WPS7238
Policy Research Working Paper 7238
Parental Human Capital
and Effective School Management
Evidence from The Gambia
Moussa P. Blimpo
David K. Evans
Nathalie Lahire
Education Global Practice Group
&
Africa Region
Office of the Chief Economist
April 2015
Policy Research Working Paper 7238
Abstract
Education systems in developing countries are often cen- of 273 Gambian primary schools were randomized to one
trally managed in a top-down structure. In environments of the three groups. The program was implemented through
where schools have different needs and where localized the government education system. Three to four years into
information plays an important role, empowerment of the the program, the full intervention led to a 21 percent reduc-
local community may be attractive, but low levels of human tion in student absenteeism and a 23 percent reduction
capital at the local level may offset gains from local informa- in teacher absenteeism, but produced no impact on stu-
tion. This paper reports the results of a four-year, large-scale dent test scores. The effect of the full program on learning
experiment that provided a grant and comprehensive school outcomes is strongly mediated by baseline local capacity,
management training to principals, teachers, and commu- as measured by adult literacy. This result suggests that, in
nity representatives in a set of schools. To separate the effect villages with high literacy, the program may yield gains on
of the training from the grant, a second set of schools received students’ learning outcomes. Receiving the grant alone had
the grant only with no training. A third set of schools served no impact on either test scores or student participation.
as a control group and received neither intervention. Each
This paper is a product of the Education Global Practice Group and the Office of the Chief Economist, Africa Region. 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 devans2@worldbank.org, nlahire@worldbank.org, and moussa.blimpo@ou.edu.
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
Parental Human Capital and Effective School Management:
Evidence from The Gambia
Moussa P. Blimpo1 David K. Evans2 Nathalie Lahire2
Keywords: Education, Management, School-based Management
JEL Classification: O15, I21, C93
The authors would like to thank the Ministry of Basic and Secondary Education in The Gambia for unceasing
collaboration on this study. The authors also acknowledge the World Bank's Africa Program for Education
Impact Evaluation, of which this study is a part, and the Education Program Development Fund for funding.
Deon Filmer, Arianna Legovini, Raja Bentaouet Kattan, Moustapha Lo, Harry Patrinos, Yasu Sawada, and Jee-
Peng Tan provided valuable guidance and feedback. The authors thank the participants of seminars at the
Stanford Institute for Economic Policy Research, New York University, Stanford University (Economics
Department), the University of Southern California, the Pacific Conference for Development Economics 2012,
and the World Bank for their comments and suggestions. Emily K. Rains and Anna Popova provided excellent
research assistance.
1
University of Oklahoma. moussa.blimpo@ou.edu
2
World Bank. devans2@worldbank.org and nlahire@worldbank.org
1. Introduction
Every year, billions of dollars are spent to provide services to the poor in low-income
countries. Unfortunately, there is a long-standing record of failures in delivery systems,
whether in education, health, or other sectors. Empowerment of local communities in
school management has received growing attention from both academics and practitioners
in developing countries as part of a broad and global program to improve service delivery
to the poor by involving them directly in the delivery process (World Bank 2004). The quality
of local school management has been shown to be strongly associated with favorable
education outcomes across countries (Bloom et al. 2014). In Africa, countries including
Ghana, Niger, Senegal, Madagascar, Kenya, Rwanda, and Mozambique have already
embraced variants of school autonomy in their education systems (Bruns, Filmer & Patrinos
2011).
In this research, we assess the medium-run impact of a program seeking to empower local
communities in school management in The Gambia. On the one hand, local leadership may
have significant additional information relative to the central authorities about local needs,
local politics, and other constraints. Local management also may increase accountability
(Bruns, Filmer & Patrinos 2011), as was observed and demonstrated with a school-based
management and accountability program in Mexico (Gertler, Patrinos, & Rubio-Codina
2012). These would suggest that the program may be effective in improving student
learning. On the other hand, local leadership or members of the community may also lack
the competency (relative to central leadership) to design or implement the processes
necessary to tackle local problems, suggesting that the program could be ineffective. The
net effect of such a policy is ambiguous.3 This paper uses a large field experiment in The
Gambia to evaluate and draw lessons from a comprehensive school management and
capacity building program – called Whole School Development (WSD). The intervention and
subsequent data collection were carried out between 2007 and 2011.
In WSD schools, principals, certain teachers, and members of the communities received
comprehensive training in school management. During this training, the schools'
stakeholders (including the community) developed school management plans addressing
short- and long-term goals in each of these areas. A national semi-autonomous WSD unit
associated with the Ministry of Education guided them. In order to help schools initiate the
implementation of their plans, the Ministry of Education provided a grant worth
approximately US$500. To separate the effect of the grant from that of the training,
another set of schools received a grant of the same size without the accompanying training
component (called Grant-only schools).
3 Retrospective evaluations of such complex programs present many challenges, but early evidence from El
Salvador’s community-managed schools program found positive impacts on participation and language skills
(Jimenez and Sawada 1999).
2
In addition, a new school constitution had been developed by the Ministry of Education as
part of its new School Management Manual (SMM) to enhance cooperation between
teachers and the community. Acceptance of the new constitution was a prerequisite for
receipt of the grant. All schools receiving grants (both schools with WSD plus grant and
Grant-only schools) were directed to use the grant towards some aspect of school
development that related directly to teaching and learning (i.e., constructing teacher
housing would not be an acceptable use). Finally, the control schools received neither a
grant nor the management training. We randomly assigned each of 273 Gambian basic
cycle schools to one of the three groups.
At the end of the 2011 school year, three to four years into the program, we found no effect
of the WSD intervention on learning outcomes, measured by scores on a comprehensive
test in Mathematics and English. However, we found that the intervention did lead to a
reduction in student absenteeism of nearly 5 percentage points from a base of 24 percent,
and a reduction in teacher absenteeism of about 3 percentage points from a base of about
13 percent. We found no effect of the Grant-only intervention, relative to the control, on
test scores or participation. If the reduction in student absenteeism in the WSD schools led
to increased attendance of students with poorer performance, then the average treatment
effect on test scores would be biased downward. To correct for this potential selection bias,
we used Lee’s (2009) trimming procedure to calculate the upper and lower bounds of the
treatment effect on test scores. Our estimates indicate that, once corrected for selection,
the average treatment effect ranges from -0.19 to 0.17 standard deviations for
Mathematics and -0.16 to 0.26 standard deviations for English. Given that the bounds are
roughly centered on zero, we take zero as our preferred and conservative point estimate.
We analyzed the importance of baseline local capacity in mediating the effect of the WSD.
As mentioned earlier, theory would predict that, all else equal, the WSD is more effective
in areas with higher baseline capabilities. We interacted the intervention dummies with
average district level adult literacy in 2006. The estimates yield a positive and significant
effect of the interaction term. The finding is qualitatively the same when we replace district
level adult literacy by the share of School Management Committee (SMC) members with
no formal education (i.e., cannot read or write): In that case, we find a negative and
significant effect of the interaction term. Our findings suggest that the WSD can work in
areas with higher adult literacy at baseline. Our point estimates suggest that a minimum of
45% adult literacy is needed for the WSD to begin showing effects on learning outcomes.
We found no interaction effect for the Grant-only intervention. In summary, we find little
to no evidence that a comprehensive intervention such as the WSD can improve learning
outcomes, except when baseline capacity is sufficiently high.
This paper adds to the literature on interventions to increase community involvement in
schools. The findings are consistent with Banerjee et al. (2010) who compare three
interventions that aim to increase community involvement in the Indian context, where the
central government is expanding the number of schools that are organized locally. They
3
found these interventions to have no effect on beneficiaries' participation or on learning
outcomes.
In contrast, a recent study in Kenya compared different interventions involving additional
resources, teacher incentives, and some level of institutional changes (Duflo et. al. 2014).
They found that training the community to specifically monitor teachers, combined with
reduced class size and teacher incentives, yielded significant gains in various education
outcomes. They also found that hiring additional teachers reduced the effort of existing
teachers. However, where communities were involved in monitoring, the negative impact
on teachers' effort dropped significantly, leading to improvement in learning outcomes.
Our findings also contrast with those of Bjorkman and Svensson (2009) who evaluated
another intervention to enhance community engagement in the health sector in Uganda.
They provided report cards (on health care providers) to members of treatment
communities and encouraged them to define monitoring strategies. One year into the
program, they found large effects on health outcomes. Why do some of these – apparently
similar – interventions seem to work whereas others – such as the WSD – did not? Beside
the specificities of the contexts and the interventions, there is at least one fundamental
difference between these two sets of interventions: the extent to which the intervention
is simple and focused on one or a few specific areas. Whereas the WSD is a comprehensive
(and relatively complex) program, these two interventions, and many similar interventions
that worked, are focused on one main dimension: monitoring.
There are other potential reasons why the WSD did not work to improve learning outcomes
on average. First, in low-income countries such as The Gambia, other inputs that enter the
educational production function such as teacher quality and content knowledge might be
low and thus constitute binding constraints that prevent other policies from functioning
well. For example, in the course of this evaluation, Gambian teachers agreed to take a sixth-
grade level content knowledge test and revealed overall poor outcomes. In addition, due
to resource constraints, a large number of schools function in double shifts and the total
instructional time is less than 80% of what is recommended.
Second, in low-income countries, the problem of local capture has often been pointed out
in the literature as one of the main drawbacks of decentralization (Bardhan and
Mookherjee 2002; Gugerty and Kremer 2008; Reinikka and Svensson 2004). However, we
find no evidence of this issue in the context of The Gambia when we analyze the school
finances and the disbursement process. The WSD program put in place a mechanism to
prevent the misuse and misappropriation of school funds. All expenses were required to be
approved by the SMC and the regional directorate. Schools were required to subsequently
submit the receipts to the regional directorate. In addition, there were officials at the
regional directorate, called “cluster monitors,” whose role was to monitor activities at the
school level and report back to the director. There is no evidence suggesting that political
economy forces, such as local capture, were at play.
4
Finally, even in an environment where local capture is limited or controlled, local capacity
to make informed decisions and effectively implement them is crucial to the success of
decentralization policies. In high-income countries such as the United States, conventional
wisdom suggests that institutional arrangements that favor and foster accountability,
competition, and autonomy are the most effective in improving schools (Hanushek and
Woessman, 2007 & 2009). Differences between the high and low-income countries, and
even between India and The Gambia, render extrapolation from existing evidence to poor
country settings difficult. The interaction effects reported earlier suggest that baseline local
capacity may constrain the benefits from local empowerment. We conclude that a
combination of low baseline local capability, the complexity of the intervention, and the
low quality of other educational inputs are the main factors explaining the limited impact
of the intervention. School-based management models will need to be appropriately
adapted to the needs of local communities.
2. The context
This section combines administrative data with our baseline data to describe the education
system in The Gambia. Basic education in The Gambia lasts nine years. The first six years
are called Lower Basic and the following three years are Upper Basic. Upon completion of
basic education, students take a national exam (9th grade exam) that determines admission
to high school. High school lasts an additional three years.
The education sector in The Gambia has been growing rapidly in recent years. The total
number of students enrolled in the formal education system doubled between 1998 and
2010. Nearly every community has its own lower basic school or has one within a five-
kilometer radius. The basic infrastructure (classrooms, tables, chairs, water) is in general
sufficient even in rural areas. However, due to the increased enrollment, many schools have
adopted a double shift system where one group of students comes in the morning and the
other group in the afternoon.
In terms of organization, there is a Ministry of Basic and Secondary Education (MoBSE) in
charge of the education system up to 12th grade. The country is organized in six
administrative regions: five regions outside the capital plus the district of Banjul (the capital
city). Each of the regions has a regional educational office with a regional director. The
regional directors are the key liaisons between the schools in their region and the ministry.
They ensure the monitoring of activities at the school level and collect key indicators on a
regular basis.
The baseline data from this research (gathered in 2008) include specific information about
Gambian schools (Adebimpe, Blimpo, and Evans, 2009). Those data demonstrate that
overall the basic infrastructure of schools was in good condition.4 The main buildings
4
These assessments are based on visual observation by the enumerators. We limited self-reported
information whenever possible. For example, when inquiring about management practices such as
5
(classrooms and staff headquarters) were overall in good condition throughout the
country. Of the 273 schools visited, 9% required some minor repairs for the walls, roofs,
floors etc. One percent of the schools was in very bad condition and needed total
rehabilitation; these schools were all located in one region. In another region, 15% of the
schools had buildings that needed minor repairs. In 97% of the 526 classrooms visited, most
of the students were seated on a chair with a table. The teaching areas were equipped with
a chair and a table in 92% of the classrooms visited. The student-teacher ratios were similar
across regions at about 40 students per teacher.
At the baseline survey, we looked at recordkeeping as one proxy for management. When
the head teacher was the respondent, 69% reported keeping financial records and were
able to show them. In the absence of the head-teacher, we interviewed the deputy head
teachers. In those cases (i.e., when the head teacher was absent), only 30% of them
reported that the school kept records of finances and were able to show them. Forty-one
percent of schools conducted classroom observation to ensure the quality of the teaching
and were able to show records that confirmed it. All the schools reported the existence of
some form of Parent-Teacher Association (PTA); however, 65% of PTAs have no funding.
Head teachers were asked to report the most important challenge that the school faced in
its effort to provide proper education to the student. The most frequent responses were
the lack of resources (34%) and the lack of proper teacher training (14%).
Absenteeism is high for both students and teachers but is comparable to other low-income
countries. Within the surveyed schools, teacher absenteeism ranged from about 12% of
teachers absent on the day of the survey in two regions to about 30% in another region. In
addition, during the classroom visits, 32% of the teachers reported having missed at least
one day of class during the previous week. Forty-eight percent of teachers had a written
lesson plan. Student absenteeism is measured as the percentage of the class that was
absent on the day of the survey in two randomly selected classes in each school:
specifically, a randomly selected classroom of classes 4 and 6 where possible; where not
possible, a randomly selected other class. In the 526 classroom visits, student absenteeism
ranged from about 20% of the total number of students enrolled in some regions to nearly
40% in another.
Learning assessments have revealed poor learning outcomes: For example, the 2007 Early
Grade Reading Assessment found that almost 50% of third graders could not correctly read
a single word (USAID et al. 2008). Hence there is strong demand to improve learning
outcomes. Within this study, in terms of both literacy and numeracy, student performance
is lower than expected (per the curriculum) in Grade 3 but improves substantially by Grade
5, indicating that – at least – students are learning in school. There was considerable
good recordkeeping, in addition to yes or no answers, enumerators recorded a third option that
consisted of visually confirming the existence of the relevant records.
6
heterogeneity in student performance within each grade, particularly in math skills. In
almost all tests, girls under-performed boys by about 3 percentage points.
On average, third graders are 10 years old and the fifth graders are 12 years old. Half of the
students live in homes with improved latrines. Only 20% of the students reported having
electricity. Ninety percent of students had a radio at home, 83% of households owned a
telephone,5 and 69% owned a bicycle.
3. Experimental design
3.1. The intervention arms
The main intervention evaluated in this paper is a holistic school management capacity
building program called Whole School Development (WSD).6 This intervention consists of
the distribution of management manuals, a comprehensive training component, and a
grant to help implement the activities in the first year. In order to be able to separate the
impact of the capacity building component from the grant, a second intervention group
received the grant but did not receive training. We compare these two interventions to a
control group that received neither the grant nor the training. Table 1 provides a summary
of the key elements of the intervention arms, and Table 2 summarizes the project timeline.
3.1.1. The Management manual
The school management manual (SMM) is a comprehensive guide to management
practices both within the school and for interactions with other stakeholders at the
community, regional, and national levels. International experts developed the manual
together with national officials and stakeholders at the local level, including teachers. The
manual addresses six specific topics pertaining to the management and functioning of
schools: school leadership and management, community participation, curriculum
management, teacher professional development, teaching and learning resources (e.g.,
textbooks and libraries), and the school environment. All these aspects are integrated in a
three-step cycle for effective school management. The first step is information gathering
and analysis. This step provides information as to what kind of data and information should
be collected by schools on a regular basis (e.g., monitoring learning outcomes and
absenteeism). It emphasizes how to analyze the data and then to create a plan for short-
term and long-term solutions to school problems. The second step is the implementation
of the resulting plan. The third step involves effective monitoring of the plan that is being
implemented and adjustments along the way. The SMM advocates for strong, broad
5
Either the household had a landline or a person in the household possessed a mobile phone.
6 The WSD intervention has previously been implemented in South Africa (Bayona & Sadiki 1999).
7
inclusiveness in school decision making. The manual was provided to all schools
participating in the study.
3.1.2. The Management Training
The management training and capacity building are the centerpiece of the WSD
intervention. The principals, teachers, and representatives of parents and students
received training in six areas of school management, also described in the school
management manual. The six areas were (1) community participation, (2) learner’s welfare
and school environment, (3) curriculum management, (4) teaching and learning resources,
(5) teachers’ professional development, and (6) leadership and management. In the course
of this training, participants developed a local school development plan addressing various
areas with guidance from the trainers and the supervision of the WSD unit within the
Ministry of Education.
The training used a cascade method. First, the experts who developed the SMM trained
twenty people (“master trainers”) at the national level. Second, the master trainers
conducted regional trainings of “cluster monitors” (school inspectors over a cluster of
schools), school directors, and some senior teachers. Then those regional trainees carried
out a local training with the school management committee, senior teachers, and – in some
cases – a student representative. The training lasted between 10 and 20 days, with sessions
split across several periods. The initial sessions of the local trainings were supervised by
experts who had developed the SMM. Since most parents do not speak, read, or write
English, the training put emphasis on local languages and drawings (See Figure 1) to convey
the messages more effectively.
3.1.3. The Grant
Some of the activities suggested in the manual and included in the school development
plans, like workshops, might require financial resources. Over time, the funding for these
activities was expected to come from the school budget and locally raised funds. However,
during the first year, the intervention schools were provided with a grant to serve as a
catalyst for school improvement. A grant of US$500 was given to all the schools in the WSD
and the Grant-only groups after a school development plan was presented. The schools
were required to spend the funds on activities pertaining broadly to learning and teaching.
The schools informed the regional office about their spending plans and submitted the
receipts. This grant represents about 16 months worth of salary for a first grade teacher
without experience or about 14.5 months worth of salary of a first grade teacher with five
years of experience. It represents less than 5% of the average annual school budget.
3.2. Sampling
The sample in this study is the census of lower basic public and government-aided schools
in regions 2, 3, 4, and 6 (276 schools) in The Gambia (Figure 2). The two regions that were
excluded from the study were Region 1, which is essentially only the capital city and was
8
excluded on the basis that it was too urban and distinct from the rest of the country, and
Region 5, because it was used extensively to pilot the WSD prior to the large randomized
experiment. Of the 276 schools, one school was excluded from the sample because it was
very small and had only a few students in grades 1 and 2. Another school was closed but
still appeared on the official list of schools. Figure 3 summarizes the sampling procedure.
Of the 273 remaining schools, 90 schools were randomly assigned to the WSD treatment,
94 schools to the Grant-only treatment, and 89 schools served as the control group. The
schools were clustered in groups of 2 or 3 schools on the basis of geographic proximity to
limit contamination while allowing useful exchange and cooperation between nearby
schools.7 Because this represents the universe of schools meeting the inclusion criteria,
rather than a sample, clustering of groups of schools is unnecessary in the subsequent
analysis.8 The randomization was further stratified by school size and accessibility.9 Each
group proved to be similar at baseline, as discussed in detail in Section 5.1. As all schools
remained in the study between baseline and endline, there is zero attrition.
4. Data
The Gambia Bureau of Statistics, under the supervision of the research team, collected the
data for this study. The baseline data were collected in 2008 at the onset of the study, the
first round of follow-up data were collected in 2009, the second round of follow-up data
were collected in 2010, and the end-line data were collected in 2011 (Table 2).
In the 2009 follow-up, data were collected in the WSD and Control schools only. The Grant-
only schools were not visited at that time because grant disbursement was delayed in one
region during the first year, and many schools that had received their grant had not yet
used it.10 This problem of slow disbursement of education grants by local committees was
also observed in Kenya (Conn et al. 2008).
At each round, teams of enumerators arrived unannounced (in order to avoid strategic
attendance by teachers and students) at each school and collected information about the
school and the students, conducted classroom observation, and gave a literacy and
7
At the regional level, schools that are close to one another are assigned a “cluster monitor” who
serves as a liaison between the regional directorate and those schools. The cluster monitor is
encouraged to promote good practices among the schools she is assigned to.
8 Furthermore, the intracluster correlation for test scores and absenteeism is much higher (55-
80% higher) at the school level than at the level of school clusters.
9
The Ministry defines accessibility through “hardship status”. Schools that are most remote receive
an allowance from the Government, as discussed in Pugatch & Schroeder (2014).
10 This information was obtained from the regional directorates who were the key intermediaries
for the grant disbursement process.
9
numeracy test.11 Unless otherwise indicated, the following data were collected at each of
the four rounds of data collection. Table 3 summarizes the data collected in each wave.
4.1. School data
The data on the school as a whole were obtained through enumerator observation and a
comprehensive interview with the head teacher or – in the absence of the head teacher –
the teacher in charge of the school at the time. The directly observed information includes
the condition of the buildings, the number of classrooms and other facilities, etc.
Information from the head teacher included school finances, record keeping, community
participation, management practices, etc. To improve the accuracy of the information
collected, we requested to see written records to substantiate responses whenever
applicable.
4.2. Classroom visits
In each school, we randomly selected two classrooms for observation. The goal of the
classroom visit was to gather information about teaching practices, the classroom
environment, and student participation. It also served to substantiate the absenteeism data
from the administrative records by comparing the student register to the number of
students present in the classroom. Each classroom visit lasted fifteen minutes, followed by
a five-minute interview with the teacher.
4.3. Student written literacy and numeracy test
Forty students were selected randomly at each school and were given a written numeracy
and literacy test. At the baseline, we tested twenty third-grade students and twenty fifth-
grade students at each school. Third- and fifth-graders were selected as these are the
earliest grades regularly evaluated by the Gambian government. At the first follow up in
2009, we gave the test to students in fourth and sixth grades to allow for tracking of the
baseline students. At the second follow-up in 2010, the test was given again to third and
fifth grade students because much of the original cohort would have completed primary
school. In total, 8,959 students were tested at baseline, roughly evenly distributed across
the three treatment groups.
4.4. Student interview and oral literacy test
Of the forty students who took the written test, ten were randomly selected to take an
orally administered reading and comprehension test and to participate in an interview
about their socio-demographic characteristics, school performance, and other information.
These students were tracked in 2009 in the WSD and Control schools, and in 2010 in all the
11
The schools were given a range of time during which a team of enumerators would visit them.
The actual dates were not disclosed.
10
schools whenever possible.12 Students for the pupil interview were selected randomly from
among those who participated in the written test. At baseline, 2,696 students were
interviewed in total: 879 from WSD schools, 920 from Grant-only, and 897 from the control
schools.
4.5. Teacher content knowledge
In 2009, we tested teacher knowledge of content: The test was similar to the students'
written test, with additional questions drawn from Gambian secondary school reading and
math textbooks. A short background interview was also administered to the teachers who
took the test.
4.6. Qualitative data
In 2010, we added many open-ended questions to the head teacher interviews to collect
some information about their views regarding school management. We addressed similar
questions to parents or caregivers in a few households whose children were in the relevant
schools. The research team was also heavily involved on the ground for the entire first year
of this program; the associated conversations with the government, the schools, and the
communities add important information that is useful for a better understanding of the
findings.
5. Identification, empirical strategy, intermediate outcomes
5.1. Identification and group comparison
In the design of a field experiment, the goal of employing random assignment to allocate
participation in the program is to achieve a situation in which each of the groups has similar
characteristics – both observed and unobserved – before the implementation of the
program. If the treatment and control groups are balanced at baseline, then differences in
teaching activities and student learning outcomes between the groups in the follow up
survey can be attributed to the WSD and Grant-only programs, rather than to some pre-
existing difference between the groups. Using the data from the baseline survey, we
examine observed characteristics across the different groups.
We first compare the outcome variable at baseline across groups. Figure 4 shows the
distribution of test scores of fifth-grade students on a written test in English, Math, and a
combined score. It shows that the baseline performance level of student, across groups,
comes from the same distribution. The t-test of comparison of means cannot reject the
hypothesis that the underlying distribution of students’ performance at the baseline has
the same mean. Similarly, the Kolmogorov-Smirnov test of comparison of distribution does
not reject the hypothesis that the distributions of students’ performance are identical
12
Most of the students in 5th grade at baseline had finished the basic cycle by the time of the
second follow up.
11
across the three groups. We reach the same conclusion on the student reading outcomes.
Fifth grade students were presented with a sixty-word text to read in one minute. Figure 5
shows the similarity of the distribution of reading outcomes across the groups. In addition
to the students’ baseline performance, we compare school and student characteristics
across groups.
A list of indicators and their means across groups are included in Table 4 (school
characteristics) and Table 5 (student characteristics). We observe no systematic differences
across the groups. For example, the average size of the schools is comparable across groups
and the average student-teacher ratio is nearly identical: There were 32 students per
teacher in the WSD and Control schools versus 34 in the Grant-only schools. Out of 17
characteristics at the school level, the only significant difference is that the WSD schools on
average reported 4.4 Parent-Teacher Association (PTA) meetings during the year prior to
the survey versus 3.7 for both the Grant-only and the Control group. WSD communities
held sensitization meetings about the program in advance of program implementation and
the survey, which is the most likely explanation for this difference. There is no significant
difference for cash and in-kind contributions across groups, which might be expected if the
difference in meetings were an indication of greater baseline involvement more generally
in WSD communities.
In terms of student characteristics, the groups are comparable as well. Third-grade students
are a little over 10 years old and fifth-graders are about 12.5 years old in all three groups.
The socioeconomic backgrounds of students, in terms of access to electricity at home,
possession of a television, and access to a telephone are also comparable across groups.
The percentage of students currently repeating a grade is identical (9%) in all three groups.
We conclude that there are no apparent systematic differences across the treatment
groups at the baseline. The random assignment to the different interventions groups
means that there are also no expected systematic differences among the three groups in
unobserved characteristics.
5.2. Main Empirical Strategy
Because of the random assignment of schools to the treatment groups, the following basic
regression model provides the estimates of the causal effect of the interventions:
������������������������������������������������������ = ������ + ������1 ������������������������ + ������2 ������������������������������������ + ������������������ (1)
where Outcomeis is the outcome of student i in school s, WSDs = 1 if school s received the
WSD intervention and 0 otherwise, and GRANTs = 1 if school s received the grant-only
intervention and 0 otherwise. The error term ������������������ is clustered at the school level to account
for intra-school correlation of outcomes. The parameters of interest are ������1 , which is the
average effect of the WSD intervention on the outcome, and ������2, which is the average effect
of the Grant-only intervention. A simple test of the null hypothesis – ������0 : ������1 = ������2 –
compares the WSD intervention to the Grant-only intervention.
12
5.3. Intermediate results
5.3.1. One year post-interventions
One year after the implementation of the WSD, we collected data in all the WSD and control
schools. The goal of this round of data collection was to ensure that the WSD was properly
implemented, to monitor the evolution of the process, and to collect some intermediate
variables to assess the early impact. The key results described in this section are reported
in Tables 6, 7, 8.
Most of the significant results at the school administration level are focused around take-
up of the WSD program in the WSD schools. We assessed take-up by looking at basic
elements that indicate whether the WSD program is functioning or not.13 There is a higher
rate of establishment of various school management committees (SMC) in WSD schools, as
recommended by the School Management Manual (Table 6). For example, 84% of the WSD
schools had set up a curriculum management committee whereas only 51% of the control
schools did so. (The committees in the control group are often different in nature and
reflect the school organization in place prior to this research.) Similarly, for each of the
other SMCs, we observed statistically significant differences in favor of the WSD. Only about
one-third of the schools in each group had adopted and actually implemented the new PTA
constitution, with a 3-percentage point edge in the WSD schools.
In terms of intermediate outcomes, the control schools appeared to perform better in
teacher preparedness one year into the program (Table 7). We observed teachers’ written
lesson notes for the day of the visit in more control classrooms (41%) than in the WSD
classrooms (32%). We also observed 11% more lesson plans in the control classrooms than
the WSD classrooms. Both of these results are significant.14 It could be that new committee
work associated with the set-up of the WSD program actually took teachers away from
classroom preparation. (This is consistent with the fact that significant differences on
teacher preparation disappear in subsequent observations, longer after the program was
established.)
Absenteeism remained pervasive (Table 7). About 25% of the students were missing, when
we compared the number of students present to the number of students listed on the
register. We also picked five days randomly from the register and found an average of
nearly 38% recorded absenteeism over those 5 days, nearly identical in both groups. More
teachers in the control group (7% more) reported having missed at least one day of class in
the previous week. Teacher absenteeism remained the same as at the baseline in the
control group (32% of teachers reported having missed a day during the previous week)
13
The control schools were given the basic manual of the WSD, but that they did not receive the
training and the grant.
14
In this context, the “lesson plan” is the weekly or monthly outline of topics to be taught,
whereas the “lesson note” is the document outlining the specific activities for a given day.
13
whereas it dropped by 6 percentage points in the WSD group, according to teacher reports.
However, the average percent of teachers absent over 5 random days, based on school
records, indicates relatively low absenteeism (6%) and no difference between across
groups (Table 7).
We found no difference between the two groups in terms of student performance (Table
8). Fourth graders read about 24 words per minute and sixth graders read 41 words.
Research suggests that about 45 to 60 words per minute are required for comprehension
(Abadzi 2008).
These findings show – unsurprisingly – a higher rate of adoption of the school organization
recommended by WSD in the WSD schools and its components within the WSD group
compared to the control group. No differences were observed regarding student
performance, although it would likely be too early to observe such an effect at that point.
At the very least, this indicates that the program was implemented as planned.
5.3.2. Two years post-interventions
In this section, we present the impact of the intervention on student learning outcomes,
teaching practices at the school level, and school management two years into the
interventions in all three groups.
The estimates of the average treatment effect (Table 9) indicate that neither the WSD nor
the Grant-only interventions had any impact on student learning outcomes two years after
their implementation. Student performance in all groups remains relatively poor and
comparable to baseline levels. This is also true for the control group, which rules out the
possibility that the control group may have improved along with the treatment groups over
the two years but due to reasons other than the intervention.
Even though we observe no average treatment effect, it is possible that the distribution of
performance may have been impacted in a way that would balance out the average effect
(e.g., improved performance at the bottom of the distribution and worse performance at
the bottom of the distribution). However, the distribution of test scores across groups
shows no significant heterogeneity by level of performance except for a small range around
the average performance (Figure 6).
Teaching practices improved slightly in the WSD group. As Table 10 shows, the probability
that the teacher frequently used the blackboard increased by 7% relative to the control
group and teachers were more likely (10%) to call on student by their names (both results
significant with 90% confidence). However, we see no evidence that the program affected
the confidence of children to participate and ask questions during class. Similarly, the
programs did not improve the likelihood that a teacher would prepare for the class with
written notes.
14
The first four columns in Table 11 indicate that the intervention groups are more likely than
the control group to consult teachers, parents, and the regional office for planning and
decisions about school expenses. The point estimates in column 4 indicate that the WSD
group relies less on the regional education authorities than the Grant-only group,
potentially due to the training component of the WSD. Moreover, the WSD group is more
likely to conduct fundraisers relative to the control group, whereas this is not the case for
the Grant-only group. The WSD treatment has a negative effect on the number of overall
PTA meetings: On average, PTAs met 0.41 less in the WSD group than in the Control group
(column 7, Table 11). The likely explanation for this finding may be the fact that the WSD
creates six sub-committees (as observed in the one-year follow-up data) within the
community to deal with different challenges pertaining to the functioning of the school.
Parents may participate in sub-committee meetings and so the school may hold fewer
overall PTA meetings. Although some of the changes observed may be expected to impact
student learning, we observe no impact on student performance (Table 12).
6. Final results
6.1. Average Treatment Effects on learning outcomes and participation
The main outcome variables of interest are the learning outcomes measured by a
comprehensive written test. Other outcomes of interest beside student test scores include
measures of absenteeism for teachers and for students, and a measure of enrollment. Table
13 presents the estimates of Equation 1 where the dependent variable is a standardized
test score in math or English. The estimates show that the interventions have no positive
effect on student math and English test scores. The point estimates are mostly negative but
small and statistically insignificant. A test comparing the mean score between the WSD and
the Grant-only does not reject the null hypothesis that the two interventions have the same
effect on test scores.15 Table 14 presents the same results, controlling for baseline test
scores: The significant patterns are identical and all coefficients of interest are within 0.02
of each other.
We run the same model where the outcome variables are student absenteeism and teacher
absenteeism. The estimates in the first column of Table 15 indicate that the WSD
intervention reduced student absenteeism by about 5 percentage points from a base of
about 23% (significant at the 5% level). This corresponds to a nearly 21% reduction in
absenteeism. The second column shows that the WSD reduced teacher absenteeism by
15 Across both treatment groups, school identified the largest budget item on which the grant was
spent: 46% reported teaching and learning materials (including stationery), 23% reported
infrastructure (e.g., furniture, building improvements), 20% reported some kind of workshop, 7%
reported a radio, while a few reported spending the grant on garden materials.
15
about 3 percentage points from a base of about 13%, which represents a 23% reduction in
teacher absenteeism. We observe no impact of student enrollment.
6.2. Discussion, Interpretation, and potential mechanisms
The Whole School Development program, over time, had a positive impact on student and
teacher school attendance. In theory, increased participation should translate into
increased learning outcomes. However, in this case we observe increased participation but
no change in test scores. We explore four potential explanations for this finding: (1)
Selection, (2) Poor teacher quality, (3) Human capital in the community, and (4)
Improvements in the control schools.
6.2.1. Selection as treatment effect
One plausible explanation could be that the increased student participation brought back
students that perform poorer than the average. If the intervention has brought in worse-
performing students in the intervention group, then the average treatment effect (ATE)
may be biased downward. The distribution of test scores shown in Figure 7 shows a shift to
the left, albeit only at the left tail. This is suggestive evidence for the hypothesis that the
WSD program attracted more low-performing students into schools. Miguel and Kremer’s
(2004) analysis of a de-worming intervention in Kenya found large effects on participation
but no effect on test scores: this same kind of selection was a potential explanation in that
context as well.
If students who attended more because of the WSD were also students who otherwise
perform more poorly, then one might expect the treatment effect to be larger at higher
percentiles of the performance distribution. To verify this, we first look at the treatment
effect in the each quantile. Figure 8 shows an upward trend, which partially supports this
story. However, for this effect to be interpreted as the effect of the intervention on the
students on the respective quintiles, the rank preserving assumption between the baseline
and the end-line needs to be true. (In other words, one must assume that students would
occupy the same rank in the test score distribution independent of the intervention: If
Student A had better scores than Student B before the intervention, she would continue to
outperform Student B after the intervention, even though one or both of them may have
improved.) This is a strong assumption and there is no way to test it.
Nevertheless, we address this selection issue by bounding the treatment effect using Lee's
trimming procedure (Lee 2009). The procedure consists of dropping a proportion of the
lower tail of the distribution in the WSD group – i.e., those low-performing students drawn
to schools by the intervention – in order to construct an upper bound of the effect of the
intervention. Then we drop a proportion of the upper tail in order to construct a lower
bound. Lee shows that the proportion to trim is given by
% ������������������������������������������������������������ − % ������������������������������������������������������������������������������������
������ =
% ������������������������������������������������������������
16
Let ������������ be the test score of student i and ������������ = ������ −1 (������) with G being the cumulative
distribution function of y conditional on being in the WSD group and being successfully
tracked. Then, the sharpest bounds of the treatment effect are given by calculating the
sample counterpart of the following:
������������������������������������ ������������������������������ = ������[������|������������������, ������������������������������������������, ������ ≥ ������������ ] − ������[������|������������������������������������������, ������������������������������������������]
and
������������������������������������ ������������������������������ = ������[������|������������������, ������������������������������������������, ������ ≤ ������1−������ ] − ������[������|������������������������������������������, ������������������������������������������]
Under the assumption of independence and monotonicity, these bounds are shown to be
the smallest upper bound and the largest lower bounds that are consistent with the data
at hand. The bounds can be calculated only using the subset of students that we tracked by
design from the baseline to the end line. These students were five third-graders per school
in 2008 who were in sixth grade at the end. At the end, we were able to find 71% of them
in the control schools versus 79% in the WSD schools. The average test scores are
comparable between the two groups, but if the extra students tracked in the WSD are
weaker on average, then this comparison will be biased in favor of not finding an effect
(Table 16).
Table 17 presents the estimates of Lee's sharp bounds, accounting for selection. The results
indicate an upper bound of 0.17 and a lower bound of -0.19 standard deviations on the
mathematics test score. The effect on English is bounded by 0.26 and -0.16 standard
deviations. These ranges are not a confidence interval for the average treatment effect, but
a range of point estimates that are all consistent with the data given the selection concern.
Given these bounds (which clearly include a zero effect), and given the underlying
assumption on the absentees, it is reasonable to lean toward an interpretation of no
significant effect. These findings suggest that the selection issue may not be pronounced.16
6.2.2. Poor complementary inputs: Teacher quality
A third explanation for a lack of learning effects in the face of attendance improvements is
that other inputs such as teacher quality are sufficiently low that increased participation
does not translate into improved learning outcomes. In 2009, we conducted a teacher
content knowledge test. The test consisted of the same test applied to students, with a few
additional questions from Gambian secondary school textbooks. Figure 9 and Figure 10
show sample questions from the test and average teacher performance on them. The
findings suggest that teacher content knowledge was indeed low: Only 2.6% of teachers
16Note that these bounds do not account for the potential peer effect from absentees that are coming
back, i.e., if poorer performing students were returning and not only bringing down the average test scores
but negatively affecting the performance of student who were previously attending. To account for this
particular aspect, one would need a structural model, which is beyond the scope of this paper.
17
scored 95% or more, and over one-third of the teachers scored below 75%. There were no
significant differences across treatment groups.
Figure 11 shows a positive correlation between matched teacher and pupil test scores.17
Sixth-grade math test scores mainly drive the correlation. In addition, the result from
classroom observation indicates that only about 45% of the instructional time is actually
focused on learning activities (Table 18), to be contrasted with estimates between 52% and
65% in a sample of Latin American countries (Bruns & Luque 2014). Taken together, these
results suggest that teacher quality and effectiveness may be so low in The Gambia that
other school improvement interventions will be ineffective.
6.2.3. Community human capital at baseline: Heterogeneity
The Gambia is characterized by a low adult literacy rate, especially in rural areas. This
characteristic was reflected in the School Management Committees. Nearly 4 out of 5
committee members from the community (i.e., not school employees) had no formal
education and only 16% had completed at least primary education. Some level of human
capital may be needed at the local level for interventions such as the WSD to build on. For
example, for parents to effectively help to run the school, the parents would need some
schooling of their own. We investigated this hypothesis by interacting the interventions
with a baseline measure of human capital.
������������������������������������������������������������ = ������ + ������1 ������������������������������ + ������2 ������������������������������������������ + ������3 ������������������������������������������������ ������������������������������ ������������������������������������������������ +
������4 ������������������������������������������ ∗ ������������������������������������������������ ������������������������������ ������������������������������������������������ + ������5 ������������������������������ ∗ ������������������������������������������������ ������������������������������ ������������������������������������������������ +
������������������������ (2)
We report estimates of equation 2 in Table 19, where BaselineHCd is the district level adult
literacy in 2006. Across the districts included in the evaluation, the average adult literacy
was 31%, ranging from 12% to 53% across the localities where the schools are located. The
interaction between WSD and adult literacy in 2006 has a significant and positive effect on
both math and English test scores. This suggests that human capital, at least measured as
adult literacy, has an amplifying effect on the WSD. The same is not found for the Grant-
only intervention.
The estimates also suggest that interventions such as WSD could potentially have
detrimental effects in places where human capital is very low. One channel for this negative
effect could be that shifting from one set of management practices to another is costly. If
existing practices are functioning at some level and new practices (which are expected to
be better) are not properly adopted, the end outcome could be negative. Furthermore,
WSD shifts some degree of decision making from school leaders to the community: If the
community has very little capacity, then the result on school management quality could be
17 This could of course in part be driven by selection, if higher-performing students are placed in the classes
of higher-performing teachers.
18
negative. This is also consistent with the multitasking literature (e.g., Holmstrom and
Milgrom 1991), which – in this case – suggests that when asked to perform many tasks
simultaneously (as in an integrated program such as WSD), schools would prioritize some
tasks over others. However, if the different tasks are complements, then improvements in
just a few may not yield a positive overall outcome. Table 20 presents the same estimates
where BaselineHCd is replaced by the percentage of the school management committee
members who have no formal education. The results are qualitatively the same.
We graphically present the results of this analysis in Figures 12 and 13. We conclude that
the WSD intervention is likely to improve learning outcomes in areas with high baseline
human capital, but it could be counterproductive in areas where the basic human capital is
very low. Our point estimates suggest that the WSD would have a positive impact on
learning outcomes if the level of adult literacy at the baseline were greater than 45%.
To further understand this human capital aspect, we also conducted qualitative analysis.
After two years of exposure to the WSD program, we asked the head teachers about their
opinion regarding shifting school management to the schools and the communities. Most
of the head teachers (75%) disapproved of this idea, 19% thought that it would be a good
idea and, 6% expressed no opinion either way. Most of the head teachers who approved
the idea supported their position with the argument that the communities and the schools
better know their problems and that it would be more effective to allow them to handle
them. Others pointed out that it would induce more accountability as the teacher can be
monitored more effectively and action can be taken in a timely fashion if they do not
deliver.
However, most head teachers disagree with that point of view in the context of Gambia.
Almost all of those who opposed the idea pointed out the lack of capacity at the local level
to manage the school. As one head teacher expressed, such decentralized decision making
would be “almost impossible because a large portion of our communities are illiterate.”
Even though standards are low, pupils are performing poorly, and teacher content
knowledge is problematic, over 90% of parents are satisfied with the school and think that
the school is doing fine in training their children. When asked to give the reason why they
make such assessments, 83% of the parents say that the child is performing well and that
the school has good teachers. Another 15% based their assessment on the fact that the
child is better behaved and disciplined at home.
Similarly, over 90% of the parents report high aspirations for their children. They reported
wanting them to study to the highest level and enter careers with high social esteem such
as doctors, ministers, etc. These responses indicate that these parents care about and value
the educational outcomes of their children, but there is a contrast between this aspiration
and their ability to assess the effectiveness of the school and hold the teachers accountable.
This large disconnect between actual student academic performance (and, consequently,
school performance) and the parents' assessment is in tune with the theoretical motivation
19
of this paper. Among the few parents who are dissatisfied by student and school
performance, most pointed out specifics about the incapacity of the child to read and write
properly, and the mismanagement of the school. It may be that those parents are more
educated and better able to assess the progress of the children and the performance of the
school. These findings confirm that the WSD intervention may be more appropriate where
local capacity is sufficiently high.
In addition to increasing the capacity of communities to hold schools accountable, district-
level literacy could be a proxy for some other effect. One possibility is that the literacy
effect, demonstrated in Table 19 and Table 20, is actually a proxy for some other, correlated
characteristic of households or districts. We test this by interacting the WSD treatment with
socio-economic status at the household and the district level in Table 21 and Table 22;
those interactions are not statistically significant. Another possibility is that adult literacy is
a proxy for some baseline level of ability among the students. We test this by interacting
WSD with baseline test scores (Table 23) and again find no significant relationship. A third
possibility is that, rather than having additional management capacity, more literate
districts are wealthier and are contributing more to schools. In Table 24 we present
evidence that higher economic status does not associate with higher likelihood of financial
or in-kind contributions to the school. Taken together, these tests further suggest that the
human capital effect is in effect a proxy for the capacity of parents within the district.
6.2.4. Improvement in the control schools?
The lack of impact on average test scores could also be due to the fact that control schools
have improved as well, through mechanisms other than increased participation. Since the
school management manuals were made available to all the schools, it is possible that the
control group would implement at least part of the practices, although it seems unlikely
that they would have adopted a similar set of practices to the WSD schools without any
support. Qualitatively, we found no evidence that they used the manual. In addition, our
test score data from 2008 and 2010 were collected at the same grade level. This allows us
to conduct a before and after analysis in each group, including the control group (25). We
find no evidence of a positive time trend in the control group between the baseline and the
2010 test scores.
8. Conclusion and future research
In this research, we evaluated a school management training program in The Gambia called
Whole School Development (WSD). Intermediate results one year post-intervention
showed some basic changes in school organization in the WSD schools but no effect on test
scores or on student and teacher absenteeism. These results served mostly as evidence of
project implementation. Two years post-intervention, we found no effect on test scores
but modest positive effects on student and teacher participation measured by the
prevalence of absenteeism.
20
Three years into the program, we found no effect of the WSD intervention on learning
outcomes measured by scores on a comprehensive test of Mathematics and English.
However, we found a large effect on participation: The intervention led to reductions in
student and teacher absenteeism respectively of nearly 5 percentage points from a base of
24%, and about 3 percentage points from a base of about 13%. We found no effect of the
Grant-only intervention relative to the control on test scores or on participation.
Since this intervention emphasized local capacity building, we analyzed the heterogeneity
of the effectiveness of the program by one dimension of initial capacity, adult literacy. Our
findings suggest that the WSD may be effective when adult literacy at baseline is sufficiently
high. The range of the estimated effects suggests that, for places where local capacity is
extremely low, this intervention could potentially be counterproductive as the reform may
shift decision making away from school leaders with relatively higher human capital.
We also observed a large disconnect between the parents’ evaluation and the actual
performance of the schools. Whereas evidence from student tests reveals poor
performance of children, over 90% of the parents are satisfied with the schools and their
children's performance. This disconnect may explain why parents do not hold the schools
accountable and participate effectively in school management. Parents have very high
professional aspirations for their children, but the evidence suggests that they may lack the
ability to evaluate the performance of their children and thus to demand accountability
from educators. That is precisely what the capacity building component of the WSD
attempted to address, but the WSD does not appear to have accomplished this, at least not
sufficiently to change test scores. While the WSD focused on concrete actions by parents
to hold schools accountable, the relevant challenge may be more related to the basic
inability of parents to read and write.
With the grant-only intervention, we found no evidence of positive effects on outcomes,
except on process variables such as community engagement in decision making. However,
there are many reasons why this should be taken with caution. First, principals found the
disbursement process cumbersome because disbursements had to be approved by the
regional directorates. This may have prevented schools from effectively addressing issues
that required immediate attention. Second, and perhaps most importantly, the one-time
grant was relatively small to expect a substantial effect three years later (although note
that no effect was observed at any point). With an increased amount or with more
sustained yearly grants, the results might differ.
Based on this study, we draw the following conclusions and policy implications. First, a
crucial feature for an effective local management program, such as the one envisioned and
studied here, is local human capital (such as literacy) in the communities. We hypothesize
that in general, the gap between capacity at the central and local levels is a key determinant
of the success of such policies. In countries where this gap is small, regardless of absolute
capacity levels, a decentralized policy may be superior because of the added value of
localized information. However, if the gap is sufficiently high in favor of the central
21
government, then localized information is less useful because communities are not well
equipped to act on it. Our findings show that The Gambia may fall in the latter group. An
intervention like this one may not be effectively by itself for the median community. Rather,
interventions to increase community involvement should seek to relax constraints on
community capacity.
Second, in The Gambia, there appear to be other binding constraints on the education
production function. Two of these constraints, explored here, are teacher capacity and
effectiveness; others are limited instructional time due to the widespread double-shift
schools, and teacher compensation. National policy shifts may need to lay the groundwork
for improvements in these areas before school-level improvement plans can be effective.
Third, our findings suggest that a mechanism to supply accurate information to
communities (about the relative performance of their children and the schools) could be
desirable. This, in essence, substitutes for baseline capacity on the part of parents to
evaluate the schools. Our data suggest that most parents – including in the rural areas –
have high aspirations for their children's professional futures and educational
achievements. However, this is juxtaposed with the sharp inability of parents to understand
the performance of their children and the functioning of the schools, even after the
intervention. If well informed, parents may seek to hold schools accountable for their
children's learning outcomes. In recent years, the government has experimented with
providing school report cards that are focused on pictograms (such as smiley faces). Such a
communication intervention that does not require high levels of literacy would be worth
testing.
Our findings call for nuance in the design of policies that decentralize school management
to communities. School-based management is gaining popularity in low-income countries
(Barrera-Osorio et al. 2009; Bruns et. al. 2011). In Africa alone, there are many ongoing field
experiments to test variants of school-based management policies. These studies will shed
much needed light on which models will help communities to keep schools accountable.
22
Works Cited
Abadzi, Helen. Efficient learning for the poor: New insights into literacy acquisition for
children. International Review of Education, 54:5-6, November 2008.
Adekanmbi, Adebimpe, Moussa P. Blimpo, and David K. Evans. The state of lower basic
education in the Gambia: A baseline survey report prepared for the ministry for basic &
secondary education, the Gambia. APEIE, World Bank. Available upon request to authors,
2009.
Banerjee, Abhijit, Rukmini Banerji, Esther Duflo, Rachel Glennerster, and Stuti Khemani.
Pitfalls of participatory programs: evidence from a randomized evaluation in education in
India. American Economic Journal: Economic Policy, 2:1–30, 2010.
Bardhan, Pranab, and Dilip Mookherjee. Relative capture of local and central government:
An essay in the political economy of decentralization. Center for International and
Development Economics Research, Institute of Business and Economic Research, UC
Berkeley, 2002.
Barrera-Osorio, Felipe, Tazeen Fasih, and Harry Anthony Patrinos with Lucrecia Santibez.
Decentralized Decision-Making in Schools: The Theory and Evidence on School- Based
Management. 2009.
Bayona, E.L.M and B. Sadiki. An investigation into appropriate ways of implementing
institutional development (whole school development). University of Venda School of
Education research report, Accessed at http://www.jet.org.za/publications/pei-research
on 31 July 2010, 1999.
Bjorkman, Martina, and Jakob Svensson. Power to the people: Evidence from a
randomized field experiment on community-based monitoring in Uganda. The Quarterly
Journal of Economics, 124:735–769, 2009.
Bloom, Nicholas, Renata Lemos, Raffaella Sadun, and John Van Reenen. Does
Management Matter in Schools? Working paper. 2014.
Bruns, Barbara, Deon Filmer, and Harry Anthony Patrinos. Making Schools Work: New
Evidence on Accountability Reforms. World Bank. 2011.
Bruns, Barbara, and Javier Luque. Great Teachers: How to Raise Student Learning in Latin
America and the Caribbean. World Bank. 2014.
Conn, Katharine, Esther Duflo, Pascaline Dupas, Michael Kremer, and Owen Ozier. Bursary
targeting strategies: Which method(s) most effectively identify the poorest primary
23
school students for secondary school bursaries? Innovations for Poverty Action Kenya,
Unpublished Report, 2008.
Duflo, Esther, Pascaline Dupas, and Michael Kremer. School governance, teacher
incentives, and pupil-teacher ratios: Experimental evidence from Kenyan primary schools.
2014.
Gertler, Paul, Harry Patrinos, and Marta Rubio-Bodina. Empowering parents to improve
education: Evidence from rural Mexico. Journal of Development Economics 99(1), 2012.
Gugerty, Mary Kay and Michael Kremer. Outside funding and dynamics of participation in
community associations. American Journal of Political Science, 52:585–602, 2008.
Hanushek, Eric A. and Ludger Woessmann. The role of education quality for economic
growth. World Bank Policy Research Working Paper No. 4122, 2007.
Hanushek, Eric A. and Ludger Woessmann. Schooling, cognitive skills, and the Latin
American growth puzzle. NBER Working Paper 15066, 2009.
Holmstrom, Bengt, and Paul Milgrom. Multitask Principal-Agent Analyses: Incentive
Contracts, Asset Ownership, and Job Design. Journal of Law, Economics, & Organization,
7:24-52, 1991.
Jimenez, Emmanuel and Yasuyuki Sawada. Do community-managed schools work? an
evaluation of El Salvador’s educo program. The World Bank Economic Review, 13:415–
441, 1999.
Lee, David S. Training, wages, and sample selection: Estimating sharp bounds on
treatment effects. The Review of Economic Studies, pages 1071 – 1102, 2009.
Miguel, Edward and Michael Kremer. Worms: Identifying impacts on educational and
health in the presence of treatment externality. Econometrica, 72:159 – 217, 2004.
Pugatch, Todd and Elizabeth Schroeder. Incentives for Teacher Relocation: Evidence from
the Gambian Hardship Allowance. Economics of Education Review, 41:120-136, 2014.
Reinikka, Ritva and Jakob Svensson. Local capture: Evidences from a central govern- ment
transfer program in uganda. The Quarterly Journal of Economics, 2004.
USAID, The World Bank, & Eddata. The Gambia Early Grade Reading Assessment (EGRA):
Results from the 1,200 Gambian Primary Students Learning to Read in English—Report for
the World Bank. January 2008.
24
World Bank. Making Services Work for Poor People. World Development Report, 2004.
25
Tables
Table 1: Key elements of intervention arms
Grant provided Management training Management manual
provided provided
WSD Schools Yes Yes Yes
Grant-only Schools Yes No Yes
Control Schools No No Yes
Table 2: Timeline of intervention and evaluation
Date Activity
10/2007 – 4/2008 Sensitization and coordination between stakeholders
4/2008 – 6/2008 Assignment to interventions and baseline data collection
5/2008 – 12/2008 Grant distribution and training in the WSD schools
5/2009 – 6/2009 Collection of first follow-up data
5/2010 – 6/2010 Collection of second follow-up data
5/2011 – 6/2011 Collection of third follow-up data
Throughout Monitoring of implementation
Table 3: Description of the data
Year Data type Respondent Obs Notes
School data Principal, deputy 273
Student test 3rd & 5th grades 8,856
2008 Classroom visit 4th & 6th grades 528
Student interview 3rd & 5th grades 2,688 Administered to a subset of tested
children
School data Principal, deputy 176
Student test 4th & 6th grades 5,660
2009 Classroom visit 3rd & 5th grades 346 No data in Grant-only schools
Student interview 4th & 6th grades 1,755
Teacher test About 6 teachers 1,049
School data Principal, deputy 276
Student test 3rd & 5th grades 9,022
2010 Classroom visit 4th & 6th grades 502
Student interview 3rd & 5th grades 2,678
Parent interview Parent or caregiver 567 Of two interviewed students
School data Principal, deputy 274
Student test 4th & 6th grades 5,230
Classroom visit 3rd & 5th grades 534
2011 Student interview 4th & 6th grades 2,579
SMC interview Committee (minus 249 Mostly PTAs, in controls and
principal) Grant
Teacher interview 4th & 6th grades 517 Teachers of tested students
26
Table 4: Baseline Group Comparison on School Characteristics
WSD Grant Control
Student Observations
Number of students 461 433 426
(59) (41) (45)
Student-teacher ratio 32 34 32
(0.89) (0.97) (1.14)
Double shift 0.33 0.49 0.41
(0.50) (0.50) (0.05)
Tap drinking water 0.23 0.20* 0.33
(0.04) (0.04) (0.05)
Student-latrine ratio 79 49 64
(15) (4) (9)
Has a library/storage for books 0.37 0.53 0.47
(0.05) (0.05) (0.05)
Received cash/in-kind from community 0.38 0.31 0.29
(0.05) (0.05) (0.05)
Number of meetings with parents 4.39** 3.70 3.69
(0.27) (0.24) (0.25)
Has mentoring system 0.86 0.82 0.81
(0.04) (0.04) (0.04)
Written staff code of conduct 0.39 0.43 0.44
(0.05) (0.05) (0.05)
Pupils per class (2006 Administrative Data) 34 33 34
(0.10) (0.10) (0.11)
Adult literacy (2003 Census) 38% 39% 38%
(0.015) (0.014) (0.012)
Primary Education or more (2003 Census) 57% 55% 55%
(0.017) (0.016) (0.014)
Years Established 24 25 24
(1.6) (1.8) (1.9)
Number of observations 90 94 89
Classroom Observations
Teacher has lesson notes 0.31 0.33 0.27
(0.04) (0.04) (0.03)
Percentage of pupils absent 0.25 0.21* 0.26
(0.06) (0.02) (0.02)
Hours/week English 3.67 3.57 3.81
(0.15) (0.15) (0.13)
Number of observations 175 180 173
Notes: Standard errors are in parentheses. *** 1% Significance Level, **5% significance Level,
*10% Significance Level. The mean comparison test contrasts each treatment group with the
control group.
27
Table 5: Baseline Group Comparison on Student Characteristics
3rd grade 5th grade
WSD Grant Control WSD Grant Control
Student age 10.20 10.20 10.10 12.73 12.59 12.64
(0.10) (0.10) (0.10) (0.08) (0.08) (0.08)
Number of siblings 4.90 4.70 4.75 4.70 4.70 4.80
(0.13) (0.13) (0.13) (0.13) (0.12) (0.12)
Ate breakfast today 0.69 0.71 0.73 0.67** 0.73 0.74
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)
Ate lunch yesterday 0.96 0.95 0.94 0.94 0.97 0.95
(0.01) (0.01) (0.01) (0.01) (0.01) (0.01)
Electricity at home 0.19* 0.21 0.24 0.20 0.17 0.20
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)
Radio at home 0.91 0.92 0.93 0.88 0.89 0.87
(0.01) (0.01) (0.01) (0.01) (0.01) (0.02)
TV at home 0.37 0.38 0.38 0.40 0.36 0.36
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)
Telephone/Mobile at home 0.83 0.81 0.82 0.81 0.86 0.83
(0.02) (0.02) (0.02) (0.02) (0.02) (0.02)
Percent repeating the Class 0.09 0.09 0.09 0.08 0.07 0.08
(0.29) (0.29) (0.29) (0.26) (0.26) (0.26)
Observations 462 462 445 423 458 447
Notes: Standard errors in parentheses. *** 1% Significance Level, **5% Significance Level, *10%
Significance Level. The mean comparison test contrasts each treatment group with the control group.
28
Table 6: Community participation, school management and characteristics (2009)
WSD Control Difference P-value
Received support/aid from the community 0.46 0.35 0.11 0.15
(0.05) (0.05) (0.07)
Does the school have a PTA 1.0 0.99 0.01 0.32
(0) (0.01) (0.01)
PTA fund raisers 0.10 0.11 -0.01 0.83
(0.03) (0.03) (0.05)
PTA member contribution 0.09 0.05 0.04 0.23
(0.03) (0.02) (0.04)
PTA not funded 0.71 0.75 -0.04 0.57
(0.05) (0.05) (0.07
Number of meetings with the parents or PTA 4.45 3.92 0.53 0.19
(0.31) (0.26) (0.4)
Mentoring system in place for junior teachers 0.47 0.53 -0.06 0.41
(0.05) (0.05) (0.08)
Mentors trained 0.7 0.57 0.14* 0.08
(0.05) (0.05) (0.08)
Leadership and Management committee in place 0.94 0.75 0.19*** 0
(0.03) (0.06) (0.06)
Community Participation committee in place 0.79 0.63 0.16** 0.04
(0.05) (0.07) (0.08)
Curriculum Management committee in place 0.84 0.51 0.33*** 0
(0.04) (0.07) (0.08)
Teachers’ professional development com. in place 0.8 0.61 0.19** 0.02
(0.05) (0.07) (0.08)
Teaching and learning resources com. in place 0.81 0.59 0.22** 0.01
(0.05) (0.07) (0.08)
Learners welfare committee in place 0.88 0.71 0.17** 0.01
(0.04) (0.06) (0.07)
School has developed school policy 0.45 0.36 0.09 0.26
(0.05) (0.05) (0.07)
First grade enrollment 91.82 76.29 15.53 0.2
(9.85) (7.02) (12.12)
Student-teacher ratio (Lower Basic) 53.18 53.18 0 1
(11.55) (7) (13.11)
Seen records of the teachers attendance 0.91 0.89 0.02 0.64
(0.03) (0.03) (0.05)
Teacher Absenteeism/ Average 5 random days 0.06 0.06 0 0.94
(0.01) (0.01) (0.01)
School has a library 0.53 0.6 -0.07 0.43
(0.05) (0.05) (0.08)
Observations 88 89
Notes: Standard deviations in parentheses. *** 1% Significance Level, **5% Significance Level, *10%
Significance level. The test of comparison of means is between each treatment group and the control group.
29
Table 7: Teaching practices and absenteeism (First follow-up in 2009)
WSD Control Difference P-value
Teacher absent (at our arrival) 0.11 0.12 0.01 0.73
(0.02) (0.03) (0.04)
Teacher missed at least one day last week 0.26 0.33 0.07 0.16
(0.03) (0.04) (0.05)
Teacher Absenteeism (Five random days average) 0.06 0.06 0 0.94
(0.01) (0.01) (0.01)
Student Absenteeism (Day of test) 0.26 0.24 0.02 0.55
(0.02 (0.01 (0.02
Student Absenteeism (Five random days average) 0.38 0.36 0.02 0.71
(0.04) (0.03) (0.05)
Teacher has written lesson plan 0.56 0.67 -0.11** 0.04
(0.04) (0.04) (0.05)
Teacher has a written lesson note for today’s lesson 0.32 0.41 -0.09* 0.08
(0.04) (0.04) (0.05)
Teacher missed at least one day last week 0.26 0.33 0.07 0.16
(0.03) (0.04) (0.05)
Call out children by their names 0.48 0.35 0.13** 0.03
(0.04) (0.04) (0.06)
Address questions to the children during class 0.69 0.75 0.06 0.27
(0.04) (0.04) (0.05)
Encourages the children to participate 0.61 0.68 0.07 0.23
(0.04) (0.04) (0.06)
The children used textbooks during the class 0.38 0.47 -0.09* 0.09
(0.04) (0.04) (0.05)
The children used workbooks during the class 0.54 0.45 0.08 0.14
(0.04) (0.04) (0.06)
The children ask questions for clarification 0.26 0.23 0.03
their doubts (0.04) (0.03) (0.05)
Observations 88/169 89/177
Notes: Standard deviations in parentheses. *** 1% Significance Level, **5% Significance Level, *10%
Significance Level. Based on data from school and classroom visits.
30
Table 8: Student performance (First follow-up in 2009)
4th Grade 6th Grade
WSD Control P-value WSD Control P-value
Reading test
Correct letters per minute 55 57 0.26 73 75 0.17
(1.23) (1.23) (1.15) (1.1)
Correct words per minute 23 25 0.33 41 41 0.75
(1.18) (1.15) (1.08) (1)
Written test
Overall 47.2 48.22 0.5 60.59 61.79 0.4
(0.46) (0.45) (0.49) (0.45)
Math 47.04 49.75 0.2 65.95 68.19 0.23
(0.65) (0.66) (0.67) (0.62)
Literacy 45.82 45.94 0.93 57.19 57.76 0.67
(0.44) (0.41) (0.47) (0.43)
Observations 411 403 431 460
Notes: Standard deviations in parentheses. *** 1% Significance Level, **5% Significance Level,*10%
Significance Level. Same students at the baseline. The score of the written test is the average score expressed
in percentage.
Table 9: Student performance (two years into intervention – 2010)
a
Test score Percentage of students who can
read b
3rd graders 5th graders 3rd graders 5th graders
WSD -0.001 -0.08 0.01 -0.05
(0.08) (0.09) (0.03) (0.04)
Grant 0.01 0.03 -0.01 -0.05
(0.08) (0.09) (0.02) (0.04)
Observations 4537 4354 1241 1202
Mean of dependent 35.32% a 52.06% a 11% b 38% b
variable in comparison
group
Notes: Standard deviations in parentheses. *** 1% Significance Level, **5% Significance Level, *10%
Significance Level. a Test score is normalized to 100 points. It is standardized only for the calculation of the
treatment effect. b Percentage of students who can read 45 or more words per minute.
31
Table 10: Teaching practices (two years into intervention – 2010)
Probability of Probability of Probability of Probability that
calling students frequent use of children asking the teacher has
by name the blackboard questions in class no lesson notes
(1) (2) (3) (4)
WSD 0.10* 0.07* 0.03 0.03
(0.07) (0.03) (0.06) (0.06)
Grant -0.001 0.02 -0.08 -0.01
(0.07) (0.04) (0.06) (0.06)
Observations 427 427 420 511
Mean of 39% 82% 33% 37%
dependent variable
in comparison
group a
Notes: Standard deviations in parentheses. *** 1% Significance Level, **5% Significance Level, *10%
Significance Level. The unit of observation is a classroom. Robust standard errors. All coefficients are
marginal probabilities. a Percent of classrooms where dependent variable is 1.
Table 11: Participation in management (two years into intervention –2010)
Marginal Probability to participate in decision-making
Teachers Parent Rely on SDP RED Fundraisers Know # Meetings
PTA parent/school
memb.
rule
(1) (2) (3) (4) (5) (6) (7)
Model Probit Probit Probit Probit Probit Probit OLS
WSD 0.42*** 0.64*** 0.18*** 0.26*** 0.11** -0.15** -0.41***
(0.08) (0.06) (0.07) (0.08) (0.06) (0.08) (0.18)
Grant 0.37*** 0.65*** 0.16** 0.37*** 0.07 -0.04 -0.26
(0.08) (0.06) (0.07) (0.08) (0.06) (0.08) (0.18)
Observations 274 274 274 274 274 505 505
Mean of 3.3% 9% 1% 2% 7% 50% 1.9
dependent
variable in
comparison
Notes: Marginal effects are reported for Probit regressions. Robust standard errors in parentheses. *** 1%
Significance Level, **5% Significance Level, *10% Significance Level. The unit of observation is the school
in the first four columns and the household in the remaining columns. RED = Regional Education Directorate.
SDP = School Development Plan.
32
Table 12: Treatment effect on student performance and learning outcomes – two
years into intervention (2010)
3rd Grade 5th Grade
Standardized Probability that a Standardized Percentage of
test score a child can read 45 or test score a students who can
more words per read 45 or more
minute words per minute
WSD group -0.001 0.01 -0.08 -0.05
(0.08) (0.03) (0.09) (0.04)
GRANT group 0.01 -0.01 0.03 -0.05
(0.08) (0.02) (0.09) (0.04)
Number of 4537 1241 4354 1202
observations
Mean of dependent 35.32% 11% 52.06% 38%
variable in
comparison group
Notes: Robust standard error clustered at school level in parenthesis. *** 1% Significance Level, **5%
Significance Level, *10% Significance Level. a Test score normalized to 100 point. It is standardized only for
the calculation of the treatment effect.
Table 13: Average Treatment Effect on 4th and 6th-Graders – Three to four years
into the intervention
Math English
WSD -0.05 0.01
(0.07) (0.08)
Grant -0.07 -0.08
(0.06) (0.07)
4th Grade Dummy -0.69*** -0.74***
(0.03) (0.03)
Constant 0.40*** 0.42***
(0.04) (0.05)
P-value WSD = Grant 0.76 0.23
Observations 4817 4817
Notes: Standard errors in parentheses. *** 1% Significance Level,
**5%. Significance Level, *10% Significance Level.
33
Table 14: Average Treatment Effect on 4th and 6th-Graders – Three to four years into
the intervention – Controlling for variables at baseline
Math English
WSD -0.06 0.00
(0.07) (0.06)
Grant -0.07 -0.06
(0.06) (0.06)
Baseline PTA Meetings -0.01 -0.01
(0.01) (0.01)
Baseline Test Scores 0.34 *** 0.46***
(0.05) (0.06)
4th Grade Dummy -0.68*** -0.73***
(0.03) (0.03)
Constant -0.21*** -0.27***
(0.07) (0.07)
P-value WSD = Grant 0.92 0.30
Observations 4716 4716
Notes: Standard errors in parentheses. *** 1% Significance
Level, **5% Significance Level, *10% Significance Level.
Table 15: Effect of the Interventions on Student and Teacher Absenteeism, and on
Enrollment
Absenteeism Log First-Grade
Students Teachers Enrollment
WSD -4.94** -3.11* -0.01
(2.24) (1.75) (0.1)
Grant -2.61 -0.22 0.03
(2.24) (1.76) (0.1)
Constant 23.35*** 13.31*** 4.16***
(1.72) (0.01) (1.26)
P-value WSD = Grant 0.25 0.11 0.62
Observations 407 274 274
Notes: Robust standard errors in parentheses. *** 1% Significance Level, **5%
Significance Level, *10% Significance Level. The dependent variable in the first
column is the percentage of student absent on the day of survey (scale of 0-100). The
dependent variable in the second column is percentage of teachers absent (scale of 0
- 100). The dependent variable in the third column is the log enrollment of first-
graders. The unit of observation in the first column is the classroom. The unit of
observation in columns 2-3 is the school.
34
Table 16: Inputs to Lee trimming procedure
Control Treatment
Number of observations 444 453
Proportion non-missing 71.0% 79.3%
Math score 73.0% 71.1%
(20) (23)
English score 61.0% 62.0%
(18) (21)
Notes: The dependent variable is a standardized test score. Standard deviations are in parentheses.
Table 17: Bounds for the average treatment effect, accounting for selection using the
trimming procedure
Lee’s upper bound Lee’s lower bound
Math 0.17 -0.19
(0.06) (0.09)
English 0.26 -0.16
(0.07) (0.11)
Notes: The dependent variable is a standardized test score. Standard errors are in parentheses.
Table 18: Classroom Stallings, instructional time allocation
Share of time* (%)
All WSD GRANT CONTROL
Learning activities 44 44 44 45
Social interaction 22 21 23 22
Student (s) uninvolved 19 20 18 19
Discipline 1 1 2 1
Classroom management 2 2 1 1
Classroom management alone 3 3 3 2
Teacher out of the room 9 8 10 10
Obs. 534 176 183 175
Notes: Based on ten two-minute snapshots of classroom activities in 534 classroom observations.
35
Table 19: Role of baseline levels of human capital
Math English
WSD -0.50*** -0.31*
(0.17) (0.17)
Grant -0.13 0.01
(0.16) (0.18)
Adult Literacy 0.54* 1.66***
(0.32) (0.37)
WSD × Adult Literacy 1.12** 0.78*
(0.46) (0.51)
Grant × Adult Literacy 0.07 -0.46
(0.43) (0.54)
Constant 0.25 -0.10
(0.11) (0.12)
Observations 2331 2331
Notes: Robust Standard errors in parentheses. *** 1%
Significance Level, **5% Significance Level,*10%
Significance Level. Adult literacy is the district level
percentage of adults who are literate. It is expressed in the
range 0-1.
Table 20: Role of human capital at the baseline
Math English
WSD 0.36 0.38
(0.24) (0.28)
Grant 0.17 0.20
(0.25) (0.32)
SMC Literacy 0.02 -0.28
(0.21) (0.24)
WSD × SMC Illiteracy -0.65** -0.57*
(0.29) (0.34)
Grant × SMC Illiteracy -0.36 -0.39
(0.30) (0.39)
Constant 0.41 0.64
(0.17) (0.21)
Observations 2035 2035
Notes: Robust Standard errors in parentheses. *** 1%
Significance Level, **5% Significance Level, *10%
Significance Level. SMC illiteracy is the percentage of the
School Management Committee members who have no
formal education. It is expressed in the range 0-1.
36
Table 21: The effect of baseline socio-economic status (using individual SES)
Math English
WSD -0.14* -0.08
(0.07) (0.08)
Grant -0.14* -0.17*
(0.08) (0.08)
Child’s SES 0.07* 0.05
(0.04) (0.04)
WSD ×Child’s 2011 SES -0.07 0.04
(0.06) (0.34)
Grant ×Child’s 2011 SES 0.01 0.03
(0.04) (0.06)
6th Grade Dummy 0.68*** 0.73***
(0.04) (0.04)
Constant - -0.18***
0.14***
(0.17) (0.06)
Observations 2289 2289
Notes: Robust Standard errors in parentheses. *** 1%
Significance Level, **5% Significance Level, *10%
Significance Level. Child’s 2011 SES is a composite
measure of the child’s socio-economic background as
measured in 2011. The variables included in the factor
analysis are the quality of the housing (floor, roof, walls,
electricity), the assets (phone, motorcycle, fridge, car), and
the occupation of the father – Higher values of the factor
associate with higher economic status. The treatment is not
correlated with the measure of SES in 2011.
37
Table 22: The effect of baseline socio-economic status (using district SES)
Math English
WSD -0.04 0.00
(0.08) (0.09)
Grant -0.08 -0.08
(0.07) (0.08)
District 2004 SES 0.14** 0.30***
(0.07) (0.09)
WSD ×District 2004 SES 0.02 -0.10
(0.10) (0.11)
Grant ×District 2004 SES 0.00 -0.13
(0.10) (0.11)
6th Grade Dummy 0.70*** 0.77***
(0.04) (0.04)
Constant -0.26*** -0.29***
(0.17) (0.06)
Observations 3659 3659
R Square 0.13 0.16
Notes: Robust Standard errors in parentheses. *** 1%
Significance Level, **5% Significance Level, *10%
Significance Level. District 2004 SES is the district level
composite measure of the socio-economic background as
measured in 2004 – Prior to the interventions. The variables
included in the factor analysis are the quality of the housing
(floor, roof, walls, electricity), the assets (phone, motorcycle,
fridge, car, TV, fan, generator, livestock), and the expenditure
on educator the past 12 months – Higher values of the factor
associate with higher economic status of the district.
38
Table 23: The effect of baseline school level test scores
Math English
WSD -0.04 -0.03
(0.08) (0.09)
Grant -0.08 -0.11
(0.07) (0.08)
Baseline Test Score 0.14** 0.30***
(0.07) (0.09)
WSD × Baseline Test Score 0.02 -0.10
(0.10) (0.11)
Grant × Baseline Test Score 0.00 -0.13
(0.10) (0.11)
6th Grade Dummy 0.70*** 0.77***
(0.04) (0.04)
Constant -0.26*** -0.29***
(0.17) (0.06)
Observations 2313 2313
R Square 0.04 0.07
Notes: Robust Standard errors in parentheses. *** 1%
Significance Level, **5% Significance Level, *10%
Significance Level. District 2004 SES is the district level
composite measure of the socio-economic background as
measured in 2004 – Prior to the interventions. The variables
included in the factor analysis are the quality of the housing
(floor, roof, walls, electricity), the assets (phone, motorcycle,
fridge, car, TV, fan, generator, livestock), and the expenditure
on educator the past 12 months – Higher values of the factor
associate with higher economic status of the district.
39
Table 24: Do wealthier district contribute more to funding the schools?
Marginal effect of 2004
District Level SES
Gave books to school -0.01
(0.04)
Cash contribution 0.04
(0.04)
Building supply -0.03*
(0.02)
Furniture contribution 0.00
(0.01)
Food contribution -0.03
(0.04)
Observations 3659
Notes: Robust Standard errors in parentheses. *** 1%
Significance Level, **5% Significance Level, *10%
Significance Level. District 2004 SES is the district level
composite measure of the socio-economic background as
measured in 2004 – Prior to the interventions. The variables
included in the factor analysis are the quality of the housing
(floor, roof, walls, electricity), the assets (phone, motorcycle,
fridge, car, TV, fan, generator, livestock), and the expenditure
on educator the past 12 months – Higher values of the factor
associate with higher economic status of the district. The
coefficients are the marginal effect of the District’s 2004 SES
on the dependent variable.
40
Table 25: Test scores before and after by intervention group
WSD Control
3rd Grade 5th Grade 3rd Grade 5th Grade
2008 2010 2008 2010 2008 2010 2008 2010
Math (0-100) 32 36 59 56 35 36 59 58
(22) (23) (25) (24) (22) (23) (25) (24)
English (0-100) 35 35 48 48 34 35 47 49
(11) (12) (18) (18) (10) (12) (17) (18)
14 - 8 (% correct) 45 45 65 66 47 47 64 66
11 + 5 (% correct) 65 67 89 84 72 71 88 88
2 × 33 (% correct) 9 11 46 38 12 11 45 41
Observations 1484 144 1359 142 143 151 136 142
5 4 1 9 7 1
Notes: Standard deviations in parentheses. *** 1% Significance Level, **5% Significance Level, *10%
Significance Level. The test of comparison of mean is between years.
41
Figures
Figure 1: Example of drawing during the training
42
Figure 2: Geographical distribution of the schools
Figure 3: Sampling procedure
43
Figure 4: Fifth-grade test scores at baseline (cumulative distribution)
Figure 5: Fifth-grade reading outcomes at baseline (cumulative distribution)
44
Figure 6: Distribution of composite test scores two years into intervention
Probability distribution Cumulative distribution
45
Figure 7: Distribution of composite test scores at endline (3-4 years into intervention)
46
Figure 8: Treatment effect on composite student test scores by quantile
47
Figure 9: Teacher Content knowledge on selected English questions
Selected Literacy Questions (Full Sample)
The children worked in ___ silence during
the test. 85.28
(Complete, Common, Company, Count )
ENORMOUS: Heavy/Hard/Huge/Rotten 54.07
EVEN: Sandy/ Level/ Rocky/Hard 69.58
STARTLED :
Began/Scattered/Frightened/Deafened
40.73
MYSTERIOUS: Pleasant/Stange/Quiet/
Frightening
48.38
0 20 40 60 80 100
% of Teachers who Answered Correctly
Figure 10: Teacher content knowledge on selected math questions
Selected Math Questions (Full sample)
1/4 + 1/2 + 1/8 45.04
1/4 x 5 29.44
1/4 x 1/6 ÷ 1/8 37.00
75% of 36 48.09
1/10 ÷ 1/5 54.86
1/2 x 1/3 50.74
1 1/2 – 3/4 36.80
864 ÷ 24 63.98
252 ÷ 7 77.82
14 + 139 + 9 83.42
0 10 20 30 40 50 60 70 80 90 100
% of Teachers who Answered Correctly
48
Figure 11: Correlation between teacher content knowledge and student test scores
Figure 12: Level of baseline adult literacy and effectiveness of the WSD on composite
student test scores
49
Figure 13: Level of baseline adult literacy and effectiveness of the WSD on composite
student test scores: Non-parametric estimate
50