Educate to Lead? The Political Economy Effects of
School Construction in Indonesia∗
Monica Martinez-Bravo†
June 9th , 2014
Abstract
The extension of mass education not only affects the level of education of the labor
force, but also raises the average education of local politicians. This paper investigates
the impact of a large program of school construction in Indonesia on local governance
and public good provision. By using a panel dataset of 10,000 villages and exploiting
the staggered timing of local elections, I isolate the effects driven by changes in local
governance. The results suggest that the school construction program lead to important
increases in the provision of public goods. Furthermore, the results are heterogeneous
with villages experiencing stronger increases in public goods for which there was greater
demand. I provide evidence that the results are driven by the higher education of the
incumbent local politician, which suggests that the level of human capital of local
politicians is a key ingredient of public good provision in developing countries.
JEL Classification: D72, P16, O12, O17
Keywords: Political Leaders, Education, Institutions, Local Elections
∗
I would like to thank Manuel Arellano, Stéphanne Bonhome, Michèle Tertilt and Tim Besley, for valuable
suggestions. I also thank seminar participants at University of Mannheim, Universidad Carlos III, CEMFI,
University of Alicante, University of Vigo, University of Namur, Paris School of Economics, Bank of Spain,
University Pompeu Fabra, IIES, Stanford University, University of California-Berkeley, and LSE-UCL Development Seminar, for their comments. Jiayi Wen, Inés Berniell and Gustavo Fajardo provided excellent
research assistance.
†
CEMFI; email: mmb@cemfi.es.
1
1
Introduction
In the last decades, numerous developing countries have experienced unprecedented increases
in school enrolment and educational attainment. A large number of studies have documented
these changes and have examined the implications for labor force productivity and economic
growth (Bills and Klenow 2000, Duflo 2001, among others). However, less is understood
about the political economy effects of a rapid increase in the educational attainment. One
important implication of the extension of mass education is that the level of education of
local politicians increases. This can have a profound impact on the effectiveness of local
governments and ultimately in the quality of public good provision.
In this paper, I investigate the impact of a large school construction program in Indonesia
on village governance and public good provision. Between the years 1974 and 1978, 61,000
new schools were constructed, which represented a doubling of the existing stock of schools.
In a first part of the paper, I document that this program was associated with a rapid increase
in the educational attainment of village heads: their average number of years of schooling
increased from 7.6 to 10.2 between the years 1986 and 2003.
A key challenge of this project is to discern to what extent changes in local public good
provision are driven by local governance and to what extent they are driven by a more
educated labor force. To isolate the effects of of changes in local governance I combine the
presence of the school construction program with other sources of variation. First, Indonesian
adults can join the labor force when they are 16 years old, but they have to wait until they
turn 25 in order to be eligible for the village head position. Hence, the effects on public good
provision that are driven by changes in local governance start taking place considerably later
in time. Second, village electoral calendars are not synchronised and, consequently, elections
for village heads take place in different years for different villages. My main identification
strategy exploits variation across villages in the timing of the first village-head election to
which newly educated cohorts are eligible to run as candidates to isolate the effects on public
good provision that are driven by changes in local governance.
In order to conduct this research design, I construct a village-level panel of over 10,000
villages in the island of Java between the years 1986 and 2003. I create this panel by merging
several waves of the Indonesia village census (Potensi Desa) which is collected every 3 or 4
years. These data contain a large set of measures of public good provision such as number
of health and educational facilities, household access to basic services, and roads. To my
knowledge, this is one of the largest panel datasets, in terms of the number of administrative
units, used to document the evolution of public good provision in a developing country.
The panel structure of the data allows me implement a Differences-in-Differences strategy
2
where I control for village fixed effects that capture any time invariant unobserved determinant of the level of public good provision, such as geographic characteristics, cultural factors,
or the quality of informal institutions. Furthermore, the presence of variation across villages
in the timing of the first election to which newly educated cohorts can run for office enables
the introduction of year fixed effects. Controlling for year fixed effects is particularly helpful
in this context since they capture changes over time that affect all villages in a similar way.
Since the year when newly educated cohorts join the labor market is the same across villages, the year fixed effects control for the secular improvement in the education level of the
labor force. In order to allow for a flexible specification of the year fixed effects, the baseline
specification includes year fixed effects interacted with province fixed effects.
A first set of results shows that the school construction program lead to important increases in the provision of key public goods: number of doctors, number of schools and
number of kindergarten. Furthermore, a number of other public goods also increase in villages where there was greater underlying demand for that specific public good. For instance,
the number of midwives increases by 10% in villages that have high level of fertility at baseline. There are also heterogeneous increases across villages in the number of health facilities,
access to purified drinking water, garbage disposal and asphalted roads. The main identifying assumption is that the timing of the village elections is quasi-random (i.e., as good
as randomly assigned conditional on controls). I provide evidence of the validity of this
assumption.
In a second set of results I show that the level of education of the village head experienced
a significant increased as a result of the school construction program. Then, I implement
an instrumental variable strategy where I use the timing of the first village-head election
to which newly educated cohorts can run as candidates as an instrument for the level of
education of the village head. The instrumental variable results are consistent with the
reduced form results and suggest that the education of the village head increases the provision
of key public goods and also has heterogeneous effects across villages based on the underlying
villagers needs. For instance, one extra year of education of the village head leads to a 17%
increase in the number of doctors and to a 14% increase in the likelihood of having an
asphalted road for villages with bad quality of roads at baseline.
The validity of the instrumental variable strategy requires that the timing of these elections only affects public good provision through inducing changes in the level of education
of the village leadership. I provide a number of robustness checks that suggest that the
results are not driven by changes in other characteristics of village heads or by changes in
the electorate.
Finally, I explore the mechanisms driving these results. First, I review the literature
3
that documents the process of provision of public goods in Indonesia in the late 80s and
90s. Based on this literature and my own conversations with experts in village governance
in Indonesia, I identify three main ways through which village heads can affect public good
provision in their villages: reallocation of the village budget towards development projects,
more efficient management of development projects, and obtaining more funds from upper
levels. The evidence suggest that the former two mechanisms are behind the increase in the
provision of public goods. In particular, I show that more educated village heads spend a
greater share of the village budget on development projects. Consistent with this, I also
find that the number of development projects undertaken in the village increases with the
education of the village head. This is driven only by projects funded by villagers and not
by upper levels. Furthermore, there is evidence that more educated leaders manage projects
more efficiently: conditional on the type of project and the completion rate, more educated
village leaders complete projects in shorter time.
This paper relates to a number of different literatures. First, it is related to the literature
that examines the role of political leaders on economic performance and policy outcomes.
One of the seminal papers in this literature is Jones and Olken (2005). The authors exploit
natural deaths in office of heads of state to document that the turnover of national leaders
in non-democracies substantially affect economic growth. Using a similar empirical strategy
Besley, Montalvo, and Reynal-Querol (2011) (BMR, henceforth) show that the replacement
of highly educated heads of state has a negative impact on economic growth. BMR is
the paper most closely related to the present project in terms of the motivating question.
However, the present project uses a different dataset and approach that contributes to address
some of the limitations imposed by the BMR data and context. First, BMR only examine
economic growth as an outcome variable. The rate of economic growth can be affected by
many factors other than the decisions of the head of state (such as the investment climate
of a country which can be in turn affected by perceptions of how competent is a head of
state). Second, the BMR study uses cross-country data which can potentially have many
confounders and is less suited to explore underlying mechanisms. Finally, the BMR study
leaves unanswered the question of whether the education level of politicians only matters
for heads of state, or whether it matters more broadly for local-level politicians. To my
knowledge, the present project is the first to estimate the causal effect of the education of
local politicians on public good provision using within-country variation.
Second, this paper also relates to the recent literature in empirical political economy that
studies the effects of institutional variations on the quality of local leaders. Some examples
are Ferraz and Finan (2011), Gagliarducci and Nannicini (2013), Brollo et. al. (2013), Beath,
et. al. (2013). These papers use the level of formal education of politicians as a proxy for
4
their level of competence. My contributes to this literature by providing evidence that the
education of local politicians is indeed a good proxy for competence, since an exogenous
increase the education of village heads leads to a substantial improvement in the quality of
a range of public goods.
Finally, this paper relates to the literature that has studied the impact of the school
construction program in Indonesia. See for instance Duflo (2001, 2004), Breierova and Duflo
(2004), Somanathan (2008). To my knowledge, this paper is the first to have studied the
impact of this program that are driven by its effect on the quality of political leaders.
The rest of the paper proceeds as follows. In section 2, I describe the institutional
background with special emphasis on the process of provision of public goods. In section
3 I describe the data. In section section 4, I present the reduced form empirical strategy
and results of the impact of the school construction program on the level of education of
village heads and on public good provision. Section 5 presents the instrumental variables
empirical strategy and results, which shows the return to more educated village heads in
terms of public good provision. Section 6 provides evidence on the mechanisms driving the
results and, finally, section 7 concludes.
2
Institutional Background
2.1
Political Context
Java is the world’s most populous island containing 135 million people (57% of the Indonesian
population). At the time of this study Java was divided in five different provinces West Java,
Central Java, East Java, Jakarta and Yogjakarta. The lowest level of the administration is
the village level also known as desa.1
The political context during the period of this study corresponds to the period of General Suharto rule (1965-1998) also know as New Order. Suharto’s government undertook
several institutional reforms with the objective to centralize power and to restrict political
opposition. During this period, the country held elections every five years for national and
local-level parliaments. However, these elections were highly controlled and Suharto’s party
always obtained wide margins of victory.
1
There is another type of villages known as kelurahan or urban wards which have different village governance structures: the kelurahan head is an appointed official. These villages are not included in the main
analysis but will be used in some of the robustness checks. See Martinez-Bravo (2014) for further details on
the differences between kelurahan and desa.
5
2.2
Village Level Institutions
Village level institutions experienced a profound transformation during the New Order.
When Suharto took power villages were considerable autonomous units governed by traditional customs. In Java, villages were governed by the village head who was elected by
villagers. The Suharto government found this heterogeneity in village governance structures
not conductive for its objectives of top-down development and political control. As a result,
in 1979 the Law number 5 on Village Government was adopted. This law imposed a uniform
structure in village institutions across the country. The village government was formed by
the village head and the village assembly (Lembaga Musyawrah Desa, LMD). These reforms
concentrated a high amount of power on the figure of the village head: the village head was
ex-officio the head of the village assembly and had the right to appoint all if its members;
village legislation and the village budget were drafted by the village head in collaboration
with the village assembly; the village head had the right to appoint all members of the village
administration. Despite the fact that village heads were very powerful in the village, they
were tightly controlled by upper levels: the central government maintained veto power on
a number of village decisions, such as village budget and the village legislation, and village
heads had to be loyal to Suharto’s party. The combination of power concentration in the
village head position and the cooptation of the village heads enabled the Suharto regime
to exert close control of activities and decisions in the villages (Evers (2000), Antlöv (2003,
2004)).
During the New Order there were very few venues of community participation in villagelevel decisions. Villagers participation was mainly restricted to voting for the village head
position every eight years.2 Candidates for the village head position were pre-screened by
upper levels. The fundamental objective of this pre-screening process was to ban entry of
any candidate that could have a communist past or that could be opposed to the Suharto
regime.3 Nevertheless, among candidates loyal to the Suharto regime there was intense
political competition in many villages and there are reports of candidates spending up to
100 million rupiah (approximately, $25,000) in campaign expenditures (Evers (2000), Antlöv
(2003)).
Eligible candidates for the village head position had to be at least 25 years old. Once
elected, the village head could serve at most two eight-year terms.
2
Village elections were non-partisan elections. All candidates were supposed to be supporters of Suharto’s
party.
3
See section 8.1 for a complete list of the requirements that candidates had to fulfil and for further
discussion.
6
2.3
The Role of the Village Head on Public Good Provision
Village-level public goods are provided by mainly two different mechanisms: they are either funded by the regular village-government budget, or they are funded by the National
Development Planning Process (PD5).4
Appendix Table 1 shows the village budget of the average village in the year 1996. Most
of village government revenue (63.8% in particular) originates from within-village sources.5
Villages also receive a lump sum transfer from the central government that amounts 6.5
million rupiah for all villages and some smaller amounts from the province and district
governments. Village revenues are then allocated to routine expenses or to development
expenses. On average, 62% of total revenue is allocated to development projects, mainly to
fund transportation facilities, social facilities, and general infrastructure expenses.
One potential way in which the village head could influence the level of public good
provision in the village is by a better management of the village government administration.
A more skilled village head could reduce routine expenditures and devote more resources to
fund development projects in the village.6
Village Heads can also influence the level of provision of public services in the village
by managing more efficiently development projects. During the time of this study, most
development projects are managed by the village, even when funded by upper levels of
government.7 Grant allocations are typically used to buy materials necessary for the project
and labor is provided by villagers. The tasks of coordinating labor and raising for cash
contributions from villagers is typically undertaken by the village head. A village head that
manages projects more efficiently and engages in less rent extraction might be more able
to attract more contributions from villagers and hence, to provide more and better public
services.
An alternative venue through which village heads can increase public good provision is by
applying to additional funds from the National Development Planning Process. Proposals
4
The discussion in this subsection is based on Evers (2000). This report was the result of the Local Level
Institution Study. This study consisted in substantial data collection and case-study analysis conducted in
48 villages in the year 1996. Evers (2000) provides in depth discussion about public good provision on this
villages and how competent village leaders could influence the allocation of public goods.
5
Revenues from village sources correspond to the sum of village original income (revenue from renting
out village communal property, e.g. titisara land) and village community income (income contributed by
households in cash, labor or goods to finance village development projects, swadaya).
6
Although the size of the village budget is not large, villages could still afford to finance relevant development projects. The total village budget revenue corresponds to $26,000 current USD-PPP adjusted.
According to Evers (2000) a full school renovation costed at that time 18 million rupiah. Hence, if a village
were to spend all its development budget (25.5 million rupiah) in school renovations, it could fully renovate
1.4 schools every year.
7
The data used in this project reveals that in 1986, 88% of all development projects undertaken in villages
were managed by villagers themselves.
7
are collected at the subdistrict level and then sent to the relevant sectoral agency at the
district level for evaluation. According to Evers (2000) many village heads are not aware of
these calls for proposals and, when aware, they submit poor quality proposals not backed
by proper cost estimates. More educated village head might be better informed about the
funding opportunities and more skilled in drafting project proposals.
2.4
The Sekolah Dasar INPRES Program
In the mid 1970s Indonesia undertook one of the fastest school construction programs in
the world to that date (World Bank, (1990)). Between the years 1974 and 1978, the stock
of schools in Indonesia doubled: over 61,000 new primary schools were constructed. This
corresponded to one new school per 500 children of ages 5 to 14 in 1971 (Duflo (2001, 2004)).
The program was named Sekolah Dasar INPRES program and it was funded by oil revenues.8
The INPRES program was designed by the central government. A number or presidential
regulations were enacted that stipulated the number of schools to be constructed in each
district. This number was approximately proportional to the number of school-aged children
not enrolled in school prior to the program.9 Each of the new schools constructed was
designed to host 3 teachers and 120 pupils of primary school age. In particular, children
between the ages of 7 and 12 could attend these new schools.
3
Data
The main data source used in this project corresponds to different waves of the Indonesian
Village Census (Potensi Desa). These data are collected by the Central Bureau of Statistics
(Badan Pusan Statistik ) every 3 or 4 years. The data comprise a large number of measures
of public goods provided in the village, such as health and educational facilities, and a host
of village characteristics. Each year, the village census has a different focus (agriculture,
economy, or population) and, as a result, several variables are not consistently reported in
all waves of the village census. In this paper I focus on public good outcomes that are
consistently reported across the different waves of the survey.
To construct the dataset I merge 6 different waves of the village census collected between
the years 1986 and 2003.10 The Central Bureau of Statistics does not keep consistent village
identifiers across time and, as a result, the merge has to be implemented based on village
8
INPRES stands for Instruksi Presiden or presidential instruction
Duflo (2004) shows that 10% increase in the number of children not enrolled in school is associated with
7.8% more schools. Hence, the program was less redistributive than intended.
10
In particular, the waves correspond to the years 1986, 1990, 1993, 1996, 2000 and 2003.
9
8
names. Given the difficulty in merging datasets with a large number of observations I focus
on the most populous island of Indonesia: Java. The final dataset contains information on
a balanced sample of 10,591 villages comprised in 4 provinces and 82 districts.11
From the year 1986, the village census incorporates information on basic characteristics
of the village head such as highest level of education achieved, age, and gender. Based on
the information on the education level of village heads I construct a measure of the number
of years of schooling of the incumbent village head at each point in time. In the year 1993
the incumbent village head reports their length of tenure. I use this variable to derive the
electoral calendar of each village.12
Table 1 shows some summary statistics. The average village on the baseline sample
comprises 3,378 people. Most of the inhabitants of these villages, 62%, are employed in
the agricultural sector and only 11% of the villages are classified as urban, according to the
Bureau of Statistics. Hence, the average village in the sample corresponds to a considerably
rural village. The data also contain the number of INPRES schools constructed in each
village. The average number of INPRES schools per village was 1.06, hence confirming
the large magnitude of the school construction program. The average number of years of
education of the village head is 9.8 and the average age of the village head is 43 years old.
97% of village heads are male. The bottom part of Table 1 contains descriptive statistics of
some of the outcomes of interest. The average village in the sample has 3.2 primary schools,
and almost half of them have a high schools. The average number of doctors in the village
is 0.16 in only 5.7% of the villages the water source was the official water provider company,
and hence had access to purified water. These results suggest that the average village in the
sample over the period of this study corresponds to a poor village with deficient access to
basic public good services.
11
The province-level administrative region of Jakarta is not included in the sample because all of its
villages are kelurahan and hence have a different governance structure. For more details see section 8.2 in
the Appendix.
12
With the exception of villages that report a length of tenure lower than 1 year in office, I use the next
scheduled election to construct my measure of the electoral timing. Using the scheduled election instead of
the actual election decreases the power of the Differences-in-Differences strategy, but increases our confidence
on the quasi-randomness assumption of the election timing. See the data appendix (section 8.2) for more
information on how this variable is constructed.
9
4
The Effects of the School Construction on Village
Head Education and Public Goods
4.1
Reduced Form Empirical Specifications
Figure 1 provides a summary of the timeline of events that is useful to illustrate the main
empirical strategy. The INPRES school construction program began in 1974. Children begin
primary school in Indonesia at the age of 7. Hence, the first cohort that could have completed
primary school in the INPRES schools corresponds to children that were 7 years old in 1974.
Nine years later, in 1983, this first treated cohort turns 16 and joins the labor market.
From that point on, every year one more educated cohort joins the labor market. In 1992
individuals that belong to the first treated cohort turn 25 and become eligible candidates for
the village head position. However, since village-head elections are held in different years,
the newly educated cohorts are not able to run for elections until the next village-head
election takes place. Since the length of tenure of village heads is 8 years, this means that
a village that held a village head election in 1991 has to wait until 2000 for the INPRES
program to affect outcomes through the village government office. In contrast, a village
that holds a village head election in 1992 can experience the effects of the INPRES program
through changes in village governance from that point in time. Hence, the year of the first
village-head election held after 1992 in each village captures the time when newly educated
cohorts in the INPRES schools can run for office, and hence the effects driven by changes
in the village government can start taking place. The shaded area of Figure 1 corresponds
to the time window where the “INPRES effects through the village office” are activated for
different villages.13
Hence, I estimate the following empirical specification:
yvpt = λ0 + λ1 postel92vpt + λ2 postel92vpt × N um IN P RESvp + αv + δt + γp × δt + εvpt (1)
where yvpt corresponds to the outcome of interest in village v in province p and year t.
For instance, yvpt could correspond to the number of doctors in the village. The variable
postel92vpt is a dummy that takes value 1 for the periods t after village v has held its first
village-head election after 1992. The next regressor is the interaction betweeen the postel92vpt
13
Although it is possible that cohorts older than 7 in 1974 benefited from the INPRES schools for some
portion of their primary school education, I do not find substantial effects on village governance of elections
that took place before 1992. A possible interpretation of this finding is that substantial changes in village
governance are only observed when cohorts that completed their whole primary education in the INPRES
villages run for elections.
10
dummy with the number of INPRES schools constructed in each village minus the sample
mean of INPRES schools. This normalization facilitates the interpretation of the coefficients
with λ1 capturing the effect of the postel92vpt dummy at mean value of the INPRES schools
and λ2 capturing the additional effect for villages with one extra INPRES school over the
sample mean. Finally αv are village fixed effects, δt are years fixed effects, and γp × δt are
province fixed effects interacted with year fixed effects.
A number of features of this specification are worth highlighting. First, since there is
variation across villages in the timing of the first election post-1992, the postel92vpt is not
collinear with the year fixed effects. Controlling for year fixed effects is particularly helpful
because the year when the first educated cohort joins the labor market is the same for all
villages. Hence, the year fixed effects will capture the secular improvement in the education
level of the labor force as well as any other factors that change over time that affect villages
similarly.14 In order to allow for a flexible specification of the year fixed effect across regions
I include province fixed effects times year fixed effects as controls. Hence, I only rely on
within province variation in the timing of the village level elections to identify the effects on
public good provision.
This reduced form specification is equivalent to a Differences-in-Differences strategy
where I compare the change in public good provision in each village before and after the first
post-1992 election is held and across villages that have already held their post-1992 elections
to those that had not yet held their post-1992 election. Furthermore, I explore whether this
effect is stronger for villages that experienced a higher level of school construction.
The identifying assumption requires that the timing of the first village-election post-1992
is quasi-random (i.e., as good as randomly assigned conditional on the controls).15 In Table
2, I test the validity of this identifying assumption by examining whether changes in the
pre-treatment value of a large number of characteristics predicts the timing of elections. In
particular, each row corresponds to a different cross-sectional regression where the dependent
variable is the year of the first election post-1992 and the independent variable the change
in the corresponding regressor between the years 1986 and 1990. Columns 1 and 2 show
the point estimates and standard errors, respectively, while column 3 shows the standadized
(beta) coefficient. Out of 50 pairwise correlations performed, only 3 of them are statistically
significant at the 10% level or lower. This represents a 6% of the regressions performed,
14
Another factor that is also captured by the fixed effects is the increase in the average education of the
population eligible to vote in village elections. The minimum voting age in Indonesia is 17 years old. Since
this is the same for all villages, the average education of the voting age population is also controlled for
by the year fixed effects. I will discuss this in further detail and present a number of robustness checks in
section 5.3.
15
See Appendix Table 2 for descriptive statistics on the election timing.
11
which is consistent with the quasi-randomness assumption of the election timing.16
In a second empirical specification, I explore the presence of heterogeneous effects across
villages. More specifically, I study whether villagers demands for particular a public good
in a pre-treatment year is associated with particular expansions of that public good. This is
captured by the following specification
yvpt = λ0 + λ1 postel92vpt + λ2 postel92vpt × Demandvp + αv + δt + γp × δt + εvpt
(2)
where Demandvp is a proxy for villagers demands measured in the year 1986.17 As predictors
of demand for health spending I use the outbreak of infectious diseases, the level of mortality
and the level of fertility. As predictors of access to safe water and garbage disposal I use the
proxies for demand for health improvements and also dummies for deficient level of public
good provision at baseline. Finally, as predictors of investment in roads I used the baseline
quality of roads and the distance to the capital of the subdistrict as a proxy of how remote
is the village.18
This specification corresponds to a Differences-in-Differences strategy where I allow the
increase in public good provision to be heterogenous across villages depending on the underlying demand for each specific public good. If the increase in public good provision responds
to underlying villagers’ needs I expect to find λ̂2 > 0.
4.2
Results: The Effect of the School Construction program on
Village Leadership
In this subsection I present the effects of the school construction program on village leadership
using variations of equation (1). Table 3 shows then main results when the dependent variable
is the number of years of education of the village head. Column 1 shows the raw difference in
years of schooling between the pre and the post first election periods. Column 2 incorporates
village fixed effects and year fixed effects. Controlling for year fixed effects greatly reduces
the magnitude of the coefficient, since they capture the secular increase in years of education
of both village leaders and the population. Column 3 additionally controls for province
16
The covariates that are found to be correlated with election timing are the availability of a horse-drawn
cart in the village, the presence of village cooperatives, and the number of banks. The full set of pairwise
correlations are available from the author upon request. All regressions include province fixed effects as
controls.
17
When the variable is not available for the year 1986, I measure it using the year 1990. See the Data
Appendix for a detailed description of how the proxies for demand are constructed.
18
See the Data Appendix for more details on how these proxies have been constructed.
12
interacted with year fixed effects and the estimated results are similar. Column 4 shows the
regression estimates that correspond to equation (1). The results suggest that the school
construction program lead to an increase of 0.53 years of education of the village head, on
average. This effect is stronger in villages where more INPRES schools were constructed.
However, the interaction is not precisely measured and the coefficient is not statistically
significant. Column 5 shows that the effect of the school construction on the education of
village heads was stronger for villages that introduce elections later in the window 1992 to
2000. This result corresponds to what we would expect since in villages that have the election
in 1992 only one newly educated in the INPRES schools cohort can run as candidates, while
in villages that have the election in 2000, eight newly educated cohorts can run as candidates.
Finally, column 6 shows that the increase in the education of village heads not only took
place in villages where one member of the newly educated cohorts took office, but it is
also present in villages where an older candidate won the elections. To show that I create
a dummy that takes value 1 for villages where the village head that took office after the
first election-post-1992 is young enough to have attended the INPRES schools.19 Seventeen
percent of the villages have a village head after 1992 who is young enough to have attended
the INPRES schools. In the empirical specification, I interact these dummy with the post1992 election dummy. Naturally, the increase in education is much stronger in villages where
the elected village head is young enough to have attended the new schools. In those villages
the increase in years of education of the village head almost 1.5. However, even in villages
where the village head after elections is old, we also observe an increase in the years of
education of the village head. This can be explained by the fact that the entry of young
and educated cohorts as candidates of these elections made the education of the candidates
a salient characteristic of the political competition. This could have made the equilibrium
village head more educated even though he was old.20
Appendix Table 3 shows the results of similar specifications when the dependent variable
is the age of the village head. The results suggest that the school construction program
lead to a decrease of 3.5 years of age. Although both factors could have an independent
impact on public good provision, I will provide evidence that the increase in the education
of the village head is the main factor for understanding the effects of the school construction
program on public good provision.
19
This dummy takes value 1 if the village head is younger than 25 and the village had elections in 1992,
takes value 1 if the village head is younger than 26 and the village had elections in 1993, etc.
20
In other words, between two old candidates, but one more educated than the other, voters might have
been indifferent before the entry of young, educated candidates. However, after this entry, voters might have
developed a preference for the more educated, older candidate.
13
4.3
Results: The Effect of the School Construction program on
Public Goods Provision
Next, I describe the effects of the school construction program on the provision of public
goods. Table 4 shows results of estimating equation (1) when the dependent variables correspond to different types of public goods. The results suggest that the school construction
program lead to increases in the number of kindergartens, the number of primary and high
schools and the number of doctors in the village. The effects seem to be small in magnitude,
but nevertheless statistically significant. Furthermore, the effects are present in villages
that experienced a considerable expansion in the number of INPRES schools, while are insignificant for villages where the average number of INPRES schools was constructed.21 For
instance, villages that had 2 INPRES schools constructed in their village experienced a 3%
increase in kindergartens, 2% increase in primary schools, 6% increase in high schools, and
11% increase in the number of doctors, over the corresponding sample mean. The results
for other public goods are not statistically significant, as I will describe when I present the
heterogeneous results. These results suggest that only key public goods such as the number
of schools and doctors increased on average (conditional on enough schools having been constructed), while most other public goods only increased if there was an underlying demand
for that specific public good.
Tables 5A and 5B present the heterogeneous results by villagers demands. For each
public good, the first column reproduces the results of estimating equation (1). In the rest
of columns I estimate equation (2) for different villagers’ demand proxies.
Columns 1 and 2 show that there are considerable heterogeneous results for the number
of primary schools. Villages with a level of primary school enrolment below the median of
their province experienced a stronger increase in the number of primary schools.22 Next, I
examine the provision of health services. Columns 3 to 5 show that the number of doctors
experienced larger increases in villages that experienced an outbreak of infectious diseases
at baseline and that have higher mortality. In columns 6 to 9, the dependent variable is
the number of midwives. Column 6 shows that, on average, there is not an increase in the
number of midwives. However, this uncovers substantial heterogeneous effects: the number
of midwives increases in villages with high demand for health spending (proxied by the
presence of infectious diseases and by high mortality). There is also a substantial increase
in the number of midwives in those villages with particular high fertility. The effects on
the number of health facilities also exhibit a similar heterogeneous pattern, and hence, are
21
This effect is captured by the fact that the coefficient of the uninteracted dummy of post-1st-election
after 1992 is not statistically significant.
22
See the Data Appendix for more precise definitions on how the proxies for demand are constructed.
14
consistent with the results on the number of health workers. The results on the number
of health posts (posyandu) are particularly interesting since these are small communitybased healthcare facilities typically organized by the village government. The results show a
particular expansion of these small facilities in places where there was stronger demand for
health investment.
The next set of results explore the heterogeneous effects on household access to basic
services and roads. The results are presented in Table 5B. The results suggest that household
access to purified drinking water (provided by the Indonesian water company) increased
specially in regions where mortality was above the median of the province or that had a bad
quality of service at baseline (the main source of drinking water was from a natural spring,
rain or river). Consistent with these results, there is a decrease in the presence of critical
land in areas where the quality of water services increased the most. Critical land is defined
as polluted land that endangers water resources of the village. Hence, this points out to a
particular mechanism through which village heads might have improved access to safe water.
Columns 22 to 24 show that households with high mortality experienced an increase in
the likelihood that the village has a system of garbage dispose through garbage bins (as
opposed to garbage disposal to a river or hole). The increase was also larger for villages with
a more deficient garbage disposal at baseline.
Finally, the effects on investment on asphalted or hardened roads are also heterogeneous.
Villages with worse roads at baseline (too narrow for a four-wheel vehicle to go through)
and far from the subdistrict capital experience greater improvement in the quality of roads.
Distance from the subdistrict capital captures how remote is the village. Remote villages
have typically poor roads and greater returns on becoming connected with the subdistrict
capital, which is where the main economic activity in the area concentrates.
4.4
Robustness Checks of the Heterogeneous Results on Public
Goods
In this section, I explore whether the results are robust to alternative econometric specifications and different sets of controls. Table 6 displays the robustness checks on the heterogeneous results presented in Tables 5A and 5B. Each estimate corresponds to a different panel
regression where the dependent variable is defined by each row. In the sake of brevity I report
only the interaction coefficient of the post-1992 election with the corresponding predictor of
demand for that public good (as defined by the row subtitle). Appendix Table 4 shows the
full results. Column 1 presents the baseline results for comparison. Column 2 controls for a
quartic in log population. The results are robust to these controls, hence, reducing the con-
15
cern that different sizes of villages could affect the results. Column 3 incorporates as controls
the pre-treatment value of the corresponding dependent variable interacted with year fixed
effects. As we can see, the results are highly robust to this alternative specification. This
mitigates the concern that villages with different levels of pre-treatment public good provision had differential underlying trends of public good provision. In other words, it mitigates
the concern of mean reversion. Column 4 adds as controls the number of INPRES schools
constructed in each village interacted with village fixed effects. Villages with a large number
of INPRES schools constructed in the 1970s could have experienced a stronger improvement
in the education of the labor force. Allowing the year fixed effects to differ by the intensity
of the INPRES program can better accommodate these differential trends. The results are
robust to this control as well. Finally, column 5 incorporates the pre-treatment value of the
3 covariates that were associated to election timing reported in Table 2, interacted with year
fixed effects. The results are robust to this specification as well.
5
Estimating the Returns to the Education of the Village Head
5.1
Two-Stage Least-Squares Empirical Specifications
The second set of empirical specifications exploits the staggered timing of the first election
post-1992 across villages to construct an instrument for the education level of the village head.
The previous literature on village governance in Indonesia and the results from this paper,
point out at the level of education as the main mechanism behind the increase in public good
provision. The entry educated cohorts as candidates for the village head position resulted in
more educated incumbents in office (as has been shown in Table 3). It is very likely that the
increase in public good provision observed is the result of having more educated village heads
in office. This suggests the possibility of implementing an instrumental variables strategy to
estimate the effect of village head education on public good provision, using the variation
generated by the timing of the 1st election post-1992 as an instrument. More formally, I
estimate the following econometric model where equation (3) presents the structural equation
while equation (4) shows the first stage:
16
yvpt = β0 + β1 educvpt + β2 educvpt × Demandvp + αv + δt + γp × δt + εvpt
educvpt = ρ0 + ρ1 postel92vpt + ρ2 postel92vpt × Demandvp + αv + δt + γp × δt + εvpt
(3)
(4)
where educvpt is the number of years of education of the village head in office in year t in
village v and the rest of variables are defined as in equation (2).
The validity of this instrumental variables strategy requires that the following assumptions are satisfied:
1. Relevance: the instruments are partially correlated to the endogenous regressor.
2. Validity of the first stage: the timing of elections is quasi-random.
3. Exclusion restriction: the timing of the first election post-1992, conditional on controls,
only affects public good provision through changes in the level of education of the village
head.
Previous sections already provided evidence of the validity of the first assumption (Table
3) and of the second assumption (Table 2). Regarding the third assumption, note that the
flexible specification of the year fixed effects capture any secular change that affects villages
within a province in a similar pattern: secular improvements on the level of education of the
labor force, or even in the population eligible to vote, will be controlled by the year fixed
effects. Hence, the exclusion restriction is a plausible assumption. Nevertheless, in section
5.3, I discuss in detail a number of potential threats to the exclusion restriction.
5.2
2SLS Results: The Effect of the Village Head Education on
Public Goods Provision
Table 7A and 7B show the results of the second stage of the instrumental variables strategy described above. The 2SLS results confirm the heterogeneous pattern of public good
provision described by the reduced form results: more educated village heads increase those
public good provision for which there is an underlying demand (as proxied by the variables
described above). Given the similarity in the pattern of the results, in this section I will focus
on discussing the magnitude of the effects.23 In panel B of Table 7A and 7B, I present the
effects in percentages of the sample mean of the dependent variable. I report the estimates
23
Appendix Table 5 shows the 2SLS results for the average results shown in Table 4.
17
of a one year increase in the education of the village head and of an increase equivalent to
one standard deviation of the education of the village head. I focus on the predicted change
for villages with the highest demands of the corresponding public good.24
The results suggest that one additional year of education increases the number of schools
by 3%. Although this result is small in magnitude, it is statistically significant at the 1
percent level. The stronger effects in terms of magnitude are seen for the number of doctors
and health facilities. One additional year of education raises the number of doctors by 17%
and the percentage change is 25% for villages that experienced high mortality at baseline.
The magnitude of the effects for midwives is also modest, with a 2% increase in the number
of midwives for each year of education of the VH, in villages with high fertility. One extra
year of education of the VH increases the number of health posts by 3%. The magnitude of
the effects on polyclinics is larger, but it seems to be driven by the relative infrequency of
the presence of polyclinics in villages—only 4% of the villages have a polyclinic.
Table 7B shows that one extra year of education of the VH increases the likelihood that
a village with poor service quality at baseline gets access to purified drinking water by 3
percentage points. This effect represents a 50% increase over the sample mean. Consistent
with this finding the likelihood of having critical (polluted) land in the village decreases by
2 percentage points. The improvements in garbage bin disposal are also relatively small in
magnitude with a 2% improvement per year of education for villages with poor service at
baseline. Finally, the results on improvement of road quality are higher in magnitude with
a 15% increase in the likelihood of having an asphalt road per year of education of the VH,
for those villages with poor quality at baseline.25,26
24
For instance, for villages with low school enrolment, one year of education of the VH increases the
number of schools by 0.033 (=-0.021+0.054). As a percentage of the sample mean this corresponds to a 1%
increase (0.01=0.033/3.30). For the same set of villages one standard deviation increase in the education
of the VH increases the number of schools by 0.1 (=0.033*3.13). As a percentage of the sample mean this
corresponds to a 3% increase (0.033=0.1/3.30)
25
The OLS results are presented in Appendix Table 6A, 6B. The comparison between the 2SLS results
and the OLS results shows that the point estimates of the 2SLS results are higher in absolute value than
the OLS estimates. This could be driven by at least two different reasons. First, the number of years of
education of village heads might be measured with error, which would generate attenuation bias. Second, in
the presence of heterogenous treatment effects, the 2SLS must be interpreted as the effect of the education
of village leaders in those villages where the INPRES program induced them to replace their village leaders
by more educated ones. This set of villages might be the ones with greater returns to the education of
village heads, and hence, higher point estimates of the effect of education of the village head on public good
provision.
26
Appendix Table 7 shows that the 2SLS results are robust to the different specifications described in
Table 6.
18
5.3
Threats to the Exclusion Restriction
The validity of the 2SLS specification relies on the validity of the exclusion restriction:
conditional on controls, the timing of the first election post-1992 should only affect public
good provision through changes in the level of education of the village head. In this section,
I examine a number of threats to this exclusion restriction.
A More Educated Electorate. As described above, the INPRES program increased
the average educational attainment of both the population and village heads. Since the population elects the village head some of the effects in public good provision can potentially be
driven by a more educated electorate. Earlier research by Milligan, Moretti, and Oreopoulos
(2004) suggests that, in the US and the UK, more educated citizens report a higher interest
in politics, are more informed about politics, and participate more actively in community
meetings. If these results apply to the Indonesian context, more educated villagers might
provide stronger oversight of the village government and demand greater provision of public
goods. However, this is unlikely to explain the results in this paper for several reasons. First,
note that the flexible year fixed effects specification not only controls for the secular improvement of the education of the labor force, but also by the improvement in the educational
attainment of the population eligible to vote.27 Since the minimum voting age is the same
across Indonesia, the fact that every year a more educated cohort joins the relevant electorate
is controlled by the year fixed effects. In other words, if village heads performance every
year is affected by the demands of the population eligible to vote that year, the year fixed
effects will also be controlling for this secular increase in these demands. However, a more
problematic situation for my econometric specification would emerge if village heads did not
respond to those eligible to vote each year, but to the group of individuals that elected them.
This would be problematic because the average education of the relevant electorate would
also change discontinuously at the time of each election.
To mitigate this concern, I examine how the results compare to the results in kelurahan
villages. In addition to the villages with elected village head that have been the object
of study so far (desa villages), in Indonesia there exist another type of village known as
kelurahan or urban ward that have an appointed village head.28 If the results presented in
this paper are driven by a more educated electorate, we expect to find lower, or no effect on
kelurahan villages, since the electorate played no role on the appointment process of their
village heads. Table 8 presents the results for a selected set of outcomes and Appendix Tables
27
The minimum voting age in Indonesia is 17 years old.
Kelurahan heads are appointed by the district mayor. See Martinez-Bravo (2014) for further details on
the differences between desa and kelurahan and for robustness checks on the potential endogeneity in their
classification.
28
19
8A and 8B show the complete set of results. More specifically, Table 8 shows the 2SLS results
in a sample that includes both type of villages but allows the relevant coefficients to differ
between desa and kelurahan. Columns with headings “VH Elec.” show the coefficients for
desa villages, while columns with headings “VH App.” show the coefficients for kelurahan
villages. The results suggest that villages that have an appointed village head also experience
substantial increases in public good provision once a more educated village head takes office.
Some of the results suggest that the increases in public good provision are less sensitive to
underlying demands: this is the case for the number of polyclinics or the number of midwives.
This is consistent with the hypothesis that elected village heads are more accountable to
villagers, and hence more responsive to underlying village needs. However, not all the results
exhibit this pattern and hence it is hard to extract general lessons. However, the results
unequivocally suggest that villages with appointed village heads also benefited from having
more educated leaders, which is difficult to reconcile with the hypothesis that the increases
in public good provision are driven by changes in the electorate.
Changes in Age Composition of Village Heads. Next, I discuss the possibility
that the changes in public good provision are driven by changes in other characteristics of
the village heads. As discussed above, 1992 is the year when members of the first educated
cohort in the INPRES schools could run as candidates for the village head position. One
could expect that this influx of young and educated cohorts reduces substantially the average
education of village heads. Appendix Table 3 suggests that there is a small decrease of 3.5
years in the age of the VH. Despite the fact that the decrease in age is small, it is still
necessary to examine whether changes in age could drive the results. I explore this in Table
9. The format of the table is analogous to the robustness checks Table 6: Column 1 shows
the baseline results where the dependent variable is defined by each row. Each coefficient
reported corresponds to a different 2SLS regression. The coefficient reported corresponds to
the interaction coefficient of years of education of the VH with the corresponding predictor
of demand, as defined by each row subtitle. Columns 2 and 3 show the coefficients of a
similar regression where the age of the VH is added as an exogenous control. As we can
see the coefficient on education interacted with demand is robust to the inclusion of this
control, while the coefficient on age is small and statistically insignificant. Columns 4 and
5 show the results when age is treated as an endogenous regressor. In the first stage of
this specification I add as additional instrumental variables the average age and education
of village heads in neighboring villages. The interaction coefficient on years of education is
robust to this specification. With the only exception of critical land and roads, age is an
insignificant predictor of public good provision. Given, how similar are the point estimates
in this specification to the baseline 2SLS specification and the lack of evidence of an effect
20
on age, I decide to keep the main specification because the reduced form estimates are easier
to interpret.
6
Mechanisms
In this section I examine the different mechanisms driving the results. There are three main
ways through which village heads could affect the level of public provision in these villages
and the types of goods provided.
1. Obtaining more funds from upper levels of government.
2. Reallocating of the village budget towards development projects
3. Using village funds and development projects more efficiently.
The more direct way of testing for mechanisms 1 and 2 is to study the effect of village
head education on the different components of the village budget. Information on the village
budget is only provided for the year 1996 and, hence, the empirical specification is modified.
In particular I estimate the following equation:
yvp = β0 + β1 educvp + γp + X0vp δ + εvp
(5)
where yvp is the corresponding budget category, educvp is the level of education of the village
head in office in village v, γp are province fixed effects, and X0vp is a vector of controls for
the population size of village v.29 I estimate equation (5) by OLS and also by 2SLS where I
instrument the education of the village head with a dummy that takes value 1 if the village
has had their first post-1992 election by the year 1996, which corresponds to the year when
the village budget is available.
Table 10a shows the result for different sources of village government revenue, and Table
10b shows the results on village government expenditures. The OLS results in Table 10a
suggest that villages with more educated village heads have higher total revenues and that
this is driven by sources of income within the village. However, the 2SLS very imprecise
and hence we this result is not conclusive. Nevertheless the evidence is at odds with the
hypothesis that more educated politicians obtain more funds from upper levels of government:
both in the OLS and 2SLS specification there is no effect of village head education on central
29
Controls include a quartic in log population size and a quartic in the percentage of households whose
main occupation is in agriculture. Note that adding these controls is more important in this specification
than in the panel specification because in the cross-sectional regression I cannot control for village fixed
effects.
21
or district government transfers and a negative effect on provincial transfers. Table 10b shows
the effect on village government expenditures. More educated governments seem to spend
more on infrastructure, expenses in goods and social facilities.
Next, I examine the characteristics of development projects undertaken in these villages.
This information is available for the years 1983 and 1986. Hence, I estimate equation (5)
by OLS and 2SLS in a cross section of villages where the dependent variable are different
characteristics of these projects. The 2SLS specification uses as instrument the average
education of the village head in other villages in the same subdistrict.30 Table 11 presents
the results. Column 1 suggests that one additional year of education of the village head
increases the number of development projects in the village by 0.13. However, the effect
is not statistically significant. The estimate becomes more precise when I study the effect
on the number of projects managed by the village (column 2). In this case, one additional
year of education increases the number of village projects by 0.19. The data also provide
information on the main source of funding for each project. Columns 3 to 5 suggest that the
increase in the number of projects is exclusively driven by projects funded by villagers and
provides suggestive evidence that the central government provides funds fewer development
projects in villages with more educated village heads. This could reflect that the higher
number of projects initiated by more educated village heads crowds out funding from upper
levels.
Overall, these results are consistent with the hypothesized mechanisms 2 and 3 but
inconsistent with 1: more educated village heads reallocate resources towards infrastructure
and social facilities, engage in more development projects managed by the village, but do
not obtain more funds from upper levels of government.
The last set of results presented in Table 11, columns 6 to 8 aims at obtaining measures
of efficiency in the management of projects. To do so, I follow an analysis inspired by Rasul
and Rogger (2013)’s approach and examine within-type-of-project variation on duration of
projects. For a subset of projects there is information on the duration of the project, the
completion rate of the project and the type of project.31 I estimate an analogous regression
to (5) but at the project level. Column 6 explores the effect of village head education on
project duration. The OLS estimates suggests that length of duration of projects is positively
associated with village head education. However, this is driven by highly educated village
heads engaging in a few ambitious projects that last very long. Column 7 shows that when I
30
Note, that information on the education of the village head is reported in the year 1986, but not 1983. I
combine both waves in order to increase the precision of the estimates, but still the analysis is cross-sectional.
31
The type of project is reported in 16 different categories. Unfortunately, the data documentation does
not provide information on what these 16 categories correspond to. At the moment I am trying to obtain
this information.
22
restrict the sample to projects with reported duration lower than 10 months and I condition
on the project completion rate, more educated village heads complete projects in shorter
time. Column 8 shows similar results when controlling also for project type. This evidence
suggests that for the same type of project, villages with more educated village heads are able
to achieve a given completion rate in less amount of time than villages with less educated
village heads. This constitutes highly suggestive evidence that better educated village heads
are better managers of development projects. This in turn could explain why villagers are
more willing to contribute to these projects and also the overall increase in the number of
village-managed projects undertaken.
7
Discussion
This paper provides evidence that the level of education of village heads has an impact on
a wide range of public goods. I find that one additional year of education of the village
head leads to an increase of 3% in the number of primary schools and of 17% increase in the
number of doctors. Furthermore, these and other public goods increases are heterogeneous
across villages with stronger increases of public goods in villages where there is at baseline
a strong underlying demand for that public good.
The results of this paper suggest that improvements in the level of education of village
heads could generate large increases in household access to basic services. In particular,
the improved access to safe water, garbage disposal and access to formally trained doctors
and midwives can have important impact on health of households and also generate positive
externalities. Although undertaking a full cost-benefit analysis is beyond the scope of this
paper, the results of this paper suggest that policies that contribute to the selection of more
educated local-level politicians have the potential of greatly contributing to the well being
of poor communities.
The increase in the provision of public services in these villages does not seem to be
driven by greater access to funds from upper levels of government. To the contrary, the
results seem to be explained by changes in within village dynamics: more educated village
heads reallocate funds towards development projects and manage them more efficiently.
An still open question is why these more educated village heads perform better. The
evidence on more efficient management of village development projects could suggest that
their skills and cognitive abilities might have been enhanced by staying extra years in school.
However, it could also be that more educated village heads have stronger preferences for specific types of public goods: the positive effect on the number of primary schools could point
in this direction. Finally, the results suggest that more educated village heads are more sen23
sitive to the demands of villagers, and hence the level of education of local politicians might
contribute to the degree local accountability. Discerning between these different hypothesis
requires further work. In any case, the results in this paper unambiguously suggest that
more educated village heads deliver more public services, hence, supporting the use of the
education of politicians as a proxy for their quality.
REFERENCES
Acemoglu, Daron, Georgy Egorov and Konstantin Sonin. 2010. “Political Selection and Persistence of Bad Governments.” Quarterly Journal of Economics, 125(4), pp.
1511-1576.
Antlöv, Hans. 2003. “Village Government and Rural Development in Indonesia: The
New Democratic Framework” Bulletin of Indonesian Economic Studies 39: 2, 193–214.
Antlöv, Hans. 2004. “National elections, local issues: The 1997 and 1999 national
elections in a village on Java.” In Elections in Indonesia: The New Order and beyond,
edited by Hans Antlöv and Sven Cederroth. London: RoutledgeCurzon.
Beath, Andrew, Fotini Christia, Georgy Egorov, and Ruben Enikolopov. 2013.
“Electoral Rules and the Quality of Politicians: Theory and Evidence from a Field Experiment in Afghanistan,” MIT Political Science Department Research Paper No. 2013-27.
Besley, Timothy. 2007. Principled Agents? The Political Economy of Good Government. The Lindahl Lectures, Oxford: Oxford University Press.
Besley, Timothy, Jose Montalvo and Marta Reynal-Querol. 2011. Do Educated
Leaders Matter for Growth?, Economic Journal, 121(5), F205-227.
Breierova, Lucia and Esther Duflo. 2004. “The Impact of Education on Fertility
and Child Mortality: Do Fathers Really Matter Less than Mothers?” NBER Working Paper
No.10513.
Brollo, Fernanda, Tommaso Nannicini, Roberto Perotti, and Guido Tabellini.
2013. “The Political Resource Curse.” American Economic Review, 103(5): 1759-96.
Caselli, Francesco and Massimo Morelli. 2004. Bad Politicians Journal of Public
Economics vol 88 (3-4).
Dal Bó, Ernesto, Pedro Dal Bó and Rafael Di Tella. 2006. ”Plata or Plomo?
Bribe and Punishment in a Theory of Political Influence” American Political Science Review,
Vol. 100, No. 1.
24
Duflo, Esther. 2001. “Schooling and Labor Market Consequences of School Construction in Indonesia: Evidence from an Unusual Policy Experiment.” American Economic
Review, 91(4): 795-813.
Duflo, Esther. 2004. “The Medium Run Consequences of Educational Expansion:
Evidence from a Large School Construction Program in Indonesia.” Journal of Development
Economics, 74(1): 163-197.
Egorov, Georgy and Konstantin Sonin. 2005. “Dictators and their Viziers: Endogenizing the Loyalty-Competence Trade-off.” Journal of European Economic Association.
9(5): 903-930.
Evers, Pieter J. 2000. “Resourceful villagers, powerless communities (Rural village
government in Indonesia).” A World Bank–Bappenas Research Project.
Ferraz, Claudio and Frederico Finan. 2011. “Motivating Politicians: The Impacts
of Monetary Incentives on Quality and Performance.” University of California - Berkeley
Mimeo.
Gagliarducci, Stefano and Tommaso Nannicini. 2013. “Do Better Paid Politicians Perform Better? Disentangling Incentives from Selection.” Journal of the European
Economic Association, 11, 369-398.
Glewwe, Paul and Michael Kremer. 2006. “Schools, Teachers, and Education
Outcomes in Developing Countries,” in Eric Alan Hanushek and Finis Welch, eds., the
Handbook on the Economics of Education, volume 2. Amsterdam and London: NorthHolland.
Jones, Benjamin and Benjamin Olken. 2005. “Do Leaders Matter? National
Leadership and Growth Since World War II.” Quarterly Journal of Economics, 120 (3):
835-864.
Martinez-Bravo, Monica. 2014. “The Role of Local Officials in New Democracies:
Evidence from Indonesia.” American Economic Review, 104(4): 1244-1287.
Milligan, Kevin, Enrico Moretti, and Philip Oreopoulos. 2004. “Does Education
Improve Citizenship? Evidence from the U.S. and the U.K.” Journal of Public Economics,
88(9-10).
Rasul, Imran and Daniel Rogger. 2013. “Management of Bureaucrats and Public
Service Delivery: Evidence from the Nigerian Civil Service” UCL Working Paper.
Somanathan, Aparnaa. 2008. “Use of Modern Medical Care for Pregnancy and Childbirth Care: Does Female Schooling Matter?” The World Bank Policy Research Working
Paper No. 4625.
25
Figure 1: Timeline of the INPRES Program and its Effects on Village Government
1st%elecEons%
aJer%1992%
1974%%
1983%%
1992%%
INPRES%
program%
begins%
1st%cohort%
1st%cohort%
affected%
turns%25%
Can%join%the%
labor%market%
Eligible%for%
village%head%
posiEon%
affected%
turns%16%
2000%
0
1,000
Frequency
2,000
3,000
4,000
5,000
Figure 2: Timeline of the INPRES Program and its Effects on Village Government
No school Incomplete PS
PS
Junior HS
1986
26
Senior HS
2003
Academy
Bachelor
Table 1. Summary Statistics
Observations
Mean
Std. Dev
Number of Population
Number of Housholds
Primary School Aged Children / Population
Primary School Enrolment Rate
Percentage of Rural HH
Urban village
Number of Inpres Schools¶
63,543
63,546
10,585
63,522
63,526
63,546
10,591
3,378
803
0.16
0.94
0.62
0.11
1.06
1,881
457
0.05
0.10
0.31
0.31
0.98
Number of Primary Schools
Number of High Schools
Number of Polyclinics
Number of Doctors*
Number of Midwives*
Most HH have garbage bin
Access to Purified Water for Drinking
Presence of Critical (polluted) Land
Main Road is asphalted
Wide Main Road
63,546
63,546
63,545
52,920
63,522
63,546
63,546
63,546
63,480
63,458
3.30
0.43
0.04
0.16
1.62
0.57
0.05
0.12
0.92
0.96
1.85
0.76
0.22
0.58
1.96
0.50
0.22
0.33
0.26
0.19
Years of Education of the Village Head
Age of the Village Head
Male Village Head
63,546
63,519
63,538
9.85
42.98
0.97
3.13
9.19
0.17
Summary statistics corresponding to 10,591 villages in 4 provinces. The years included are 1986, 1990, 1993,
1996, 2000, and 2003. § Year 1996 not reported. * Year 2000 not reported. ¶ Only reported in 1980.
27
Table 2. Determinants of Timing of Village Elections
(1)
(2)
(3)
Dependent Variable: Year of the 1st election after 1992
Coefficient
Standard Error
Standardized Effect
Number of Primary Schools
-0.113
(0.080)
-0.013
Number of High Schools
-0.028
(0.019)
-0.012
Number of Doctors
-0.024
(0.032)
-0.008
Number of Midwives
0.000
(0.017)
0.000
Number of Polyclinics
0.075
(0.055)
0.011
Number of Health Post
0.005
(0.025)
0.002
Purified Water for drinking
0.024
(0.044)
0.006
Critical Land
-0.017
(0.016)
-0.010
Most HH have garbage bin
-0.040
(0.032)
-0.015
Asphalt/Hardened road
-0.009
(0.011)
-0.007
Horse-drawn cart (pedati)
-0.016*
(0.009)
-0.014
Log Population
-0.439
(1.058)
-0.004
Percentage of Rural HH
-0.011
(0.051)
-0.002
Urban village
-0.018
(0.020)
-0.009
Village Cooperative
-0.008
(0.025)
-0.003
Other type of Village Cooperative
-0.026*
(0.014)
-0.019
Village Group Shop
0.022
(0.027)
0.009
Number of Churches
-0.066
(0.041)
-0.016
Number of Mosques
0.028
(0.027)
0.008
Number of Markets
-0.011
(0.040)
-0.003
-0.068**
(0.029)
-0.032
Number of Banks
Notes: Bivariate regressions estimated in the cross-section of villages. The dependent variable is the year of the 1st election
post-1992. The regressor of interest is defined by each row. In particular it corresponds to the percentage change of the
covariate between two pre-treatment years: 1986 and 1990. When information for 1990 is missing, I use the percentage change
between 1983 and 1986. Column 3 shows the standardized (beta) coefficient. All regressions include province fixed effects as
controls. 10,591 villages included.
28
Table 3. The Effects of School Construction on VH Education
(1)
(2)
(3)
(4)
(5)
(6)
Dependent Variable: Years of Education of the Village Head
Dep. Var. Mean
post 1st Election after 1992
9.84
9.84
9.84
9.84
9.84
9.84
1.642***
(0.068)
0.422***
(0.087)
0.534***
(0.088)
0.530***
(0.087)
0.052
(0.040)
0.344**
(0.134)
0.415***
(0.090)
post 1st Election after 1992*Num INPRES schools§
post 1st Election after 1992*Election in 93-94
0.137
(0.234)
0.497***
(0.144)
0.298**
(0.134)
0.443**
(0.213)
post 1st Election after 1992*Election in 95-96
post 1st Election after 1992*Election in 97-98
post 1st Election after 1992*Election in 99-00
post 1st Election after 1992* Young VH
Village FE
Year FE
Province FE * Year FE
Observations
R-squared
1.059***
(0.121)
N
N
N
Y
Y
N
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
63,546
0.528
63,546
0.582
63,546
0.596
63,546
0.596
63,546
0.597
63,546
0.600
Notes: Robust Standard errors clustered at the district level in parenthesis. The sample includes 82 districts. The unit of observation is the villageyear level. The sample includes 10,591 villages. Each column corresponds to a panel regression. The dependent variable is the numer of years of
education of the village head in office in the corresponding village and year. The years included in all regressions are 1986, 1990, 1993, 1996,
2000 and 2003. § The number of INPRES schools is defined in deviations from its sample mean. *** p<0.01, ** p<0.05, *p<0.1.
29
Table 4. The Effects of School Construction on Public Good Provision
(1)
Number of
Kindergarten
Dep. Var. Mean
post 1st Election after 1992
post * Num INPRES schools§
Observations
Number of villages
R-squared
(2)
(3)
Dependent Variables:
Number of Primary
Number of High
Schools
Schools
(4)
Number of doctors
1.229
3.299
0.428
0.158
-0.005
(0.009)
0.001
(0.008)
0.002
(0.006)
-0.000
(0.008)
0.048***
(0.011)
0.069***
(0.015)
0.025***
(0.005)
0.019***
(0.006)
63,546
10,591
0.892
63,546
10,591
0.960
63,546
10,591
0.914
52,920
10,584
0.740
Notes: Robust Standard errors clustered at the district level in parenthesis. The sample includes 82 districts. The unit of observation is the
village-year level. The dependent variable is defined by the column headings. The years included in all regressions are 1986, 1990, 1993,
1996, 2000 and 2003, except for number of doctors which is not reported in the year 2000. § The number of INPRES schools is defined in
deviations from its sample mean. *** p<0.01, ** p<0.05, *p<0.1.
30
Table 5A. Heterogeneous Effects of School Construction on Public Good Provision
(1)
(2)
(3)
Num Primary Sch
Dep Var Mean
post 1st Elec after 1992
post * INPRES schools§
(4)
(5)
(6)
Number of Doctors
(7)
(8)
(9)
Dependent Variables:
Number of Midwives
(11)
(12)
Number of Polyclinics
(13)
(14)
(15)
Number of Health Posts
3.30
3.30
0.16
0.16
0.16
1.62
1.62
1.62
1.62
0.04
0.04
0.04
4.30
4.30
4.30
0.001
(0.008)
0.069***
(0.015)
-0.039***
(0.013)
-0.000
(0.008)
0.019***
(0.006)
-0.008
(0.008)
-0.060***
(0.012)
-0.003
(0.020)
0.020
(0.016)
-0.011
(0.020)
-0.123***
(0.031)
-0.025
(0.021)
-0.002
(0.003)
0.009***
(0.003)
-0.004
(0.004)
-0.021***
(0.004)
-0.138
(0.102)
0.136
(0.125)
-0.131
(0.104)
-0.319**
(0.125)
post*low schooling rate
0.089***
(0.020)
post*infectious disease
0.106***
(0.034)
31
post*mortality 25th-50th
0.109**
(0.049)
0.054***
(0.012)
0.081***
(0.016)
0.126***
(0.021)
post*mortality 50th-75th
post*mortality >75th
0.032***
(0.010)
0.104***
(0.032)
0.176***
(0.043)
0.234***
(0.048)
post*fertility 25th-50th
0.044
(0.220)
0.013***
(0.004)
0.023***
(0.004)
0.048***
(0.007)
0.120
(0.105)
0.205
(0.123)
0.497**
(0.205)
0.076
(0.063)
0.018
(0.061)
0.197**
(0.087)
post*fertility 50th-75th
post*fertility >75th
Observations
R-squared
Number of Villages
(10)
63,546
0.960
10,591
63,546
0.960
10,591
52,920
0.740
10,584
52,920
0.740
10,584
52,920
0.740
10,584
63,522
0.845
10,587
63,522
0.845
10,587
63,522
0.846
10,587
63,522
0.845
10,587
63,545
0.673
10,591
63,545
0.673
10,591
63,545
0.674
10,591
41,608
0.639
10,402
41,608
0.639
10,402
41,608
0.640
10,402
Notes: Robust Standard errors clustered at the district level in parenthesis. The sample includes 82 districts. The unit of observation is the village-year level. The dependent variable is defined by the column headings. The
years included in all regressions are 1986, 1990, 1993, 1996, 2000 and 2003, except for number of doctors which is not reported in the year 2000 and number of health posts which is not reported in years 1986 and 1996. §
The number of INPRES schools is defined in deviations from its sample mean. *** p<0.01, ** p<0.05, *p<0.1.
Table 5B. Heterogeneous Effects of School Construction on Public Good Provision
(16)
(17)
(18)
Safe Drinking Water
Dep Var Mean
post 1st Elec after 1992
post * INPRES schools§
(19)
(20)
(21)
(22)
Dependent Variables:
Critical (polluted) land
(23)
(24)
(25)
Bin Garbage Disposal
(26)
Asphalt / Hard Road
0.05
0.05
0.05
0.12
0.12
0.12
0.57
0.57
0.57
0.92
0.92
0.92
-0.003
(0.002)
0.004
(0.003)
-0.010***
(0.004)
-0.006**
(0.003)
-0.002
(0.006)
0.002
(0.005)
-0.003
(0.007)
0.000
(0.007)
0.002
(0.007)
-0.015*
(0.008)
-0.012
(0.010)
-0.001
(0.008)
0.001
(0.003)
-0.004
(0.004)
-0.024***
(0.004)
-0.012***
(0.004)
32
post*mortality > 50th
0.017***
(0.006)
post*bad baseline service
0.001
(0.006)
0.067**
(0.029)
0.027**
(0.012)
-0.040**
(0.017)
0.082***
(0.020)
0.235***
(0.016)
post*far from subdistrict
Observations
R-squared
Number of villages
(27)
0.044***
(0.008)
63,546
0.740
10,591
63,546
0.740
10,591
63,546
0.741
10,591
63,546
0.588
10,591
63,546
0.588
10,591
63,546
0.588
10,591
63,546
0.802
10,591
63,546
0.802
10,591
63,546
0.803
10,591
63,480
0.578
10,591
63,480
0.596
10,591
63,480
0.579
10,591
Notes: Robust Standard errors clustered at the district level in parenthesis. The sample includes 82 districts. The unit of observation is the village-year level. The dependent variable is
defined by the column headings. The years included in all regressions are 1986, 1990, 1993, 1996, 2000 and 2003. § The number of INPRES schools is defined in deviations from its
sample mean. *** p<0.01, ** p<0.05, *p<0.1.
Table 6. Robustness Checks of the Effects of the School Construction on Public Goods
(1)
Baseline
Specification
(2)
(3)
(4)
Different Set of Controls:
Pre-treatment
INPRES
Controlling for
Dep Var*Year schools * Year
population
FE
FE
(5)
Pre-treatment
Covariates
* Year FE
Number of Primary Schools
(interaction with low enrolment)
0.089***
(0.020)
0.085***
(0.019)
0.085***
(0.018)
0.074***
(0.017)
0.089***
(0.019)
Number of Doctors
(interaction with dengue outbreak)
0.106***
(0.034)
0.101***
(0.032)
0.101***
(0.033)
0.104***
(0.034)
0.093***
(0.033)
Number of Doctors
(interaction with >75th perc mortality)
0.126***
(0.021)
0.101***
(0.021)
0.116***
(0.020)
0.119***
(0.021)
0.095***
(0.021)
Number of Midwives
(interaction with dengue outbreak)
0.109**
(0.049)
0.094**
(0.047)
0.120**
(0.049)
0.107**
(0.050)
0.090*
(0.050)
Number of Midwives
(interaction with >75th perc fertility)
0.197**
(0.087)
0.206**
(0.086)
0.164*
(0.085)
0.184**
(0.086)
0.188***
(0.048)
Number of Polyclinics
(interaction with dengue outbreak)
0.032***
(0.010)
0.030***
(0.009)
0.032***
(0.010)
0.032***
(0.010)
0.029***
(0.010)
Number of Polyclinics
(interaction with >75th perc mortality)
0.048***
(0.007)
0.040***
(0.007)
0.049***
(0.007)
0.045***
(0.006)
0.040***
(0.007)
Number of Health Post
(interaction with >75th perc mortality)
0.497**
(0.205)
0.403*
(0.204)
1.062***
(0.226)
0.440**
(0.170)
0.531**
(0.225)
Purified Water for drinking
(interaction with >50th perc mortality)
0.017***
(0.006)
0.016***
(0.005)
0.017***
(0.006)
0.016***
(0.006)
0.014**
(0.006)
Purified Water for drinking
(interaction with bad pre-treatment service)
0.067**
(0.029)
0.068**
(0.028)
0.066**
(0.029)
0.067**
(0.028)
0.070**
(0.028)
Critical Land endangering Water Sources
(interaction with bad pre-treatment service)
-0.040**
(0.017)
-0.040**
(0.017)
-0.031**
(0.014)
-0.040**
(0.017)
-0.039**
(0.017)
Garbage bin
(interaction with >50th perc mortality)
0.027**
(0.012)
0.024**
(0.011)
0.027**
(0.011)
0.034***
(0.012)
0.017
(0.011)
Garbage bin
(interaction with bad pre-treatment service)
0.082***
(0.020)
0.082***
(0.020)
0.057***
(0.021)
0.081***
(0.020)
0.078***
(0.019)
Asphalt/Hardened road
(interacted with distance to the subdistrict)
0.044***
(0.008)
0.044***
(0.008)
0.014***
(0.004)
0.045***
(0.008)
0.037***
(0.008)
Asphalt/Hardened road
(interaction with bad pre-treatment service)
0.235***
(0.016)
0.235***
(0.016)
0.039***
(0.007)
0.235***
(0.016)
0.232***
(0.015)
Notes: Robust Standard errors clustered at the district level in parenthesis. The unit of observation is the village-year level. Each
cell corresponds to a different panel regression. The dependent variable is defined by each row. The reported coefficient
corresponds to the coefficient of post 1st election after 1992 interacted with the predictor of demand defined by each row
subtitle. All regressions include village fixed effects, year fixed effects and province-year fixed effects. See the Online
Appendix for the full regression results. *** p<0.01, ** p<0.05, * p<0.1.
33
Table 7A. Heterogeneous Effects of VH Education on Public Good Provision (2SLS Results)
Num Primary Sch
(1)
(2)
Dep Var Mean
years of educ VH
Number of Doctors
(3)
(4)
(5)
(6)
Dependent Variables:
Number of Midwives
(7)
(8)
(9)
A. Regression Results
1.62
1.62
3.30
3.30
0.16
0.16
0.16
1.62
1.62
0.093***
(0.025)
-0.021
(0.017)
0.054***
(0.010)
0.027*
(0.016)
-0.000
(0.015)
-0.033**
(0.015)
0.021
(0.032)
-0.008
(0.033)
edu*low schooling rate
edu*infectious disease
0.063***
(0.018)
edu*mortality 25th-50th
34
edu*mortality >75th
0.04
0.04
0.04
4.29
4.29
4.29
0.007
(0.006)
-0.004
(0.006)
-0.015***
(0.006)
0.003
(0.152)
-0.142
(0.099)
-0.276***
(0.098)
0.019***
(0.006)
0.058***
(0.017)
0.104***
(0.026)
0.139***
(0.026)
edu*fertility 25th-50th
0.037
(0.165)
0.007***
(0.002)
0.013***
(0.003)
0.029***
(0.004)
0.074
(0.074)
0.137
(0.095)
0.398***
(0.146)
0.021**
(0.010)
0.017*
(0.010)
0.036***
(0.013)
edu*fertility 50th-75th
edu*fertility >75th
Observations
Number of villages
Cragg-Donald F-Stat
-0.002
(0.032)
Number of Health Posts
(13)
(14)
(15)
0.066**
(0.028)
0.029***
(0.006)
0.047***
-0.01
0.073***
(0.012)
edu*mortality 50th-75th
-0.074**
(0.033)
Number of Polyclinics
(10)
(11)
(12)
63,546
10,591
22.94
63,546
10,591
22.76
52,920
10,584
11.10
52,920
10,584
10.71
52,920
10,584
5.273
63,522
10,587
23.04
63,522
10,587
22.43
63,522
10,587
11.05
63,522
10,587
11.37
63,545
10,591
22.95
63,545
10,591
22.33
63,545
10,591
11.00
0.03
0.01
0.17
0.40
0.25
0.01
0.04
0.04
0.02
0.16
0.35
0.33
0.09
0.03
0.53
1.24
0.79
0.04
0.11
0.13
0.07
0.51
1.09
1.02
41,608
10,402
56.16
41,608
10,402
56.69
41,608
10,402
27.96
0.00
-0.02
0.03
0.00
-0.08
0.09
B. Magnitude of the Effects: Effect of X extra years of education of the VH as a percent of the sample mean (for top demand villages)
1 extra year of
education
1 Stand Dev (=3.13
years of educ)
Notes: Robust Standard errors clustered at the district level in parenthesis. The sample includes 82 districts. The unit of observation is the village-year level. The dependent variable is defined by the column headings. The
years included in all regressions are 1986, 1990, 1993, 1996, 2000 and 2003, except for number of doctors which is not reported in the year 2000 and number of health posts which is not reported in years 1986 and 1996.
Each column corresponds to the 2nd stage of a 2SLS estimation where the education of the VH and its corresponding interactions are instrumented using the post-1st election after 1992 dummy and the corresponding
interactions. See the Online Appendix for the First Stage of these regressions. *** p<0.01, ** p<0.05, *p<0.1.
Table 7B. Heterogeneous
Effects of VH Education on Public Good Provision (2SLS Results)
Table 7B. The Effect of Village Head on Access to Basic Services (2SLS results)
Dependent Variables:
(16)
Safe Drinking Water
(17)
(18)
Critical (polluted) land
(19)
(20)
(21)
(22)
Bin Garbage Disposal
(23)
(24)
(25)
Asphalt / Hard Road
(26)
(27)
A. Regression Results
Dep Var Mean
years of educ VH
0.05
0.05
0.05
0.12
0.12
0.12
0.57
0.57
0.57
0.92
0.92
0.92
-0.000
(0.005)
-0.009**
(0.005)
0.010***
(0.003)
-0.008
(0.005)
-0.002
(0.013)
-0.004
(0.011)
0.000
(0.004)
-0.002
(0.011)
-0.017
(0.016)
-0.006
(0.013)
0.017**
(0.007)
0.002
(0.013)
-0.003
(0.006)
-0.007
(0.008)
-0.008
(0.006)
edu*mortality > 50th
35
edu*bad baseline service
0.035**
(0.017)
-0.020**
(0.009)
0.008***
(0.002)
0.145***
(0.018)
edu*far from subdistrict
Observations
Number of villages
Cragg-Donald F-Stat
1 extra year of education
1 Stand Dev (=3.13 years of educ)
0.026***
(0.005)
63,546
10,591
22.94
63,546
63,546
63,546
63,546
63,546
63,546
63,546
63,546
63,480
63,480
63,480
10,591
10,591
10,591
10,591
10,591
10,591
10,591
10,591
10,591
10,591
10,591
22.37
22.00
22.94
22.37
22.00
22.94
22.37
22.29
22.91
21.91
21.95
B. Magnitude of the Effects: Effect of X extra years of education of the VH as a percent of the sample mean (for top demand villages)
0.02
0.51
-0.02
-0.03
-0.18
-0.03
0.02
0.02
0.00
0.15
0.02
0.06
1.68
-0.05
-0.11
-0.60
-0.10
0.06
0.06
-0.01
0.49
0.06
0.00
0.00
Notes: Robust Standard errors clustered at the district level in parenthesis. The sample includes 82 districts. The unit of observation is the village-year level. The dependent variable is defined
by the column headings. The years included in all regressions are 1986, 1990, 1993, 1996, 2000 and 2003. Each column corresponds to the 2nd stage of a 2SLS estimation where the education
of the VH and its corresponding interactions are instrumented using the post-1st election after 1992 dummy and the corresponding interactions. See the Online Appendix for the First Stage of
these regressions. *** p<0.01, ** p<0.05, *p<0.1.
Effect top demand
over sample mean
Times 1 Std Dev
-0.000
0
0
0.001
0.018868
0.062264
0.027
0.509434
1.681132
-0.002
-0.01653
-0.05455
-0.004
-0.03306
-0.10909
-0.022
-0.18182
-0.6
-0.017
-0.03009
-0.09929
0.011
0.019469
0.064248
0.01
0.017699
0.058407
-0.003
-0.00325
-0.01071
0.138
0.149351
0.492857
0.018
0.019481
0.064286
Table 8. Heterogeneous Effects of VH Education by Method of Selection of the VH (2SLS Results)
(1)
Dep Var Mean
years of educ VH
edu*low schooling rate
(2)
(3)
(4)
Num Primary Sch
VH Elec.
VH App.
Number of Doctors
VH Elec.
VH App.
3.33
0.23
-0.024
(0.016)
0.054***
(0.010)
0.030
(0.019)
0.016
(0.027)
(5)
(6)
(7)
(8)
Dependent Variables:
Number of Midwives
Number of Polyclinics
VH Elec.
VH App.
VH Elec.
VH App.
1.74
-0.047**
(0.019)
-0.094
(0.092)
0.028***
(0.006)
0.044***
(0.009)
0.071***
(0.011)
0.266**
(0.125)
0.321***
(0.114)
0.340***
(0.105)
-0.001
(0.034)
(9)
(10)
-0.022***
(0.006)
0.026*
(0.014)
0.006**
(0.002)
0.012***
(0.003)
0.028***
(0.004)
0.014
(0.021)
0.035
(0.022)
0.031
(0.021)
(12)
Safe Drinking Water
VH Elec.
VH App.
Bin Garbage Disposal
VH Elec.
VH App.
0.07
0.57
0.05
0.210***
(0.069)
(11)
-0.007
(0.005)
0.062***
(0.011)
-0.006
(0.014)
-0.052***
(0.019)
0.035**
(0.017)
0.200***
(0.053)
0.008***
(0.002)
0.018***
(0.003)
edu*infectious disease
edu*mortality 25th-50th
edu*mortality 50th-75th
36
edu*mortality >75th
edu*fertility 25th-50th
0.017
(0.010)
0.013
(0.010)
0.032**
(0.013)
edu*fertility 50th-75th
edu*fertility >75th
0.017
(0.022)
0.006
(0.013)
-0.009
(0.025)
edu*bad baseline service
Observations
Number of villages
Cragg-Donald F-Stat
66,042
11,007
12.29
54,965
10,993
2.895
65,976
10,996
6.028
66,041
11,007
5.954
66,042
11,007
11.85
66,042
11,007
12.14
Notes: Robust Standard errors clustered at the district level in parenthesis. The sample includes 82 districts. The unit of observation is the village-year level. The dependent variable is defined by
the column headings. The years included in all regressions are 1986, 1990, 1993, 1996, 2000 and 2003, except for number of doctors. Each pair of columns corresponds to the 2nd stage of a 2SLS
estimation where the education of the VH and its corresponding interactions are instrumented using the post-1st election after 1992 dummy and the corresponding interactions. The interaction
coefficients are allowed to differ depending on whether the village has an elected village head (desa) or an appointed village head (kelurahan). See the Online Appendix for the complete set of
results. *** p<0.01, ** p<0.05, *p<0.1.
Table 9. Robustness Checks of the Effects of VH Education to Controlling for VH Age
(1)
(2)
(3)
Different Set of Controls:
Baseline,
Years Educ*
Predictor of Demand
Years Educ*
Predictor of
Demand
Age
(Exogenous
Regressor)
(4)
Years Educ*
Predictor of
Demand
(5)
Age
(Endogeneous
Regressor)
Number of Primary Schools
(interaction with low enrolment)
0.054***
(0.010)
0.049**
(0.021)
0.015
(0.045)
0.054***
(0.010)
0.004
(0.006)
Number of Doctors
(interaction with dengue outbreak)
0.063***
(0.018)
0.066***
(0.024)
0.011
(0.035)
0.063***
(0.018)
0.004
(0.004)
Number of Doctors
(interaction with >75th perc mortality)
0.073***
(0.012)
0.073***
(0.017)
0.001
(0.026)
0.073***
(0.011)
0.004
(0.004)
Number of Midwives
(interaction with dengue outbreak)
0.066**
(0.028)
0.068**
(0.033)
0.009
(0.059)
0.063**
(0.027)
-0.008
(0.010)
Number of Midwives
(interaction with >75th perc fertility)
0.139***
(0.026)
0.140***
(0.036)
0.002
(0.060)
0.134***
(0.029)
-0.008
(0.010)
Number of Polyclinics
(interaction with dengue outbreak)
0.019***
(0.006)
0.019***
(0.006)
-0.002
(0.011)
0.020***
(0.006)
0.001
(0.002)
Number of Polyclinics
(interaction with >75th perc mortality)
0.029***
(0.004)
0.027***
(0.007)
-0.004
(0.012)
0.030***
(0.004)
0.001
(0.002)
Number of Health Post
(interaction with >75th perc mortality)
0.398***
(0.146)
0.239
(0.265)
-0.252
(0.224)
0.431***
(0.134)
0.034
(0.057)
Purified Water for drinking
(interaction with >50th perc mortality)
0.010***
(0.003)
0.008
(0.006)
-0.005
(0.012)
0.010***
(0.003)
0.001
(0.002)
Purified Water for drinking
(interaction with bad pre-treatment service)
0.035**
(0.017)
0.038**
(0.018)
-0.005
(0.014)
0.036**
(0.016)
0.001
(0.002)
Critical Land endangering Water Sources
(interaction with bad pre-treatment service)
-0.020**
(0.009)
-0.015
(0.022)
-0.009
(0.028)
-0.018
(0.011)
-0.006*
(0.003)
Garbage bin
(interaction with >50th perc mortality)
0.017***
(0.007)
0.017
(0.011)
-0.000
(0.023)
0.018***
(0.007)
-0.005
(0.004)
Garbage bin
(interaction with bad pre-treatment service)
0.008***
(0.002)
0.008***
(0.002)
0.008
(0.027)
0.008***
(0.002)
-0.006
(0.005)
Asphalt/Hardened road
(interacted with distance to the subdistrict)
0.026***
(0.005)
0.026***
(0.005)
-0.001
(0.011)
0.026***
(0.005)
-0.007**
(0.003)
Asphalt/Hardened road
(interaction with bad pre-treatment service)
0.145***
(0.018)
0.155***
(0.040)
0.020
(0.042)
0.141***
(0.015)
-0.007**
(0.003)
Notes: Robust Standard errors clustered at the district level in parenthesis. The unit of observation is the village-year level. Each cell of
column (1) corresponds to a different panel regression. In particular to the 2nd stage of a 2SLS estimation. The dependent variable is defined
by each row and the reported coefficient corresponds to the coefficient of years of education of the VH interacted with the predictor of
demand defined by each row subtitle. The instruments used in this regression are the post 1st election after 1992 interacted with the
corresponding predictor of demand. Columns (2) and (3) report coefficients of an analogous 2SLS regression where age of the VH is added
as an exogenous regressor. Columns (4) and (5) report the coefficients of an analogous 2SLS regression where age of the VH is added as an
endogenous regressor. In this case the additional instruments are the average age and education of VH in neigboring villages. All regressions
include village fixed effects, year fixed effects and province-year fixed effects. See the Online Appendix for the full regression results. ***
p<0.01, ** p<0.05, * p<0.1.
37
Table 10a. The Effect of VH Education on the Village Budget (Revenues)
Dep. Var. Mean
(not logged)
(1)
(2)
(3)
(4)
Dependent Variable
(5)
(6)
(7)
Log Total
Income
Log Surplus
from last year
Log Income
from Village
Sources
Log Transfer
from Central
Government
Log Transfer
from
Provincial
Government
Log Transfer
from District
Government
Log Other
Income
40595
262
25878
6489
747
762
6456
-0.076***
(0.025)
-0.012
(0.028)
0.067**
(0.029)
Panel A. OLS
years of educ VH
0.020***
(0.004)
0.063***
(0.018)
0.027***
(0.006)
0.015
(0.011)
Panel B. 2SLS
years of educ VH
-0.006
(0.039)
-0.251
(0.181)
-0.056
(0.082)
-0.119
(0.141)
-0.043
(0.293)
0.016
(0.261)
-0.017
(0.217)
Observations
R-squared OLS
10,589
0.158
10,589
0.090
10,589
0.055
10,589
0.011
10,589
0.243
10,589
0.102
10,589
0.030
Table 10b. The Effect of VH Education on the Village Budget (Expenditures)
Dep. Var. Mean
(not logged)
(8)
(9)
Log Total
Expenditures
Log
Employee
Expenses
40,502
10,225
(10)
(11)
(12)
Dependent Variable
Log
Log Expenses Log Expenses
Infraestructure
in Goods
in Maintenance
Expenses
898
632
2,880
(13)
(14)
(15)
Log
Production
Facilities
Expenses
Log
Transportation
Facilities
Expenses
Log Social
Facilities
Expenses
2,558
7,091
6,554
0.074***
(0.025)
0.062***
(0.023)
0.048
(0.030)
0.298
(0.276)
-0.068
(0.247)
0.427*
(0.231)
Panel A. OLS
years of educ VH
0.020***
(0.004)
0.036***
(0.005)
0.012
(0.008)
0.052***
(0.014)
0.065**
(0.031)
Panel B. 2SLS
years of educ VH
-0.005
(0.039)
-0.011
(0.058)
0.127**
(0.065)
0.036
(0.119)
0.686**
(0.342)
Observations
10,589
10,589
10,589
10,589
10,589
10,589
10,589
10,589
R-squared OLS
0.157
0.096
0.036
0.018
0.031
0.046
0.025
0.028
Notes: Robust Standard errors clustered at the district level in parenthesis. Each column corresponds to a cross-sectional regression for the year
1996, when village government budget information is available. The unit of observation is the village-level. All regressions include province
fixed effects, a quartic on log population, and a quartic on the percentage of rural households as controls. Panel A shows the OLS results while
Panel B shows the 2SLS results. A dummy for having had elections between 1992 and 1996 is used as an instrument for the years of education of
the VH. *** p<0.01, ** p<0.05, * p<0.1.
38
Table 11. Effect of VH Education on Development Projects
(1)
(2)
(3)
(4)
(5)
Dependent Variable
(6)
(7)
(8)
Project
duration
Project
duration
(controlling
for project
completon)
Project
duration
(controlling
for project
completon
and type)
7.44
3.57
3.57
0.0527
(0.045)
-0.0382***
(0.012)
-0.0357***
(0.012)
-0.2849
(0.293)
-0.2122***
(0.059)
-0.1666***
(0.060)
Number of pojects by main source of funding:
Dep. Var. Mean
Number of
projects
Number of
projects
managed by
the village
by the
villagers
by central
government
grants
by central
government
budget
allocation
5.73
4.60
2.42
1.52
1.47
Panel A. OLS
years of educ VH
0.0100
(0.012)
0.0243**
(0.012)
0.0262**
(0.012)
-0.0156**
(0.006)
0.0025
(0.006)
Panel B. 2SLS
years of educ VH
0.1301
(0.102)
0.1869**
(0.089)
0.2240**
(0.100)
-0.1048*
(0.056)
0.0299
(0.048)
Observations
10,591
10,591
10,591
10,591
10,591
10,546
8,396
8,396
R-squared (OLS)
0.134
0.077
0.099
0.061
0.109
0.003
0.032
0.073
Notes: Robust Standard errors clustered at the district level in parenthesis. Columns 1 to 5 correspond to cross-sectional regression where the unit
of observation is the village level. The dependent variable in column 1 corresponds to number of development projects undertaken in the
corresponding village between the years 1983 and 1986. The dependent variable in columns 2 to 5 corresponds to the number of projects that had
the characteristic defined by the column heading. The regressor of interest corresponds to the years of education of the village head in office in
1986. Columns 6 to 8 corresponds to regressions at the project level. Each observation corresponds to a development project undertaken in the
years 1983 or 1986. The regressor of interest corresponds to the years of education of the village head in office in 1986 in the village where the
project was undertaken. Columns 7 and 8 restrict the sample to projects with total duration lower than 10 months. Panel A shows the OLS results
while Panel B shows the 2SLS results where the average education of village heads in a particular subdistrict is used as instruments for the
education of the village head. All regressions include province fixed effects and a quartic on log population as controls, and a quartic on the
percentage of rural households as controls. *** p<0.01, ** p<0.05, * p<0.1.
39
8
Appendix [For Online Publication Only]
8.1
Further Details on the Historical Background
Pre-screening
Before every village election, a Village Election Board was established. This Board
comprised the appointed members of the LMD (except for the incumbent village head),
the sub-district head, one army officer, one police officer and other sub-district officials
(Antlöv (2003)). Hence, the Board was closely controlled by village elites and upper-levels
of government.
The Board was in charge of supervising the election process and to verify that all candidates for the village head position satisfied the requirements stipulated in the Law no. 5 of
1979 on Village Governance. The set of requirements are the following:
1. Devoted to God Almighty.
2. Faithful and loyal to the Pancasila and the 1945 Constitution.
3. Well-behaved, honest, fair, intelligent, and authoritative.
4. No direct of indirect involvement with the communist party.
5. No voting rights revoked.
6. Not a criminal record.
7. Registered as a permanent resident and living in the village in question for at least 2
years or born in the village.
8. Being at least 25 years old and at most 60 years old.
9. Junior high school education or equivalent knowledge or experience.
Note that the last requirement mentioned a minimum education level: junior high school.
However, the wording of the law was explicitly ambiguous to allow a lax implementation of
the requirement: many candidates without a junior high school degree could claim that
they had an “equivalent knowledge or experience”. Not surprisingly, the junior high school
requirement was not strictly enforced, as shown in Figure 2.
Overall, there is some consensus among scholars that the fundamental reason of the prescreening process was to limit entry in village politics to opponents of the Suharto regime.
40
The list of requirements were used as a set of reasons to ex-post justify decisions of disqualifying candidates for political reasons.32
8.2
Data Appendix
The dataset used in this project is the result of combining different waves of the Village
Census of Indonesia or PODES. The Central Bureau of Statistics does not keep consistent
village identifiers across time and, as a result, the merge has to be implemented based on
village names. Among the 22,000 villages in the island of Java, I am able to consistently
match 13,819 across the 8 waves included in this study. Among these villages, 300 of them
have some missing information on the years of education of the village head and 2,700 of them
do not report consistent term lengths of the village head for years 1996, 2000, and 2003. I also
drop 633 villages that are kelurahan villages and hence have a different village governances
structure.33 The resulting sample contains information on 10,156 villages. The baseline
specifications in the paper focus on the period 1986-2003 when information of the education
of the village head is available. There are 6 census waves in this period. Consequently, the
number of village-year observations in the baseline specification is 60,936.
Construction of Variables
Years of Education of the Village Head: I follow the standard of the literature to
define the number of years of education as follows: no schooling = 0 years of schooling; not
completed primary school = 2 years; primary school = 6 years; junior high school = 9 years;
senior high school = 12 years; academy = 14 years; college = 16 years.
Village-level electoral calendar: The electoral calendar in each village is derived from
the reported length of tenure of the village head in 1993. In particular, the year of the first
election post-1992 is inputed using the following procedure.
• If the village head reports having been in office between 0 and 1 years in 1993, I know
there was an election in that village between 1992 and 1993. I assume that election
was the first election post-1992.
• If the village head reports having been in office for 2 or 10 years in 1993, I know there
was an election in 1991. Since this election was prior to 1992, the first election after
1992 is scheduled to take place in 6 years. Hence, in 1999.
32
33
Based on personal interviews.
See Martinez-Bravo (2014) for further details on the differences between kelurahan and desa.
41
• If the village head reports having been in office for 3 or 11 years in 1993, I know there
was an election in 1990. Since this election was prior to 1992, the first election after
1992 is scheduled to take place in 5 years. Hence, in 1998.
• Similarly for the rest of villages.
Proxies for Demands of Public Goods:
• Proxies for the demand in education spending:
– Schooling rate. This variable takes value 1 for villages where the fraction of
primary school aged kids not enrolled in school is higher than the median schooling
rate in their province. The year when the enrolment rate is measured is 1986.
• Proxies for demand in health spending:
– Outbreak of infectious diseases. This variable takes value 1 if there is an outbreak
of dengue fever in the year 1990.
– Mortality. Dummies for the different quantiles of level of mortality in the province,
as measured by the number of deaths. The year when the mortality is measured
is 1986.
– Fertility. Dummies for the different quantiles of level of fertility in the province,
as measured by the number of births. The year when the fertility is measured is
1986.
• Proxies for demand of improved drinking water:
– The village census records the main source of drinking water for households in
the village: the listed options are: provision by the main water company (PAM),
electric pump, manual pump, well, spring water, river, rain, or other.
– The dependent variable of safe drinking water is a dummy that takes value 1 if
access to water is provided by the water company and hence, it is purified.
– Bad baseline level of drinking water is a dummy that takes value 1 if in the year
1986 the main source of drinking water in the village was water from a river, rain,
a spring or other.
• Proxies for demand of improved garbage disposal:
– Bad baseline service of garbage bad disposal is a dummy that takes value 1 if
most households dispose their garbage to a hole or to a river.
42
• Proxies for demand of improved roads:
– Bad baseline service of roads is a dummy that takes value 1 if a four-wheel motor
could not drive through the main road in the village in the year 1986.
– Far from subdistrict is a dummy that takes value 1 if the village is further
away from the subdistrict than the 75th percentile distance of all villages in the
province.
43
Appendix
Table 1. Village
Government
Balance Balance
Sheet in year
1996 for year 1996
Appendix Table
1. Average
Village
Government
Sheet
As a fraction of
total revenue
Mean
Std. Dev
Total Revenue (in 1,000 IDR)
Surplus previous year
Village Original Income
Village Community Income (swadaya)
Transfer from Central Government
Transfer from Provincial Government
Transfer from District Government
Other revenues
40,600
262
12,696
13,188
6,489
747
762
6,456
32,936
2,445
11,441
16,542
4,058
3,127
2,980
20,175
0.006
0.313
0.325
0.160
0.018
0.019
0.159
Total Expenditures (in 1,000 IDR)
Routine expenses
Deficit previous year
Employee Expenses
Expenses in Goods
Expenses in Maintenance
Travelling Expenses
Other routine expenses
Miscellaneous expenses
40,507
15,486
43
10,224
899
632
596
2,188
904
32,701
11,255
784
7,643
1,475
1,532
1,143
3,610
3,202
0.998
0.381
0.001
0.252
0.022
0.016
0.015
0.054
0.022
25,022
2,881
2,559
7,091
648
6,555
5,288
27,681
5,674
5,395
10,232
4,093
11,882
12,839
0.616
0.071
0.063
0.175
0.016
0.161
0.130
Development Expenses
Infraestructure Expenses
Production Facilities
Transportation Facilities
Marketing Facilities
Social Facilities
Other facilities expenditure
The number of observations are 10,591
44
Appendix Table 2. Timing of the First Election after 1992
Year
Number of villages having their
1st election post-1992 in the
corresponding year
Percent
1992
1993
1994
1995
1996
1997
1998
1999
2000
435
756
624
253
123
1,375
3,011
3,757
257
4.11
7.14
5.89
2.39
1.16
12.98
28.43
35.47
2.43
Total
10,591
100
45
Appendix Table 3. The Effects of School Construction on VH Age
(1)
Dep. Var. Mean
post 1st Election after 1992
(2)
(3)
(4)
(5)
Dependent Variable: Years of Education of the Village Head
43
43
43
43
43
43
-1.191***
(0.164)
-3.135***
(0.280)
-3.509***
(0.290)
-3.512***
(0.290)
0.030
(0.119)
-3.040***
(0.422)
-2.558***
(0.295)
post 1st Election after 1992*Num INPRES schools§
post 1st Election*Election in 93-94
0.423
(0.630)
-1.966***
(0.445)
-1.001**
(0.448)
0.021
(0.767)
post 1st Election*Election in 95-96
post 1st Election*Election in 97-98
post 1st Election*Election in 99-00
post 1st Election after 1992* Young VH
Village FE
Year FE
Province FE * Year FE
Observations
R-squared
Number of villages
Number of districts
(6)
-8.411***
(0.446)
N
N
N
Y
Y
N
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
63,519
0.369
10,665
82
63,519
0.455
10,665
82
63,519
0.476
10,665
82
63,519
0.476
10,665
82
63,519
0.477
10,665
82
63,519
0.502
10,665
82
Notes: Robust Standard errors clustered at the district level in parenthesis. The sample includes 82 districts. The unit of observation is the villageyear level. The sample includes 10,591 villages. Each column corresponds to a panel regression. The dependent variable is the age of the village
head of the corresponding village and year. The years included in all regressions are 1986, 1990, 1993, 1996, 2000 and 2003. § The number of
INPRES schools is defined in deviations from its sample mean. *** p<0.01, ** p<0.05, *p<0.1.
46
Appendix Table 4A. Robustness Checks of the Effects of the School Construction on Public Goods (Full Results)
Dependent Variables
Number of Doctors
Num Primary Schools
Baseline
(1)
post92
post_lowsch86_50
-0.039***
(0.013)
0.089***
(0.020)
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(2)
(3)
(4)
(5)
-0.037***
(0.013)
0.085***
(0.019)
-0.037***
(0.013)
0.085***
(0.018)
-0.034***
(0.012)
0.074***
(0.017)
-0.039***
(0.012)
0.089***
(0.019)
post_deng
Baseline
(6)
Number of Doctors
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(7)
(8)
(9)
(10)
-0.008
(0.008)
-0.007
(0.008)
-0.009
(0.008)
-0.007
(0.008)
-0.006
(0.008)
0.106***
(0.034)
0.101***
(0.032)
0.101***
(0.033)
0.104***
(0.034)
0.093***
(0.033)
post92_mort25_50
post92_mort50_75
post92_mort75_100
Observations
R-squared
DepVar Mean
63,546
0.960
3.297
63,543
0.960
3.297
63,546
0.960
3.297
63,546
0.961
3.297
63,546
0.960
3.297
52,920
0.740
0.159
52,917
0.743
0.159
47
(16)
post92
post_deng
-0.011
(0.020)
0.109**
(0.049)
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(17)
(18)
(19)
(20)
-0.008
(0.020)
0.094**
(0.047)
-0.015
(0.019)
0.120**
(0.049)
52,920
0.741
0.159
52,920
0.744
0.159
-0.012
(0.020)
0.107**
(0.050)
-0.010
(0.020)
0.090*
(0.050)
post92_mort25_50
post92_mort50_75
post92_mort75_100
Baseline
(21)
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(22)
(23)
(24)
(25)
-0.123***
(0.031)
-0.084***
(0.029)
-0.139***
(0.030)
-0.124***
(0.031)
-0.102***
(0.030)
0.104***
(0.032)
0.176***
(0.043)
0.234***
(0.048)
0.067**
(0.030)
0.121***
(0.042)
0.166***
(0.045)
0.111***
(0.032)
0.190***
(0.044)
0.264***
(0.051)
0.105***
(0.032)
0.175***
(0.044)
0.229***
(0.048)
0.088***
(0.031)
0.148***
(0.042)
0.188***
(0.048)
post92_fert25_50
post92_fert50_75
post92_fert75_100
Observations
R-squared
DepVar Mean
63,522
0.845
1.617
63,519
0.846
1.617
63,522
0.846
1.617
(11)
63,522
0.846
1.617
63,522
0.847
1.617
63,522
0.846
1.617
63,519
0.847
1.617
63,522
0.846
1.617
63,522
0.846
1.617
63,522
0.847
1.617
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(12)
(13)
(14)
(15)
-0.060***
(0.012)
-0.048***
(0.012)
-0.058***
(0.011)
-0.057***
(0.012)
-0.046***
(0.011)
0.054***
(0.012)
0.081***
(0.016)
0.126***
(0.021)
0.044***
(0.012)
0.065***
(0.017)
0.101***
(0.021)
0.052***
(0.012)
0.077***
(0.016)
0.116***
(0.020)
0.053***
(0.012)
0.078***
(0.016)
0.119***
(0.021)
0.043***
(0.011)
0.063***
(0.016)
0.095***
(0.021)
52,920
0.741
0.159
52,917
0.743
0.159
52,920
0.743
0.159
52,920
0.741
0.159
52,920
0.744
0.159
Dependent Variables
Number of Midwives
Number of Midwives
Baseline
52,920
0.742
0.159
Baseline
Baseline
(26)
Number of Midwives
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(27)
(28)
(29)
(30)
-0.025
(0.021)
-0.024
(0.021)
-0.024
(0.021)
-0.026
(0.022)
-0.025
(0.021)
0.076
(0.063)
0.018
(0.061)
0.197**
(0.087)
0.072
(0.062)
0.017
(0.062)
0.206**
(0.086)
0.069
(0.063)
0.004
(0.062)
0.164*
(0.085)
0.080
(0.061)
0.015
(0.061)
0.184**
(0.086)
0.077
(0.063)
0.013
(0.062)
0.187**
(0.086)
63,522
0.845
1.617
63,519
0.846
1.617
63,522
0.846
1.617
63,522
0.846
1.617
63,522
0.847
1.617
Appendix Table 4B. Robustness Checks of the Effects of the School Construction on Public Goods (Full Results)
Baseline
(17)
post92
post_deng
-0.004
(0.004)
0.032***
(0.010)
Number of Polyclinics
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(18)
(19)
(20)
(21)
-0.004
(0.004)
0.030***
(0.009)
-0.004
(0.004)
0.032***
(0.010)
-0.004
(0.003)
0.032***
(0.010)
-0.004
(0.003)
0.029***
(0.010)
post92_mort25_50
post92_mort50_75
post92_mort75_100
48
Observations
R-squared
DepVar Mean
63,545
0.673
0.0429
Baseline
(32)
post92
post92_mort25_50
post92_mort50_75
post92_mort75_100
-0.319**
(0.125)
0.120
(0.105)
0.205
(0.123)
0.497**
(0.205)
63,542
0.675
0.0429
63,540
0.673
0.0429
63,545
0.674
0.0429
63,545
0.676
0.0429
Number of Health Posts
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(33)
(34)
(35)
(36)
-0.264**
(0.122)
0.072
(0.105)
0.131
(0.126)
0.403*
(0.204)
-0.555***
(0.117)
0.286***
(0.105)
0.484***
(0.117)
1.062***
(0.226)
-0.295**
(0.128)
0.113
(0.105)
0.185
(0.130)
0.440**
(0.170)
-0.342***
(0.124)
0.143
(0.103)
0.228*
(0.119)
0.531**
(0.225)
post92_mort50_100
Baseline
(22)
Dependent Variables
Number of Polyclinics
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(23)
(24)
(25)
(26)
-0.021***
(0.004)
-0.017***
(0.004)
-0.022***
(0.004)
-0.020***
(0.004)
-0.018***
(0.004)
0.013***
(0.004)
0.023***
(0.004)
0.048***
(0.007)
0.009**
(0.004)
0.017***
(0.005)
0.040***
(0.007)
0.013***
(0.004)
0.024***
(0.004)
0.049***
(0.007)
0.012***
(0.004)
0.022***
(0.004)
0.045***
(0.006)
0.011***
(0.004)
0.019***
(0.004)
0.040***
(0.007)
63,545
0.674
0.0429
63,542
0.676
0.0429
63,540
0.674
0.0429
63,545
0.675
0.0429
63,545
0.677
0.0429
41,608
0.640
4.293
41,607
0.640
4.293
41,608
0.650
4.293
41,608
0.640
4.293
41,608
0.640
4.293
(27)
-0.131
(0.104)
0.044
(0.220)
-0.125
(0.103)
0.022
(0.219)
-0.134
(0.103)
0.130
(0.215)
-0.126
(0.105)
0.048
(0.218)
-0.135
(0.102)
0.057
(0.207)
41,608
0.639
4.293
41,607
0.640
4.293
41,608
0.648
4.293
41,608
0.640
4.293
41,608
0.640
4.293
Dependent Variables
Safe Drinking Water
Baseline
(37)
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(38)
(39)
(40)
(41)
-0.010***
(0.004)
-0.010***
(0.004)
-0.010***
(0.004)
-0.010***
(0.004)
-0.009**
(0.004)
0.017***
(0.006)
0.016***
(0.005)
0.017***
(0.006)
0.016***
(0.006)
0.014**
(0.006)
post_bdrink
Observations
R-squared
DepVar Mean
Baseline
63,546
0.740
0.0530
63,543
0.741
0.0530
63,546
0.740
0.0530
63,546
0.740
0.0530
63,546
0.742
0.0530
Number of Health Posts
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(28)
(29)
(30)
(31)
Baseline
(42)
Safe Drinking Water
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(43)
(44)
(45)
(46)
-0.006**
(0.003)
-0.006**
(0.003)
-0.006**
(0.003)
-0.006**
(0.003)
-0.007**
(0.003)
0.067**
(0.029)
0.068**
(0.028)
0.066**
(0.029)
0.067**
(0.028)
0.070**
(0.028)
63,546
0.741
0.0530
63,543
0.741
0.0530
63,546
0.741
0.0530
63,546
0.741
0.0530
63,546
0.743
0.0530
Appendix Table 4C. Robustness Checks of the Effects of the School Construction on Public Goods (Full Results)
Baseline
(47)
post92
post92_mort50_100
-0.003
(0.007)
0.001
(0.006)
Critical (polluted) land
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(48)
(49)
(50)
(51)
-0.002
(0.008)
0.000
(0.006)
0.001
(0.007)
-0.007
(0.006)
-0.003
(0.008)
0.001
(0.007)
-0.002
(0.008)
-0.000
(0.007)
post_bdrink
Observations
R-squared
DepVar Mean
63,546
0.588
0.121
63,543
0.588
0.121
63,546
0.648
0.121
63,546
0.588
0.121
63,546
0.588
0.121
Baseline
(52)
Dependent Variables
Critical (polluted) land
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(53)
(54)
(55)
(56)
0.000
(0.007)
0.000
(0.007)
-0.001
(0.007)
-0.000
(0.007)
0.000
(0.007)
-0.040**
(0.017)
-0.040**
(0.017)
-0.031**
(0.014)
-0.040**
(0.017)
-0.039**
(0.017)
63,546
0.588
0.121
63,543
0.588
0.121
63,546
0.648
0.121
63,546
0.588
0.121
63,546
0.588
0.121
Baseline
(57)
Bin Garbage Disposal
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(58)
(59)
(60)
(61)
-0.012
(0.010)
0.027**
(0.012)
-0.010
(0.010)
0.024**
(0.011)
-0.013
(0.010)
0.027**
(0.011)
-0.014
(0.010)
0.034***
(0.012)
-0.008
(0.009)
0.017
-0.011
63,546
0.802
0.565
63,543
0.803
0.565
63,546
0.838
0.565
63,546
0.803
0.565
63,546
0.806
0.565
49
Bin Garbage Disposal
Baseline
(62)
post92
post_bgarb
-0.001
(0.008)
0.082***
(0.020)
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(63)
(64)
(65)
(66)
-0.001
(0.008)
0.082***
(0.020)
-0.002
(0.007)
0.057***
(0.021)
0.000
(0.007)
0.081***
(0.020)
-0.001
(0.007)
0.078***
(0.019)
post_farkec75
Baseline
(67)
Dependent Variables
Asphalt / Hard Road
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(68)
(69)
(70)
(71)
-0.012***
(0.004)
-0.012***
(0.004)
-0.003
(0.003)
-0.012***
(0.004)
-0.010***
(0.004)
0.044***
(0.008)
0.044***
(0.008)
0.014***
(0.004)
0.045***
(0.008)
0.037***
(0.008)
post_broad
Observations
R-squared
DepVar Mean
63,546
0.803
0.565
63,543
0.803
0.565
63,546
0.838
0.565
63,546
0.804
0.565
63,546
0.807
0.565
63,480
0.579
0.924
63,477
0.580
0.924
63,411
0.813
0.924
63,480
0.580
0.924
63,480
0.582
0.924
Baseline
(72)
Asphalt / Hard Road
INPRES
Pre-treatment
Controlling Pre-treat Dep
schools * Year Covariates
for population Var*Year FE
FE
* Year FE
(73)
(74)
(75)
(76)
-0.024***
(0.004)
-0.024***
(0.004)
-0.003
(0.002)
-0.024***
(0.004)
-0.024***
(0.004)
0.235***
(0.016)
0.235***
(0.016)
0.039***
(0.007)
0.235***
(0.016)
0.232***
(0.015)
63,480
0.596
0.924
63,477
0.596
0.924
63,411
0.813
0.924
63,480
0.597
0.924
63,480
0.598
0.924
Appendix Table 5. The Effects of VH Education on Public Goods
(1)
Number of
Kindergarten
Dep. Var. Mean
(2)
(3)
Dependent Variables:
Number of Primary
Number of High
Schools
Schools
1.229
3.299
0.428
(4)
Number of doctors
0.158
Panel A. Reduced Form
post 1st Election after 1992
post * Num INPRES schools§
R-squared
-0.005
(0.009)
0.001
(0.008)
0.002
(0.006)
-0.000
(0.008)
0.048***
(0.011)
0.069***
(0.015)
0.025***
(0.005)
0.019***
(0.006)
0.892
0.960
0.914
0.740
Panel B. OLS
years of educ VH
R-squared
0.0017
(0.001)
0.0000
(0.001)
0.0012*
(0.001)
0.0005
(0.001)
0.891
0.960
0.913
0.740
0.037***
(0.009)
0.027*
(0.016)
63,546
10,591
22.94
52,920
10,584
11.10
Panel C. 2SLS
years of educ VH
0.056***
(0.020)
0.093***
(0.025)
Observations
63,546
63,546
Number of villages
10,591
10,591
Cragg-Donald F-Stat
22.94
22.94
§ The number of INPRES schools is defined in deviations from its sample mean.
50
Appendix Table 6A. Heterogeneous Effects of VH Education on Public Good Provision (2SLS Results)
Num Primary Sch
(1)
(2)
Dep Var Mean
years of educ VH
Number of Doctors
(3)
(4)
(5)
(6)
Dependent Variables:
Number of Midwives
(7)
(8)
(9)
Number of Health Posts
(13)
(14)
(15)
3.30
3.30
0.16
0.16
0.16
1.62
1.62
1.62
1.62
0.04
0.04
0.04
4.30
4.30
4.30
0.0000
(0.001)
-0.005**
(0.002)
0.010***
(0.003)
0.0005
(0.001)
-0.001
(0.001)
-0.005***
(0.001)
-0.0004
(0.002)
-0.002
(0.002)
-0.018***
(0.004)
0.007***
(0.002)
-0.0003
(0.000)
-0.000
(0.000)
-0.002***
(0.000)
0.0142
(0.009)
0.014
(0.010)
-0.006
(0.018)
edu*low schooling rate
edu*infectious disease
0.013***
(0.004)
51
edu*mortality 25th-50th
0.018**
(0.008)
0.007***
(0.002)
0.007***
(0.002)
0.011***
(0.003)
edu*mortality 50th-75th
edu*mortality >75th
0.001
(0.001)
0.015**
(0.006)
0.026***
(0.007)
0.033***
(0.007)
edu*fertility 25th-50th
-0.001
(0.039)
0.001
(0.001)
0.003***
(0.001)
0.004***
(0.001)
0.007
(0.020)
0.022
(0.025)
0.062*
(0.035)
-0.010***
(0.003)
-0.020***
(0.004)
-0.028***
(0.005)
edu*fertility 50th-75th
edu*fertility >75th
Observations
R-squared
Number of villages
Number of Polyclinics
(10)
(11)
(12)
63,546
0.960
10,656
63,546
0.960
10,656
52,920
0.740
10,649
52,920
0.740
10,649
52,920
0.740
10,649
63,522
0.845
10,652
63,522
0.845
10,652
63,522
0.845
10,652
63,522
0.846
10,652
63,545
0.673
10,656
63,545
0.673
10,656
63,545
0.673
10,656
41,608
0.639
10,461
41,608
0.639
10,461
41,608
0.639
10,461
Appendix Table 6B. Heterogeneous Effects of VH Education on Public Good Provision (2SLS Results)
Dependent Variables:
Safe Drinking Water
(1)
(2)
(3)
Dep Var Mean
years of educ VH
Critical (polluted) land
(4)
(5)
(6)
(7)
Bin Garbage Disposal
(8)
(9)
(10)
Asphalt / Hard Road
(11)
(12)
52
0.05
0.05
0.05
0.12
0.12
0.12
0.57
0.57
0.57
0.92
0.92
0.92
-0.001
(0.000)
-0.002***
(0.000)
0.002***
(0.001)
-0.001*
(0.000)
0.001
(0.001)
0.001
(0.001)
-0.001
(0.001)
0.001
(0.001)
0.001
(0.001)
-0.000
(0.001)
0.002
(0.002)
0.012***
(0.001)
-0.001
(0.001)
-0.004***
(0.001)
-0.001**
(0.001)
edu*mortality > 50th
edu*bad baseline service
0.003
(0.002)
-0.005**
(0.002)
-0.025***
(0.001)
0.027***
(0.003)
edu*far from subdistrict
Observations
R-squared
Number of villages
0.003*
(0.001)
63,546
0.740
10,591
63,546
0.740
10,591
63,546
0.740
10,591
63,546
0.588
10,591
63,546
0.588
10,591
63,546
0.588
10,591
63,546
0.802
10,591
63,546
0.802
10,591
63,546
0.820
10,591
63,480
0.578
10,591
63,480
0.584
10,591
63,480
0.578
10,591
Appendix Table 7. Robustness Checks of the Effects of the VH Education on Public Goods
(2SLS Results)
Table 7. Robustness Checks to Alternative Econometric Specifications (2SLS results)
(1)
(2)
(3)
Different Set of Controls:
(4)
Baseline
Specification
Controlling for
population
Number of Primary Schools
(interaction with low enrolment)
0.054***
(0.010)
0.051***
(0.010)
0.051***
(0.009)
0.045***
(0.009)
0.049***
(0.010)
Number of Doctors
(interaction with dengue outbreak)
0.064***
(0.018)
0.061***
(0.017)
0.061***
(0.018)
0.062***
(0.018)
0.054***
(0.018)
Number of Polyclinics
(interaction with dengue outbreak)
0.019***
(0.006)
0.018***
(0.005)
0.019***
(0.006)
0.015***
(0.005)
0.018***
(0.006)
Number of Midwives
(interaction with dengue outbreak)
0.066**
(0.028)
0.057**
(0.026)
0.073***
(0.027)
0.065**
(0.028)
0.062**
(0.030)
Number of Midwives
(interaction with >75th perc fertility)
0.038***
(0.013)
0.040***
(0.013)
0.033**
(0.013)
0.037***
(0.013)
0.039***
(0.015)
Purified Water for drinking
(interaction with >75th perc mortality)
0.010***
(0.003)
0.010***
(0.003)
0.010***
(0.003)
0.010***
(0.003)
0.008**
(0.003)
Purified Water for drinking
(interaction with bad pre-treatment service)
0.035**
(0.017)
0.036**
(0.017)
0.035**
(0.017)
0.035**
(0.017)
0.037**
(0.018)
Critical Land endangering Water Sources
(interaction with bad pre-treatment service)
-0.020**
(0.009)
-0.020**
(0.009)
-0.015**
(0.007)
-0.020**
(0.009)
-0.013*
(0.008)
Garbage bin
(interaction with bad pre-treatment service)
0.008***
(0.002)
0.008***
(0.002)
0.006***
(0.002)
0.008***
(0.002)
0.008***
(0.002)
Asphalt/Hardened road
(interacted with distance to the subdistrict)
0.026***
(0.005)
0.026***
(0.005)
0.008***
(0.002)
0.027***
(0.005)
0.021***
(0.005)
Asphalt/Hardened road
(interaction with bad pre-treatment service)
0.144***
(0.018)
0.144***
(0.018)
0.025***
(0.006)
0.145***
(0.018)
0.135***
(0.017)
Width of the main road
(interacted with distance to the subdistrict)
0.036***
(0.004)
0.036***
(0.004)
0.003**
(0.002)
0.037***
(0.004)
0.031***
(0.004)
Width of the main road
(interaction with bad pre-treatment service)
0.258***
(0.029)
0.258***
(0.029)
0.147
(0.129)
0.258***
(0.029)
0.260***
(0.032)
Pre-treatment Dep INPRES schools
Var*Year FE
* Year FE
(5)
Pre-treatment
Covariates
* Year FE
Notes: Robust Standard errors clustered at the district level in parenthesis, except for column 6 where robust standard errors are reported.
The unit of observation is the village-year level. Each cell corresponds to the 2SLS coefficient on village head years of education of a
different specification where the dependent variable is defined by each row and the set of controls included is defined by each column. All
regressions include village fixed effects, year fixed effects and province-times-year fixed effects. *** p<0.01, ** p<0.05, * p<0.1.
53
Appendix Table 8A. Heterogeneous Effects of VH Education by Method of Selection of the VH (Full Results)
Num Primary Sch
Desa
Kelur
Dep Var Mean
years of educ VH
edu*low schooling rate
Desa
Number of Doctors
Kelur
Desa
Kelur
Desa
Dependent Variables:
Number of Midwives
Kelur
Desa
Kelur
Number of Polyclinics
Kelur
Desa
Kelur
Desa
Number of Health Posts
Kelur
Desa
Kelur
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
(13)
(14)
(15)
(16)
(17)
(18)
3.331
3.331
0.227
0.227
0.227
0.227
1.737
1.737
1.737
1.737
0.0521
0.0521
0.0521
0.0521
4.346
4.346
4.346
4.346
-0.024
(0.016)
0.054***
(0.010)
0.030
(0.019)
0.016
(0.027)
-0.010
(0.020)
0.148***
(0.038)
-0.047**
(0.019)
-0.094
(0.092)
-0.076**
(0.036)
0.107
(0.077)
-0.001
(0.034)
0.210***
(0.069)
0.026*
(0.014)
-0.184*
(0.111)
-0.096
(0.106)
-0.312***
(0.107)
-0.011
(0.154)
0.062***
(0.018)
0.177*
(0.104)
0.038
(0.165)
0.270
(0.186)
0.072
(0.075)
0.128
(0.096)
0.397***
(0.147)
-0.113
(0.147)
-0.072
(0.151)
0.072
(0.202)
43,188
10,797
15.24
43,188
10,797
15.24
edu*infectious disease
54
edu*mortality 25th-50th
0.028*** 0.266**
(0.006)
(0.125)
0.044*** 0.321***
(0.009)
(0.114)
0.071*** 0.340***
(0.011)
(0.105)
edu*mortality 50th-75th
edu*mortality >75th
0.058***
(0.017)
0.104***
(0.026)
0.138***
(0.027)
edu*fertility 50th-75th
edu*fertility >75th
66,042
11,007
12.29
66,042
11,007
12.29
54,965
10,993
5.854
54,965
10,993
5.854
54,965
10,993
2.895
54,965
10,993
2.895
65,976
10,996
12.18
-0.012* 0.048*** -0.022***
(0.007)
(0.010)
(0.006)
0.019***
0.015
(0.006)
(0.015)
-0.055
(0.105)
0.230**
(0.091)
0.202*
(0.123)
edu*fertility 25th-50th
Observations
Number of villages
Cragg-Donald F-Stat
Desa
65,976
10,996
12.18
0.017
(0.010)
0.013
(0.010)
0.032**
(0.013)
0.017
(0.022)
0.006
(0.013)
-0.009
(0.025)
65,976
10,996
6.028
65,976
10,996
6.028
66,041
11,007
5.954
66,041
11,007
5.954
0.006**
(0.002)
0.012***
(0.003)
0.028***
(0.004)
0.014
(0.021)
0.035
(0.022)
0.031
(0.021)
66,041
11,007
5.954
66,041
11,007
5.954
43,188
10,797
30.75
43,188
10,797
30.75
Appendix Table 8B. Heterogeneous Effects of VH Education by Method of Selection of the VH (Full Results)
Table 8B. The Effect of Village Head on Access to Basic Services (2SLS results)
(1)
Desa
Dep Var Mean
55
years of educ VH
edu*mortality >50th
(2)
Safe Drinking Water
Kelur
Desa
(2)
(4)
Kelur
Desa
(4)
(5)
Dependent Variables:
Critical (polluted) land
Kelur
Desa
(5)
(6)
Kelur
Desa
(7)
(8)
Bin Garbage Disposal
Kelur
Desa
(9)
(1)
Kelur
Desa
0.0692
0.0692
0.0692
0.0692
0.0692
0.118
0.118
0.118
0.118
0.574
0.574
0.574
0.574
-0.008
(0.005)
0.010***
(0.003)
0.074***
(0.017)
-0.011
(0.016)
-0.007
(0.005)
0.062***
(0.011)
-0.005
(0.011)
0.000
(0.004)
-0.003
(0.010)
0.000
(0.007)
-0.003
(0.011)
-0.003
(0.008)
-0.013
(0.013)
0.016**
(0.007)
-0.047**
(0.019)
-0.019
(0.018)
-0.006
(0.014)
-0.052***
(0.019)
0.035**
(0.017)
0.200***
(0.053)
-0.020**
(0.009)
-0.003
(0.011)
0.008***
(0.002)
0.018***
(0.003)
66,042
11,007
11.85
66,042
11,007
11.85
66,042
11,007
11.85
66,042
11,007
11.85
66,042
11,007
12.14
66,042
11,007
12.14
edu*bad baseline service
Observations
Number of villages
Cragg-Donald F-Stat
(1)
66,042
11,007
12.07
66,042
11,007
12.07
66,042
11,007
12.07
66,042
11,007
12.07
66,042
11,007
12.07
66,042
11,007
12.07
(2)
Asphalt / Hard Road
Kelur
Desa
0.0692
-0.006 -0.030***
(0.006)
(0.005)
0.026***
0.009
(0.005)
(0.011)
65,970
11,007
11.83
(1)
65,970
11,007
11.83
(2)
Kelur
0.0692
0.0692
-0.006
(0.007)
-0.024***
(0.006)
0.036***
(0.004)
0.003
(0.003)
65,970
11,007
11.83
65,970
11,007
11.83