WPS8125 Policy Research Working Paper 8125 Willing but Unable? Short-Term Experimental Evidence on Parent Empowerment and School Quality Elizabeth Beasley Elise Huillery Development Economics Vice Presidency Operations and Strategy Team June 2017 Policy Research Working Paper 8125 Abstract Giving power over school management and spending increased, and there was no measured impact on test decisions to communities has been a favored strategy to scores. An analysis of heterogeneous impacts and spending increase school quality, but its effectiveness may depend decisions provides additional insight into these dynam- on local capacity. Grants are one form of such a transfer ics. Overall, the findings suggest that programs based of power. Short-term responses of a grant to school com- on parent participation should take levels of community mittees in Niger show that parents increased participation capacity into account: even when communities are will- and responsibility, but these efforts did not improve qual- ing to work to improve their schools, they may not be ity on average. Enrollment at the lowest grades increased able to do so. The short-term nature of the experiment and school resources improved, but teacher absenteeism reduces the extent to which the results can be generalized. This paper is a product of the Operations and Strategy Team, Development Economics Vice Presidency. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at elise.huillery@sciencespo.fr. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Willing but Unable? Short-Term Experimental Evidence on Parent Empowerment and School Quality Elizabeth Beasley and Elise Huillery JEL codes: 015, C93, I21 Keywords: Community participation, education, parental involvement, randomized trial, school-based management. Elizabeth Beasley is the coordinator of the CEPREMAP Well-Being Observatory; her email address is elizabeth.beasley@cepremap.com. Elise Huillery (corresponding author) is a professor in the Sciences Po Department of Economics and J-PAL; her email address is elise.huillery@sciencespo.fr. Cornelia Jesse led the implementation of this project and contributed substantially to its design, and we are deeply indebted to her for her leadership. We thank Yann Algan, Bruno Crépon, Esther Duflo, Pascaline Dupas, Pierre de Galbert, Emeric Henry, Cornelia Jesse, Florian Maynéris, Miguel Urquiola, and the seminar participants at J-PAL Europe, Sciences Po, Columbia University, Oxford University, UCL, Paris I, and the Journées d'Economie Publique Louis-André Gérard-Varet for helpful comments and discussions. We also thank several anonymous referees for thoughtful and detailed suggestions. We thank Adama Ouedraogo for his support throughout the project and are grateful to Pierre de Galbert for excellent project management and Gabriel Lawin for data collection and management and also Elizabeth Linos, Andrea Lepine, and Hadrien Lanvin for research assistance. We thank the Government of Niger and the staff of the Ministry of Education for their collaboration, in particular Amadou Tchambou, Yacouba Djibo Abdou, Salou Moussa, and Damana Issaka. Mathieu Brossard was central in the initial conception and design of the project. Much of the project was carried out when Elizabeth Beasley was at J-PAL Europe and she thanks J-PAL for their support. Finally, and most importantly, we gratefully acknowledge the parents, staff, and pupils of the schools for the time and information they shared with us. This work was supported by the World Bank and the donor partners of the Education for All Fast Track Initiative through the Education Program Development Fund. All errors remain our own, and the opinions expressed in this paper are ours alone and should not be attributed to the institutions with which we are affiliated, the World Bank, or the Government of Niger. The dramatic expansion of access to schools in the last two decades is the result of an unprecedented effort to increase education in poor countries. However, the quality of education is often low. One common strategy to improve quality is through improved management and oversight and in particular by increasing involvement of parents and the community (World Bank 2003). Community-based management policies have been widely adopted throughout the world over the past decade (see Barrera-Osorio et al. 2009 for an overview).1 Grants to school committees, that is, putting money under the control of parents, are one potential way to increase school quality directly, by increasing school resources, and indirectly, by spurring parent participation. For this to work, parents must have the time, energy, and capacity to participate in school management effectively. Given the heavy investment in such programs, it is important to understand whether, and under which circumstances, they can actually work. This paper provides evidence from a field experiment on the short-term impact of a program to encourage parent participation in school management through grants to school committees in a context of low parent authority and capacity. In Niger, levels of education among adults are extremely low: 70% of Nigeriens aged 15–44 in 2010 had no education,2 and the system for education is very hierarchical and centralized. In a pilot program to improve school quality, the Ministry of Education of Niger, in partnership with the World Bank, gave grants to school committees that had been trained in school management with the aim of increasing parent involvement. A randomized evaluation was incorporated into the pilot project to provide information for scale-up. Detailed data from one thousand schools (split into five hundred treatment and five hundred control schools) were collected to assess the impact of the grant on parent empowerment, school management, and school quality. An important limitation of the study is that it provides only short-term evidence on behavioral responses: the first grant arrived in late 2007 and was meant to continue several years, but the evaluation ended in 2009. 2 The survey was administered during April and May of 2008, and administrative data were collected at the beginning of the 2008–2009 school year. This paper thus documents the short- term dynamics of an anticipated long-term program. On average, parents were willing to increase their participation in school management, but educational quality did not improve in a meaningful way as a result of this participation. There is an overall positive impact of the grant program on parents’ involvement and responsibility: communities with the grant participated more and took on more responsibilities than those without the grant, although the average community did not engage in supervising teacher presence. Parents did not reduce their own contributions in response to the grant. The impact on school management is mixed: cooperation between school stakeholders improved, but overall accountability did not change, and spending shows both expected and unexpected changes: there was more spending in infrastructure but also school festivals, playground equipment, and, most unexpectedly, investment in agricultural projects, which were probably noneducational but intended to make a profit. Finally, school quality did not improve with these changes, at least in the short term. There were subsequent improvements in infrastructure and health resources, as well as an increase in participation at the lowest grades: fewer dropouts in 2007/2008 and increased enrollment in grade two in 2008/2009, but there is no evidence of a change in test scores (note that we cannot exclude the possibility of a downward bias in the estimate of test score impact due to differential dropouts, but the lack of change in test scores at levels that had no participation changes supports the finding of no impact on test scores). Teachers decreased their effort in response to the grant, which can be attributed to the fact that some teachers have a preference 3 for a centralized government and might be reluctant to collaborate with parents, especially when parents do not spend the money on projects that make the teacher’s life easier. The paper then examines heterogeneous effects along several different dimensions and highlights three interesting patterns. First, in situations where the school committee is educated or has experience in another community organization—both of which we take as proxies for real authority—parents increased monitoring of teacher attendance in response to the grant (though this did not mitigate the negative effect of the grants on teacher attendance). Second, in small (one-teacher) schools, school committees spent on items that benefited the teacher, and teacher attendance increased in response to the grant in these schools. These results together suggest that teachers’ responses to parent participation depend on whether parents are acting in opposition to, or alliance with, the teachers. Third, rural schools used some of the grant to invest in agricultural opportunities, while urban schools did not but invested in school infrastructure instead. This study is related to two strains of the economics literature: parent participation and school resources. Previous evidence on the effectiveness of programs to increase quality via increased parent participation is mixed. Banerjee et al. (2010) report that providing information to parents about the school committee and training the community to measure educational performance in India had no impact on the activity of school committees and, therefore, no impact on education outcomes. Duflo et al. (2015) find that a training to empower parents helped mitigate the negative response of regular teachers to the addition of a contract teacher. In Madagascar, Lassibille et al. (2010) found that facilitating community/school interactions, combined with streamlining management practices had positive impacts on attendance and learning. Other studies have supplied evidence that empowering the community to manage schools improves school quality, though these papers generally do not include random variation 4 in treatment assignment, and so the identification is weaker. Bryk et al. (1998) and Hess (1999) have argued that student achievement improved in Chicago after the implementation of reform involving the community in school management and Di Gropello (2006) overviews four school- based management programs in Latin America and concludes that school-based management models have led generally to greater community empowerment and teacher effort. Participation in school management may also be linked to social capital more generally: Sawada and Ishii (2012) employ matching and instrumental variable approaches to measure the impact of the COGES program itself in neighboring Burkina Faso and find increases in social capital measured using several different tools, including field experiments. Another group of studies point to heterogeneity in the performance of participatory programs, and in the effect of decentralization more generally. Blimpo et al. (2015) find that training school committees had no impact on learning except in schools where school committee members were educated. Pradhan et al. (2014) find that an intervention to empower parents was effective only when combined with an intervention fostering the ties between the school committee and a local governing body. Decentralization of secondary school management in Argentina led to higher test scores in provinces with higher managerial capacity and lower test scores in provinces with lower managerial capacity (Galiani and Schargrodsky 2002). Galiani and Perez-Truglia (2013) review the empirical literature on school decentralization on educational outcomes and find that the better-off communities tend to profit more from decentralization than poor communities. Using panel estimation on PISA data, Hanushek et al. (2013) estimate that increasing school autonomy is associated with lower student performance in countries with generally lower performance and higher student performance in countries with generally higher performance. While the context of rural Niger is likely to be substantially different from these 5 contexts, there is good reason to anticipate that there may be heterogeneous impacts of parent participation. Previous studies on increasing school resources have found that it may crowd out the contributions of other actors. For example, parents in Romania decreased time spent on homework when their child was admitted to a better school (Pop-Eleches and Urquiola 2013). In Zambia and India, households decreased spending for education when they anticipated an increase in school funding (Das et al. 2013). In Kenya, civil-servant teachers decreased presence at school when school committee hired an extra teacher (Duflo et al. 2015). This paper contributes specifically to the literature on heterogeneity by showing that authority and capacity are important prerequisites for parents to undertake the more difficult aspects of management and that cooperation between parents and teachers (rather than confrontation) may be key. An overall message is that parents will not always or even generally make optimal spending and management decisions to increase quality. It may be costly and time- consuming, parents may not have good information about how schools work and thus may not make optimal decisions, and it may be very difficult to put pressure on teachers to improve service quality. It may be particularly difficult since capacity depends on parent power vis-à-vis teachers, or “real authority” in the terms of Aghion and Tirole (1997), who underscore the fact that formal authority (the right to make decisions) need not imply real authority (effective control over decisions).3 A major limitation of the paper is the short-term nature of the findings. Long-term follow-up was impossible, so it is possible that different results would have emerged after one or two more years. However, the results presented here are still useful: first, they give evidence about the barriers that communities may face at the beginning of participatory programs, and second, the richness of the data on spending decisions, contributions, involvement and 6 responsibility, and link to community characteristics gives some insight into the mechanisms at work within communities when making school management decisions. The remainder of the paper is as follows. Section 2 presents some background information on education in Niger and describes the school grant experiment. Section 3 presents the data and estimation strategy and section 4 the empirical results. Section 5 concludes. I. BACKGROUND AND EXPERIMENTAL DESIGN The grant program sought to empower parent school committees in a context where parents traditionally had very little control over their children’s schooling and where overall levels of learning were quite low. The experimental design was incorporated to give information on program effectiveness prior to an intended scale-up. Background on Education in Niger Niger made remarkable progress in access to education in the decade prior to this evaluation: the number of children enrolled in primary school had more than doubled from 656,000 in 2000 to 1,554,102 in 2008, and net enrollment had risen from 27% to 49% in the same period. However, only 44% of children who begin primary school finished all grades, and only 43% of the sixth graders who took the national exam at the end of primary school passed it.4 The education system in Niger has traditionally been fairly hierarchical and rigid. Inherited from French colonization, the system replicates the French education system: highly centralized, with little, if any, room for local community participation. Unlike other systems, where the school might be supervised by a local governmental body, at the time of the evaluation there was generally no way for the local community to determine school policy or practice. Schools depended entirely on the hierarchical chain that originated in the Ministry of Education 7 (except for some local fundraising, but these efforts were undertaken only when needs were not provided for by the Ministry). In 2006 the Ministry of Education in Niger introduced school committees in all primary public schools in order to improve quality. These school committees (called the COGES) were designed to involve parents and community members in the school, improve accountability, improve management, and thus enhance access to and quality of education.5 As discussed in the introduction, the establishment of local community groups for the purpose of improving public service provision via community participation is a strategy that many country governments and civil society organizations advocate. In many respects, the circumstances of Niger make a strong case for school-based management: low population density, vast distances and limited transportation, information, and communications infrastructure make supervision of primary schools by the central government (or its regional structures) very costly, and the timely transmission of information to and from the central authorities for planning purposes is challenging. In the districts where this program was carried out, the COGES were trained by several different organizations in financial management, governance (elections), and project planning. In 2006, many of the newly created and trained school committees were not actively engaged in school matters, nor did they develop a school improvement plan for the year. To spur school committee involvement and activity, the Ministry of Education introduced school grants to give the committees an incentive to meet, plan, and undertake activities. The grants were expected to improve school management through increased parental participation and accountability, to improve school infrastructure and the quality of education, and to potentially increase enrollment rates and learning. The pilot project was carried out as a randomized evaluation in order to provide reliable information on impact prior to national scale-up. 8 The Ministry selected the regions of Zinder and Tahoua because the COGES there were already functional and had received basic training on planning and financial management, whereas COGES in the other six regions of Niger had not been trained yet. However, the context of these two regions is specific, even relative to the rest of Niger. The Zinder region is culturally similar to Northern Nigeria, with a relatively conservative Muslim population that has lower rates of formal schooling. On the other hand, the Tahoua region is a nomadic region where formal education poses a challenge because the nomadic population (the Tuareg and the Fulani) may often rely on children for herding. In both cases, one may expect parents to adhere less to formal schooling than in other regions in Sub-Saharan Africa. Experimental Design The evaluation design included 1,000 schools in Tahoua and Zinder, randomly selected out of the 2,609 total public primary schools in those districts. Once these 1,000 schools were determined to be representative of the total pool of schools in those districts, half were randomly assigned to receive the grants and became the treatment group. The other 500 schools served as a control group. Both randomizations were stratified on inspection (a geographical administrative unit), existing support for the school committee (e.g., existing programs or sponsorship by NGOs), and whether the school was indicated as being in a rural or urban area in administrative data. Strata were constructed by grouping the schools into inspections, then within each inspection into whether or not the school had existing support, and then within each of those groups, whether the school was in a rural or urban area. This gave fifty strata. Schools were assigned a random number between zero and one, and within each stratum they were sorted by this random number, with the first half being assigned to treatment and the second to control. Data from the Administrative School Census in 2005–2006 (the school census is described 9 below) were used to confirm balance between control and treatment schools along various observable characteristics (data from 2006–2007 were not yet available at the time of sampling in August 2007). The balance checks for the randomization and p-values for the test of equality of means across control and treatment are presented in table 1, and show no statistically significant differences. The original project plan called for recurrent grants to schools for three consecutive school years, to be distributed at the beginning of each school year to support COGES activities. The Ministry of Education and the Ministry of Finance jointly worked out the grant transfer mechanism, consisting of a direct release of funds from the national treasury into the accounts of the two regional education authorities (i.e., one hierarchical level down from the national government). The funds were then transferred to the inspection level and then to the COGES. The transfers from the regional authorities and below took place as transfers of cash, which were recorded using signed receipts, which were submitted to the Ministry of Finance. In the first year, rather than receiving the grants at the beginning of the year as planned, the five hundred COGES received the grants during December 2007 and January 2008, with the school year already in full swing, due to logistical difficulties with the transfer. The grants were not immediately distributed during the 2008–2009 school year, due to problems with the transfer mechanism.6 Due to these issues and political disruptions in 2009, the evaluation had ended after only one year. As a consequence, this evaluation evaluates only one year of the grant (the 2008/2009 grant was eventually distributed to some schools, after the evaluation was terminated). The size of the grant was based on the size of the school (the number of classrooms), and the average was 209 USD per school, or 1.83 USD per student. The grant was a relatively modest amount that was determined by considerations of financial sustainability in view of a potential 10 extension of the program by the government. For the purposes of comparison, the control schools raised a little over 0.60 USD per year per student from the parents on average and had an overall budget of around 199 USD including donations from private NGOs, and so the grant is relatively large compared to the usual fundraising and about equivalent to the annual amount of money available for school projects (note that in principle most school inputs such as teachers and books were provided in kind by the central government and so not included in this 199 USD —if they were, the grant would be smaller than the overall operating budget of the schools). For an idea of the practical scale, the amount of the grant was not, except in the very largest schools, sufficient to build an additional classroom. This grant amount is smaller than grants provided to school committees in most other evaluations: Blimpo and Evans (2014) use a grant of 500 USD per school in Gambia. Gertler, Patrinos, and Rodríguez-Oreggia (2010) use grants of 500 USD to 700 USD per school in Mexico, and Pradhan et al. (2014) evaluate a grant of 326 USD (to be followed with another grand of 544 USD) per school in Indonesia. About a month before the grant arrived, all five hundred treatment schools (and school committees) received a letter informing them of the grant program and its objectives, and the grant amount allocated to their school. It also included general guidelines on the use of the grants, but the specific activity to be supported by the grants was to be decided on by the school committee.7 One copy of this letter was distributed to the school director and a second copy to the president of the school committee before the arrival of the grants. Compliance in this respect was satisfactory: the grants arrived in 498 schools of the 500 program schools, 492 in the exact amount allocated to them and six in a different amount (see appendix S2, for further details on compliance). 11 II. DATA AND EMPIRICAL STRATEGY Multiple sources contain rich information on potential treatment outcomes and community characteristics that can be used to generate estimates of treatment impact and heterogeneous impact using a simple ITT framework. Data Data come from three sources: (i) administrative data on primary schools (the Ministry of Education’s annual school census, also called administrative data), (ii) an evaluation survey administered to school staff and two members of the school committee at treatment and control schools (the 2008 School Survey), and (iii) a financial control survey administered to one member of the school committee on a subset of treatment schools. The Ministry of Education in Niger administers an annual census of all primary schools, including community schools and madrassas (Koranic schools), which provides data on enrollment, teacher characteristics, school facilities and resources, and community characteristics. This paper uses the 2006/2007, 2007/2008, and 2008/2009 censuses. Each census is collected in the fall of the school year (for example, the 2008/2009 census contains the information reported by the schools in fall of 2008). In addition to the administrative data, the Ministry and the World Bank worked with a local NGO to prepare a detailed school survey (the 2008 School Survey) to be administered to the one thousand schools included in the experiment in April/May 2008, five to six months after grant distribution, to understand the immediate effects of the grant. This questionnaire included information on school infrastructure and resources, pupil enrollment and attendance, school improvement plan, school committee functioning and membership, and school activities. It also asked detailed questions about the level of education and personal wealth of the school 12 committee members. Three tests were also administered at this time: a math test, a French test, and an oral exam. The oral exam was administered to the youngest (grades one and two) pupils. Teacher’s physical presence at that visit was also recorded. The visit was on a day when the school was supposed to be open but was not announced in advance. Finally, a financial survey was administered to eighty-five randomly selected treatment schools in January/February 2009, asking detailed questions about the receipt and spending of the grants, any problems with the administration of the grant, and use of the grants (including the existence of a receipt for each expense). Use of the Grants The school committees used the grants in a variety of ways. Eighty-five schools were randomly selected for a detailed questionnaire on grant arrival and spending. The most common use was for material inputs such as construction and office supplies, and other uses included investment projects, health and sanitation projects, and transportation. Overall, the largest share of spending of the grant was in construction, representing about a third of the total amount spent (figure 1). Construction activities included building classrooms, but communities also constructed lodging for teachers, latrines, school enclosures, and other buildings. Other projects included electrification or producing copies of exams. About fifteen percent of schools surveyed used at least part of the grant on some sort of agricultural investment project. It is unclear whether the loans or small business projects were profitable. Outcomes 13 The analysis uses many different indicators of parent participation to draw general conclusions about the experiment’s impact. In order to simplify interpretation and to guard against cherry-picking of results, it presents results for indices that aggregate information over multiple outcome variables (following Kling et al. 2007). The aggregation also improves statistical power to detect effects that go in the same direction within a domain. The summary index Y is defined to be the equally weighted average of z-scores of its components, with the sign of each measure oriented so that more beneficial outcomes have higher scores. The z-scores are calculated by subtracting the control group mean and dividing by the control group standard deviation. Thus, each component of the index has mean 0 and standard deviation 1 for the control group. The index is the average of the nonmissing components, as long as the school has a valid response to at least two components. If only one component is available (or if no components are available), the school is dropped. Different types of outcomes are calculated in this way: parent participation, school management, and school quality. For each outcome, several indices are constructed. The details and full list of component variables for each index are given in appendix S3. For parent participation in school, the paper uses indices of parent contributions (e.g., school fees), involvement (e.g., going to meetings), responsibility (e.g., in charge of supplies), and teacher oversight (e.g., monitoring teacher attendance). School management is measured by two indices, accountability (e.g., keeping records) and cooperation (e.g., reported conflicts), and also by total spending across eight possible spending categories (infrastructure, supplies and textbooks, pupil educational support (e.g., remedial courses), pupil health, teacher support (e.g., housing), COGES expenses (e.g., travel to regional meetings), school festivals and playground, and investments in agriculture). Finally, the effect of the grant on school quality is measured by four indices: infrastructure (e.g., number of desks), materials (e.g., textbooks), health resources 14 (e.g., first aid kit), and teacher effort (e.g., teacher attendance). Data for infrastructure, materials, and health resources come from the 2008/2009 annual administrative database, collected in the fall of 2008, and so reflect changes between eight and ten months after receipt of the grants. The paper also uses data on dropouts, enrollment, and test scores in order to examine the ultimate objective of increasing pupil participation and learning. Participation in education is measured by the number of dropouts reported by the school to surveyors at the April/May 2008 questionnaire and the change in enrollment from fall 2007 to fall 2008 reported to the Ministry of Education in the annual administrative censuses. The paper uses two limited measures of actual learning. First, test scores are obtained from a test administered to pupils during the April/May 2008 questionnaire. The test was administered to three grades, ten pupils per grade. The pupils were supposed to be sampled from those who were enrolled at the beginning of the year, but in practice the ten pupils appear to have been selected from the pupils present on that day. There are further quality problems with the test scores—including identical copies submitted by some grades in some schools—that raise concerns about the quality of the test score data. There is no evidence that the problems are correlated with treatment and appear instead to be related to insufficient oversight of the examiners, so it is possible that the quality problems only add noise. However, as discussed below, the fact that participation is higher in the treatment schools and test takers were sampled from those present on that day leads to concerns of attrition bias in the test scores (if more children stayed in school in the treatment group, then the impact on test scores may be biased downwards). The results are therefore considered as second-order evidence. The overall results are nonetheless informative about the general level of education in rural Niger, and some examples are provided here to help give the reader a better idea of the context. In general, after 15 discarding duplicate and suspect observations, pupils got about one third of questions correct. For example, the following questions were asked:  Grade one: The interviewer asked the pupils to pick up a red crayon and a blue crayon out of a pile containing pieces of chalk of different colors: three white, one red, one blue, one yellow, and one green. 45% of pupils were able to do this.  Grade four: Pupils were asked to place the following numbers in order, from smallest to largest: 807; 708; 788; 800. 24% of pupils were able to do this.  Grade six: Pupils were asked to change an adjective from the masculine to the feminine form (Un nouveau maitre ==> Une ____________ maitresse). 29% of pupils were able to do this. The second measure of learning comes from the annual administrative censuses which report the number of candidates for the national end-of-primary school exam and the number who passed. Results for the end of the 2007/08 school year were reported on the 2008/09 census.8 On average, slightly over half of the schools presented at least one student for the end of sixth grade test (recall that most schools do not have all grades). Interaction Variables The sample size was chosen to be large enough to allow testing for heterogeneous treatment effects by community characteristics, and this was one of the initial objectives of this the study.9 The dimensions chosen for measurement of heterogeneous effects are those that are likely to affect parent response to the grant or that have policy relevance: education, experience in other organizations, wealth of the COGES, whether the school is in an urban or rural area, and 16 whether it is a one-teacher school. Descriptive statistics and balance information for the interaction variables are given in table 2. Our intuition is that COGES with higher levels of education and experience in other organizations are likely to have higher capacity to manage schools. To make sure that these dimensions are not merely proxying for wealth, wealth is also included as an interaction term (and it is not impossible that wealthy communities might react differently, either because they have more real authority or because they can leverage a larger supplemental contribution from the community). The distinction between urban and rural schools is important for education planners in general, and it is also important to ensure that the other interaction terms are not just proxies for the urban rural divide. Finally, one-teacher schools present a unique situation in terms of the power dynamics between the teachers and the parents, and these very small schools are also of relevance to education planners. Further details on the construction of these interaction terms are given in appendix S3. Attrition There is some attrition in the datasets. Each year, a handful of schools do not return the administrative data questionnaire, or the questionnaires are improperly filled out, leading to missing data for 3% of the schools for the infrastructure index and 1.4% of the schools for 2008/09 enrollment. The April/May 2008 survey was conducted on the basis of unannounced visits, which meant that many schools were closed. In addition, some schools were not visited due to security concerns, and still others closed early that year because the summer rainy season began early and many children went to the fields with their parents to work. As a result, data from the evaluation questionnaire are available for only 814 of the 1000 schools. 17 Differences in the proportion of schools with missing outcome variables are tested by treatment group as a whole and subdivided by district, urban and rural, and whether the school had external support (for example, NGO sponsorship) prior to the project. Results are reported in appendix S2, table A1. Eighty-four tests on treatment and interaction between treatment and subgroups yield one statistically significant difference (at the 10% level or higher), which is well within the amount that would be expected with random attrition. The comparability between treatment and control groups is thus intact. As to external validity, there are more schools missing in the region where security was a concern (Tahoua, in the north). Empirical Strategy The estimations present intent-to-treat effects as measured by the differences in the means of school outcomes between schools initially assigned to the treatment group and schools initially assigned to the control group. Let T be an indicator for treatment group assignment and let X be a vector of covariates. Estimation of the intent-to-treat effect β is from the following equation: Y j = βT j + X j γ + ε j (1) where Yj is the outcome of school j. The covariates Xj are included to improve estimation precision and include whether the school is urban, the total proportion of girls in 2007/08, the total enrollment in 2007/08, whether the school was supported by an outside NGO in 2006/07, and the inspection (a geographic/administrative unit). All regressions use robust standard errors.10 The absolute magnitudes are in units of the outcome’s standard deviation (based on the control group), so the estimate shows the treatment effect in terms of standard deviations. 18 Heterogeneous Treatment Effects Along Community Characteristics In the second step, intent-to-treat effects are estimated with an interaction term to determine whether the average treatment effect on parent and teacher behavior varies with community characteristics, using the following regression specification: Y j = βT j + θ(C j T j)+ σ C j + X j γ + ε j (2) where C j denotes a given community characteristic. In this case θ is the additional (or reduction of) impact for schools with characteristic C j . We include an indicator for urban schools and the interaction of this indicator with the treatment assignment for each characteristic whose correlation with being located in an urban area is above 0.1, to disentangle the effect of this characteristic from the effect of being located in an urban area. III. RESULTS On average, parents did not reduce their own contributions in response to the grant and increased their involvement in and responsibility over school management, although they did not go so far as to enforce rules on teacher attendance. At the same time, school committees increased investment in infrastructure (buildings and the school enclosure) and school festivals and invested in agricultural projects. Accountability did not change, but reported cooperation with a number of school stakeholders improved as a result of the grant. All these effects did not create a path to school quality improvement. While infrastructure and health resources improved and pupil participation increased a bit among the youngest, teacher attendance declined on average, perhaps because of resentment over parent empowerment, and no impact is found on 19 test scores. Particular impacts on the detailed components of each index are given in appendix S4. Appendix S5 provides a model that explains the results of this paper and the existing results in the literature. Parent Participation The grants did not change parent contributions to schools (table 3, column 1). The contribution index mean of the treatment group is statistically and economically similar to the mean of the control group. The analysis of the component variables (funds collected per pupil, in kind donations, and official fees charged) shows that neither financial nor in-kind contributions were affected by the grant (table A2). This result contrasts with previous studies showing that parents decreased their contributions in response to an increase in school resources (Das et al. 2013; Pop-Eleches and Urquiola 2013).11 Note that in general the amount of cash income available to schools is obtained through parental contributions. An important consequence of this is that, due to the increase in cash from the grant, cash on hand for schools increased and thus so did the possibility for investment. The parent involvement index increased (table 3, column 2), as did all of the individual components, although no change in any individual component is significant: the number of meetings was higher, time elapsed since the last meeting was smaller, the number of topics addressed in the meetings is larger, and the presence at the last meeting is larger (table A3). Overall, the mean of the parent involvement index in the treatment group is 0.06 standard deviations larger than the mean of the control group, and this effect is significant at the 10% level. 20 The impact of grants on parent responsibility in school management is reported in table 3, column 3. The overall effect of the grants is positive: the mean of the index of the treatment group is almost 0.06 standard deviations above the mean of the control group. The analysis of detailed variables composing the index shows some small increases in the proportion of school committees in charge of infrastructure, collecting financial contribution and spending financial contributions, although none of these increases are statistically significant (although some of p- values are close to conventional significance), while the effect on the index itself is significant at the 10% level (Appendix S4, table A4). There is no overall impact on parent supervision of teachers (table 3, column 4). Changes in the proportion of school committees which discuss teacher behavior in school committee meetings, declare that they are active in increasing teacher attendance and improving education quality, declare that they monitor teacher attendance, and take remedial actions against teachers are small and insignificant (Appendix S4, table A5). No trend emerges from these variables, and so there is no change in the teacher oversight index. School Management While there is no impact of the grant on school accountability overall (table 3, column 5), the analysis of the detailed components shows a 13% increase in the proportion of schools that could present a register for fund collection for examination and a 21% increase in the proportion of schools that could present a register for fund expenses for examination, which might be simply the direct consequence of the fact that schools in the treatment group received money from the government and had something to record, rather than an overall change in accountability (Appendix S4, table A6). However, the grant did not change the use of other 21 registers nor the frequency of minutes, which suggests that the increased involvement and responsibility of parents did not lead to a higher demand for transparency and record keeping. Overall, the cooperation between the school committee and different actors improved (table 3, column 6): school committees are significantly more likely to report support from the community (+5 percentage points), from the teachers (+3 percentage points), and from the parent committee (+5 percentage points) (Appendix S4, table A7). The proportions of school committees reporting support from local authorities, school administration, educational advisors, and inspection are also consistently larger, although these differences are not significant. As a result, mean of the cooperation index for the treatment group is almost 0.07 standard deviations above the mean of the control group, significant at the 5% level. One explanation for the positive effect of grants on cooperation between school stakeholders and school committees is that giving resources under the control of the school committee increased respect for its activities. The positive effect of the grant on the cooperation between the school committee and the different actors may be important when considering the short term nature of the experiment. It echoes the short term effect of a similar program on social capital observed in Burkina Faso (Sawada and Ichii 2012). Treatment schools increased spending on infrastructure, festivals and playground, and investments in agriculture. The absolute and percent differences in amounts budgeted for a given type of project in treatment schools compared to comparison schools are presented in figure 2 (significant differences in dark grey, nonsignificant in light grey). The amount budgeted for a given type of project was significantly larger for infrastructure, festivals and playgrounds, and investments in agriculture (table 4): the amount budgeted for infrastructure was 20% larger in the treatment group (107,705 FCFA (215 USD) versus 86,119 (172 USD) significant at the 5% level), the amount budgeted for festivals and playgrounds was sixfold greater than in the control group 22 (1031 FCFA (about 2 USD) versus 166 FCFA (0.33 USD), significant at the 1% level), and the amount budgeted for investments in agriculture was fourfold greater (2,416 FCFA (5 USD) versus 583 FCFA (about 1 USD), significant at the 1% level). Note that the difference, while large relative to the amount spent in control schools on these activities, is small compared to the entire amount of the grant, so the bulk of the grant was not used on school festivals, playground and agricultural investments. The size of the increase in infrastructure spending in absolute terms (19,659 FCFA, or 40 USD) is much larger than the increases in agriculture and festival and playground expenses (1,833 FCFA (a bit less than 4 USD) and 865 FCFA (almost 2 USD), respectively). The investments in agriculture do not seem to have been done in the interest of one person, which might be considered a theft of resources, but rather as an investment on the part of the school (since they were recorded in the school ledger). One interpretation of the investment in agricultural projects is that credit in many areas of Niger is severely constrained. There may be profit opportunities from investment in agriculture (either in terms of raising crops or arbitraging prices for inputs or food products), but since isolated areas suffer from low levels of credit, these profitable opportunities are unexploited. If the COGES is aware of these opportunities, and they are patient, it may be most optimal for the long-term interest of the school to invest the windfall cash grant rather than spend it on educational inputs immediately. However, one cannot be sure that these investments were made for the profit of the school and they may not have benefited the pupils in any way.12 Finally, school committees had spent just above a quarter of the grant at the time of the April/May 2008 questionnaire: the average increase in the total spending amount is 28,512 FCFA (57 USD), while the average grant is 104,500 FCFA (209 USD). This finding indicates that about five months after the grants arrived in treatment schools, the school committees had not yet used 23 the remaining three-quarter grant. Together with the types of spending induced by the grant, these results suggest that the school budget constraint is not immediately binding: a large part of the grant is still unused, and some money is spent on leisure and agricultural spending, which seem nonessential for pure educational purposes. Also, the amount budgeted for teacher support is unchanged (the average amount in the treatment schools is even lower than in the control schools, although the difference is not significant), which is striking in a context where teachers suffer from long delays in the payment of their salary. Similarly, it seems surprising that the grant did not change the amount of money spent on supplies and textbooks, pupil educational support like remedial courses, or pupil health expenses, in a context where school equipment is very poor and pupils do not perform well at the primary school final exam. Overall, the impact of the grant on school expenses suggest that in the context of Niger, parents might not have sufficient information to make investments that are likely to improve school quality. Other explanations, which may simultaneously be true, are that parents were saving the grant in the face of uncertain future cash flows (see Sabarwal et al. 2014), that they were saving money in order to offset fees in the following year, or that they were saving money for lumpy investments School Quality Improvements are observed only for infrastructure and health resources, alongside small increases in participation at the lowest grades. There is no improvement on materials nor on teacher effort. On the contrary, there is a small decrease in teacher attendance. There is no evidence that test scores increased in response to the program. In the slightly longer term (one year after the treatment) there is a small improvement in the infrastructure index of schools: a 0.04 standard deviation increase in the index for 24 infrastructure quality (table 3, column 7), significant at the 10% level. This is largely driven by increases in the number of classrooms and the construction of walls around the compound (Appendix S4, table A8).13 The increase in the number of new classrooms amounts to 0.12 of a standard deviation, representing an additional 0.08 new classrooms per school in the treatment group over 0.28 new classrooms per school in the control group (a 29% increase). The increase in the proportion of schools with walls around the compound (enclosure) amounts to 0.18 of a standard deviation, with 9 percentage points more in the treatment group over 34% in the control group (a 26% increase). There is no overall impact on the materials available at the schools (books and classroom materials such as rulers, protractors, and maps) (table 3, Appendix S4, table A9). There is a small (0.05 standard deviations) increase in the index of health resources (table 3, column 9), significant at the 10% level. This increase is driven by increases in health information sessions (34% versus 30% of schools), first aid kits (12% versus 9% of schools), micronutrient supplementation (25% versus 22% of schools), and deworming (64% versus 62% of schools), though none of the individual components of the health index are significant alone (Appendix S4, table A10). There is no effect of the grant on the number of days when class was cancelled because teachers were on strike, nor on the opinion of the school committee on teacher assiduousness and punctuality, but a decrease in teacher presence is observed in the treatment group: around 4 percentage points less than the average of 76% presence in the control group, significant at the 10% level (table 5). Teachers thus responded to increased resources under the control of parents with a reduction in their own inputs. Informal feedback from the field suggested that those teachers who felt the central government should make education decisions disliked that the communities were in charge of the grant, and they may have felt resentful that the grants 25 undermined their authority (as representatives of the central government). In addition, the decreased teacher presence might also be related to the fact that the average school committee did not spend the grant on expenses supporting the teachers (teacher housing, furniture, supplies, guide books, and salary), even though school committees had not spent the entire grant at the time of the survey. As a consequence, teachers might have had the impression that parents were not capable of wisely investing the money allocated to them and might have been resentful. Any such resentfulness might have been exacerbated by the ongoing pay disputes between the teachers and the government at that time (in many cases, teachers’ salaries had been substantially delayed or teachers had not been paid). There is no change in enrollment or dropout overall (table 6), but there is a positive impact at the lowest grade levels. The grant program reduced dropouts from grade one at the end of the 2007/2008 school year (2% versus 3% in the control schools) (column 4 of table 6b), a finding that is matched by an increase in enrollment in grade two at the beginning of the 2008/2009 school year (thirty-three versus thirty pupils in the controls schools) (column 5 of table 6A). The fact that participation increases only for the youngest pupils suggests that participation is more elastic when the child is young. This might be because the opportunity cost of time is higher for older children.14 The number of candidates presented for the end of primary school exam at the end of the 2007/2008 school year, the pass rate for the end of primary school exam, and the math, French, or oral tests administered during the April/May 2008 questionnaire visit were not affected (table 7). Since participation increased (or fewer children dropped out) in the lowest grades, one cannot rule out a downward bias due to attrition. However, the fact that test scores remained unchanged 26 in the higher grades where there was no change in participation supports the finding that there was no improvement in learning. Heterogeneous Treatment Effects The paper now examines the different dimensions identified above to identify heterogeneous effects. Due to space limitations, we do not present the detailed regression tables in the paper, but they are available from the authors upon request. There are two overall messages from this analysis. The first is that the most difficult management task—monitoring teachers—was undertaken only by educated COGES or those with experience in other organizations, that is, those with higher capacity. The second is that, in one-teacher schools, there was a greater threat of teacher strikes, more of the grant was spent on items that benefited the teachers in some way, and, perhaps as a consequence of spending on items that benefited teachers, teacher presence increased slightly. Education of the COGES Communities where the school committees were educated increased their supervision of teacher attendance in response to the grant. Educated school committees are 9 percentage points more likely to supervise teacher presence if the school was treated, significant at the 10% level. However, the increased monitoring did not reduce teacher absenteeism, suggesting that parents were not able to effectively confront teachers. In terms of spending, educated COGES who received grants focused investments on infrastructure, perhaps to the detriment of other types of spending.15 COGES without educated members, on the other hand, increased spending on Health Resources and Pupil Educational Support.16 27 The negative impact of the grant on money for Pupil Educational Support and the health resources index might reflect that educated COGES increased expenses in infrastructure, which are generally lumpy investments, and might have required the school to spend less on other items. There is also a negative impact of the grant on math and French test scores in schools with educated COGES (about one-third of a standard deviation, significant at the 5% level for math and 10% level for French). This negative impact of the grant on learning in schools with educated COGES, who focused spending on infrastructure, echoes the findings in the literature that providing more-of-the-same educational inputs typically has no impact on learning, whereas interventions such as remedial education and rewards are more effective at increasing learning (Kremer et al. 2013). Educated COGES may not have made the optimal choice because they decreased spending on pupil educational support, perhaps to finance the lumpy infrastructure investments. Experienced COGES Schools where the COGES has at least one member who is also a member of another community organization increased monitoring of teacher attendance in response to the grant. These schools are also those that enjoyed the increases in the cooperation index, whereas schools with no member that is also a member of another community organization had no increases.17 Wealth of the COGES 28 Parent responsibility increased more in wealthy communities.18 We note that the results for wealth are different from the results for educated and experienced COGES, showing that the effects we find for education and experience are not merely proxies for wealth. One-Teacher Schools One-teacher schools seem to have made a different choice than larger schools, with important effects: they budgeted more money for expenses related to Teacher Support.19 This may be because there was more threat of striking from the teachers: one-teacher schools in the treatment group lost 1.3 days more to teacher strikes than one-teacher schools in the control schools (significant at the 10% level). Perhaps as a result, one-teacher schools are the only schools to not suffer from the negative impact of the grants on teacher attendance on the day of the visit.20 In fact, the size of the coefficient on the interaction term suggests that teacher attendance actually increased in one- teacher schools. This suggests that by transferring some of the grant to teachers—or at least to investments that benefit teachers—the one teacher schools limited the reduced teacher attendance associated with the grant in other schools. However, infrastructure in one-teacher schools did not improve, in contrast to other schools.21 Urban and Rural Schools Increases in in-kind contributions are driven by parents in urban schools.22 The increase in the parent responsibility index is also driven by increases in urban rather than rural schools.23 Only schools located in rural areas increased their spending on agricultural investments.24 29 This may be because credit constraints may be less severe in urban areas, but no data are available to confirm this. IV. CONCLUSION The short run impact of grants to school committees in Niger was to increase cooperation and participation along several dimensions without crowding out parent financial contributions. The implication of this finding is that one way to potentially avoid the crowding out due to increased inputs found in other experiments is to involve parents in the management of the funds. Increased parent participation also came with a small increase in young pupil participation. However, more pessimistically, while the parents were willing to try to improve quality by participating, they were not able to do so, at least in the short run. One possible reason for this is that, in this context, parents (the majority of whom did not go to school) do not have sufficient information to make investments that are likely to improve quality. In particular, most investments focused on buildings, rather than extra lessons or materials, and these investments did not translate into improved learning. On average, teachers decreased their effort in response to the grant to the COGES. This finding reinforces other evidence in the literature of negative teacher reactions to participatory programs and highlights the importance of taking this potential reaction into account in policy planning. The heterogeneous impact analysis, while second-order, yields potentially helpful insights for understanding the impact of the program and considering future programs. The most difficult type of participation—monitoring teachers—was attempted only by educated or experienced school committees. This suggests that participation initiatives need to take the capacity and 30 authority of the intended participators into account. In addition, one-teacher schools that invested in the teacher’s working conditions and/or made some type of transfer to the teacher, actually increased teacher attendance. This finding suggests that teachers’ negative reaction to parent participation might be reversed when parents are “on the side” of the teachers. Finally, rural school committees as well as noneducated school committees invested a small part of the grant in agriculture, perhaps because they did not prioritize education or because they invested the money in order to get more funds for the school in the future. We highlight this finding so that future programs might be aware of it and collect more data to understand what schools might do with grants and the role that education preferences and credit constraints play in those decisions. These findings are from an evaluation that ended prematurely. As such, their generalizability is limited even as they do give us some insight into what may be the immediate barriers to a community’s ability to effectively leverage grant programs. There are four key policy implications of the findings in this paper. First, on some measures, participatory programs can be successful: parents increased their participation in school management in response to the grant without immediately reducing their contributions. Second, on the other hand, there is no reason to assume that parents will make wise spending and management decisions. Third, capacity matters for difficult tasks, as in this case the parents with education or experience were those able to supervise teacher attendance. Finally, teachers may respond to parent empowerment by reducing effort, and avoiding this may require ensuring that teachers also benefit in some way. 31 REFERENCES Aghion, P., and J. Tirole. 1997. “Formal and Real Authority in Organizations.” Journal of Political Economy 105 (1): 1–29. Banerjee, A., R. Banerji, E. Duflo, R. Glennerster, and S. Khemani. 2010. “Pitfalls of Participatory Programs: Evidence from a Randomized Evaluation in Education in India.” American Economic Journal: Economic Policy 2 (1): 1–30. Barrera-Osorio, F., T. Fasih, H. Patrinos, and L. Santibanez. 2009. “Decentralized Decision- Making in Schools, The Theory and Evidence on School-Based Management.” Washington, DC: The World Bank. Blimpo, M., D. Evans, and N. Lahire. 2015. “Parental Human Capital and Effective School Management: Evidence from the Gambia.” World Bank Policy Research Working Paper N° 7238. Bryk, A., Y. M. Thum, J. Easton, and S. Luppescu. 1998. “Academic Productivity of Chicago Public Elementary Schools: A Technical Report Sponsored by the Consortium on Chicago School Research,” Mimeo. University of Chicago. Das, J., S. Dercon, J. Habyarimana, P. Krishnan, K. Muralidharan, and V. Sundararaman. 2013. “School Inputs, Household Substitution, and Test Scores.” American Economic Journal: Applied Economics 5 (2): 29–57. Di Gropello E. 2006. “A Comparative Analysis of School-based Management in Central America.” World Bank Working Paper, 72. Duflo, E., P. Dupas, and M. Kremer. 2015. “School Governance, Teacher Incentives, and Pupil- Teacher Ratios: Experimental Evidence from Kenyan Primary Schools.” Journal of Public Economics 123: 92–110. Galiani, S., P. Gertler, and E. Schargrodsky. 2008. “School Decentralization: Helping the Good Get Better, but Leaving the Poor Behind.” Journal of Public Economics 92 (10): 2106–20. Galiani, S., E. Schargrodsky. 2002. “Evaluating the Impact of School Decentralization on Educational Quality.” Economia 2 (2): 275–314. Galiani, S., and R. Perez-Truglia. 2013. “School Management in Developing Countries,” CEDLAS, Working Papers 0147, Universidad Nacional de La Plata. Hanushek, E. A., S. Link, and L. Woessmann. 2013. “Does school autonomy make sense everywhere? Panel estimates from PISA.” Journal of Development Economics 104: 212–32. Hess, A. 1999. “Expectations, Opportunities, Capacity and Will: The Four Essential Components of Chicago School Reform.” Educational Policy 13 (4): 494–517. King, E., and B. Ozler. 2005. “What’s Decentralization Got to Do With Learning? School Autonomy and Student Performance.” Discussion Paper No. 054, Interfaces for Advanced Economic Analysis, Kyoto University. Kling, J. R., J. B. Liebman, and L. F. Katz, 2007. “Experimental Analysis of Neighborhood Effects.” Econometrica 75 (1): 83–119. 32 Kremer, M., C. Brannen, and R. Glennerster. 2013. “The Challenge of Education and Learning in the Developing World.” Science 340 (6130): 297–300. Lassibille, G., J. Tan, C. Jesse, and T. Van Nguyen. 2010. “Managing for Results in Primary Education in Madagascar: Evaluating the Impact of Selected Workflow Interventions,” World Bank Economic Review 24 (2): 303–29. Pop-Eleches, C., and M. Urquiola. 2013. “Going to a Better School: Effects and Behavioral Responses.” American Economic Review 103 (4): 1289–324. Pradhan, M., D. Suryadarma, A. Beatty, M. Wong, A. Alishjabana, A. Gaduh, and R. P. Artha. 2014. “Improving Educational Quality through Enhancing Community Participation: Results from a Randomized Field Experiment in Indonesia.” American Economic Journal: Applied Economics 6 (2): 105–26. Sabarwal, S., D. Evans, and A. Marshak. 2014. “The Permanent Input Hypothesis: The Case of Textbooks and (No) Student Learning in Sierra Leone.” World Bank Policy Research Working Paper 7021. Sawada, Y., and T. Ishii. 2012. “Do Community-Managed Schools Facilitate Social Capital Accumulation? Evidence from the COGES Project in Burkina Faso.” JICA-RI Working Paper 42. World Bank. 2003. World Development Report 2004: Making Services Work for Poor People. New York, NY: Oxford University Press. 33 Figure 1: Reported Use of Grant Money, by Total Amount Spent Source: Financial Control Questionnaire in eighty-five randomly selected treatment schools. 34 Figure 2: Conditional Differences in Spending between Treatment and Control Groups Source: 2008 School Survey. Conditional differences show the size of the coefficient on treatment from a regression including controls for whether the school is in a rural or urban area, total enrollment in 07/08, proportion of girls in 07/08, whether the school had NGO support prior to the grant, and inspection fixed effects. Light bars indicate that the difference is not significant. 35 Table 1: Balance Check of Pre-program School Characteristics (1) (2) (3) (4) (5) (6) Control Treatment p-value Difference of in means difference N mean N mean (C-T) in means Pupil characteristics Enrollment 07/08 500 149.6 500 141.72 -7.88 0.28 % Girls in 07/08 500 0.38 500 0.38 -0.01 0.26 % Passed exam in 07/08 262 0.45 224 0.42 -0.03 0.28 Teacher characteristics Number of teachers 490 3.87 494 3.55 -0.32 0.13 % of teachers civil servants 490 0.2 494 0.2 0 0.91 Physical infrastructure Number of buildings in 07/08 490 3.91 494 3.68 -0.23 0.17 Number of latrines in 07/06 500 0.89 500 0.82 -0.08 0.55 Water Access in 06/07 500 0.09 500 0.11 0.01 0.53 Electricity in 06/07 500 0.01 500 0.02 0.01 0.22 COGES characteristics COGES sponsored in 07/08 500 0.57 500 0.55 -0.01 0.70 COGES exists in 06/07 500 0.88 500 0.9 0.02 0.32 Location Tahoua 500 0.52 500 0.51 -0.01 0.85 Distance to inspection 500 41.1 500 38.59 -2.5 0.17 Distance to health center 476 8.24 461 8.95 0.7 0.61 Source: Ministry of Education Administrative Data. The data from 07/08 are reported in November (prior to the intervention) and are used when available; otherwise data from 06/07 are used. “Sponsored” COGES are those that have some sort of official sponsor or support group (such as an NGO). 36 Table 2. Community Characteristics used for Heterogeneous Treatment Effect Analysis p-value of Difference p-value of Control Treatment difference Control Treatment in means difference obs. obs. in attrition mean mean (C-T) in means Educated COGES member 369 370 0.94 0.317 0.305 0.012 0.73 Experienced COGES member 369 370 0.94 0.209 0.227 -0.018 0.55 Average wealth of COGES (PCA) 360 358 0.89 -0.586 -0.674 0.088 0.42 One-teacher school 499 497 0.32 0.122 0.145 -0.023 0.29 Urban school 500 500 0.108 0.110 -0.002 0.92 Source: Ministry of Education Administrative Data and 2008 School Survey. Observations at school level. Educated COGES member = 1 if at least one member of the COGES completed primary school. Experienced COGES member = 1 if at least one member is also the member of another community organization. Average wealth is negative because the PCA was carried out with the sample including teachers, who tend to be richer than the parents. The p-value of the difference in means is calculated by creating a dummy variable equal to 1 if the data are missing for a particular school and then calculating the p-value of the difference in this variable between groups. 37 Table 3. Grant Impact on Participation, Management, and Quality Indices (1) (2) (3) (4) (5) Parent Parent Parent Teacher contribution involvement responsibility oversight Accountability index index index index index Treatment -0.0117 0.0600* 0.0586* 0.0266 0.0127 (0.0490) (0.0321) (0.0353) (0.0389) (0.0351) Constant -0.141 -0.00756 -0.0889 0.335** -0.219* (0.167) (0.117) (0.129) (0.159) (0.124) Observations 782 922 780 778 806 R-squared 0.056 0.059 0.051 0.110 0.124 Control group mean -0.00709 -0.0355 -0.0191 0.00229 0.00325 (6) (7) (8) (9) (10) Cooperation Infrastructure Materials Teacher Index Index Index Health Index Effort Index Treatment 0.0661** 0.0414* -0.0439 0.0469* -0.0237 (0.0306) (0.0236) (0.0350) (0.0270) (0.0435) Constant -0.220** -0.454*** -0.402** -0.396*** 0.484*** (0.103) (0.0936) (0.171) (0.114) (0.158) Observations 777 978 826 933 784 R-squared 0.078 0.164 0.174 0.238 0.213 Control group mean -0.00756 -2.98e-09 -0.00411 1.26e-08 -0.00712 Source: Ministry of Education Administrative Data and 2008 School Survey. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the school is in a rural or urban area, total enrollment in 07/08, proportion of girls in 07/08, whether the school had NGO support prior to the grant, and inspection fixed effects. Details on the component variables and the impact of treatment on each component variable for each index are given in the appendix S3. 38 Table 4. Impact on Spending Decisions Dependent Variable: Amount of money Spent on…. (1) (2) (3) (4) (5) (6) (7) (8) (9) Infrastructure Pupil School and Supplies and educational Pupil Teacher COGES festivals and Investments equipment textbooks support health support expenses playground in agriculture Total amount Treatment 21,586** 3,222 1,435 1,253 -1,086 32.14 864.8*** 1,833*** 28,512*** (9,121) (1,981) (1,369) (2,154) (1,331) (300.6) (285.5) (658.5) (9,993) Constant -24,197 836.7 -763.1 -13,404* 1,489 524.5 -1,599** -861.4 -34,994 (38,103) (8,622) (4,031) (8,062) (4,576) (1,046) (765.0) (1,098) (41,928) Observations 726 733 734 734 734 738 736 731 698 R-squared 0.127 0.156 0.087 0.051 0.019 0.039 0.039 0.047 0.157 Control group mean 86,119 11,631 6,058 8,711 4,352 782.7 165.8 582.9 115,898 Source: 2008 School Survey. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the school is in a rural or urban area, total enrollment in 07/08, proportion of girls in 07/08, whether the school had NGO support prior to the grant, and inspection fixed effects. Dependent variable is the amount in FCFA spent by COGES in the corresponding category of activities, as declared by the president of COGES in the April/May 2008 survey. Infrastructure and Equipment includes expenses related to classrooms, desks, chairs, blackboards, school enclosure and security, and cleaning. Supplies and textbooks include expenses for notebooks, pens, and textbooks. Pupil Educational Support includes expenses like additional courses, awareness campaigns to increase enrollment, and academic rewards. Pupil health includes expenses related to nutrition and health like drinkable water, meals, latrines and drugs. Teacher support includes expenses benefitting to teachers like teacher housing, furniture, supplies, guide books, and salary. COGES Expenses includes expenses related to COGES meetings, contributions to “COGES communal” and inspector visits. Schools festivals and Playground includes expenses like graduation ceremonies, parties, and soccer balls. Investments in Agriculture includes fields, crops and livestock, unrelated to education activities. 39 Table 5. Impact on Teacher Effort (1) (2) (3) (4) COGES Days on Teacher is opinion of Teacher strike present teacher effort effort index Treatment -0.541 -0.0382* -0.0220 -0.0237 (0.490) (0.0227) (0.0253) (0.0435) Constant -2.071 0.937*** 3.656*** 0.484*** (2.292) (0.0738) (0.0932) (0.158) Observations 706 799 734 784 R-squared 0.127 0.248 0.134 0.213 Control group mean 4.592 0.760 3.617 -0.00712 Source: 2008 School Survey. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the school is in a rural or urban area, total enrollment in 07/08, proportion of girls in 07/08, whether the school had NGO support prior to the grant, and inspection fixed effects. Days on strike is the number of days that the school was closed due to teachers striking in 2007/2008. Teacher is present is the school average of the dummy variable indicating 1 if a teacher is physically present at the day of visit (on a day when the school was supposed to be open). If the school was closed, all teachers were counted as absent. The Teacher effort index is the average of the z-scores of the variables in columns (1) to (3), oriented so that more beneficial outcomes have higher values. 40 Table 6. Impact on Dropout and Enrollment A: Dependent Variable: Dropout as reported at school visit in Spring 2008 (1) (2) (3) (4) (5) (6) (7) (8) (9) Total Total girls Total boys Grade 1 Grade 2 Grade 3 Grade 4 Grade 5 Grade 6 Treatment -0.00559 -0.206 -0.00469 -0.0136* -0.00646 -0.00791 -0.00778 0.00264 0.00139 (0.00520) (0.212) (0.00609) (0.00758) (0.0107) (0.00582) (0.0100) (0.00849) (0.00987) Constant 0.0723*** 0.775 0.0908*** 0.0366** 0.0613** 0.0678*** 0.143** 0.115** 0.0891** (0.0165) (0.662) (0.0224) (0.0183) (0.0291) (0.0240) (0.0570) (0.0455) (0.0384) Observations 748 754 753 531 434 525 454 381 466 R-squared 0.059 0.036 0.055 0.038 0.042 0.046 0.090 0.068 0.104 Control group mean 0.0359 0.366 0.0379 0.0296 0.0328 0.0295 0.0364 0.0313 0.0508 B: Dependent Variable: Enrollment as reported in 2008/09 Administrative Data (1) (2) (3) (4) (5) (6) (7) (8) (9) Total Total girls Total boys Grade 1 Grade 2 Grade 3 Grade 4 Grade 5 Grade 6 Treatment 1.366 0.505 0.862 -0.604 3.256** -0.471 -0.541 0.366 -0.639 (2.445) (1.254) (1.654) (1.502) (1.376) (1.174) (1.190) (1.019) (0.962) Constant 37.56** -21.01*** 58.57*** 34.47*** -1.052 5.214 1.546 -1.388 -1.225 (15.14) (7.562) (9.652) (6.267) (6.441) (4.881) (4.534) (3.911) (3.925) Observations 988 988 988 988 988 988 988 988 988 R-squared 0.901 0.880 0.866 0.470 0.545 0.546 0.484 0.520 0.540 Control group mean 160.3 65.70 94.63 40.09 29.95 23.87 26.22 20.98 19.22 Source: Ministry of Education Administrative Data and 2008 School Survey. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the school is in a rural or urban area, total enrollment in 07/08, proportion of girls in 07/08, whether the school had NGO support prior to the grant and inspection fixed effects. Table 7A gives the impact of the treatment on dropout rates in the spring of 2008. Schools without a particular grade level are missing. Some schools did not provide breakdowns by sex. Table 7B gives the impact of treatment on enrollment in the fall of 2008 (the academic year following the treatment). Schools that have zero pupils at a given grade level (because they are missing a particular level) are counted as zeros. 41 Table 7. Impact on Test Scores (1) (2) (3) (4) End Primary Oral Math French Pass Rate Treatment -0.101 -0.0351 -0.0338 -0.0244 0.0749 0.0588 0.0586 0.0227 Constant -0.0252 -0.159 0.0648 0.525*** 0.261 0.209 0.221 0.0706 Observations 499 763 739 557 R-squared 0.200 0.200 0.251 0.177 Control Group Mean 0.00828 0.00545 0.0145 0.614 Source: 2008 School Survey. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the school is in a rural or urban area, total enrollment in 07/08, proportion of girls in 07/08, whether the school had NGO support prior to the grant and inspection fixed effects. Oral, Math and French test scores come from normalized test scores from the World Bank administered exam in the spring of 2008. Oral test scores were given only to pupils in grades 1 and 2. The End Primary Pass Rate is the percent of students from the school who passed the exam at the end of grade 6 at the end of 2008 (administrative data). 1. School-based management programs have been implemented in Argentina, Australia, Bangladesh, Canada, Guatemala, Honduras, Hong-Kong, India, Lebanon, Lesotho, Macedonia, Madagascar, Mexico, Nicaragua, the Philippines, Senegal, Serbia, Sri Lanka, the Gambia, the United Kingdom, and the United States (Duflo et al. 2015). 2. World Development Indicators, World Bank, source: International Institute for Applied Systems Analysis (IIASA) 3. Policies of de jure autonomy do not always lead to de facto autonomy (King and Ozler 2004), and so participation may not be meaningful if communities have no actual power and even increase inequality by “leaving the poor behind” (Galiani et al. 2008). 4. The situation has continued to improve in terms of access to education: in 2011, net enrollment in primary school was 62%, and primary completion rates had risen to 46%. 5. These school committees consist of six representatives, including the school director, who serves as secretary, and parent representatives. The parents are supposed to elect the representatives, who may also be the leaders of the Parent Association (APE), which includes all parents, and the Mother’s Association (AME), which includes all mothers. In practice, the composition of the COGES varies by school. School committees are supposed to be responsible for the management of people working at the school (e.g., monitoring of teacher attendance and performance), financial resources (e.g., school meal funds) and material resources (e.g., purchase and management of textbooks, supplies and supplies). One of the school committee’s central tasks is to draft an annual school improvement plan that includes its projects, activities, budget, and timelines to guide its work for the school year. The school committee works parallel to the APE and AME. Additional details and background are given in appendix S1). 6. The regional authorities were unable to obtain the actual funds from the local treasury due to a liquidity issue at the local treasury level. 7. One randomly selected group of schools received a slightly more restrictive list of potential expenditures, and another group received a warning that their projects might be audited. Analysis of spending patterns did not show any difference between these groups. 42 8. Schools choose which of their sixth grade students would sit for the exam. There is no evidence that schools were punished in any way for a low pass rate. 9. The analysis plan was not registered in a secure independent register in 2007 when the experiment was designed, as is best practice today. 10. An alternative specification uses dummy variables for the strata used in random selection, which were defined using a dummy variable for urban, the total enrollment in 2005/06, and support by an outside NGO in 2005/06. This specification does not substantially change the results, but increases precision of some coefficient estimates and decreases precision of others. 11. An alternative interpretation would be that this result derives from the fact that we measure only the first year of the grant, and so parents did not have time to change their own contribution of inputs (see Das et al. 2013, where crowding out was greater when a school grant was anticipated than when it was unanticipated). We think this is unlikely since the parents were notified in advance of the grants arrival. 12. We would urge that future researchers examining local school management and activities collect data on school festivals, as well as school business investments, as potential targets of school spending. These expenditures were not foreseen and so detailed questions on these expenditures (for example, the number and type of school festivals, or the anticipated return of investment projects) were not included in the questionnaire, nor were questions about the local credit market. 13. These items were also projects that were frequently reported by the schools as projects undertaken using the grant money. 14. We also take the fact that only younger grades were impacted as evidence that the change in enrollment is not due to intentional misreporting by grant schools. In addition, the finding is replicated across two different types of data collections and at two different periods. 15. Note that while educated COGES budgeted more money for infrastructure (58,755 FCFA (117 USD), significant at the 5% level), the increases in infrastructure in the following year were felt primarily in schools with noneducated COGES: the coefficient on the interaction term of treatment and education is negative (-0.08 SD) and significant at the 5% level. One possible reason, if the data on spending are accurate, is that the projects undertaken by educated COGES in response to the grant might have been larger and taken more time, so that they were not yet completed at the time that data on infrastructure was collected. 16. For Health Resources, the treatment coefficient for the noneducated COGES is 0.06 SD, significant at the 10% level, while the coefficient for the interaction term is -0.12, significant at the 10% level, suggesting zero or negative impact of the grants on health resources in the educated COGES. For Pupil Educational Support, schools with noneducated COGES increased spending (3,639 FCFA (7 USD), significant at the 5% level), but no impact (or a possibly negative impact) for schools with educated COGES (the coefficient on the interaction term is -8,215 FCFA (16 USD), significant at the 5% level). 17. For monitoring teacher attendance, the coefficient on the interaction term is 0.11, significant at the 5% level, and for cooperation, the coefficient is 0.07, significant at the 10% level. 18. Each standard deviation increase in wealth is associated with an additional 0.05 standard deviation increase in the parent responsibility index in response to the grant, significant at the 5% level 19. The coefficient on the interaction term is 8,985 FCFA (18 USD), significant at the 5% level. 20. The coefficient on the interaction term is 0.17, significant at the 5% level, and the coefficient on the treatment term is -0.06, significant at the 5% level. 21. Infrastructure may have even degraded—the coefficient on the interaction term is -0.17, significant at the 1% level, while coefficient on the treatment variable is 0.06, significant at the 5% level. Note that since the grant was based on the size of the school, one-teacher schools received smaller grants. They may then have been pushed away from investment in infrastructure since the lump sum was not enough to start a project. 22. Urban schools were 17% more likely to have made in kind contributions, significant at the 10% level. 23. The coefficient on the interaction term is almost 0.3 standard deviations, significant at the 1% level, whereas the coefficient on treatment alone in the interaction specification is near zero. 24. Rural areas increased spending on agricultural investments by 2,046 FCFA (4 USD), significant at the 1% level, and the interaction term for urban schools is -1,755 FCFA (3.5 USD), significant at the 5% level. 43 Willing but Unable: Short-Term Experimental Evidence on Parent Empowerment and School Quality Elizabeth Beasley and Elise Huillery∗ November 19, 2015 S1 Supplementary Information on the COGES As part of a 10-year education sector plan, the Programme Décennal de Développement de l’Education (PDDE) 2003-2013, the government of Niger established school-based management committees, the COGES. These committees consist of representatives from the parents’ associations, the school di- rectors, representatives of the teachers and other community leaders. The COGES was intended to engage parents in their children’s education, improve school management and accountability, generate community support for the school, and, through these, improve access to and quality of education. However, at the beginning of this project, the COGES were not initially very active, and many had not fulfilled their primary task (planning the school’s activities for the year). S1.1 Members of the COGES The COGES typically include six members, of which four are the school principal, a teacher repre- sentative and parents’ representatives from the AME (mother’s associate) and APE (parent asso- ciation). Other members would be made up of community representatives (generally leaders in the community) and schools also report having student representatives. The structure of the COGES includes a president, treasurer, and secretary (the school principal is generally the secretary). These rules are not always followed to the letter: in our sample, schools report an average 6.5 COGES members. Two-thirds of COGES members are male, and overall they have low levels of ∗ Elizabeth Beasley: CEPREMAP (elizabeth.beasley@cepremap.org), Elise Huillery: Sciences Po, Department of Economics and J-PAL (elise.huillery@sciencespo.fr) 1 education: only about 21% of COGES members report completing primary school (in urban areas this is higher, and for the smaller schools it is much lower). S1.2 Responsibilities of the COGES One of the principal purposes of the COGES is to get parents involved in the educational process. Concretely, the COGES are supposed to manage the people working at the school (more specifically, the teachers), which can include monitoring attendance and evaluating performance, the material conditions of the school (such as repairs of buildings, purchasing and managing supplies), and the school budget (for example, there might be a school meal fund to be managed). Some schools have community-hired teachers, and the COGES could in principle exercise more control over these teachers. A primary task for each COGES is to write a school action plan each year, that details the expected activities and projects of the COGES, as well as a budget, timeline, and allocation of responsibilities. It is thought that through these concrete tasks, the COGES will accomplish not only improved management but also parent’s investment and implication in the school and their children’s education. S1.3 COGES Training All of the COGES in the Tahoua and Zinder regions were supposed to have received training in 2006. The training included how the COGES should be set up and run (that is, how to run an election, hold meetings, the responsibilities of the COGES and so on), and how to write a school action plan. In 2007, the COGES were also supposed to be trained on financial management and accounting. Not all COGES members were trained, but at least two per school were supposed to have been trained. In addition, turnover would result in some members being untrained. About 73% of schools report that they have received training. S2 Supplementary Information on Internal Validity S2.1 Compliance Two representatives from the COGES signed a document confirming effective receipt of the grant in the intended amount. These receipts were first collected at the regional level and the information was then entered into a database at the Ministry of Education as a way to verify the actual receipt of 2 the grants at the school level. An additional survey was conducted in 85 randomly selected schools asking detailed questions about the receipt and spending of the grants, and financial management. This questionnaire also included information about any problems with the administration of the grant and qualitative feedback and suggestions from the COGES. The use of the grants was recorded in detail, including the existence of a receipt for each expenditure. Grants were distributed as follows: the Ministry of Education issued an order to the District level, which allowed the district to withdraw cash from the Treasury to distribute to the schools. The grants were distributed first to the inspectors, and then either directly to the COGES or to other officials who brought the grants to the schools. The vast majority of schools do not have bank accounts, and other mechanisms of distribution were infeasible. The collection of grant receipts, financial questionnaires, and information from the Ministry indicated that of the 498 of the 500 treatment schools received the grant. Of the two that did not receive their grant, one school had closed, and so their grant was allocated to a school outside of the 1,000 school sample, and the other’s grant was mistakenly given to a control school. Of the schools receiving the grant, our information indicates that four schools received less money than had been allocated to them (in 3 cases the schools received 500 FCFA (1 USD) less than the assigned grant amount of 73,500 FCFA (147 USD), and in one case 10,500 FCFA (21 USD) less than the assigned grant amount of 120,500 FCFA (241 USD)), while two schools reported receiving more than had been allocated (one school received 2,000 FCFA (4 USD) more than the assigned amount of 122,500 FCFA (245 USD), and the other received 27,000 FCFA (54 USD) more than the assigned amount of 167,500 FCFA (335 USD)). All in all, the data indicate that 492 out of 500 schools received the exact amount allocated to them, and six others received the grant but not in the correct amount. This is a reasonably high compliance rate.1 Data from the qualitative questionnaire administered to the 85 randomly selected schools indi- cate that the majority of those schools received the intended grant amount. Among the 85 schools, one school that had been selected for the grant had been closed at the time that the grant arrived. In another case, the grant was accidentally given to another school. In a third case, a school re- ported receiving 500 FCFA (1 USD) less than the intended amount. Two schools reported paying 1 Note that this program was publicized within the administration and careful records were required at each step of transfer of the money. In addition, the government of Niger had recently prosecuted several corrupt officials, so there was a heightened awareness of corruption. It is possible that this transfer mechanisms to other contexts might not be so effective. 3 some money to cover transport costs to the person who delivered the grant. S2.2 Attrition There is some attrition in the datasets. Each year, a handful of schools do not return the adminis- trative data questionnaire or the questionnaires are improperly filled out, leading to missing data for 3% of the schools for the infrastructure index and 1.4% of the schools for 2008/09 enrollment. The April/May 2008 survey was conducted on the basis of unannounced visits, which meant that many schools were closed. In addition, some schools were not visited due to security concerns, and still others closed early that year because the summer rainy season began early and so many children went to the fields with their parents to plant. As a result, data from the evaluation questionnaire is available for only 814 schools (81.4%). We test for differences by treatment group in the proportion of schools with missing outcome variables as a whole and sub-divided by district, urban and rural, and whether the school had external support (for example, NGO sponsorship) prior to the project. Results are reported in the Appendix, Table A1. Eighty-four tests on treatment and interaction between treatment and sub- groups yield one statistically significant differences (at the 10% level or higher), which is well within the amount that would be expected with random attrition. The comparability between treatment and control groups is thus intact. As to external validity, there are more schools missing in the region where security was a concern (Tahoua, in the north). 4 Table A1. Attrition, by Treatment Group and Pre-Program School Characteristics (1) (2) (3) (4) (5) (6) Parent Parent COGES COGES COGES Teacher effort contrib responsib education wealth experience index index index Treatment -0.101 -0.128 -0.101 -0.104 -0.0225 -0.0984 (0.133) (0.135) (0.133) (0.126) (0.126) (0.127) Treat*Enroll 0.0395 0.0524 0.0395 0.0591 0.0625 0.0522 (0.0519) (0.0527) (0.0519) (0.0490) (0.0481) (0.0523) Treat*% girls 0.0334 0.0356 0.0334 0.0305 0.0107 0.0257 (0.0320) (0.0324) (0.0320) (0.0308) (0.0308) (0.0306) Treat*Urban 0.00436 -0.109 0.00436 -0.0933 -0.107 -0.0604 (0.0956) (0.105) (0.0956) (0.0812) (0.0816) (0.0815) Treat*Sponsor -0.0121 -0.0180 -0.0121 -0.00630 -0.0227 -0.0266 (0.0258) (0.0261) (0.0258) (0.0245) (0.0244) (0.0249) Treat*Num Teach -0.0457 -0.0322 -0.0457 -0.0348 -0.0290 -0.0142 (0.0551) (0.0573) (0.0551) (0.0523) (0.0525) (0.0542) Treat*Tahoua -0.00777 0.00526 -0.00777 -0.0186 -0.0342 -0.0136 (0.0284) (0.0286) (0.0284) (0.0274) (0.0272) (0.0280) Observations 1,000 1,000 1,000 1,000 1,000 1,000 R-squared 0.163 0.183 0.163 0.140 0.168 0.130 Control Group Mean 0,262 0,28 0,262 0,214 0,222 0,224 (7) (8) (9) (10) (11) (12) Teacher Parent involv oversight Cooperation Accountabilit index index Index y index Enrollment Test scores Treatment 0.00109 -0.0978 -0.0810 -0.0360 -0.00933 -0.147 (0.0772) (0.127) (0.127) (0.123) (0.0312) (0.124) Treat*Enroll 0.00542 0.0411 0.0429 0.0633 0.00882 0.00727 (0.0323) (0.0500) (0.0500) (0.0447) (0.0103) (0.0461) Treat*% girls 0.000701 0.0272 0.0229 0.00859 0.00543 0.0318 (0.0193) (0.0311) (0.0311) (0.0304) (0.00689) (0.0302) Treat*Urban 0.00163 -0.0714 -0.0695 -0.0674 0.0171 0.0675 (0.0590) (0.0819) (0.0820) (0.0718) (0.0274) (0.0991) Treat*Sponsor 0.00475 -0.00133 -0.00339 -0.0152 -0.00213 0.0354 (0.0164) (0.0248) (0.0248) (0.0234) (0.00697) (0.0272) Treat*Num Teach 0.00146 -0.0284 -0.0284 -0.0297 -0.0170 -0.0120 (0.0404) (0.0528) (0.0530) (0.0452) (0.0137) (0.0478) Treat*Tahoua -0.0317* -0.00607 -0.00977 -0.0277 -0.00577 -0.00675 (0.0189) (0.0276) (0.0277) (0.0263) (0.00714) (0.0283) Observations 1,000 1,000 1,000 1,000 1,000 1,000 R-squared 0.062 0.132 0.134 0.149 0.009 0.133 Control Group Mean 0,084 0,22 0,222 0,198 0,014 0,278 Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1. Dependent variable in each column is a dummy variable equal to one when the value is missing for a given school. All regressions include the level variables from the interaction terms. 5 S3 Supplementary Information on Data S3.1 Outcome Indices Parent Contributions Parent contributions reflect the material resources that parents provide to the school: the amount of fees, in-kind contributions and additional financial contributions (fundraising).2 Fundraising is measured through the question “How much money has the school raised since the beginning of the school year?”. The answer to this question was divided by the number of pupils registered in 2007/2008. The variable we use is thus funds raised per pupil3 . Finally, the amount of fees charged per pupil is declared by the school director. Parent Involvement Parent involvement measures the volume of parent participation in school management. We use 11 variables to measure parent involvement, of which nine use information from the April/May 2008 questionnaire: the number of school, communal, and parent committee meetings, the time elapsed since the last school or parent committee meeting, the number of topics which were discussed at the last school or parent committee meeting, and two measures of presence at the last school or parent committee meeting. From the 2008/2009 administrative data, we also use the number of school committee meetings and a dummy variable indicating whether the mother’s association was active or not. Parent Responsibility The parent responsibility index measures the extent to which parents take some responsibility and exercise authority in making decisions. We use eight variables to mea- sure parent responsibility, all dummy variables for whether the school committee is in charge of monitoring pupil attendance, sanctioning pupils for poor attendance, collecting financial contribu- tions, spending financial contributions, purchasing supplies, investing in infrastructure, and setting up the action plan. Teacher Oversight One possible consequence of more empowered parents would be that parents engage in teacher oversight. We use six separate variables to measure teacher oversight: dummy variables indicating whether the COGES is in charge of monitoring teacher attendance, whether the 2 Since we do not have information on the local market price of the items contributed, we use a simple dummy variable indicating whether the community provided some in-kind contributions to the school or not. 3 In the context of Niger, the funds are essentially provided by parents since there are very few NGOs. But it is important to note that our measure of financial contributions encompasses parental contributions and any other potential donors. 6 COGES has taken remedial action against a teacher for repeated absence (remedial actions include talking to the teacher, giving a warning, or complaining to the teacher’s supervisor). We also use dummy variables for whether the school or parent committee discusses teacher behavior in school or parent committee meetings. Finally, we use dummy variables for whether the school committee spontaneously mentions being active to reduce teacher absenteeism or improving education quality4 . Accountability The accountability index reflects the capacity to keep track of facts, decisions and expenses, as well keeping receipts, although having receipts for everything is generally difficult for the communities. While most schools claim to keep registers for fundraising and expenses, fewer are usually able to produce registers to be seen by the interviewers. The same is true for the school action plan, seen by many as the key activity of the school committees. The government encour- ages schools to keep records on a number of subject matters: inspector visits, pupil attendance, teacher attendance, weekly activities, supplies, fundraising, and expenses. School committees and parent associations are also supposed to take minutes at each meeting. The accountability index is composed of 10 dummy variables for the presentation to the interviewer of a written school action plan, registers for pupil attendance, inspector visits, weekly activities, supplies, fund collection, fund expenditure, and teacher attendance, and minutes for the last school or parent committee meeting. The data for this index comes from the April/May 2008 survey. Cooperation We use information about cooperation among school stakeholders, in particular between the school committee and the other actors, to assess whether the grant affected the quality of stakeholder relationships within schools. Our measures of cooperation are all self-reported by school stakeholders (the school director or the school committee president), which implies that we measure perceived cooperation which might differ from actual cooperation. The cooperation index is composed of 11 components. First, three variables indicate teachers’ attitudes as reported by the school director and the school committee president: teachers’ cooperation with COGES, with each other, and with the community - these variables are coded from 1 (not cooperative at all) to 4 (very cooperative). Second, seven dummy variables indicate whether the school committee president reports good support from the community, local authorities, the school administration, 4 The school committee president was asked to list the domains in which the committee is active, without suggesting any particular domain in order to avoid prompting responses. We thus elicit activities that come naturally to the top of school committee president’s mind. 7 teachers, parent committee, and two different levels of the Ministry of Education hierarchy, the conseil pedagogique (education advisors) and inspectors. Finally, we also use a dummy variable for whether the school committee president reports that disagreements among school committee members are rare or inexistent (as opposed to occasional or frequent disagreements). Expenses and Investments The April/May 2008 questionnaire asked the COGES to list projects included on the school action plan, as well as the amount budgeted for them. These data are ana- lyzed as total amount spent rather than indices. We coded these items into eight categories: • Infrastructure includes expenses related to classrooms, desks, chairs, blackboards, school en- closure and security, and cleaning. • Supplies and Textbooks includes expenses for notebooks, pens, and textbooks. • Pupil Educational Support includes expenses like remedial courses, awareness campaigns to increase enrollment, and academic rewards. • Pupil Health includes expenses related to nutrition and health like drinkable water, meals, latrines and drugs. • Teacher Support includes expenses benefitting teachers such as teacher housing, furniture, supplies, guide books, and salary. • COGES Expenses includes expenses related to COGES meetings, contributions to and travel expenses for the "COGES communal" (a regional grouping of the COGES) and inspector visits. • Schools Festivals and Playground includes expenses such as graduation ceremonies, parties, and soccer balls. • Investments in Agriculture includes seeds, fields, crops and livestock, unrelated to education activities. For example, one school reported the purchase of a field of peanut plants, and another purchased a herd of goats. 8 Infrastructure We create an index of infrastructure quality using data on the number of buildings and their condition, the number of blackboards, the number of desks for children, the number of teacher’s desks, the number of teacher’s chairs, the number of shelves, and whether or not the school has an enclosure (this is a fence or wall around the school grounds that separates the school from other public space). For the infrastructure index, the classroom, desks, blackboard and books figures are changes from year to year. Materials Materials include textbooks, dictionaries, geography materials (such as maps and at- lases) and math materials (such as rulers, protractors, and compasses). Since there is only one variable for textbooks but multiple variables for geography and math materials, if all variables are included in the index in the same way, we would weight the importance of each type of math mate- rial the same as the importance of textbooks, which seems to give too much weight to each kind of math or geography material. To avoid this, we first construct an index of math materials and then of geography materials and include those indices with the same weight as textbooks to construct the overall material index. We use the change in the amount of material between 2007/2008 and 2008/2009. Health Health resources include vitamin or micronutrient supplementation, school deworming, health information sessions (for example, on preventing malaria), availability of a first aid kit, trash disposal, number of latrines and access to water. Teacher Effort We use three variables to assess teacher effort. First, the unannounced school visit in 2008 recorded how many teachers were present on the day of the visit, and the number of teachers are employed at the school. We use a simple percentage of the number of teachers who were physically present at the school. To accurately represent the loss of classroom time and avoid reporting inconsistencies, this figure includes both excused and unexcused absences. If the school was closed at the unannounced visit (which was on a day the schools were supposed to be open) all teachers were counted as absent. Second, we use the number of days when class was cancelled because teachers were on strike (this information was provided by the school director). Finally, we use a variable reflecting the opinion of the school committee on teacher assiduity and punctuality, coded from 1 (not satisfactory at all) to 4 (very satisfactory). 9 S3.2 Interaction terms Education We define a community as “educated” if one or both of the two interviewed members of the school committee completed primary school (excluding the director).5 Only 31% of school committees from control schools contain at least one member who completed primary school, which indicates an important heterogeneity across communities. These data come from the Spring 2008 6 Questionnaire. Experience in other organizations Parents who have experience in other organizations may be better able to engage in management of the school because they might have more experience doing so. We use data from the April/May 2008 Questionnaire to construct a variable equal to 1 if either of the COGES members interviewed belonged to another community organization that was not engaged in the school (for example, religious communities, cooperatives, savings groups, and so on). About one-fifth (21%) of the control school COGES fall into this category. The correlation between education and experience is 0.15. Wealth We assume that the wealthier a community is, the more real authority parents will have because they will have a higher social status relative to the teachers (note that teachers are relatively homogenous in terms of wealth). These data come from the Spring 2008 Questionnaire. The wealth of school committee members is the first component of a principal component analysis of durable goods possessed by the two interviewed school committee members and the school director. Durable goods include means of transportation, animals and housing equipment. The wealth of school committee is then the average of this wealth index for the two interviewed school committee members. Note that the average wealth index does not have any material meaning in itself since the scale is one that measures individual’s wealth relative to one another. The average wealth index is 5 When information for one of the two interviewed members is missing, we impute the value of the member for whom information is available, in order to avoid dropping observations. We thus assume that the observed member is representative of the two sampled members. Results do not vary substantially when these schools are excluded but the sample size and power is reduced. 6 Wealth and education are measured in the April/May 2008 questionnaire, after the treatment was implemented. We note that it would have been better to measure these characteristics before the treatment was implemented, though this was not possible in the context of program implementation. However, these characteristics would be changed by the treatment only if the grant induced a change in the composition of the school committee, with former members replaced by new ones with different characteristics. In the data, we observe a proportion of 20% of school committee members who took their position in 2008 both in the intervention and in the control group (so no more renewal due to the grant - result not shown), and school committee members exhibit the same characteristics in both groups (Table 2). We are thus confident that the grant did not affect the composition of the school committee and that observed community characteristics are not endogenous to the grant. 10 negative since the two school committee members are poorer, on average, than the school directors, whose data was included in the construction of the wealth index. Urban or Rural Urban schools are more connected to the central government, and the students are likely to live somewhat closer to the school. Rural schools are schools where there is no village or settlement around. Urban schools are a marked minority: just over 10% of schools are located in urban areas. We do not hypothesize about the relationship of authority to urban or rural location, but we maintain this variable as an interaction term to help understand spending patterns and because we believe it is of more general interest to policymakers and planners. One-Teacher Schools Schools with only one teacher present a unique situation. One way that teachers have power in negotiations with communities is that they can leave if they are unhappy with conditions. When there is only one teacher, this threat may be even stronger (because there is no possibility of absorbing his or her students into another class). Seen from a different point of view, it may be easier for the parents to negotiate and work with a single teacher rather than a group of teachers, especially because if there is only one teacher that teacher is de facto a member of the COGES. In the fall of 2007, prior to the arrival of the grant, 12% of schools were one-teacher schools. 11 S4 Supplementary Tables S4.1 Impact estimates on all components of outcome indices Table A2. Parent Contributions (1) (2) (3) (4) Amount of User Parent Money Parent Fees Charged by Any Inkind contributed Contribution school contributions (per pupil) Index Treatment 35.48 0.0223 -72.86 -0.0117 235.3 0.0252 54.23 0.0490 Constant 1,856** 0.634*** 486.5** -0.141 772.3 0.102 194.7 0.167 Observations 745 758 719 782 R-squared 0.038 0.077 0.066 0.056 Control Group Mean 631.6 0.836 500.2 -0.00709 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the school is in a rural or urban area, total enrollment in 07/08, proportion of girls in 07/08, whether the school had NGO support prior to the grant and inspection fixed effects. Any in kind contributions includes food, building supplies, wood, and so on. The Parent Contribution Index is the average of the z-scores in columns (1) to (3). 12 Table A3. Parent Involvement (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Time Time Number Number Number elapsed elapsed of of Topics Number Number of Presence of since last Number since last COGES AME is in of Topics COGES Presence at Parent COGES COGES of APE APE meetings Active, COGES in APE Communal at APE COGES Involv meetings meeting meetings meeting 08/09 08/09 meeting meeting meetings meeting meeting Index Treatment 0.151 -0.135 0.101 -0.182 0.256 0.0492 0.0614 0.0530 0.0709 0.0800 0.00652 0.0600* 0.182 0.124 0.147 0.200 0.176 0.0299 0.104 0.104 0.116 0.0830 0.0881 0.0321 Constant 4.421*** 3.955*** 2.890*** 4.776*** 4.786*** 0.113 2.838*** 2.035*** 2.519*** 1.607*** 1.643*** -0.00756 0.612 0.526 0.479 0.782 0.868 0.119 0.455 0.454 0.419 0.359 0.375 0.117 Observations 747 549 727 465 803 888 739 649 714 743 746 922 R-squared 0.043 0.137 0.029 0.051 0.056 0.066 0.085 0.074 0.071 0.074 0.066 0.059 13 Control Group Mean 3.758 2.653 2.470 3.691 4.601 0.272 2.363 1.918 2.518 2.057 1.757 -0.0355 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the school is in a rural or urban area, total enrollment in 07/08, proportion of girls in 07/08, whether the school had NGO support prior to the grant and inspection fixed effects. APE is the Parent Association of the school. The Parent Action Index is the average of the z-scores of the variables in columns (1) to (11). Table A4. Parent Responsibility (1) (2) (3) (4) (5) (6) (7) (8) (9) COGES has COGES taken Parents are responsible remedial COGES is COGES is COGES is responsible for monitoring action on COGES is responsible Number of responsible responsible for drafting Parent pupil pupil responsible for COGES for fee for fee the Action Responsib. attendance attendance for supplies infrastructure tasks collection expenses Plan Index Treatment -0.0100 -0.00289 -0.0106 0.0451 0.0914 0.0767 0.0615 0.0308 0.0586* 0.0222 0.0375 0.0353 0.0307 0.120 0.0476 0.0481 0.0204 0.0353 Constant 0.807*** 0.643*** 0.593*** 0.571*** 3.835*** 0.213 0.131 0.873*** -0.0889 0.0866 0.141 0.125 0.109 0.491 0.169 0.179 0.0868 0.129 Observations 754 581 752 749 735 401 401 732 780 R-squared 0.063 0.062 0.063 0.064 0.070 0.065 0.074 0.029 0.051 Control Group Mean 0.769 0.713 0.603 0.739 3.441 0.301 0.335 0.900 -0.0191 14 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the school is in a rural or urban area, total enrollment in 07/08, proportion of girls in 07/08, whether the school had NGO support prior to the grant and inspection fixed effects. Parent Responsibility Index is the average of the z-scores of the variables in columns (1) to (8). Table A5. Teacher Oversight (1) (2) (3) (4) (5) (6) (7) COGES COGES COGES takes COGES works COGES works discusses APE discusses monitors remedial to reduce to improve Teacher teacher teacher teacher action on teacher education Oversight behavior behavior attendance teacher absence quality Index Treatment 0.0371 -0.00646 -0.0126 -0.0120 0.0319 0.0166 0.0266 0.0255 0.0276 0.0219 0.0337 0.0293 0.0303 0.0389 Constant 0.343*** 0.259* 0.844*** 0.378*** 0.372*** 0.814*** 0.335** 0.124 0.133 0.0820 0.128 0.117 0.111 0.159 Observations 649 573 758 758 720 731 778 R-squared 0.070 0.045 0.064 0.049 0.099 0.088 0.110 Control Group Mean 0.103 0.121 0.766 0.329 0.271 0.621 0.00229 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the school is in a rural or 15 urban area, total enrollment in 07/08, proportion of girls in 07/08, whether the school had NGO support prior to the grant and inspection fixed effects. APE is the Parent Association of the school. The Teacher Oversight Index is the average of the z-scores of the variables in columns (1) to (6). Table A6. Accountability Index (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Book for Weekly Supply PV last PV last Teacher Pupil Funds Expenses Visitor Planning Register COGES APE Register Register Register Register Action Plan Account. Exists Seen Seen meeting meeting Seen Seen Seen Seen Seen Index Treatment -0.0123 -0.0185 0.0413 -0.0278 -0.0493 -0.0151 -0.0221 0.0619* 0.101*** -0.00642 0.0127 0.0356 0.0279 0.0341 0.0344 0.0354 0.0214 0.0222 0.0353 0.0355 0.0292 0.0351 Constant 0.380*** -0.0764 0.369*** 0.351*** 0.473*** 0.126 0.111 0.506*** 0.447*** 0.490*** -0.219* 0.136 0.103 0.122 0.126 0.131 0.0869 0.0791 0.128 0.130 0.121 0.124 Observations 784 784 777 743 735 705 692 651 648 937 806 R-squared 0.043 0.095 0.124 0.130 0.113 0.032 0.021 0.218 0.213 0.223 0.124 Control Group Mean 0.519 0.224 0.509 0.594 0.520 0.0986 0.103 0.494 0.476 0.484 0.00325 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the school is in a rural or urban area, total enrollment in 07/08, proportion of girls in 07/08, whether the school had NGO support prior to the grant and inspection fixed effects. PV is proces verbal (minutes) taken in the meetings. The APE is the Parent Association of the School. The Accountability Index is the average of the z-scores of the variables in columns (1) to (10). 16 Table A7. Cooperation Index (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) COGES COGES has COGES COGES has has support has COGES COGES has Rare or no Teachers' Teachers' Teachers' support support from the support has has support fights coop with coop with coop with from the from local school from support support from within Coop COGES each other comm. comm. auth. admin. teachers from APE from CP Inspect. COGES Index Treatment 0.0234 -0,00037 0.00870 0.0482* 0.0312 0.0394 0.0315* 0.0509** 0.00846 0.0250 -0.0210 0.0661** 0.0341 0.0329 0.0331 0.0271 0.0288 0.0240 0.0178 0.0203 0.0284 0.0297 0.0350 0.0306 Constant 3.517*** 3.596*** 3.400*** 0.586*** 0.656*** 0.739*** 0.842*** 0.786*** 0.652*** 0.610*** 0.473*** -0.220** 0.130 0.121 0.127 0.113 0.114 0.0884 0.0665 0.0742 0.0965 0.0985 0.136 0.103 Observations 737 735 736 757 755 755 757 754 755 754 752 777 R-squared 0.048 0.109 0.112 0.084 0.070 0.077 0.084 0.057 0.113 0.071 0.065 0.078 Control Group Mean 3.329 3.485 3.329 0.797 0.776 0.854 0.918 0.889 0.792 0.770 0.610 -0.00756 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the school is in a rural or urban area, total enrollment in 07/08, proportion of girls in 07/08, whether the school had NGO support prior to the grant and inspection fixed effects. APE is the Parent Association of the school. Columns (8) and (9) refer to support from representatives of the Ministry of Education: CP is Conseiller Pedagogique (Education Advisor) and is from the 17 hierarchical level just above the school, the Inspection refers to the level above the CP. The Cooperation Index is the average of the z-scores of the variables in columns (1) to (11). Table A8. Infrastructure Index (1) (2) (3) (4) (5) (6) (7) (8) (10) Change in Change in Change in Change in the number the Wall around Number of Number of the number the condition of number of school teacher's teacher's Number of Infrastructure of buildings of buildings blackboards desks grounds desks chairs bookcases Index Treatment 0.0766* -0.0155 0.138 0.187 0.0866*** 0.0123 0.00454 -0.0106 0.0414* 0.0452 0.0207 0.117 0.678 0.0316 0.0590 0.0332 0.0395 0.0236 Constant 0.187 0.480*** -0.352 -0.0640 -0.135 -0.0798 -0.104 -0.161 -0.454*** 0.183 0.0796 0.550 2.934 0.114 0.239 0.145 0.170 0.0936 Observations 947 988 905 597 847 896 891 894 978 R-squared 0.046 0.057 0.038 0.054 0.132 0.032 0.026 0.029 0.164 Control Group Mean 0.274 0.670 0.654 -0.724 0.343 0.0316 0.0618 0.0461 -2.98e-09 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the school is in a rural or urban area, total enrollment in 07/08, proportion of girls in 07/08, whether the school had NGO support prior to the grant and inspection fixed 18 effects. Columns (1) to (4) use change in the variable between 2007/2008 and 2008/2009. Columns (5) to (8) use the variable from 2008/2009 because of missing data in the earlier year. The Infrastructure Index is the average of the z-scores of the variables in columns (1) to (8). Table A9. Materials Index (1) (2) (3) (4) (5) Change in number of Geography Math Tool Materials books Dictionary Tool Index Index Index Treatment 1.267 -0.0162 -0.0422 -0.0705 -0.0439 3.266 0.0281 0.0486 0.0470 0.0350 Constant -0.438 -0.176 -0.365 -0.413** -0.402** 13.52 0.121 0.249 0.192 0.171 Observations 749 858 735 841 826 R-squared 0.086 0.108 0.037 0.256 0.174 Control Group Mean -10.75 0.0472 -0.0165 -3.26e-05 -0.00411 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the school is in a rural or urban area, total enrollment in 07/08, proportion of girls in 07/08, 19 whether the school had NGO support prior to the grant and inspection fixed effects. Data are administrative data from the 2008/2009 school questionnaire. Change in the number of books is in comparison to the previous years administrative data. Dictionary is whether or not the school has a dictionary. Geography tool index is composed of whether the school has an atlas, map, or globe. Math tool index is whether the school has a protractor, ruler, or compass. Materials index is the average of the z-scores of the variables in columns (1)-(4). Table A10. Health Index (1) (2) (3) (4) (5) (6) (7) (8) Vitamin / Health micronut Information Trash Health supplement Deworming Session First Aid Kit Collection Latrines Water Index Treatment 0.0252 0.0220 0.0406 0.0283 -0.00198 -0.116 0.0133 0.0469* 0.0272 0.0285 0.0293 0.0196 0.0209 0.134 0.0199 0.0270 Constant 0.131 0.474*** 0.378*** -0.0331 0.00380 0.298 -0.0846 -0.396*** 0.115 0.118 0.113 0.0909 0.0809 0.665 0.0987 0.114 Observations 933 933 933 933 933 933 898 933 R-squared 0.061 0.219 0.096 0.062 0.048 0.343 0.220 0.238 Control Group Mean 0.223 0.615 0.305 0.0909 0.119 1.600 0.129 1.26e-08 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the school is in a rural or urban area, total enrollment in 07/08, proportion of girls in 07/08, whether the school had NGO support prior to the grant and 20 inspection fixed effects. All data are administrative data from the 2008/09 school questionnaire. For columns (1), (2), and (3) the dependent variable is an indicator for whether the action happened at least once. First aid kit indicates whether the school had a first aid kit or not. Trash collection indicates whether a system is in place to dispose of trash. Latrines is the number of latrines. Water is a dummy variable for whether there is clean water available at the school or not (piped or well). The health index is the average of the z-scores of these variables. S4.2 Selected Heterogeneous Impact Results Table A11. Educated COGES Education (1) (2) (3) (4) (5) Spending on COGES Spending on Pupil monitors teacher Infrastructure Educational Normalized Normalized attendance and Equipment Support Math Scores French Scores Treatment -0.0313 4,783 3,639** 0.132 0.0758 0.0263 9,689 1,538 0.0834 0.0803 Treatment x Educated COGES 0.0857* 58,775*** -8,215** -0.297** -0.279* 0.0508 22,737 3,452 0.146 0.146 Constant 0.804*** -23,084 -615.4 -0.227 0.0571 0.0847 38,431 4,052 0.235 0.242 Observations 720 693 700 601 581 R-squared 0.073 0.150 0.097 0.183 0.247 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the COGES is Educated or not, whether the school is in a rural or urban area, total enrollment in 07/08, proportion of girls in 07/08, whether the school had NGO support prior to the grant and inspection fixed effects. Infrastructure and Equipement includes expenses related to classrooms, desks, chairs, blackboards, school enclosure and security, and cleaning. Pupil Educational Support includes expenses like additional courses, awareness campaigns to increase enrollment, and academic rewards. Math and French test scores come from the World Bank administered exam in the spring of 2008. (1) (2) (3) (4) COGES Parent monitors teacher Cooperation Responsib. Investments in tt d attendance Index I d I Index d A i lt Agriculture Treatment -0.0308 0.0134 0.0913** 2,046*** 0.0251 0.0389 0.0384 740.4 Treatment x Experienced COGES 0.112** 0.141* 0.0534 0.0805 Treatment x Wealth 0.0530** 0.0264 Treatment x Urban -1,775** 805.2 Constant 0.804*** -0.102 -0.0387 -733.5 0.0847 0.119 0.130 1,106 Observations 720 739 707 731 R-squared 0.077 0.087 0.063 0.048 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the COGES is experienced or not and the wealth of the COGES when those are the interaction terms , whether the school is in a rural or urban area, total enrollment in 07/08, 21 proportion of girls in 07/08, whether the school had NGO support prior to the grant and inspection fixed effects. The Cooperation Index is the average of the z-scores of the variables on cooperation with the COGES. Parent Responsibility Index is the average of the z-scores of the variables on the responsibilities of the COGES. Investments in Agriculture includes fields, crops and livestock, Table A12. Experienced, Wealthy, and Urban COGES (1) (2) (3) (4) COGES Parent monitors teacher Cooperation Responsib. Investments in attendance Index Index Agriculture Treatment -0.0308 0.0134 0.0913** 2,046*** 0.0251 0.0389 0.0384 740.4 Treatment x Experienced COGES 0.112** 0.141* 0.0534 0.0805 Treatment x Wealth 0.0530** 0.0264 Treatment x Urban -1,775** 805.2 Constant 0.804*** -0.102 -0.0387 -733.5 0.0847 0.119 0.130 1,106 Observations 720 739 707 731 R-squared 0.077 0.087 0.063 0.048 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the COGES is experienced or not and the wealth of the COGES when those are the interaction terms , whether the school is in a rural or urban area, total enrollment in 07/08, proportion of girls in 07/08, whether the school had NGO support prior to the grant and inspection fixed effects. The Cooperation Index is the average of the z-scores of the variables on cooperation with the COGES. Parent Responsibility Index is the average of the z-scores of the variables on the responsibilities of the COGES. Investments in Agriculture includes fields, crops and livestock, unrelated to education activities. 22 Table A13. One-teacher Schools (1) (2) (3) (4) Spending on Teacher is Days teachers Teacher Present at Spot Infrastructure on strike Support Check Index Treatment -0.673 -2,018 -0.0563** 0.0634** 0.547 1,441 0.0238 0.0257 Treatment x One Teacher 1.324* 8,985** 0.168** -0.169*** 0.802 3,752 0.0742 0.0631 Constant -1.150 1,771 0.914*** -0.473*** 2.267 4,655 0.0745 0.0936 Observations 705 733 797 978 R-squared 0.138 0.025 0.261 0.169 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, *p<0.10. Regressions control for whether the COGES is Educated or not, whether the school is in a rural or urban area, total enrollment in 07/08, proportion of girls in 07/08, whether the school had NGO support prior to the grant and inspection fixed effects. Days on strike is the number of days that the school was closed due to teachers striking in 2007/2008. Teacher is present is the school average of the dummy variable indicating 1 if a teacher is physically present at the day of visit (on a day when the school was supposed to be open). If the school was closed, all teachers were counted as absent. Spending on teacher support includes expenses benefitting to teachers like teacher housing, furniture, supplies, guide books, and salary. The Infrastructure Index is the average of the z-scores of the variables concerning school infrastructure. 23 S5. An Explanatory Model Our results demonstrate, first, that complete crowding out is not inevitable: parents responded to grants to the COGES by increasing participation and did not immediately change their financial contribution (with some evidence that some schools increased in kind contributions). This contrasts with the results in Das et al. (2013) and Pop-Eleches and Urquiola (2013) where additional resources to the school (not under parent control) decreased parental effort. We also find that only schools where parents are educated or are members in other community organizations respond to the grant by increasing teacher monitoring - the type of parent participation that is arguably the most difficult to do. Note, however, that there was no positive impact on teacher presence due to this increased monitoring. Pupil participation, at the lowest grades, increased. However, the ultimate impact on education quality, at least in the short run, is unclear: teachers were absent more frequently (which echoes the result in Duflo, Dupas and Kremer (2012) in Kenya where giving parents the responsibility over an extra-teacher led to a reduction in effort from civil-service teachers), except in the smaller schools where the grant was at least partly spent in a way that benefited them directly. Taking into account that Niger is, in general, an environment where parents have little authority, our results contribute to the literature from many other countries, and supplement the existing results with new data. There are several experiments showing that the effectiveness of participatory programs to improve school quality is related to the level of authority or empowerment of the parents: as King and Ozler (2004) demonstrate, policies of de jure autonomy do not always lead to de facto autonomy, and so participation may not be meaningful if communities have no actual power. Empowerment might be through education (Blimpo and Evans, 2011), ties with local government (Pradhan et al, 2014), training (Duflo, Dupas and Kremer, 2012), human capital (Gunnarsson et al, 2009), school-community relations and community organizational capacity (Gershberg and Shatkin, 2007), and pre-existing levels of poverty (Galiani et al, 2008). In this appendix section, we consider a model that formalizes the behavioral responses of parents and teachers to a change in school resources and the resulting effect on school quality. The moti- vation for this model is two-fold. First, the model helps to clarify how the grant program studied in this paper can affect parent participation, teacher effort and school quality. Second, we want to show that reasonable assumptions on school dynamics are able to produce predictions consistent 24 with the evidence found in the literature on both participatory programs and school inputs. Albornoz et al (2011) model the interaction between student, parent and teacher investments and school resources, to explain the ambiguous effect of resources on parent involvement at home. This model suggests that under some circumstances, an increase in school resources generate a decline in parent investment in education at home. Das and al (2013) also provide a model to explain the decrease in parental effort at home in response to an increase in school resources. But none of these theoretical frameworks take into account parental participation in school. The model proposed in this section enriches our understanding of school dynamics by taking into account parents’ effort both at home and at school, and the difference between giving more resources to school staff versus parents. It enlarges the set of interventions of interest and adds to our understanding of the effects of educational policies. It is an ex post exercise designed to make sense of existing evidence with the hope it can be tested in subsequent analyses. S5.1 Set-Up The model involves three participants: parents, teachers and the government. Teachers decide how much time they put in teaching tt . Parents decide how much time they invest for education at home th , as well as how much time they participate in school management tp . Finally, the government chooses the level of governmental resources for the school, which decompose in two parts, Gt + Gp , where Gt is resources in the hands of school staff (principals and teachers), while Gp is governmental resources for the school under the control of parents (typically, resources handled by the school committee). Here, “participation” in school management refers to the many different kinds of participation that policy makers envision, where beneficiaries might be organized into committees, undertake projects themselves, such as construction or sanitation, raise funds, provide personal contributions, supervise, hire, and even fire teachers, engage in awareness campaigns, provide advice to staff, and so on. Participation is expressed in time units (financial participation is converted in time through hourly wage). Children’s Learning Children’s learning E is the addition of learning produced at home and learning produced at school. 25 Learning produced at home Learning produced at home is assumed proportional to the num- ber of hours parents devote to education at home, th (making sure kids get up on time and go to school or investing in private lessons, for instance). How much each hour spent on education translates into learning depends on parent’s productivity at producing learning, denoted e, reflect- ing for instance parents’ level of education (more educated parents produce more learning for each hour spent on helping with homework) or parents’ hourly wage (a higher wage can pay for a higher amount of private lessons for each working hour invested in education). Learning produced at home is thus eth . Learning produced at school Learning produced at school is proportional to the time teachers spend at school, tt . How much each hour spent at school translates into learning depends on teach- ers’ productivity, which results from school resources. Indeed, school resources encompass salaries (which should reflect both class size and teachers’ quality) and school materials (infrastructure, textbooks, flip charts, blackboard, etc.) that allow teachers for producing more learning for the same amount of time spent with the children. So we assume that the level of resources is a factor of teachers’ productivity. Furthermore, we assume that parents’ participation in school management interfere with school resources in the determination of teachers’ productivity. Indeed, parents’ participation is additional resources: parents raise funds from the community, and do administrative tasks that allow teachers for focusing on teaching and producing more learning for the same amount of time spent at school. Moreover, parents’ participation should increase the allocative efficiency of school resources by preventing rent capture and making expenses closer to educational needs and common interest (Barrera-Osorio et al, 2009). We thus assume that parents’ time spent on school management, ts , is a factor that increase the effect of resources on teachers’ productivity7 . This factor apply to resources under parental control, Gp , but not on Gt in which parents do not have a say. We thus assume that learning produced at school is given by (Gt + tp Gp )tt and total learning is given by E = eth + (Gt + tp Gp )tt . 7 A richer model could take into account the idea that parents’ participation might not translate into greater teachers’ productivity because “pushy” parents might be disruptive to teachers. For the simplicity of the model, we make the assumption that parents are not aware of this fact and would not participate if they would know that their participation would decrease teachers’ productivity. This assumption is equivalent to the idea that parents do not get utility from participating per se (for example, due to reputation effect or some sort of hedonic payoff). 26 Parents’ Utility Parents’ utility is the difference between the benefit they derive from children’ learning E , and the opportunity cost of the time they spend on producing learning, th + tp . The benefit from children’ learning is assumed concave in E (for instance ln(1 + E )), so that learning produced at home and learning produced at school are substitutes.8 The cost of time is assumed linear (for instance th + tp ), so parents’ utility is given by: Up = ln(1 + eth + (Gt + tp Gp )tt ) − th − tp (1) Teachers’ Utility Similarly, teachers’ utility is the difference between the benefit they derive from children’ learning and the opportunity cost of their time9 . We assume that teachers’ benefit and cost take the same form as parents’ ones, except that their welfare is also influenced by parents’ participation in school management: teachers derive a benefit from parents’ satisfaction towards their production of learning when parents can observe this production. This benefit takes the form of a social reward10 that is proportional to the time teachers devote to school, with a factor of parents’ participation: the more parents participate, the more they observe and reward each unit of time teachers devote to education. However, teachers who have a preference for a centralized government might resent being mon- itored by parents because of the resulting loss of autonomy and leadership in school decisions. In this case, the effect of parents’ participation on teachers’ welfare can be negative, teachers’ loss of welfare being also proportional to the time they devote to education, with a factor of parents’ participation (the more devoted they are and the more parents participate, the more teachers are resentful). We denote δ teachers’ taste for community participation in school management. A negative δ reflects a preference for a centralized government, whereas a positive δ reflects openness to collaborate with parents (a δ close to zero would reflect teachers’ indifference). 8 To the extent that parent participation is a contribution to a public good, free-riding may be a problem. For simplicity we do not include this in the model, but a more complete model might address this issue. 9 In this model, teachers are intrinsically motivated. A richer model could incorporate a broader view which would incorporate both intrinsic and extrinsic motives. We do not incorporate extrinsic motives here since it would not add to the ability of the model to explain what we seek to explain. 10 An equivalent way to put it is that teachers incur a social sanction from the community if they shirk and if parents can observe it. 27 As a result, teachers’ utility is given by: Ut = ln(1 + eth + (Gt + tp Gp )tt ) − tt + δtp tt (2) S5.2 Parents’ and Teachers’ Choices The first-order condition for the teachers’ problem is sufficient (Ut is infinitively differentiable and Ut (tt ) < 0) and gives the optimal choice of teachers: 1 eth + 1 tt = max − ,0 (3) 1 − δtp Gt + tp Gp For the parents, the first-order conditions are also sufficient and give the optimal choices: 1 + (Gt + tp Gp )tt th = max 1 − ,0 (4) e Gt eth + 1 tp = max 1 − − ,0 (5) Gp tt Gp From the expression of tp , we see that parents invest more time in school management when resources under their control increase and when teachers make more effort. In contrast, parents invest less time in school management when resources in the hands of teachers increase, when they spend more time for education at home and when their efficiency with education at home increase. Symmetrically, parents devote more effort for education at home when their efficiency at home increases, whereas they reduce effort at home when school resources increase, or when teachers’ or their own effort at school increase. Finally, teachers increase time they spend at school when school resources increase, whereas they reduce it when parents’ effort or efficiency at home increase. However, the response of teachers to an increase in parents’ participation in school management is ambiguous: if δ is positive, the response is clearly positive too: teachers spend more time at school. But in the region where δ is negative, for large absolute value of δ , the response is negative, meaning that teachers who have a strong preference for a centralized government reduce time at school when parents’ participation in school management increases. 28 S5.3 Heterogenous Best-Responses For the best clarity and simplicity, the model above just includes the main dynamics in the school system. In this paper, we also explore the possibility that power imbalances are likely to induce different choices. This section explicit how parental real authority influences parents’ and teachers’ decision. In our model, real authority of parents over the school would be captured by a parameter θ multiplying parents’ time spent in school management: learning produced at school is given by (Gt + θtp Gp )tt , reflecting the fact that more powerful parents make better use of resources under their control, therefore extracting more learning from teachers for each hour invested in school than weak parents. Also, it should be noticed that real authority of parents θ is unlikely to be orthogonal to teachers’ preference for a centralized government δ . On the one hand, teachers are more likely to resent being monitored by parents when teachers enjoy a high social status relative to parents, for instance when parents have a low if not no education, which is likely to coincide with parents’ lack of real authority. On the other hand, teachers’ preference for a centralized government largely determines the extent to which parents entitled to participate in school (have formal authority) are involved in decision making (have real authority). We thus posit that δ = δ (θ) with δ > 0. The 1+(Gt +θtp Gp )tt best-responses with a parameter θ reflecting real authority are: th = max 1 − e ,0 , Gt eth +1 1 eth +1 tp = max 1 − θGp − θtt Gp , 0 and tt = max 1−δ (θ )tp − Gt +θtp Gp , 0 . Effect of real authority on parents’ and teachers’ decisions Since their participation at school is more productive, parents with higher θ invest more time in school management and less time at home than parents with low θ. Teachers also invest more time at school because their productivity is fostered by parents’ real authority. Moreover, the likelihood of δ being negative is lower when parents’ real authority is larger, which adds to the general positive effect of parents’ real authority on teachers’ effort. Effect of real authority on parents’ and teachers’ responses Parents’ response to an in- crease in teachers’ effort or in resources under their control is amplified by real authority, just as teachers’ response to an increase in resources under parental control. Moreover, teachers’ response to an increase in parent’s participation in school management is reduced in the negative region and 29 amplified in the positive region by real authority. These predictions are consistent with the evidence presented in section 4.1 that the benefits of community-based interventions are larger when parents are more powerful. S5.4 The Effect of an Increase in School Resources In the light of this model, what is the effect of an increase in governmental resources to schools? In the short run, parents won’t take into account the fact that teachers will also react to the changing conditions (and reciprocally). We thus consider that parents take the teachers’ actions as given (fixed at their past value) and vice-versa, and we determine the comparative statics and discuss the predicted behavioral trajectories. Our focus on short-term responses that do not take account of others’ responses comes, first, because most empirical framework in the literature addresses such responses, and, second, because real-life behavioral adjustments to others’ responses seem slow. Pop-Eleches and Urquiola (2013) show that responses after one year are different from responses in the longer run, reflecting the fact that it takes quite a long time for parents to adjust their behavior to others’ responses. There are multiple reasons for slow adjustment to others’ response. For instance, parents may not realize that teachers incur a loss of welfare from collaborating with them (formally, they have a imperfect perception of teachers’ δ ) because teachers do not disclose their reluctance to collaborate with parents in front of them. Even if teachers do give signals that they do not want parents to participate, it is also possible that parents do not take what they observe in the short run into account because they expect continuous collaboration to make δ become positive in the future. Effect in the absence of parents’ participation in school In the short run, an increase in school resources increase teachers’ time at school and decrease parents’ time for education at home. The fact that parents devote less time for education at home tends to reinforce teachers’ response, which comfort parents with investing less time at home, etc. The long-term effect of an increase in school resources is thus clear-cut: teachers respond positively while parents respond negatively. The final impact on school quality is a mixed bag: the increase in school resources and teachers’ response tend to improve education outcomes, while parents’ response tend to reduce this effect. Pop-Eleches and Urquiola (2013) and Das et al. (2013) confirm that 30 an increase in school resources reduced parents’ effort (they do not observe teachers’ response). In our framework, this policy is appropriate in contexts where (i) teachers actually use resources for educational purpose, and (ii) the effect of additional resources on teachers’ productivity is large. The conditions of success of this policy are thus a sound institutional environment preventing rent capture and an initial level of school resources at which marginal gains of productivity are steep11 . Effect in contexts where parents participate in school Teachers’ and parents’ responses to an increase in school resources are the same as above, but now parents’ re-optimize their level of participation in school management too. Increase in Gt If the additional resources fall in the hands of teachers, parents decrease their participation at school. This in turn affects teachers’ effort in a way which depends on teachers’ preference for a centralized government: if teachers prefer a centralized government, the decrease in parents’ participation in school management amplifies teachers’ positive response to the increase in school resources, so teachers make unambiguously more effort. In contrast, if teachers are motivated by the collaboration with parents, the decrease in parents’ participation reduces their incentive to work hard and the policy brings a smaller benefit. The conditions of success of this policy are thus (i), (ii), plus the condition that (iii) teachers prefer a centralized government. France is an example of countries where this policy is likely to work well. Increase in Gp If the additional resources fall in the hands of parents, parents increase their participation at school, which leads to the opposite situation in which teachers invest unambiguously more effort when teachers are motivated by the collaboration with parents through three positive effects: the effects of additional resources on their productivity, the effect of parents’ participation in the management of the resources on their productivity, and the incentive produced by the social reward. When teachers prefer a centralized government, parents’ participation creates a burden for teachers which reduces teachers’ effort in a way that might be strong enough to offset teachers’ positive response to school resources and to parents’ management of the resources. In the long-run, this should eventually discourage parents to participate at school and encourage investment for 11 This analysis would benefit from evidence on the shape of teachers’ productivity as a function of school resources to know which kind of regions would experience the larger gains in teachers’ productivity. If this function is concave (resp. convex, S-shaped), gains in teachers’ productivity are larger at the bottom (resp. top, middle) part of school resource distribution. 31 education at home back up, but in the short-run concurrent increase in parents’ participation in school management and decrease in teachers’ effort can be observed, as Duflo, Dupas and Kremer (2012) find in Kenya. Our empirical results are also consistent with the situation where teachers’ preference for a centralized government is strong and parents’ real authority is weak, resulting in a negative short-term impact of parents’ participation in school on teachers’ effort. The impact on school quality can be at risk since the positive effects of school resources and parents’ management of the resources are mitigated by a double decrease in parents’ effort at home and teachers’ effort at school. When parents have a large real authority θ, the positive effect of parents’ management of the resources is larger so parents’ response is larger too, which is consistent with our empirical findings that parents contribute more and participate more in school management when they have more authority. The larger effect on parents’ participation combined with the smaller likelihood of a preference for a centralized government leads to a more favorable teachers’ response. Our data do not confirm this prediction, but Duflo, Dupas and Kremer (2012) do since they observe that parents’ empowerment through school committee training reduced the negative response of civil- servant teachers. According to this framework, the conditions of success of this policy are thus (i), (ii), plus the conditions that (iii) teachers are keen to collaborate with parents, and (iv) parents have real authority on teachers. These conditions are more likely to hold in countries where the social gap between parents and teachers is small and where the education system is decentralized. The USA is an example of countries where this policy is likely to work well, whereas Niger and Kenya are not the ideal contexts for encouraging parental control over school management since (iii) and (iv) do not hold. However, one might argue that the short-term negative impact on teachers’ effort is the price to pay for potential longer-term positive effect -which our results cannot exclude. The general picture supported by existing empirical evidence and explained by our model is three-fold: first, an increase in school resources out of parental control tends to decrease parental effort. Second, an increase in school resources under parental control tends to increase parental effort. Finally, the size of the increase in parental effort and of the resulting effect on teachers’ effort depends on power imbalances in school: the higher parents’ real authority, the larger their response and the resulting increase in teacher’s effort, with a risk of adverse effects in contexts 32 where parents are weak. This paper is a first step that uses both formal tests and intuition to build a narrative about community participation in resource management. Our hope is that future work might build on this model to provide additional insights and rigorous empirical tests. 33