The REG Procedure Model: MODEL2 Dependent Variable: LY2 Analysis of Variance Sum of Mean Source DF Squares Square F Value Pr > F Model 7 1526.85634 218.12233 291.20 <.0001 Error 517 387.26190 0.74906 Corrected Total 524 1914.11824 Root MSE 0.86548 R-Square 0.7977 Dependent Mean 0.77515 Adj R-Sq 0.7949 Coeff Var 111.65270 Parameter Estimates Parameter Standard Variable Label DF Estimate Error t Value Pr > iti Intercept Intercept 1 -3.26865 1.01060 -3.23 0.0013 LX2 1 1.11838 0.05573 20.07 <.0001 POPTOTAL Pop Total 2000 1 0.32054 0.08255 3.88 0.0001 PIB96 1 -0.04387 0.06388 -0.69 0.4925 DUM_N 1 0.63989 0.45707 1.40 0.1621 DUM_SE 1 -0.05264 0.17219 -0.31 0.7600 DUM_S 1 0.01344 0.17580 0.08 0.9391 DUM_CW 1 -0.07288 0.23572 -0.31 0.7573 93 Source of Table 2.4: (C) Regression of Log Change in Transfer Revenues The REG Procedure Model: MODEL3 Dependent Variable: LY3 Analysis of Variance Sum of Mean Source DF Squares Square F Value Pr > F Model 7 471.55365 67.36481 192.84 <.0001 Error 517 180.60244 0.34933 Corrected Total 524 652.15609 Root MSE 0.59104 R-Square 0.7231 Dependent Mean 1.89404 Adj R-Sq 0.7193 Coeff Var 31.20525 Parameter Estimates Parameter Standard Variable Label DF Estimate Error t Value Pr > Itl Intercept Intercept 1 -9.06710 0.84263 -10.76 <.0001 LX3 1 0.00327 0.07382 0.04 0.9647 POPTOTAL Pop Total 2000 1 0.71715 0.06305 11.37 < 0001 PIB96 1 0.16368 0.04863 3.37 0.0008 DUM_N 1 -0.16870 0.31216 -0.54 0.5891 DUM_SE 1 -0.06739 0.11215 -0.60 0.5482 DUM_S 1 -0.16863 0.11463 -1.47 0.1419 DUM_CW 1 0.01118 0.15922 0.07 0.9440 94 Source of Table 2.4: (D) Regression of Log Change in Total Expenditures The REG Procedure Model: MODELl Dependent Variable: LYS Analysis of Variance Sum of Mean Source DF Squares Square F Value Pr > F Model 7 4173.52293 596.21756 719.19 <.0001 Error 2539 2104.85331 0.82901 Corrected Total 2546 6278.37624 Root MSE 0.91050 R-Square 0.6647 Dependent Mean 0.27728 Adj R-Sq 0.6638 Coeff Var 328.36361 Parameter Estimates Parameter Standard Variable Label DF Estimate Error t Value Pr > Iti Intercept Intercept 1 -0.51379 0.31293 -1.64 0.1007 LX5 1 0.95003 0.02204 43.10 <.0001 POPTOTAL Pop Total 2000 1 0.33368 0.03719 8.97 <.0001 PIB96 1 -0.20421 0.03077 -6.64 <.0001 DUM_N 1 0.12080 0.14339 0.84 0.3996 DUM_SE 1 -0.27663 0.06030 -4.59 <.0001 DUM_S 1 -0.56983 0.06430 -8.86 <.0001 DUM_CW 1 -0.19556 0.09020 -2.17 0.0302 Source of Table 2.4: (E) Regression of Log Change in Education Expenditures The REG Procedure Model: MODEL4 Dependent Variable: LY4 Analysis of Variance Sum of Mean Source DF Squares Square F Value Pr > F 95 Model 7 397.40344 56.77192 93.20 <.0001 Error 517 314.92445 0.60914 Corrected Total 524 712.32789 Root MSE 0.78047 R-Square 0.5579 Dependent Mean 1.04570 Adj R-Sq 0.5519 Coeff Var 74.63624 Parameter Estimates Parameter Standard Variable Label DF Estimate Error t Value Pr > Itl Intercept Intercept 1 -8.19908 1.09121 -7.51 <.0001 LX4 1 0.09329 0.08751 1.07 0.2869 POPTOTAL Pop Total 2000 1 0.54232 0.08160 6.65 <.0001 PIB96 1 0.15998 0.06143 2.60 0.0095 DUM_N 1 -0.06725 0.41250 -0.16 0.8706 DUM_SE 1 0.16009 0.14871 1.08 0.2822 DUM_S 1 -0.24032 0.15228 -1.58 0.1151 DUM_CW 1 0.29742 0.21012 1.42 0.1575 96 Appendix Table 2.1 4:' Slmulation of Change In Per StudentFloori for UNDEF (Year, 2000), STATE MUNI Federal Federal Contributi Contributi FUNDEF Transfers Transfers on to on to Number of Available Floor of 500 Floor of 335 FUNDEF FUNDEF Students per Student (R$ mililion) (R$ million) ACRE (AC) 81.5 15.3 137,393 704.5 _ _ AMAZONAS (AM) 207.5 78.7 629,298 454.8 30.5 AMAPA (AP) 80.0 11.2 115,928 786.7 _ PARAP( 269.0 126.6 1,540,872 256.7 379.9 120 6 RONDONIA (RO) 107.2 35.8 301,511 474.3 8.8 RORAIMA (RR) 59.5 12.8 78,258 923.9 TOCANTINS (TO) 117.6 42.1 322,630 495.0 2.7 ALAGOAS (AL) 140.4 69.3 666,213 314.8 125.6 13.5 BAHIA (BA) 606.3 333.3 3,524,162 266.6 834.1 241.0 CEARA(CA 348.2 177.9 1,697,470 309.9 328.2 42.6 MARANHAO(MA) 219.5 111.3 1,541,984 214.5 445.3 185.8 PARAIBA (PB) 175.8 95.4 812,768 333.7 137.9 1.1 PERNAMBUCO (PE) 375.0 188.3 1,575,484 357.5 229.6 PIAUr (Pl) 135.0 68.8 723,214 281.8 160.2 38.5 R.G. DO NORTE (RN) 166.5 80.2 593,698 415.5 52.1 SERGIPE (SE) 138.1 48.6 394,958 472.7 12.1 GOIAS (GO) 289.5 157.6 1,042,882 428.7 77.8 MATO GROSSO DO SUL (MS) 144.2 71.8 419,627 514.7 MATO GROSSO 196.0 91.2 575,475 499.1 2.4 ESPIRITO SANTO (ES) 268.4 118.2 541,817 713.5 _ MINAS GERAIS (MG) 968.6 575.4 3,423,729 451.0 179.2 RIO DE JANEIRO (RJ) 969.1 377.9 2,000,490 673.3 SAO PAULO (SP) 3,351.4 1,396.1 5,461,201 869.3 PARANA (PA) 556.2 313.3 1,562,491 556.5 RIO GRANDE DO SUL (RS) 715.9 369.5 1,590,434 682.5 SANTA CATARINA (SC) 350.0 192.9 907,552 598.2 BRASIL 11,036.3 5,159.6 32,181,539 503.3 3,006.4 642.9 97 APPENDIX TO SECTION 3 Table.3.1. Primary Education Expenditures, 1999 _ _ ; - A.iBrazil, Municipal State All Brazil ^ .,^ s% % 1. Recurrent = (2)+(11)+(17) 86.67 80.87 2. Personnel = (3)+(6)+(7) 54.99 62.57 3. Teachers = (4)+(5) 39.50 50.62 4. In-Classroom Teachers 37.28 47.58 5. Out of Classroom Teachers 2.22 3.04 6. Administrative Personnel 9.14 5.88 7. Pedagogical Support Personnel 6.33 6.06 8. Training= (9)+(10) 1.29 1.25 9. Training of Lay Teachers 0.66 0.58 10. Training of Non-lay Teachers 0.62 0.67 Il. Non-Personnel Recurrent = (12)+(13)+(14)+(15)+(16) 23.80 14.50 12. Transportation 6.57 1.23 13. Repair and Maintenance 3.68 3.17 14. Didactic Materials 2.66 1.14 15. Other Consumption 5.24 5.33 16. School Lunch 5.63 3.62 17. Miscellaneous Recurrent 7.88 3.78 18. Capital = (19)+(20)+(21)+(22)+(23) 8.95 7.01 19. Civil Works 4.96 3.97 20. Acquisition of Vehicles 0.68 0.21 21. Acquisition of Equipment 1.68 1.47 22. Amortization of Debt 0.01 0.00 23. Miscellaneous Capital 1.61 1.34 24. Out of Budget 0.51 0.17 25. Inativos 1.37 6.60 26. Miscellaneous Cap or Rec 2.48 5.33 27. Total Primary Education = (1)+(1 8)+(24)+(25)+(26) 100 100 Source: FIPE 2000 98 Table,3;2.a. Municipal Primarry Education Expendituriesm 'rReion,i 1999- ' . e-1<: Centre-West i tNor f-&th -East- South _South-East 1. Recurrent (2)+(11)+(17) 84.67 87.53 88.40 87.51 84.30 2. Personnel = (3)+(6)+ (7) 49.75 58.60 56.67 59.05 50.11 3. Teachers = (4)+(5) 33.96 41.64 41.31 43.76 35.23 4. In-Classroom Teachers 32.35 39.41 40.05 38.48 33.72 5. Out of Classroom Teachers 1.60 2.23 1.25 5.28 1.51 6. Admninistrative Personnel 10.33 10.39 9.71 7.93 8.26 7. Pedagogical Support Personnel 5.44 6.56 5.64 7.35 6.61 8. Training= (9)+(10) 1.21 1.11 2.68 0.42 0.39 9. Training of Lay Teachers 0.35 0.93 1.49 0.16 0.00 10. Training of Non-lay Teachers 0.86 0.17 1.19 0.25 0.39 11. Non-Personnel Recurrent = 26.66 25.33 23.14 19.99 25.33 (12)+(13)+( 14)i-(15)+(16) 12. Transportation 8.51 6.44 4.13 7.08 8.49 13. Repair and Maintenance 3.75 3.50 4.59 2.20 3.68 14. Didactic Materials 3.15 4.89 2.79 1.63 1.86 15. Other Consumption 6.31 4.21 5.64 4.09 5.75 16. School Lunch 4.92 6.27 5.97 4.96 5.54 17. Miscellaneous Recurrent 8.25 3.59 8.58 8.46 8.85 18. Capital = (19)+(20)+(21)+(22)+(23) 7.93 11.39 8.85 7.52 9.02 19. Civil Works 4.05 6.69 4.72 4.76 4.72 20. Acquisition of Vehicles 1.50 0.59 0.37 0.45 0.96 21. Acquisition of Equipment 1.38 2.14 1.92 1.11 1.61 22. Amortization of Debt 0.00 0.00 0.00 0.06 0.00 23. Miscellaneous Capital 0.99 1.96 1.83 1.12 1.71 24. Out of Budget 0.56 0.14 1.05 0.46 0.09 25. Inativos 0.31 0.32 0.41 2.11 2.87 26. Miscellaneous Cap or Rec 6.51 0.60 1.26 2.37 3.70 27. Total Primary Education = 100 100 100 100 100 (1 )+ 18 +(24)+ 25)+(26_ 99 Table 3.2i-b:,Stite PrimaryEducation,'Exjenditur.es- per. Rgi"on, 1.99g- ry pz . 4Centre.- North North-East South South- T q -: West ---. . %. t~ r; < °/° 1 ~ % East 1. Recurrent= (2)+(11)+(17) 94.26 85.47 72.06 81.32 82.28 2. Personnel = (3)+(6)+(7) 70.10 68.77 55.55 67.42 58.25 3. Teachers = (4)+(5) 58.19 55.18 44.79 51.25 49.65 4. In-Classroom Teachers 57.33 50.89 40.35 50.72 48.39 5. Out of Classroom Teachers 0.86 4.28 4.43 0.52 1.25 6. Administrative Personnel 4.88 7.79 4.27 7.64 5.60 7. Pedagogical Support Personnel 7.01 5.80 6.49 8.52 2.99 8. Training = (9)+(10) 5.00 0.55 0.99 0.51 0.81 9. Training of Lay Teachers 1.50 0.48 0.32 0.42 0.76 10. Training of Non-lay Teachers 3.50 0.07 0.66 0.08 0.05 11. Non-Personnel Recurrent = 20.52 13.95 13.49 7.22 18.70 (12)+(13)+(14)+(15)+(16) 12. Transportation 1.58 0.61 0.79 1.48 2.84 13. Repair and Maintenance 3.89 3.97 3.70 1.14 1.57 14. Didactic Materials 0.55 1.48 1.18 1.52 0.63 15. Other Consumption 9.10 2.46 4.52 1.56 12.18 16. School Lunch 5.38 5.41 3.30 1.51 1.45 17. Miscellaneous Recurrent 3.63 2.74 3.01 6.67 5.32 18. Capital = (19)+(20)+(21)+(22)+(23) 5.73 12.62 5.98 3.00 3.46 19. Civil Works 2.31 7.63 3.62 1.91 1.14 20. Acquisition of Vehicles 0.00 0.73 0.01 0.00 0.05 21. Acquisition of Equipment 0.93 1.79 1.67 0.97 1.27 22. Amortization of Debt 0.00 0.01 0.00 0.00 0.00 23. Miscellaneous Capital 2.48 2.44 0.66 0.12 0.99 24. Out of Budget 0.00 0.64 0.00 0.00 0.00 25. Inativos 0.00 1.25 9.50 8.66 12.83 26. Miscellaneous Cap or Rec 0.00 0.00 12.44 7.00 1.41 27. Total = (1)+(1 8)+(24)+(25)+(26) 100 100 100 100 100 100 Table 3.3: Prima,ry Education Expendictures,er Munici' al'Size, 1999. - -w v . - r - {~V ery Small' , ~ . Small ^Medium. -; Large Very Large I (POP<.20,000) (20-50,000) (50- (100- > 100,000 . _______ 100,000) 500,000) I. Recurrent= (2)+(11)+(17) 8935 84.13 87.74 85.96 86.38 2. Personnel = (3)+(6)+(7) 48.89 49.90 58.73 61.10 64.22 3. Teachers = (4)+(5) 34.04 35.73 42.20 44.68 47.54 4. In-Classroom Teachers '__32.52 33.93 39.38 42.17 43.82 5. Out of Classroom Teachers 1.52 1.79 2.81 2.50 3.71 6. Administrative Personnel 8.60 8.57 10.42 9.27 9.66 7. Pedagogical Support Personnel 6.24 5.59 6.11 7.14 7.01 8. Training = (9)+(10) _______ __________03 8. Training- (9)+(10) ~~~~ ~ ~~~~ ~~1.21 1.50 2.30 0.75 0.36 9. Training of Lay Teachers 0.42 0.78 1.69 0.27 0.08 10. Training of Non-lay Teachers 0.79 0.72 _ 0.61 0.48 0.28 11. ~~Non-Personnel Recuffent = 33.14 24.37 21.55 17.55 16.23 (I12)+(13)+(14)+(15)+(16) 12. Transportation 11.72 7.15 6.18 2.55 1.45 13. Repair and Maintenance 5.15 3.65 2.55 3.03 3.57 14. Didactic Materials 3.99 3.15 2.11 1.79 0.76 15. Other Consumption 6.19 4.71 4.71 5.19 5.21 16. School_Lunch 6.07 5.70 5.98 4.98 5.23 . ,~~~~~~~~~~~~~~~~~~~~~~~59 17. Miscellaneous Recuffent 7.30 9.85 7.44 7.31 5.91 18. Capital = (19)+(20)+(21)+(22)+(23) 7.89 10.89 9.34 8.80 5.60 19. Civil Works ___4.19 6.21 4.61 5.43 2.59 20. Acquisition of Vehicles 1.46 0.95 0.21 0.10 0.02 21. Acquisition of Equipment 1.51 1.80 2.13 1.45 1.51 22. Amortization of Debt 0.00 0.04 0.00 0.00 0.00 23. Miscellaneous Capital 0.71 1.87 2.38 1.79 1.47 24. Out of Budget 0.81 0.59 0_36 0.33 0.13 25. Inativos _0.29 1.13 0.70_ 2.00 5.04 26. Miscellaneous Cap or Rec 1.63 3.24 1.84 2.89 2.83 27. Total Primary Education = (1)+(18)+(24)+(25)+(26) 100 100 100 100 100 101 Table 3.4: FUNDEF Education Expenditures, 1999 All Brazil _ % I . Recurrent = (2)+(11)+(17) 93.05 2. Personnel = (3)+(6)+(7) 78.99 3. Teachers = (4)+(5) 66.03 4. In-Classroom Teachers 64.36 5. Out of Classroom Teachers 1.66 6. Administrative Personnel 7.38 7. Pedagogical Support Personnel 5.57 8. Training= (9)+(10) 1.54 9. Training of Lay Teachers 0.85 10. Training of Non-lay Teachers 0.68 11. Non-Personnel Recurrent = (1 2)+( 1 3)+( 1 4)+( 1 5)+( 16) 11.19 12. Transportation 4.31 13. Repair and Maintenance 2.43 14. Didactic Materials 1.93 15. Other Consumption 2.24 16. School Lunch 0.25 17. Miscellaneous Recurrent 2.86 18. Capital = (19)+(20)+(21)+(22)+(23) 6.30 19. Civil Works 3.06 20. Acquisition of Vehicles 0.37 21. Acquisition of Equipment 1.40 22. Amortization of Debt 0.00 23. Miscellaneous Capital 1.45 24. Out of Budget 0.00 25. Inativos 0.00 26. Miscellaneous Cap or Rec 0.64 27. Total = (1)+(18)+(24)+(25)+(26) 100 102 Table 3.5:.FUNDEF Expenditures per Region, 1999 _ _ Centre .North North- -South South- West % East % East .. %~ % , I. Recurrent = (2)+(11)+(17) 94.56 93.25 91.19 95.37 93.07 2. Personnel = (3)+(6)+(7) 84.47 77.50 72.00 87.83 80.30 3. Teachers = (4)+(5) 68.26 59.45 56.34 80.97 70.21 4. In-Classroom Teachers 67.96 57.22 55.47 78.55 67.98 5. Out of Classroom Teachers 0.29 2.22 0.87 2.42 2.22 6. Administrative Personnel 9.96 11.51 9.52 2.76 4.92 7. Pedagogical Support Personnel 6.24 6.53 6.12 4.09 5.17 8. Training = (9)+(10) 1.08 1.20 3.14 0.82 0.47 9. Training of Lay Teachers 0.42 0.96 1.90 0.43 0.00 10. Training of Non-lay Teachers 0.66 0.24 1.24 0.38 0.46 I1. Non-Personnel Recurrent = (12)+(13)+(14)+(15)+(16) 8.30 14.14 14.30 6.54 9.96 12. Transportation 3.53 5.00 4.46 3.21 4.74 13. Repair and Maintenance 2.05 2.90 3.59 0.80 2.04 14. Didactic Materials 1.57 4.26 2.25 0.76 1.20 15. Other Consumption 1.12 1.67 3.32 1.75 1.96 16. School Lunch 0.01 0.29 0.67 0.00 0.00 17. Miscellaneous Recurrent 1.78 1.60 4.88 0.98 2.80 18. Capital = (19)+(20)+(21)+(22)+(23) 5.03 6.57 8.18 3.74 6.06 19. Civil Works 2.00 3.89 4.20 2.27 2.14 20. Acquisition of Vehicles 1.34 0.33 0.16 0.07 0.54 21. Acquisition of Equipment 0.78 1.57 1.85 0.63 1.48 22. Amortization of Debt 0.00 0.00 0.00 0.00 0.00 23. Miscellaneous Capital 0.90 0.76 1.95 0.76 1.89 24. Out of Budget 0.00 0.00 0.00 0.00 0.00 25. Inativos 0.00 0.00 0.00 0.00 0.00 26. Miscellaneous Cap or Rec 0.39 0.16 0.61 0.88 0.85 27. Total = (1)+(18)+(24)+(25)+(26) 100 100 100 100 100 103 APPENDIX TO SECTION 5 Table 5.1- Variable description-and-datA source Variable Variable definition Construction (and unit of measure) Time Source span Outcomes AGEGRADIS14 Age-for-grade distortion in grades I to 4 (State and Measured as a weighted average over the I to 4 2000, INEP Municipal sub-systems). grade range of the proportion of students with age 1996 higher than the official one in each single grade (%) AGEGRADIS58 Age-for-grade distortion in grades 5 to 8 (State and Same over the 5 to 8 range (%) 2000, INEP Municipal sub-systems) 1996 PASSRATE14 Passing rate in grades I to 4 (State and Municipal Measured as a weighted average over the 1 to 4 2000, INEP sub-systems) grade range of the proportion of students 1996 successfully completing each single grade (%) PASSRATE58 Passing rate in grades 5 to 8 (State and Municipal Same over the 5 to 8 range (%) 2000, INEP sub-systems) 1996 DROPOUT14 Drop-out rates in grades I to 4 (State and Municipal Measured as a weighted average over the 1 to 4 2000, INEP sub-systems) grade range of the proportion of students dropping- 1996 out in each single grade (%) DROPOUT58 Drop-out rates in grades 5 to 8 (State and Municipal Same over the 5 to 8 range (%) 2000, INEP sub-systems) 1996 104 Table 5.1-continued: Variable description and data source Variable Variable definition Construction (and unit of measure) Time Source Expenditure STATEXPEND State expenditure per student in basic education Expenditure in basic education per State was 2000,1996 Finance obtained applying the enrollment share of basic Ministry, education in the State system per State to total INEP education expenditure. This amount was then divided by the number of students in basic education in the State system. (Reales) MUNEXPEN Municipal expenditure per student in basic education Expenditure in basic education per municipality was 2000, Finance obtained applying the enrollment share of basic 1996 Ministry, education in the municipal system per municipality INEP to total education expenditure. This amount was then divided by the number of students in basic education in the municipal system. (Reales) Expenditure composition SFUNDSHARE FUNDEF share in State expenditure FUNDEF expenditure as a proportion of total 2000, Ministry of expenditure in basic education in the State system 1996 Education (%) (;FUNDEF), Finance Ministry, INEP MFUNDSHARE FUNDEF share in municipal expenditure FUNDEF expenditure as a proportion of total 2000, Ministry of expenditure in basic education per municipality in 1996 Education the municipal system (%) (FUNDEF), Finance Ministry, INEP 105 Table 5.1-continued: Variable description and data source Variable Variable definition Construction (and unit of measure) Time span Source Municipalization MENROLSH14 Municipal enrollment share in grades I to 4 Enrollment in the municipal system as a proportion 2000, 1996 INEP of public sector enrollment (municipal + state) per municipality in grades I to 4 of basic education (%) MENROLSH58 Municipal enrollment share in grades 5 to 8 Same as above for grades 5 to 8 (%) 2000, 1996 INEP Control variables SFISCAP State fiscal capacity State own revenues as a proportion of total revenues 2000, 1996 Finance (%) Ministry MFISCAP Municipal fiscal capacity Municipal own revenues as a proportion of total 2000, 1996 Finance revenues (%) Ministry POPULATION Total Population Total Population per municipality 2000, 1997 IBGE URBPOP Degree of urbanization Urban Population as a proportion of total population 2000 IBGE per municipality (%) PIBPC PIB per-capita (municipality and state) Total PIB divided by total Population per 1996 IBGE municipality and state (Reales) (municipality); 2000, 1996 (State) PREBASENR Enrollment in pre-school education Total enrollment in pre-school education across the 2000,1996 INEP municipal and state system per municipality divided by total Population (%) 106 Table 5.1-continued:. Variable description and data source Variable Variable definition Construction (and unit of measure) Time Source span Other inputs CLASSIZE14 Class size in grades I to 4 (Statc and Municipal Weighted average of single grades (nb of students) 2000, INEP systems) 1996 CLASSIZE58 Class size in grades 5 to 8 (State and Municipal Weighted average of single grades (nb of students) 2000, INEP systems) 1996 GRADTEAC14 Graduate teachers in grades I to 4 (State and Weighted average of single grades (%) 2000, INEP Municipal systems) 1996 GRADTEAC58 Graduate teachers in grades 5 to 8 (State and Weighted average of single grades (%) 2000, INEP Municipal) 1996 CLASSHOUR14 Class hours per day in grades I to 4 (State and Weighted average of single grades (nb of hours) 2000, INEP Municipal systems) 1996 CLASSHOUR58 Class hours per day in grades 5 to 8 (State and Weighted average of single grades (nb of hours) 2000, INEP Municipal systems) 1996 107 Table 5.2: Some summary statistiks Variables Mean (SD) Variables Mean (SD) Variables Mean (SD) 2000 1996 2000-1996 difference Municipal sub-system AGEGRADIS14 38.2 (22.1) AGEGRADIS14 49.8 (24.7) AGEGRADIS14 -10 1 (11.54) AGEGRADIS58 64 (22.2) AGEGRADIS58 63.9 (24.3) AGEGRADIS58 -1.31 (13.6) PASSRATE14 75 2 (15.5) PASSRATE14 65.4 (16.2) PASSRATE14 8.8 (13 2) PASSRATE58 74.5 (13.5) PASSRATE58 69.4 (14.6) PASSRATE58 49 (15) DROPOUT14 10.9 (10.5) DROPOUTI4 156 (11.7) DROPOUTI4 -4.1 (108) DROPOUT58 15.9 (11.6) DROPOUT58 18 (12.8) DROPOUT58 -2.4 (127) CLASSIZE14 24.7 (5.7) CLASSIZE14 23 (7) CLASSIZE14 1.7 (5.5) CLASSIZE58 29.7 (7 8) CLASSIZE58 27.1 (9) CLASSIZE58 2.6 (6.9) GRADTEAC14 15 (20.3) GRADTEAC14 7.7 (14.5) GRADTEAC14 5.7 (12.4) GRADTEAC58 37.2 (34.3) GRADTEAC58 33.5 (33.5) GRADTEAC58 8.3 (22.4) CLASSHOUR14 4.2 (0.35) CLASSHOUR14 4 (0.33) CLASSHOUR14 0.15 (0.34) CLASSHOUR58 4.2 (0.6) CLASSHOUR58 4.1 (0.8) CLASSHOUR58 0.13 (0.83) MUNEXPEN 1551(2189) MUNEXPEN 1205(1916) MUNEXPEN 144 (1947) MFUNDSHARE 50.3 (23) FUNDSHARE 0 (0) FUNDSHARE 50.3 (23) MISCAP 74 (83) MFISCAP 5.6 (7.9) MFISCAP -0 9 (3 96) State sub-system AGEGRADIS14 34 3 (21.8) AGEGRADIS14 41.8 (21.6) AGEGRADIS14 -9.1 (11) AGEGRADIS58 52 (20.4) AGEGRADIS58 58.5 (19.3) AGEGRADIS58 -69 (11 5) PASSRATE14 79 (14.8) PASSRATE14 74.8 (13.9) PASSRATE14 5.2 (11.7) PASSRATE58 76.5 (11.9) PASSRATE58 70.3 (11.9) PASSRATE58 6.1 (11 9) DROPOUT14 10.3 (108) DROPOUT14 12.3 (9.5) DROPOUTI4 -2.2 (9.7) DROPOUT58 15 (9.9) DROPOUT58 17.6 (9.5) DROPOUT58 -2.6 (9 8) CLASSIZE14 26.7 (5.4) CLASSIZE14 29 (5.3) CLASSIZE14 -1.76 (42) CLASSIZE58 32.8 (6) CLASSIZES8 33.3 (5.7) CLASSIZE58 -0.3 (4 8) GRADTEAC14 24.7 (26.6) GRADTEAC14 23 (25.6) GRADTEAC14 4.4 (16.3) GRADTEAC58 54.6 (33.1) GRADTEAC58 52 (34 2) GRADTEAC58 4.5 (18.6) CLASSHOUR14 4.3 (0.36) CLASSHOUR14 4.2 (045) CLASSHOURI4 0.11 (0.35) CLASSHOUR58 4.3 (0.33) CLASSHOUR58 4.2 (051) CLASSHOUR58 0.13 (0.41) STATEXPEND 1055 (577) STATEXPEND 708 (277) STATEXPEND 347 (397) SFUNDSHARE 60 (21) SFUNDSHARE 0 (0) SFUNDSHARE 60 (21) SFISCAP 69 (17.8) SFISCAP 72 (16.8) SFISCAP -1.7 (8.3) Shared variables MENROLSHI4 73 (24) MENROLS14 52 (26.5) MENROLS14 20 (24) MENROLSH58 30 (34) MENROLS58 17 (27) MENROLS58 12.6 (25) POPULATION 31064(187256) POPULATION 29222 (177157) POPULATION 1820 (10703) URBPOP 58.3 (23.3) PIBPC(96) 2962(2842) PIBPC 2962(2842) PIBPC(State) 4391(2447) PEBPC(State) 3684(2042) PEBPC(State) 706 (501) PREBASENR 3.6 (10.4) PREBASENR 2.3 (2 2) PREBASENR 1.4 (106) 108 -Table 5.3: OILSestimates-Dependent variable: AGEGRADISi,4; DRPOUT14; PASSRATE14 Dependent variable: Dependent variable: Dependent variable: AGEGRADIS14 DROPOUT14 PASSRATE14 CS (OLS FD (OLS- CS (OLS FD (OLS- CS (OLS FD (OLS- robust Robust robust Robust robust Robust estimates- estimates) estimates- estimates) estimates with esti mates) with State with State State fixed fixed-effects) fixed effects) effects) Coefficients Coefficients Coeff. Coeff. Coeff. Coeff. (t-ratios) (t-ratios) (t-ratios) (t-ratios) (t-ratios) (t-ratlos) Variables _ Inputs CLASSIZE14 0.20 -0.03 0.05 0.02 -0.13 -0.01 (6.30)*** (-1.00) (1.94)* (0.55) (-3.76)*** (-0.41) GRADTEAC14 -0.07 0.00 -0.03 -0.01 0.05 0.01 (-11.30)*** (-0.01) (-6.79)*** (-0.99) (7.34)*** (1.28) CLASSHOUR14 -4.63 -0.02 -1.77 -0.44 3.19 0.21 (-6.52)*** (-0.05) (1.56) (-0.69) (2.35)** (0.28) Control variables SFISCAP -0.06 0.033 -0.005 -0.018 0.056 0.032 (-13.03)*** (1.99)** (-1.28) (-1.05) (10.56)*** (1.75)* MFISCAP -0.12 -0.03 -0.016 -0.0002 0.068 -0.04 (-6.50)*** ( 0.66) (-0.92) (-0.60) (2.82)*** (-1.01) POPULATION 0.00 -0.00 -0.00 -0.00 0.00 0.00 (0.58) (-2.59)** (-0.36) (-0.40) (0.84) (0.97) URBPOP -0.059 0.016 0.00 (-9.12)*** (3.12)*** (0.14) PIBPC -0.0001 0.004 -0 00 0.002 0.0001 -0.002 (-3.15)*** (10.84)*** (-2.58)** (5.47)*** (2.63)*** (-4.64)*** PREBASENR -0.029 0.012 -0.011 -0.007 0.021 0.024 (-2.03)** (0.51) (-0.81) (-0.47) (1.39) (1.22) R2 0.78 0.022 0.44 0.006 0.50 0.004 N of observations 7284 6199 7283 6194 7283 6181 *Significant at 10%; ** Significant at 5%; *** Significant at 1% 109 Table 5.4: OLS estimates-s ependent variable: AGEGRADIS58,; DROPOUT58; PASSRATE5S Dependent variable: Dependent variable: Dependent variable: AGEGRADIS58 DROPOUT58 Tap_f58 CS (OLS FD (OLS- CS (OLS FD (OLS- CS (OLS FD (OLS- robust Robust robust Robust robust Robust esti mates- estimates) estimates- estimates) estimates with esti mates) with State with State State fixed fixed-effects) fixed effects) effects) Coefficients Coefficients Coeff. Coeff. Coeff. Coeff. Variables (t-ratios) (t-ratios) (t-ratios) (t-ratios) (t-ratios) (t-ratios) Inputs CLASSIZE58 0.18 0.19 0.07 0.10 -0.15 -0.14 (5.83)*** (5.26)*** (2.58)** (2.84)*** (-5.02)*** (-3.35)*** GRADTEAC58 -0.05 -0.03 -0.02 -0.02 0.01 -0.01 (-7.95)*** (-3.20)*** (-3.79)*** (-2.49)** (1.03) (-0.65) CLASSHOUR58 -2.82 -1.34 -2.33 -0.46 2.18 1.02 (-2.58)** (-3.31)*** (-4.08)*** (-1.12) (3.79)*** (2.09)** Control variables SFISCAP -0.067 -0.007 0.028 0.048 -0.019 -0073 (-10.44)*** (-0.36) (4.89)*** (3.05)*** (-2.78)*** (-3.15)*** MFISCAP -0.078 -0.002 0.031 -0.05 -0.038 -0.00 (-2.46)** (-2.49)** (0.94) (-0.76) (-1.01) (-0.37) POPULATION 0.00 -0.00 -0.00 -0.00 0.00 0.00 (0.76) (-3.99)*** (-1.81)* (-2.77)*** (1.08) (2.94)*** URBPOP -0.041 0.038 -0.064 (-5.43)*** (6.27)*** (-8.84)*** PEBPC -0.0002 -0.00 -0.00 0.001 0.00 -0.00 (-2.51)** (-0.69) (-0.87) (2.25)** (1.19) (-1.75)* PREBASENR -0.047 0.008 -0.028 -0039 0.040 0.073 (-2.11)** (0.17) (-2.64)*** (-2.71)*** (2.35)** (4.36)*** R2 0.75 0.017 0.31 0.009 0.29 0.012 N of observations 5490 4393 5500 4381 5500 4378 *Significant at 10%; ** Significant at 5%; *** Significant at I% 110 Table 5.5: OLS estimates. Dependent variable: AGEGRADIS14; DROPOUT14; PASSRATE14 Dependent variable: Dependent variable: Dependent variable: AGEGRADIS14 (a) DROPOUT14 (a) PASSRATE14 CS (OLS FD (OLS- CS (OLS FD (OLS- CS (OLS FD (OLS- robust Robust robust Robust robust Robust estimates- estimates) estimates- estimates) estimates Estimates) with State with State with State fixed-effects) fixed fixed effects) effects) Coefficients Coefficients Coeff. Coeff. Coeff. Coeff. Variables (t-ratios) (t-ratios) (t-ratios) (t-ratios) (t-ratios) (t-ratios) Expenditure STATEXPEND -0.0002 -0.0009 -0.004 -0.005 0.004 0.0002 (-0.21) (-1.06) (-4.74)*** (-7 34)*** (4.20)*** (0.25) MUNEXPEN -0.0002 -0.00 -0.0001 -0.0013 0.0004 0.001 (-4.31)*** (-0.14) (-1.82)* (-3.05)*** (2.64)*** (3.09)*** Composition of expenditure _ SFUNDSHARE 0.020 -0.047 -0.047 -0.024 0.054 0.049 (1.32) (-5.60)*** (-3.88)*** (-3.19)** (3.41)*** (5.54)*** MFUNDSHARE 0.088 -0.085 0.004 -0.092 -0.023 0 10 (9.58)*** (-9.85)*** (0.50) (-10.37)*** (-2.27)** (10.34)*** Municipalization MENROLSH14 -0.007 -0.047 -0.010 -0.010 0.011 0.018 (-1.22) (-6.42)*** (-2.16)** (-1.70)* (1.76)* (2.30)** Other inputs CLASSIZE14 0.15 0.025 0.050 0.063 -0.11 -0.10 (4,56)*** (0.72) (1.95)* (2.02)** (-3.17)*** (-2 73)*** GRADTEAC14 -0.071 -0.002 -0.030 -0.011 0.049 0 020 (-I11.08)*** (-0.26) (-6.73)*** (-1.42) (7.27)*** (1.88)* CLASSHOUR14 -4.07 0.13 -1.32 -0.28 2.57 0.21 CLASSHOUR14 (-5.84)*** (0.28) (-1.17) (-0.44) (1*91)* (0.28) Control variables SFISCAP -0.013 0.034 0.083 0.051 -0 050 0.032 (-0.70) (1.74)* (5.22)*** (2.33)** (-2.52)** (1.43) MFISCAP -0.083 -0.067 -0.021 -0.065 0.060 0 061 (-4.36)*** (-1.68)* (-1.13) (-1.98)** (2.44)** (1.31) POPULATION 0.00 -0.00 -0.00 -0 00 0.00 0 00 (0.09) (-2.33)** (-0.45) (-0.12) (0 94) (0.75) URBPOP -0.061 0.014 0.0038 (-9.53)*** (2.73)*** (0.54) PIBPC -0.0001 0.003 -0.00 0.002 0.00 -0.001 (-2.55)** (9.48)*** (-2.23)** (5.52)*** (2.1 1)** (-3 35)*** PREBASENR -0.028 0.011 -0.010 -0.0055 0.021 0.024 (-2.09)** (0.42) (4079) (-0.38) (1-35) (1.21) R2 0.79 0.049 0.45 0.039 0.51 0.038 N of observations 7284 6127 7283 6122 7283 6110 111 Table 5.6: OLS estimates- Dependent variable: AGEGRADIS58;-DROPOUT58; PASSRATE58 Dependent variable: Dependent variable: Dependent variable: AGEGRADIS58 (a) DROPOUT58 (a) PASSRATE58 CS (OLS FD (OLS- CS (OLS FD (OLS- CS (OLS FD (OLS- robust Robust robust Robust robust Robust estimates- estimates) estimates- estimates) estimates Estimates) with State with State with State fixed-effects) fixed fixed effects) effects) Coefficients Coefficients Coeff. Coeff. Coeff. Coeff. Variables (t-ratios) (t-ratios) (t-ratios) (t-ratios) (t-ratios) (t-ratios) Expenditure STATEXPEND -0.002 -0.0009 -0.003 -0.0009 0.002 0.002 (-1.77)* (-096) (-2.88)*** (-1.11) (2.41)** (2.37)** MUNEXPEN -0.002 0.000 -0.0003 -0.002 0.001 0.003 (-2.92)*** (0.02) (-0.81) (-2.65)*** (2.86)*** (3.95)*** Composition of expenditure SFUNDSHARE -0.070 -0.10 -0.040 -0.067 0.051 0.010 (-3.04)*** (-8.45)*** (-2.00)** (-5.57)*** (2.26)** (7.26)*** MFUNDSHARE 0.044 -0.034 0.022 -0.064 -0.010 0.065 (2.40)** (-2.69)*** (1.65)* (-5.12)*** (-0.65) (4.3)**** Municipalization MENROLSH58 -0.033 -0.014 0.0039 0.0084 -0.011 -0.028 (-4.87)*** (-1.28) (0.62) (0.63) (-1.49) (-1.85)** Other inputs CLASSIZE14 0.15 0.12 0.060 0.097 -0.13 -0.11 (5.11)*** (3.33)*** (2.24)** (2.72)*** (-4.41)*** (-2.61)*** GRADTEAC14 -0.057 -0.036 -0.021 -0.020 0.008 -0.001 (-8.30)*** (-3.86)*** (-3.95)*** (-2.47)** (1.22) (-0.I1) -2.65 -1.35 -2.17 -0.40 1.97 0.87 CLASSHOUR14 (-2.46)** (-3.32)*** (-3.92)*** (-0.97) (3.48)*** (1.81)* Control variables SFISCAP -0.099 -0.075 0.11 0.018 -0.084 -0.030 (-0.44) (-3.48)*** (5.25)*** (0.96) (-3.46)*** (-1.26) MFISCAP -0.023 0.031 0.048 -0.046 -0.065 -0.001 (-0.69) (0.41) (1.30) (-0.68) (-1.62)* (-1.51) POPULATION 0.00 -0.00 -0.00 -0.00 0.00 0.00 (0.38) (-3.69)*** (-1.85)* (-2.46)** (1.19) (2.61)*** URBPOP -0.043 0.039 -0.064 (-5.84)*** (6.46)*** (-8.95)*** PIBPC -0.00 -0.003 0.00 0.001 0.00 -0.001 (-1.94)* (-0.65) (-0.71) (2.16)** (0.95) (-2.49)** PREBASENR -0.046 0.072 -0.030 -0.037 0.042 0.070 (-2.19)** (0.17) (-2.76)*** (-2.63)*** (2.49)** (4.15)*** R2 0.76 0.046 0.32 0.020 0.30 0.028 N of observations 5490 4370 5500 4358 5500 4355 *Significant at 10%; ** Significant at 5%; *** Significant at I%; (a) A negative sign indicates a positive contribution of the independent variables to the outcome. 112 Table'5.7: OLS estimates- Dependent variables: AGEGRADIS14; DROPOUT14; PASSRATE14 (all municipalities excluding North/Northeast) Dependent Dependent variable: Dependent variable: DROPOUT14 variable: AGEGRADIS14 PASSRATE14 FD (OLS robust FD (OLS- FD (OLS robust estimates) Robust estimates) estimates) Coefficients Coefficients Coefficients Variables (t-ratios) (t-ratios) (t-ratios) Expenditure STATEXPEND 0.005 -0.003 -0.045 (4.96)*** (-3.03)*** (-3.75)*** MUNEXPEN 0.00002 -0.0001 0.0002 (0.18) (-2.40)** (2.25)*** Composition of expenditure SFUNDSHARE -0.09 -0.023 0.026 (-10.57)*** (-0.29) (2.41)** MFUNDSHARE 0.044 -0.038 0.044 (-2.95)*** (-3.05)*** (2.58)*** Municipalization MENROLSH14 -0.048 -0.013 0.041 (-6.47)*** (-2.40)*** (5.19)*** Control variables SFISCAP -0.37 0.25 -0.85 (-3.80)*** (2.84)*** (-6.74)*** MFISCAP 0.027 -0.046 0.039 (0.57) (-1.50) (0.84) POPULATION -0.00 0.00 0.00 (-1.04) (1.85)* (-0.72) PEBPC -0.002 -0.0005 0.002 (-5.06)*** (-1.17) (2.90)*** PREBASENR 0.059 0.054 -0.043 (2.63)*** (1.14) (-0.73) R2 0.06 0.01 0.04 N of observations 4110 4104 4082 *Significant at 10%; ** Significant at 5%; *** Significant at 1% 113 Table 5.8: OLS estimates (a)- Dependent variables: AGEGRADIS14; DROPOUT14; PASSRATE14 (Northeast municipalities) Dependent Dependent variable: Dependent variable: DROPOUT14 variable: AGEGRADIS14 PASSRATE14 FD (OLS robust FD (OLS- FD (OLS robust estimates) Robust estimnates) estimates) Coefficients Coefficients Coefficients Variables (t-ratios) (t-ratios) (t-ratios) Municipalization MENROLSH14 -0.11 -0.063 0.065 (-4.93)*** (-2.87)*** (2.62)*** R2 0.15 0.11 0.14 N of observations 2394 2392 2393 *Significant at 10%; **Significant at 5%; ***Significant at 1% ; (a) Only municipalization coefficient reported. Table 5.9: OLS estimates- Dependent variable: MENROLSH14 Cs FD coefficients coefficients Variables (t-ratios) (t-ratios) CLASSIZE14 0.65 0.99 (14.6)*** (17.8)*** GRADTEAC14 -0.17 0.29 (-16.2)*** (10.8)*** CLASSHOUR14 -13.2 3.5 (-17.8)*** (3.9)*** 112: 0.1 0.1 *Significant at 10%; ** Significant at 5%; * Significant at 1% Table 5.10: OLS estimates- Dependent vaiiable: MENROLSH58 Variables CS lD CLASSIZE58 0.24 0.76 (4.8)*** (10.5)*** GRADTEAC58 -0.21 0.015 (-20.3)*** (0.7) CLASSHOUR58 -9.9 -0.73 (-12.6)*** (-1.2) 112: 0.09 0.06 *Significant at 10%; ** Significant at 5%; *** Significant at 1% 114 References Abreu, Mariza and Balzano, Sonia. (2001) "Progressao na Carreira do Magist6rio e Avaliacdo de Desempenho" in PRASEM HI: Guia de Consulta, organized by M. Marques and M. Giagio. FUNDESCOLA/MEC. Brasilia Afonso, Jose Roberto R., and Mello, Luiz de. (2000) "Brazil: An evolving Federation" Working Paper presented at the IMF/FAD Seminar on Decentralization. Washington, D.C. Boardman, A., and R. Murname. (1979) "Using Panel Data to Improve Estimates of the Determinants of Educational Achievement." Sociology of Education 52(2): 113-21. Castro, Jorge A. de (2001) "Financiamento da Educacdo no Brasil." IPEA Working Paper. IPEA, Rio de Janeiro. Deaton, A. (1995) "Data and Econometric Tools for Development Analysis", Handbook of Development Economics, Volume HI, edited by J. Behrmann and T.N. Srinivasan, Elsevier Science Filmer, D., and G. Eskeland. (2002) "Autonomy, Participation and Learning in Argentinian Schools: Findings and Implications for Decentralization." Policy Research Working Paper No. 2766. World Bank, Washington, D.C. Fuller, B., and P. Clarke (1994) "Raising School Effects While Ignoring Culture? Local Conditions and the Influence of Classroom Tools, Rules and Pedagogy." Review of Educational Research 64(1): 119-157 Fried, H.O., Lovell, C.A.K. and Shimdt, S. (Eds.) (1993) "The Measurement of Productive Efficiency: Techniques and Applications". Oxford, New York. Greene, W. (1993) "Econometric Analysis", McMillians Hanushek, E. (1986) "The Economics of Schooling: Production and Efficiency in Public Schools." Journal of Economic Literature 24(3): 1141-1177 Jimenez, E., and V. Paqueo (1996) "Do Local Contributions Affect the Efficiency of Public Primary Schools?" Economics of Education Review 15(4):377-386 Jimenez E., and Sawada, Y. (1999) "Do community managed schools work? An evaluation of El Salvador's EDUCO program" in The World Bank Economic Review, 13 (3):415-441 King, E. and B. Ozler (1998) "What's Decentralization Got to Do with Learning? The Case of Nicaragua's School Autonomy Reform." Working Papers on the Impact 115 Evaluation of Education Reforms 9. World Bank, Development Research Group, Washington, D.C. MEC (1999) Balango do Primeiro Ano do FUNDEF. Brasilia. MEC (2001) Balango do FUNDEF 1998-2000. Brasilia. Ministry of Education of Brazil, UNICEF and The World Bank (1997) "Brazil: A Call to Action: Combating School Failure in the Northeast of Brazil' Report No. 18358-BR. World Bank, Washington, D.C. Mora, M. and Varsano, R. (2001) "Fiscal Decentralization and Sub-national Fiscal Autonomy in Brazil: Some facts of the Nineties" Working Paper. IPEA, Rio de Janeiro. Moulton, B.(1990) "An illustration of a pitfall in estimating the effects of aggregate variables on micro units" in The Review of Economics and Statistics 72 (2):334-338. Newey W, Holtz-Eakin, D and H, Rosen. (1988) "Estimating Vector Autoregressions with Panel Data". Econometrica 56(6):1371-1395 Oliviera, Weder de (2000) "Lei de Responsabilidade Fiscal: Principais Aspectos Concernentes aos Municipios". Camara dos Deputados, Brasilia. Paes de Barros, R., and R. Mendonca (1998) "The Impact of Three Institutional Innovations in Brazilian Education." In William Savedoff, ed., Organization Matters: Agency Problems in Health and Education in Latin America. Washington, D.C.: Inter- American Development Bank. Pitt, M., M. Rosenzweig, and D. Gibbons (1993) "The Determinants and Consequences of the Placement of Government Programs in Indonesia." World Bank Economic Review 7(3):319-348 Tendler, Judith. (1997) "Good Government in the Tropics" JHU Press, Baltimore. World Bank. (2000) "Brazil: Higher Education Sector Study" Report No. 19392-BR. World Bank, Washington, D.C. World Bank. (2001) "Brazil: Teachers Development and Incentives: A Strategic Framework" Report No. 20408-BR. World Bank, Washington, D.C. World Bank. (2000) "Secondary Education in Brazil: Time to Move Forward" Report No. 19409-BR. World Bank, Washington, D.C. 116 FReport No.: 24413 BR Typs: S