Democratic Institutions and Social Capital: Experimental ...Jimenez, Harounan Kazianga, Mushfiq Mobarak, Keijiro Otsuka, Albert Park, Menno Pradhan, Nancy Qian, Imran Rasul, Halsey

Democratic Institutions and Social Capital:Experimental Evidence on School-Based

Management from Burkina Faso

Yasuyuki Sawada∗ Takeshi Aida Andrew S. GriffenEiji Kozuka Haruko Noguchi Yasuyuki Todo

September 18, 2019

Abstract

We estimate the effects of school-based management (COGES) on social cap-ital formation in Burkina Faso using a large-scale RCT of COGES combined withlab-in-the-field experiments to measure social capital. We find that the implemen-tation of COGES significantly increased social capital in the form of public goodsgame contributions. Several novel aspects of the social experiment, field experi-ments, and data collection provide insight into the mechanisms behind the impact.

∗Sawada: Chief Economist, Asian Development Bank, 6 ADB Avenue, Mandaluyong City, 1550Metro Manila, Philippines, and University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-0033, Japan(e-mail: [email protected]); Aida: Institute of Developing Economies, Japan External Trade Or-ganization 3-2- 2, Wakaba, Mihama-ku, Chiba-shi, Chiba 261-8545 (e-mail: [email protected]);Griffen: University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-0033, Japan (e-mail: [email protected]); Kozuka: Japan International Cooperation Agency, Nibancho Center Building 5-25, Niban-cho, Chiyoda-ku, Tokyo 102-8012, Japan, (e-mail: [email protected]); Noguchi: Waseda Univer-sity, 1-6-1 Nishi-Waseda, Shinjuku-ku, Tokyo 169-0051 (e-mail: [email protected]); Todo: WasedaUniversity, 1-6-1 Nishi-Waseda, Shinjuku-ku, Tokyo 169-0051 (e-mail: [email protected]). This paper,previously titled “Election, Implementation, and Social Capital in School-Based Management: Evidencefrom a Randomized Field Experiment on the COGES Project in Burkina Faso,” has been prepared asa part of a research project of the Japan International Cooperation Agency Research Institute (JICA-RI) entitled “Impact Evaluation Analyses for the JICA Projects,” led by Yasuyuki Sawada. We thankour collaborators at the Institut National de la Statistique et de la Dmographie (INSD), JICA BurkinaFaso office, and the JICA Research Institute for their valuable cooperation in implementing our surveyand experiments. We also thank Oriana Bandiera, Abhijit Banerjee, Moussa Blimpo, Nazmul Chaud-hury, Gordon Dahl, Esther Duflo, Deon Filmer, Jun Goto, Hidehiko Ichimura, Hideshi Itoh, EmmanuelJimenez, Harounan Kazianga, Mushfiq Mobarak, Keijiro Otsuka, Albert Park, Menno Pradhan, NancyQian, Imran Rasul, Halsey Rogers, Chika Yamauchi, and the participants of the COGES-SABER seminarheld in Ouagadougou on February 9, 2015 for useful comments. The usual disclaimers apply.

1 Introduction

A large literature spanning several areas of social science has sought to understand the

roles of social capital and institutions in affecting outcomes in society. Social capital has

been linked with democracy, growth, health, happiness, education, and the provision of

public goods (Putnam et al., 2001; Putnam, 2000; Knack and Keefer, 1997; Szreter and

Woolcock, 2004) and good institutions are believed to be a fundamental driver of long-

run economic performance (North, 1991; Acemoglu et al., 2001, 2019; Dell, 2010; Dell

et al., 2018). These findings have naturally led to calls for policies to strengthen both so-

cial capital (Putnam, 1993) and institutions (World Bank, 2004). However, the design of

external interventions to improve social capital is difficult (Ostrom, 2000) and evidence

on the death of institutions suggests that simply copying institutional successes with-

out accompanying strong norms is not guaranteed to work (Levitsky and Ziblatt, 2018).

There are also important theoretical and empirical links between social capital and in-

stitutions (Alesina and Giuliano, 2015): high levels of social capital are hypothesized

to be able to correct institutional failures (Hayami, 2009; Mansuri and Rao, 2013) and

there are concerns that formal, externally imposed institutions may crowd-out informal,

evolved solutions to collective-action problems (Ostrom, 1990, 2000).

A recent wave in development policy designed to improve institutions has focused

on decentralization, which is motivated by the theory that more local control will deliver

more responsive institutions (Mansuri and Rao, 2013). Such decentralization policies

fall under the rubric of “community driven development” for more general projects

and “school-based management” (SBM) for reforms that target schools (Mansuri and

Rao, 2004). However, the details and results of such interventions vary tremendously

across projects and contexts and the evidence on their effectiveness is mixed (Casey,

2018; Kremer and Holla, 2009). In contrast to the substitutes view of institutions and

social capital, some empirical studies have actually found complementarities between

social capital and the introduction of formal institutions (Putnam et al., 2001; Dell et

al., 2018; Martinez-Bravo et al., 2017; Pradhan et al., 2014) suggesting that institutions

both increase social capital and higher social capital increases the effectiveness of in-

2

stitutions. These studies point to the importance of understanding the conditions under

and mechanisms through which external interventions designed to affect social capital

and institutions actually work, which is critical for policy design.

To investigate this interplay between SBM institutions and the dynamics of social

capital in rural communities, we conducted a randomized control trial (RCT) of a SBM

program in Burkina Faso called COGES1. The village level treatment consisted of a se-

cret community-wide democratic election of COGES members followed by implemen-

tation of an action plan for proposed changes within the local school by the COGES

using feedback from the community. A unique aspect of the intervention is the ran-

domization of a complete institutional structure including democratic mechanisms for

electing SBM members. Previous experimental papers in the SBM literature have eval-

uated policy variation within existing SBM institutions (Pradhan et al., 2014; Barr et

al., 2012; Beasley and Huillery, 2017; Blimpo et al., 2015). As an outcome, we col-

lected data from lab-in-the-field public goods game contributions, which we interpret as

a measure of social capital (Anderson et al., 2004).

We exploit the timing of the randomization and data collection to separately iden-

tify effects of the election of COGES committee members and the implementation of

projects by COGES. According to our pre-analysis plan, we also intentionally con-

ducted the public goods game experiment with different configurations of community

members to investigate how COGES affected bonding, bridging, and linking forms of

social capital (Woolcock, 2001). These different aspects of the experiment and data

collection give a rich characterization of the channels through which the introduction of

COGES influenced social capital in the villages.

The experimental results show that implementation of COGES had a large and sta-

tistically significant impact on social capital. In the schools treated with COGES, aver-

age contributions in the public goods game increased between 8.1% and 9.7%. We did

not find any variation in impact by group configuration, suggesting the impacts did not

operate through a specific type of social capital but more generally. We complement

1This acronym is derived from the French name of the project: projet d’appui COmits de Gestion dansdes EcoleS primaires.

3

our main experimental findings by examining impacts on attitudinal measures of trust,

fairness, and help borrowed from the General Social Survey (GSS), and by examining

the impact of COGES on real-world decisions observed in the schools. We find impacts

along both dimensions.

Our paper is related to several literatures spanning development economics, insti-

tutional economics, and experimental economics. In development economics, there is

an emerging literature on the effects of decentralization policies in both schools and

communities (Casey, 2018). For SBM, the literature typically examines test scores and

school enrollment given the aim of SBM to improve schools. However, some papers

have considered more intermediate outcomes such as the composition of the SBM com-

mittees or involvement of parents within schools (Kremer and Holla, 2009). Although

SBM is pushed as an important development strategy to improve schools, the estimated

impacts of SBM are mixed; some studies have found positive impacts (Barr et al., 2012;

Barrera-Osorio et al., 2009; Beasley and Huillery, 2017; Blimpo et al., 2015; Bruns et

al., 2011; Gertler et al., 2012; Duflo et al., 2015; Pradhan et al., 2014; Kozuka et al.,

2016), while others report negligible effects (Banerjee et al., 2010; De Laat et al., 2008;

Kremer and Holla, 2009). Reviewing this literature, Kremer and Holla (2009) suggest

effects are likely modulated by important contextual factors such as control of funding,

firing and hiring, and nepotism. In that sense, our paper investigates the impact of a

particular SBM design (democratic elections, community input for projects, and no ex-

ternal transfer of resources) on social capital as one such important contextual factor.

In addition, there are only a few rigorous evaluations of SBM in developing countries

(Westhorp et al., 2014) so our paper makes an important contribution to understanding

of the design of these programs.

By examining social capital, our paper is connected to the literature on institutions

and society, which is quite broad (Alesina and Giuliano, 2015). However, despite evi-

dence about the importance of institutions for long-run development (Acemoglu et al.,

2001; Dell, 2010; Dell et al., 2018), there are very few randomized evaluations of insti-

tutions and especially of elections, which is a function of the difficulty of implementing

4

such types of RCTs. The closest paper is Pradhan et al. (2014) who use an RCT to

examine the impact of within SBM policy changes (including introducing elections) in

Indonesia.2 They find large, positive impacts on test scores through election of com-

mittee members and through reforms that provide “linkage” of SBM institutions with

powerful village committees. They infer the impact operated through social capital type

mechanisms based on treatments that impacted intermediate outcomes measuring social

capital and that connected the schools more directly to the community.

The contribution of our work is to examine the impact of such institutions in a dif-

ferent setting (Burkina Faso), using more direct measures of social capital (experimen-

tal and attitudinal), and by examining the role of institution vs. no-institution impacts

instead of policy impacts within an existing institutional framework. Both types of

impacts are important for understanding policy design but answer different policy ques-

tions. Environments without existing institutions may have less capacity to implement

such institutions. Or, conversely, setting up institutions from scratch may provide an

opportunity to “get it right” from the beginning. In addition, the positive impacts found

in our study in a completely different context provide evidence on the external valid-

ity of the policy (Deaton, 2010). Relative to the quasi-experimental evidence on his-

torical reforms that examine long-run impacts, these RCT papers provide evidence on

more modern reforms that may be implemented by current policymakers (e.g., related to

schools) and also allow a richer characterization of the mechanisms through the collec-

tion of contemporaneous measures. The downside is the typically short-run measured

impacts of such reforms, which may take many years to operate.

Our work also fits within studies that have examined interactions between institu-

tions and traditional societies in Africa (Ensminger, 1996) and an experimental eco-

nomics literature on cross-culture determinants of lab-in-the-field outcomes (Henrich et

al., 2001), which is important given the bias in the literature towards nonrepresentative

WEIRD samples (Henrich et al., 2010). Our work is also connected to the effects of

2Another important paper related to our work is Martinez-Bravo et al. (2017), which examines theimpact of elections using quasi-experimental introduction of elections in China and finds complementarybetween elections and existing social capital in the form of temples. Although our paper finds directimpacts of the institutions on social capital, we do not find any evidence of such complementary.

5

institutions on behavior in laboratory experiments (Dal Bo et al., 2010; Dal Bo, 2014)

except with lab-in-the-field experiments and expanded to encompass real-world institu-

tions.

The remainder of the paper is organized as follows. Section 2 discusses the COGES

project, the social experiment, the lab-in-the-field experiments, the data, and the identi-

fication strategy. Section 3 presents the empirical results. Section 4 concludes.

2 School-Based Management

2.1 Background

Burkina Faso lags behind much of the rest of the world in achieving universal pri-

mary education.3 To address this deficiency, the government of Burkina Faso adopted a

Poverty Reduction Strategy in 2000 in which an important goal was to “guarantee that

the poor have access to basic social services”. To achieve this goal in the education sec-

tor, the Ministry of Basic Education and Literacy (MEBA) drew up a Basic Education

Ten-Year Development Plan (PDDEB), which was divided into phase I (2000-2006) and

phase II (2007-2010).4 During phase II, strong emphasis was placed on improving ac-

cess to and the quality of education through a decentralization process, which delegated

additional authority to the lowest administrative level for education, the Circonscription

d’Education de Base (CEB). Each CEB oversaw approximately 14 elementary schools

in which they facilitated teacher training programs and received the right to manage

preschool infrastructure, basic education, and literacy programs. Although post-reform

enrollments increased by 9.7% annually at public primary schools, additional problems

still remained including widening gender enrollment gaps, dropouts, and grade repeti-

tion. To tackle these problems, the government, with technical assistance from the Japan

International Cooperation Agency (JICA), started the “Support for the Improvement of

3The education system of Burkina Faso comprises three years of preschool, six years of primary, fouryears of lower secondary, and three years of upper secondary education, followed by tertiary education.Multi-grade classrooms are also common, especially in rural schools.

4The official acronyms are based on the French names, which we have translated into English. MEBArefers to Ministre de l’Enseignement de Base et de l’Alphabtisation and PDDEB refers to Plan Decennalde Dveloppement de l’Education de Base.

6

School Management through a Community Participation Project” in 2008, which in-

cluded the COGES project.

2.2 COGES

COGES involved setting up a SBM committee in each primary school. The COGES

members would have a central role in setting and implementing an annual school action

plan and the idea was for COGES to use input from the local community about how

to improve the schools. A distinctive feature used to facilitate this process was that

new COGES members would be democratically elected through secret ballot voting by

the community. Although some COGES members had been previously appointed by

government decree, the newly elected COGES members had important roles including

the presidency of COGES as well as members in charge of community participation,

girls’ enrollment, monitoring, accounting, and auditing5.

To help facilitate the COGES elections and the development and implementation

of the action plan, several types of training were conducted for stakeholders. The se-

quence of training is described in Figure 1. School principals initially attended two days

of training on how to organize community meetings and hold elections. Two community

meetings were subsequently held in the same month: the first for sharing information

about the upcoming COGES and the second for the election of COGES members. After

the election, the school principals, the COGES president and accountant, and repre-

sentatives from the municipal offices participated in two additional days of training on

making an action plan for the schools, including its implementation, monitoring, and

evaluation. Typical action plans included things like providing separate toilets for fe-

male students, constructing or repairing school facilities (e.g., classrooms, desks, and

chairs), providing school lunch for students, arranging housing for teachers, and pur-

chasing learning materials. After the action plan was proposed, another community

meeting was then held to discuss and approve the action plan. Because most schools

5Previously appointed COGES members included the local mayor, the Presidents of the Parents’ andMothers’ Associations, the school principal, as well as teacher, NGOs and union representatives. Parents’Associations (APE) and Mothers’ Associations (AME) among parents of students have also existed asschool councils in Burkina Faso since the 1960s but they had limited roles in actual school management.

7

Figure 1: Timing of the COGES Project.

could not expect external resources from the central government, COGES needed to

mobilize financial and non-financial resources within the community. Further meetings

were held to monitor the ongoing action plan and then to evaluate the previous year’s

action plan. The same cycle would then be repeated every year; at the beginning of the

new school year COGES and the community members would make a new action plan,

including a procedure to implement, monitor, and evaluate the action plan using their

own resources.

2.3 Randomized Controlled Trial

To identify the causal effect of the COGES project, we conducted an RCT in the form of

a randomized “roll-out” of the COGES project in all elementary schools in Ganzourgou

Province, Burkina Faso. Using a list of all schools in the province provided by MEBA,

we first partitioned the 279 schools into 30 strata: 10 CEB by 3 school types (public

schools, private Islamic schools, and private Catholic schools). Using random assign-

ment within each stratum, 141 schools were assigned to be first-year COGES schools

(treatment group) and 138 schools were assigned to be second-year COGES schools

(control group).6 The COGES project was offered to the first-year COGES schools dur-

ing the 2009-10 academic year and the second-year COGES schools received a delayed

offer of treatment during the subsequent 2010-11 academic year.

We conducted detailed surveys of all the major stakeholders in the school: the school

principal, a randomly selected teacher from each grade, five randomly selected students

6During data collection, we discovered that some schools did not exist or had been closed, whichreduced the number of the schools to 134 and 132 for the first-year and second-year COGES schools.

8

Table 1: Pre-treatment balance in observables

Treatment ControlAge (years) 40.27 40.94Male % 0.58 0.57Education (years) 2.40 2.45Director % 0.05 0.05Teacher % 0.05 0.06AME % 0.04 0.04APE % 0.04 0.04Public school % 0.67 0.71Private school % 0.07 0.03Islamic school % 0.27 0.26Individuals 898 847Schools 41 40

Notes: * denotes a statistically significant differ-ence in means between treatment and control at the5% level.

of each randomly selected teacher, and the household head of each of the five randomly

selected students. The first round baseline surveys were conducted December 2009 -

January 2010 and the second round endline surveys were conducted in January - Febru-

ary 2011.

For the lab-in-the-field experiments discussed below, we selected a random subset

of the 279 baseline schools, which gave us 41 first-year COGES schools and 40 second-

year COGES schools at baseline. We then recruited participants within the schools

belonging to 5 different groups: elected COGES members, teachers, parents, and com-

munity members. Table 1 reports tests of pre-treatment balance in observables for the

participants by treatment status. We cannot reject the null hypothesis of no mean dif-

ference between treatment and control in any pre-treatment covariate. Because of bud-

getary reasons, we further selected a random subset of the 81 baseline schools for the

endline experiments in November and December of 2010, which gave us 21 first-year

and 21 second-year COGES schools at endline.

9

2.4 Lab-in-the-field experiments

For our main outcome data, we conducted a public goods game among school princi-

pals, teachers, parents, community members, and elected COGES members. The public

goods game is a standard laboratory experiment used to measure voluntary cooperation

among subjects (Camerer and Fehr, 2004; Cardenas and Carpenter, 2008; Levitt and

List, 2007). However, contributions are also regarded as a measure of social capital

(Anderson et al., 2004) and this is the interpretation we use in this paper.

In our public goods games, each participant was placed in a group containing N

nonanonymous members and given an initial endowment, E. Each participant then had

to decide on an amount Yi of their endowment to secretly contribute to the public good.

The contributions were then totaled, multiplied by a factor ρ with 1 < ρ < N, and

divided equally among the N group members. Group members did not observe the

contributions of the other members but only their individual payoff, which is given by:

πi = E−Yi +ρ

N

N

∑i=1

Yi (1)

When 1 < ρ < N, ∂πi∂Yi

= −1+ ρ

N < 0 so that Yi = 0 is a dominant strategy for each

participant. Therefore, a pure-strategy Nash equilibrium is Yi = 0 for all i and any

amount Yi > 0 represents a deviation from the individually rational Nash equilibrium.

Following the literature we interpret Yi as a measure of participant i’s social capital.

In the experimental implementation, we designated groups of four members N = 4,

an endowment E = 500 FCFA, and set ρ = 2 so that the combined individual contribu-

tions were doubled before dividing them.7 We formed five different types of groups: the

fathers of students (fathers group), the mothers of the students (mothers group), either

four men or four women from the community who did not send children to the school

7On January 21, 2016, 1 US dollar was equivalent to 602 FCFA. FCFA refers to the Franc CommunautFinancire Africaine, which is a currency backed by the French Treasury and used in Burkina Faso andmany other West African Francophone countries. To understand the magnitude of these transfers notethat the official minimum wage rate in Burkina Faso is 1,050 FCFA per day. However, it is common toset a daily wage rate at 300 to 500 FCFA in rural agricultural and urban service sectors. So keeping theentire transfer and contributing nothing would be the equivalent of approximately one day of work formany individuals in our sample.

10

(community group), a group consisting of the school principal, one teacher, one father,

and one mother (mixed group), and a group with the four elected COGES members

(COGES group).8 Each group played the public goods game twice with an immediate

monetary reward after each round. The repeated play was to check whether, similar to

existing experimental findings, public goods contributions would decline towards the

free riding Nash equilibrium over time (Andreoni, 1988).

A potential concern with using public goods game contributions is that they could

instead be driven by altruism instead of capturing social capital or the propensity for

voluntary cooperation. To separate out potential effects of altruism, we followed An-

derson et al. (1998) and used the dictator game contributions among our subjects as an

additional control9. The dictator game was conduct as a hypothetical (without monetary

incentives) survey question among the public goods game participants. Each participant

was asked to imagine that they were randomly matched with another group member

from their public goods game experimental session and asked for a hypothetical trans-

fer amount out of a hypothetical endowment of 500 FCFA. They were assured their

answers would be confidential. The choice set for the transfer was {0,100,200,300,400,

500} FCFA.

2.5 Data

Table 2 displays summary statistics. We use three data sets in our analysis. The first

is individual level data from the field experiment data combined with survey measures

on GSS questions and community participation. The second and third are school-level

panel data sets on activities within the schools (school director survey) and on COGES

activities (project records data). These data come from the larger set of 279 schools

(with some schools missing data).

In the field experiment data, individuals contributed on average 321.2 FCFA in the

8For the community group, if the school id number was even, we chose four women and otherwisewe chose four men.

9Although a positive transfer in the dictator game is usually interpreted as a measure of altruism,other potential interpretations such as image self-construction are again possible (Camerer and Fehr,2004; Levitt and List, 2007). These interpretations of the dictator game are fine in our context as long asit nets out other drivers of public goods game contributions.

11

Table 2: Summary Statistics

Field experiment Mean SDPublic goods game contribution (FCFA) 321.2 137.4Hypothetical dicator game contribution (FCFA) 277.0 109.8GSS trust 0.60 -GSS fair 0.74 -GSS help 0.74 -COGES random assignment % 0.51 -COGES implementation % 0.47 -Father group % 0.21 -Mother group % 0.23 -Community group % 0.21 -Mixed group % 0.22 -COGES group % 0.14 -2nd round % 0.50 -Observations 4,460Number of individuals 1,745Number of schools 81

School director survey Mean SDTuition fee (FCFA per year) 1,352 3,964Textbook fee (FCFA per year) 31.07 326.0Financial contribution (FCFA per year) 3,791 7,815School meal frequency (days per month) 18.15 4.51School meal (%) 0.69 -Functional toilet (%) 0.72 -Observations 531Number of schools 271

Project records data Mean SDNumber of projects 2.79 3.09Total budget (FCFA per year) 32,768 142,073Observations 523Number of schools 270

Notes:

public goods game (64% of the endowment) and 277.0 FCFA in the hypothetical dictator

game (55% of the endowment). Although these are higher than average contributions

as a percent of the endowment reported in meta-analyses: 37.7% for the public goods

game (Zelmer, 2003) and 28% in the dictator game (Engel, 2011), they are not outside

the range of contributions, especially in non-WEIRD societies (Henrich et al., 2004).

Levels of GSS trust, fair, and help are also high. In our survey, 60% think most peo-

12

ple can be trusted, 74% think people would try to be fair, and 74% think people would

try to be helpful.10 Compared to the 5th wave of the World Values Survey 2005-2009,

our sample would be at the 71st percentile for trust11. Our sample also reports sub-

stantially higher values than US respondents in the 2010 GSS with analogous responses

35% for trust, 58% for fair, and 52% for help12.

From the school director survey, children were required to pay tuition fees (1,352

FCFA per year), textbook fees (31 FCFA per year), and the parents also needed to

make a financial contribution to the school (3,791 FCFA per year). Most schools of-

fered school meals (69%) on average 18 days per month and most had functional toilets

(72%). From the project records data, schools on average implemented 2.79 projects

per school year and spent more than 32,000 FCFA total on average. However, there is

substantial between school variation in all of these variables.

2.6 Econometric Model

We estimate the impact of the COGES project on social capital Y as measured by public

goods game contributions. We first conducted the public goods game in February 2010

after the COGES elections in the first-year COGES (treated) schools. The second public

goods experiments were conducted in November and December of 2010 after the 2010

school year during which treated schools had implemented their COGES action plans.

This was also after the COGES elections in the second-year COGES schools (Figure 1).

Because the COGES project involved a particular sequence of experimental intervention

and data collection, the timing of events is important for interpreting what is being

identified in the econometric model.

The data from the public goods games can be classified into four cases according

to treatment status (D = 1 or D = 0) and whether the data were collected at time t

10We used standard questions from the GSS. The GSS trust question reads “Generally speaking, wouldyou say that most people can be trusted or that you can’t be too careful in dealing with people?”, the GSSfair question reads “Do you think most people would try to take advantage of you if they got a chance, orwould they try to be fair?” and the GSS help question reads “Would you say that most of the time peopletry to be helpful, or that they are mostly just looking out for themselves?”. For each question, we codedthe answer as 1 if it was “positive” (people can be trusted, are helpful, or are fair), and 0 otherwise.

11Authors’ calculations.12Authors’ calculations.

13

“before” (t = b) or after (t = a). If we employ the “before” data collected in February

2010, the average outcome difference between the first-year and second-year COGES

schools, Y D=1b − Y D=0

b , identifies the impact of the COGES election. This is because

the election had occurred in the first-year COGES (treated) schools but it had not yet

occurred in the second-year COGES (control) schools. The COGES project itself had

also yet to be implemented in either the treatment or the control schools. We call this

an “election effect”, which is defined as the effect arising from the randomization of

the democratic elections. With the after data from November and December of 2010,

the outcome difference between the first-year and second-year COGES schools, Y D=1a −

Y D=0a , identifies the impact of the implementation of the COGES action plan in the first-

year COGES schools. This is because the second-year schools had then been exposed

to the election, while the first-year schools had been exposed to both the election and to

the implementation of the school action plan. We call this the “implementation effect”,

which is defined as the accumulated impact of the COGES implementation net of the

direct election effect13. The total impact of the COGES project can be estimated by

summing the election and the implementation effects.

We use the following linear regression model to estimate the Intent to Treat (ITT)

impact of the COGES project,

Yirst = αt +βtDs +uist , (2)

where Yirst is public goods game contribution for individual i in school s in round r at

time t. Given that Ds was randomly assigned, when t=b the treatment effect βb identifies

the effect generated by the election. Alternatively, when t=a, the treatment effect βa

identifies the effect generated by the implementation of the COGES project. Because

of the timing of the data collection in our context, the difference-in-difference estimator

13An additional possibility is that there is fade-out of the election effects in the first-year COGESschools. In this case, the impacts in the after data are estimating the difference between the implementeffect in the first-year COGES schools and the election effects in the second-year COGES schools. Inthis situation the after data impact serves as a lower bound on the true implement effect because it netsout the (presumably nonnegative) election effect. However, in our empirical results the election effectsare mostly zero in the first-year COGES schools, so if the second-year COGES schools also have a zeroelection effect then the impact on the after data identifies the implementation effect.

14

captures the difference between these two effects.14 We also show the estimation results

with covariates because their inclusion can potentially help increase the precision of the

treatment impact estimate. 15

3 Estimation Results

3.1 Social Capital

Table 3 summarizes the estimation results for the election and implementation effects

on public goods game contributions (social capital). In columns (1), (2), and (3), we

estimated Equation 2 using the “before” data, which identifies the election effect. Al-

though positive, the coefficients on the treatment variable, D, are insignificant, which

indicates that a community-wide democratic election of COGES members did not in-

crease social capital on average. However, when we estimate equation (2) using the

after data to identify the implementation effect of COGES on social capital, the esti-

mates are positive and statistically significant. These results are shown in columns (4),

(5), and (6) in which each column adds additional controls. With the implementation

of the COGES project, the average amount of voluntary contributions to public goods

increased between 8.1% and 9.7%. These effects are large. In a survey of public goods

game results, Ledyard (1995) reports that average contributions across studies ranged

between 40% to 60% of the endowment so the effect size of our intervention moves

approximately halfway between the range of contributions reported in the literature.

Because each participant played the public goods game twice, we report the estima-

tion results using data pooled from the two rounds of the game. In all specifications,

the second round public goods game actually stimulated a significantly larger amount

of voluntary contribution to the public goods than the first round.16 Although there

14(Y D=1

a − Y D=1b

)−(Y D=0

a − Y D=0b

)=(Y D=1

a − Y D=0a

)−(Y D=1

b − Y D=0b

)= βa−βb

15Because of the existence of some second-year COGES schools that implemented COGES projectsduring the first-year (always-takers) and some second-year COGES schools implemented COGESprojects during the first-year (always-takers), we also estimated IV models using COGES randomiza-tion as instrument for COGES implementation, which identifies a local average treatment effect (LATE)on the subpopulation of compliers (Imbens and Angrist, 1994). However, the magnitude of impacts isvery similar to the ITT estime so we relegated these results to the appendix.

16Estimating the models separately for the first and second round public goods game data gives thesame overall pattern of results.

15

Table 3: COGES Election and Implementation Effects on PGG (ITT)

Election Effect Implementation Effect(1) (2) (3) (4) (5) (6)

D 10.77 3.65 4.65 28.91† 28.59† 22.54*(9.07) (8.87) (8.18) (13.81) (13.63) (11.55)

Mothers group -14.58 -17.48 19.11 7.77(17.03) (15.82) (26.59) (23.61)

Community group -20.59 -23.25* 15.28 5.61(13.35) (12.40) (22.95) (20.24)

Mixed group 32.58† 25.41* 40.12* 32.69(15.04) (14.02) (22.93) (21.02)

COGES group 44.46† 30.21* 35.87 30.94(17.15) (16.32) (22.13) (20.02)

Dictator game contribution 0.37‡ 0.42‡

(0.03) (0.04)Round 2 24.17‡ 24.17‡ 24.17‡ 13.06‡ 13.06‡ 13.06‡

(4.83) (4.83) (4.83) (3.64) (3.64) (3.64)Constant 294.9‡ 296.8‡ 199.6‡ 360.7‡ 354.3‡ 232.1‡

(16.15) (22.74) (23.65) (20.75) (34.50) (34.43)

Controls No Yes Yes No Yes YesStrata FE Yes Yes Yes Yes Yes YesEstimation OLS OLS OLS OLS OLS OLSTime period Before Before Before After After After

N 2822 2822 2822 1638 1638 1638R2 0.07 0.12 0.20 0.07 0.09 0.22

Notes: The dependent variable is the amount contributed in the public goods game from an initial stakeof 500 FCFA. Robust standard errors clustered at the school x group level are reported in parentheses.Control variables are age, years of schooling, and indicator variables for male, private school, Islamicschool, school director, teacher, AME member, and APE member. ‡ p<0.01, † p<0.05, * p<0.1.

was a publicly announced fixed ending time of the game, this finding is not necessarily

in conflict with theoretical possibilities such as learning about free-riding or voluntary

contribution arising from an infinitely repeated game (Andreoni, 1988).

As discussed, our results may also be driven by social norms or other-regarding

preferences such as altruism. To address this possibility, in columns (3) and (6) of

Table 3, we add the amount sent in the dictator game as measure of altruism. Although

the dictator game contribution is a strong predictor of public goods game contributions,

the impact estimate of COGES changes very little compared to the baseline specification

in columns (1) and (4).

16

Table 4: COGES Election and Implementation Effects on GSS Questions (ITT)

Election Effect Implementation EffectTrust Fair Help Trust Fair Help

D -0.01 0.01 -0.03 0.01 0.06* 0.06†

(0.03) (0.03) (0.03) (0.03) (0.03) (0.03)Mothers group -0.02 -0.03 -0.03 0.07 0.03 -0.03

(0.06) (0.05) (0.06) (0.07) (0.06) (0.06)Community group -0.01 0.02 -0.01 0.06 0.01 -0.04

(0.05) (0.05) (0.05) (0.06) (0.05) (0.05)Mixed group 0.00 0.00 -0.02 0.01 -0.01 -0.05

(0.05) (0.05) (0.05) (0.07) (0.07) (0.06)COGES group 0.11 0.08 0.06 -0.01 0.02 -0.04

(0.07) (0.06) (0.06) (0.06) (0.06) (0.05)Constant 0.54‡ 0.74‡ 0.76‡ 0.77‡ 0.80‡ 0.75‡

(0.08) (0.08) (0.08) (0.09) (0.10) (0.08)

Controls Yes Yes Yes Yes Yes YesStrata FE Yes Yes Yes Yes Yes YesEstimation OLS OLS OLS OLS OLS OLSTime period Before Before Before After After After

N 1410 1411 1409 814 819 818R2 0.06 0.07 0.06 0.11 0.09 0.12

Notes: The dependent variable is the amount contributed in the public goods gamefrom an initial stake of 500 FCFA. Robust standard errors clustered at the school xgroup level are reported in parentheses. Control variables are age, years of schooling,and indicator variables for male, private school, Islamic school, school director, teacher,AME member, and APE member. ‡ p<0.01, † p<0.05, * p<0.1.

3.2 GSS Questions

In Table 4, we estimate ITT impacts on the GSS measures using the same specification

as Equation 2. We find positive implementation effects of COGES on the GSS fair and

GSS help measures with both increasing by 6 percentage points. The COGES demo-

cratic institution seems to be engendering fairer and more community oriented behavior

(e.g., most people are not just looking out for themselves) but not necessarily individual

trust. This is also interesting because only GSS trust and not GSS help or fair predict

PGG contributions. Although a complete theory of how or why this particular institu-

tion affected these different measures of social capital is beyond the current paper, we

report these impact estimates to give a broad characterization of what we think happened

17

Table 5: COGES Impacts on School Inputs (ITT)

Table 9: COGES Impacts on School InputsTuition Textbook Financial School School Meal Functional

Fee Fee Contribution Meal Frequency ToiletD -102.8 6.30 65.16 -0.02 0.13 0.06

(141.1) (12.46) (571.4) (0.05) (1.01) (0.04)After 568.3 70.30 1,562 0.37‡ 0.22 0.07†

(484.1) (69.31) (1,616) (0.09) (1.25) (0.03)D x After

Constant -314.4 -29.13 1,057 0.59‡ 19.14‡ 0.60‡

(214.1) (22.28) (896.8) (0.05) (0.82) (0.03)

N 505 515 431 522 497 520R2 0.52 0.13 0.17 0.28 0.08 0.40

Notes: ‡ p<0.01, † p<0.05, * p<0.1.

within the communities.

3.3 Real-World Decisions

Although the public goods game gives a measure that is comparable across studies, one

concern is that it might be an artificial situation for participants from these rural com-

munities. Therefore, we also checked the consistency of our public goods game results

using real-world decisions related to the schools from the school director and project

records data. From the school director data, we considered the following variables;

tuition fee paid per year in FCFA, annual textbook fee per student or family, annual

financial contributions to the school, whether the school provided school meals, the fre-

quency of school meals per month, and the availability of functional toilets at the school.

From the project records data, we use the number of projects implemented at the school

and the total budget for projects. Because the timing of the data collection for these data

sets differs from the lab-in-the-field experiments, we need to set up slightly different

econometric models to analyze these data.

18

3.3.1 School Director Data

Unlike the public goods game data, the school director data have a true baseline col-

lected prior to both the election and implementation of COGES so we use a canonical

difference-in-differences model,

Yst = γ0 + γ1Ds + γ2At + γ3DsAt +ust , (3)

where At = 1{t = a} is an indicator that the observation comes from the after data.

The coefficient on the COGES indicator now represents the combined election and im-

plementation effect because we can no longer utilize the timing of data collection to

untangle the two effects. These estimation results are displayed in Table 5. The results

show that COGES increased both the amount of tuition fee payments (467.1 FCFA)

and the availability of school meals (8 percentage points). For these two variables, the

experimental results are consistent with real-world behavior.

3.3.2 Project Records Data

From the project records we considered the number of projects implemented in the

schools and the amount of the project budget as outcome variables. Note that even

schools without COGES still had the option to implement projects, just that they lacked

the formal mechanism of an SBM to facilitate such projects.

We have project records data from both the first school year (2009-10) during which

only the treated schools implemented COGES and from the second school year (2010-

11) during which both treated and control schools implemented COGES (because of

the randomized roll-out). However, we do not have pre-treatment data on any project

records. For this reason, we use a “difference-in-differences in reverse” model (Kim

and Lee, 2018; Lee and Sawada, 2019) in which the post-treatment period (instead of

the pre-treatment period) serves to net out permanent unobserved differences between

19

Table 6: COGES Impacts on School Projects (ITT)

Number of projects Project budget(1) (2) (3) (4)

1−D -3.61‡ -110,383‡

(0.28) (23,113)After -0.05 -115,780‡

(0.25) (23,034)(1−D) x After 2.38‡ 107,042‡

(0.35) (23,351)D 3.61‡ 110,383‡

(0.28) (23,113)Constant 3.98‡ 0.37‡ 119,122‡ 8,739†

(0.25) (0.12) (22,790) (3,821)

Data years 2009-11 2009-10 2009-11 2009-10Estimation DND Reverse OLS DND Reverse OLS

N 523 254 523 254R2 0.22 0.40 0.12 0.08

Notes: ‡ p<0.01, † p<0.05, * p<0.1.

treatment and control. Such a model can be set-up as follows:

Yst = ζ0 +ζ1(1−Ds)+ζ2At +ζ3At(1−Ds)+ust . (4)

The estimation results are displayed in Table 6 in columns (1) and (3). The results

show that COGES generated 2.38 school projects on average and also caused a large

and statistically significant increase in the annual school budget of FCFA 107,042 (ap-

proximately 180 USD), an almost 400% increase in the school budget compared to

second-year COGES schools. In addition to the difference-in-difference model a sim-

ple mean difference between treatment and control in columns (2) and (4) using only

the first-year data delivers similar impact estimates of 3.27 (number of projects) and

110,039 FCFA (project budget). This is intuitive given the randomization. In terms of

interpretation, since all children attending school arguably benefit from improved school

services, these results indicate that COGES improved contributions to public goods not

only in the laboratory setting.

20

3.3.3 Tontine

A final piece of supporting evidence comes from a complementary paper in our re-

search group. Using the same RCT, Todo et al. (2016) find that the COGES project

increased the use of village Tontines, which are rotating savings and credit associations

(ROSCAs) common in Burkina Faso. Because ROSCAs require social capital to self-

select reliable participants and enable mutual monitoring (Armendariz and Morduch,

2010; Zeller, 1998), the results are consistent with the idea that COGES generated real-

world increases in social capital in a broader sense beyond both the laboratory and the

school setting.

4 Conclusion

This paper evaluates the impact of the randomized introduction of a democratic insti-

tution in the form of SBM on social capital in rural Burkina Faso. We found large

impacts on social capital as measured by public goods game contributions and the main

experimental result was complemented by impacts on attitudinal measures of fairness

and help within the communities and on real-world outcomes observed in the schools.

These findings are important in identifying how promoting democratic structures and

community participation in a country with otherwise weak governance (House, 2009)

can improve the local provision of services. A potential implication of these findings

is that SBMs may improve cost recovery by increasing community members’ willing-

ness to contribute to local public goods. This could lead to better fiscal sustainability in

schools in which local stakeholders have more control. More broadly the results speak

to important positive interactions between institutions and social capital and suggests

that some forms of institutions can be successfully externally introduced. Successful

institutions seem not to be those that simply receive money but those that have demo-

cratic mechanisms and that operate through existing social capital networks (Pradhan et

al., 2014), which is consistent with the idea that institutional structures function better

when resources are raised and used with the consent of the governed (Deaton, 2013).

Our findings also add to work that seeks to understand the formation of social capital

21

(Glaeser, 2001) and the impact we found running from institutions to social capital is

consistent with positive feedback loops and multiple equilibria that characterize models

of social capital formation (Bowles and Gintis, 2002). Unfortunately understanding the

cumulative impact of COGES is not possible because of the randomized roll-out and

this is an important limitation of our study because some authors have found only short-

run impacts of community-based interventions (Kremer and Miguel, 2007; Casey et al.,

2012; Casey, 2018). Although our results are consistent with the COGES pilot study

(Sawada and Ishii, 2012) and results from Indonesia (Pradhan et al., 2014), further val-

idation in different contexts is important. JICA has been supporting other COGES-like

projects in West Africa (in Niger from 2004, in Senegal from 2007, and in Mali from

2008) so careful investigation of their effectiveness can generate important evidence on

SBM more generally.

22

References

Acemoglu, Daron, Simon Johnson, and James A Robinson, “The colonial origins ofcomparative development: An empirical investigation,” American Economic Review,2001, 91 (5), 1369–1401.

, Suresh Naidu, Pascual Restrepo, and James A. Robinson, “Democracy doescause growth,” Journal of Political Economy, 2019, 127 (1), 000–000.

Alesina, Alberto and Paola Giuliano, “Culture and institutions,” Journal of EconomicLiterature, 2015, 53 (4), 898–944.

Anderson, Lisa R, Jennifer M Mellor, and Jeffrey Milyo, “Social capital and con-tributions in a public-goods experiment,” American Economic Review, 2004, 94 (2),373–376.

Anderson, Simon P, Jacob K Goeree, and Charles A Holt, “A theoretical analysisof altruism and decision error in public goods games,” Journal of Public Economics,1998, 70 (2), 297–323.

Andreoni, James, “Why free ride?: Strategies and learning in public goods experi-ments,” Journal of Public Economics, 1988, 37 (3), 291–304.

Armendariz, Beatriz and Jonathan Morduch, The Economics of Microfinance, MITpress, 2010.

Banerjee, Abhijit V, Rukmini Banerji, Esther Duflo, Rachel Glennerster, and StutiKhemani, “Pitfalls of Participatory Programs: Evidence from a Randomized Evalu-ation in Education in India,” American Economic Journal: Economic Policy, 2010, 2(1), 1–30.

Barr, Abigail, Truman Packard, and Danila Serra, “Participatory accountability andcollective action: evidence from field experiments in Albanian schools,” World BankPolicy Research Working Paper 6027, Impact Evaluation Series No.55, 2012.

Barrera-Osorio, Felipe, Tazeen Fasih, and Harry Anthony Patrinos, DecentralizedDecision-Making in Schools: The Theory and Evidence on School-Based Manage-ment, The World Bank, 2009.

Beasley, Elizabeth and Elise Huillery, “Willing but Not Able: Short-Term Experimen-tal Evidence on Parent Empowerment and School Quality,” World Bank EconomicReview, 2017, 31 (2).

Blimpo, Moussa P, David K Evans, and Nathalie Lahire, “Parental human capitaland effective school management: evidence from The Gambia,” World Bank PolicyResearch Working Paper 7238, Impact Evaluation Series No.55, 2015.

Bo, Pedro Dal, “Experimental evidence on the workings of democratic institutions,”Institutions, Property Rights, and Economic Growth: The Legacy of Douglass North,2014, 266.

23

, Andrew Foster, and Louis Putterman, “Institutions and behavior: Experimentalevidence on the effects of democracy,” American Economic Review, 2010, 100 (5),2205–29.

Bowles, Samuel and Herbert Gintis, “Social capital and community governance,” Theeconomic journal, 2002, 112 (483), F419–F436.

Bruns, Barbara, Deon Filmer, and Harry Anthony Patrinos, Making schools work:New evidence on accountability reforms, The World Bank, 2011.

Camerer, Colin F and Ernst Fehr, “Measuring social norms and preferences usingexperimental games: A guide for social scientists,” Foundations of human sociality:Economic experiments and ethnographic evidence from fifteen small-scale societies,2004, 97, 55–95.

Cardenas, Juan Camilo and Jeffrey Carpenter, “Behavioural development eco-nomics: Lessons from field labs in the developing world,” The Journal of Devel-opment Studies, 2008, 44 (3), 311–338.

Casey, Katherine, “Radical decentralization: Does community-driven developmentwork?,” Annual Review of Economics, 2018, 10, 139–163.

, Rachel Glennerster, and Edward Miguel, “Reshaping institutions: Evidence onaid impacts using a preanalysis plan,” The Quarterly Journal of Economics, 2012,127 (4), 1755–1812.

Deaton, Angus, “Understanding the mechanisms of economic development,” Journalof Economic Perspectives, 2010, 24 (3), 3–16.

, The great escape: health, wealth, and the origins of inequality, Princeton UniversityPress, 2013.

Dell, Melissa, “The persistent effects of Peru’s mining mita,” Econometrica, 2010, 78(6), 1863–1903.

, Nathan Lane, and Pablo Querubin, “The historical state, local collective action,and economic development in Vietnam,” Econometrica, 2018, 86 (6), 2083–2121.

Duflo, Esther, Pascaline Dupas, and Michael Kremer, “School governance, teacherincentives, and pupil–teacher ratios: Experimental evidence from Kenyan primaryschools,” Journal of Public Economics, 2015, 123, 92–110.

Engel, Christoph, “Dictator games: A meta study,” Experimental Economics, 2011, 14(4), 583–610.

Ensminger, Jean, Making a market: The institutional transformation of an Africansociety, Cambridge University Press, 1996.

Gertler, Paul, Harry Anthony Patrinos, and Eduardo Rodrıguez-Oreggia,“Parental Empowerment in Mexico: Randomized Experiment of the” Apoyos a LaGestion Escolar (Age)” Program in Rural Primary Schools in Mexico.,” Society forResearch on Educational Effectiveness, 2012.

24

Glaeser, Edward L, “The formation of social capital,” Canadian Journal of PolicyResearch, 2001, 2 (1), 34–40.

Hayami, Yujiro, “Social capital, human capital and the community mechanism: To-ward a conceptual framework for economists,” The Journal of Development Studies,2009, 45 (1), 96–123.

Henrich, Joseph Patrick, Robert Boyd, Samuel Bowles, Ernst Fehr, Colin Camerer,Herbert Gintis et al., Foundations of human sociality: Economic experiments andethnographic evidence from fifteen small-scale societies, Oxford University Press onDemand, 2004.

Henrich, Joseph, Robert Boyd, Samuel Bowles, Colin Camerer, Ernst Fehr, Her-bert Gintis, and Richard McElreath, “In search of homo economicus: behavioralexperiments in 15 small-scale societies,” American Economic Review, 2001, 91 (2),73–78.

, Steven J Heine, and Ara Norenzayan, “The weirdest people in the world?,” Be-havioral and brain sciences, 2010, 33 (2-3), 61–83.

House, Freedom, “Annual survey of freedom country scores 1972-73 to 2009-10,”Database on-line available from Freedom House. Washington, DC: Freedom House,2009.

Imbens, Guido W and Joshua D. Angrist, “Identification and Estimation of LocalAverage Treatment Effects,” Econometrica, 1994, 62 (2), 467–475.

jae Lee, Myoung and Yasuyuki Sawada, “Review on difference in differences,” Ko-rean Economic Review, 2019.

Kim, Kimin and Myoung jae Lee, “Difference in differences in reverse,” EmpiricalEconomics, 2018, pp. 1–21.

Knack, Stephen and Philip Keefer, “Does social capital have an economic payoff?A cross-country investigation,” The Quarterly journal of economics, 1997, 112 (4),1251–1288.

Kozuka, Eiji, Yasuyuki Sawada, Yasuyuki Todo et al., “How Can Community Par-ticipation Improve Educational Outcomes? Experimental Evidence from a School-Based Management Project in Burkina Faso,” Technical Report 2016.

Kremer, Michael and Alaka Holla, “Improving education in the developing world:what have we learned from randomized evaluations?,” Annu. Rev. Econ., 2009, 1 (1),513–542.

and Edward Miguel, “The illusion of sustainability,” The Quarterly journal of eco-nomics, 2007, 122 (3), 1007–1065.

Laat, Joost De, Michael Kremer, and Christel Vermeersch, “Teacher incentives andlocal participation,” Working Paper Harvard University, 2008.

Ledyard, John O., “Public Goods: A Survey of Experimental Research,” in J. Kageland A. Roth, eds., Handbook of Experimental Economics, 1995.

25

Levitsky, Steven and Daniel Ziblatt, How democracies die, Crown, 2018.

Levitt, Steven D and John A List, “What do laboratory experiments measuring socialpreferences reveal about the real world?,” Journal of Economic perspectives, 2007,21 (2), 153–174.

Mansuri, Ghazala and Vijayendra Rao, “Community-based and-driven development:A critical review,” The World Bank Research Observer, 2004, 19 (1), 1–39.

and , “Localizing Development,” 2013.

Martinez-Bravo, Monica, Gerard Padro i Miquel, Nancy Qian, Yiqing Xu, andYang Yao, “Making Democracy Work: Formal Institutions and Culture in RuralChina,” Unpublished manuscript, 2017.

North, Douglass C, “Institutions,” Journal of economic perspectives, 1991, 5 (1), 97–112.

Ostrom, Elinor, Governing the commons: the evolution of institutions for collectiveaction, Cambridge: Cambridge University Press, 1990.

, “Social capital: a fad or a fundamental concept,” Social capital: A multifacetedperspective, 2000, 172 (173), 195–98.

Pradhan, Menno, Daniel Suryadarma, Amanda Beatty, Maisy Wong, Arya Gaduh,Armida Alisjahbana, and Rima Prama Artha, “Improving Educational Qualitythrough Enhancing Community Participation: Results from a Randomized Field Ex-periment in Indonesia,” American Economic Journal: Applied Economics, 2014,pp. 105–126.

Putnam, Robert, “The prosperous community: Social capital and public life,” Theamerican prospect, 1993, 13 (Spring), Vol. 4.).

Putnam, Robert D, Bowling alone: Americas declining social capital, Springer, 2000.

Putnam, Robert D., Robert Leonardi, and Raffaella Y. Nanetti, Making DemocracyWork: Civic Traditions in Modern Italy, Princeton University Press Princeton, NJ,2001.

Sawada, Yasuyuki and Takaharu Ishii, “Do Community-Managed Schools FacilitateSocial Capital Accumulation? Evidence from the COGES Project in Burkina Faso,”JICA Research Institute Working Paper42, 2012.

Szreter, Simon and Michael Woolcock, “Health by association? Social capital, socialtheory, and the political economy of public health,” International journal of epidemi-ology, 2004, 33, 650–667.

Todo, Yasuyuki, Eiji Kozuka, Yasuyuki Sawada et al., “Can School-Based Manage-ment Generate Community-Wide Impacts in Less Developed Countries? Evidencefrom Randomized Experiments in Burkina Faso,” Technical Report 2016.

26

Westhorp, Gill, Bill Walker, Patricia Rogers, Nathan Overbeeke, Daniel Ball, andGraham Brice, “Enhancing community accountability, empowerment and educationoutcomes in low and middle-income countries: A realist review,” EPPI-Centre, SocialScience Research Unit, Institute of Education, University of London, 2014.

Woolcock, Michael, “The place of social capital in understanding social and economicoutcomes,” Canadian Journal of Policy Research, 2001, 2 (1), 11–17.

World Bank, World Development Report 2004 : Making Services Work for Poor People2004.

Zeller, Manfred, “Determinants of repayment performance in credit groups: The roleof program design, intragroup risk pooling, and social cohesion,” Economic develop-ment and cultural change, 1998, 46 (3), 599–620.

Zelmer, Jennifer, “Linear public goods experiments: A meta-analysis,” ExperimentalEconomics, 2003, 6 (3), 299–310.

27