Educational Assistance and Education Quality in in

Educational Assistance and
Education Quality in
Indonesia: The Role of
Decentralization
VIRGI A. SARI
THE EQUITABLE PROVISION of quality education is important for growth and

inclusive development. Human capital facilitates the creation of knowledge,
fosters innovation, and enhances productivity. Its importance for growth
has been acknowledged since work such as that of Becker (1962) and Romer
(1990). In addition, education is the key factor in explaining gaps in eco-
nomic inclusion. For example, returns to education have been conclusively
shown to have important implications for earnings and broader aspects of
labor market inequality (Autor 2014). Declining returns to education and
skills mismatch in the labor market contribute to lower relative growth in
the average earnings of low-income workers over time. Both returns to ed-
ucation and skills mismatch are linked not only to trends on the demand
side of the labor market, but also to the quality of the education sector; thus,
it is essential to study the provision of high-quality education. The quality of
education is a highly politicized issue. Improving access to high-quality edu-
cation has continued to be a pressing agenda for governments in developing
economies, where the challenge has shifted from increasing the quantity of
education to improving its quality. Inequality in learning outcomes persists:
the 75th percentile of children in developing countries performs less well
than the 25th percentile of children in Organization for Economic Coop-
eration and Development (OECD) countries, and more than 60 percent of
primary school graduates leave school without basic competencies (World
Bank 2018, 6–8). Achieving equality in quality education is now the fourth
goal of the Sustainable Development Goals (SDGs).
Nevertheless, gaps remain in our understanding of what works best to
improve learning outcomes. Studies have examined many education ini-
tiatives to identify best practices for improving learning outcomes in devel-
oping countries (Ganimian and Murnane 2016; Masino and Niño-Zarazúa
2016; García and Saavedra 2017). Education initiatives have had a posi-
tive effect on learning outcomes when designed around at least two deter-
mining factors of quality of learning and when the program design takes
local norms into account (Masino and Niño-Zarazúa 2016). However, poli-
tics also matter in setting the quality of education. The multiple interests of
P O P U L AT I O N A N D D E V E L O P M E N T R E V I E W 4 5 ( S 1 ) : 1 2 3 – 1 5 4 ( D E C E M B E R 2 0 1 9 ) 123

C 2019 UNU-WIDER. Population and Development Review published by Wiley Periodicals, Inc.
This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-ShareAlike License, which
permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and the
content is offered under identical terms.
124 EDUCATIONAL ASSISTANCE AND EDUCATION QUALITY IN INDONESIA
political actors may jeopardize an education system, eventually contributing

to a low quality of learning outcomes (World Bank 2018, 189–195). Clearly,
it is crucial to examine the extent of how institutions and local norms and
conditions affect the delivery of education interventions, and thereafter ed-
ucation quality.
Various education initiatives and policies have aimed to improve edu-
cation outcomes. Among others, educational assistance, here broadly de-
fined as any in-kind or cash transfers for educational purposes, plays a
key role in improving education outcomes by directly benefiting the key
actors—students, schools, parents—and addressing educational challenges
from both the demand and the supply sides. Programs such as scholar-
ships for the poor, school vouchers, and subsidies for obtaining school re-
sources help to improve participation in schooling by reducing the asso-
ciated costs (Sparrow 2007; Afridi 2011; Filmer and Schady 2014). More
importantly, some programs focus on preparing students from early child-
hood by improving nutrition. Education has also become the main element
of human development conditional cash transfer (CCT) programs such as
Brazil’s Bolsa Familia, Indonesia’s Program Keluarga Harapan, and Mexico’s
Oportunidades, which aim to improve health and education outcomes. Nev-
ertheless, the results of such interventions are often mixed and heteroge-
neous, depending on program design (García and Saavedra 2017). The links
between educational assistance and learning outcomes can also be weak
(World Bank 2013).
This article aims to enhance understanding of the mechanisms through
which decentralization—the transfer of certain functions of central to
smaller independent government units (Faguet and Sanchez 2008, 1296)—
affects education quality by unpacking the role that a decentralized educa-
tion system can play in facilitating progress in education quality through
the provision of educational assistance. The analysis focuses on both direct
and indirect impact channels. I investigate several indirect channels, with
a particular focus on the provision of educational assistance, which is often
provided at the level of subnational governments (e.g., municipalities). In
the context of education, a decentralized education system has been found
to improve education by adapting it to local educational needs (Di Gropello
and Marshall 2011) and by enhancing the sense of responsibility across ac-
tors (De Grauw 2005). In relation to this, the effect of educational assis-
tance depends on the capacity of the local institutions that administer the
program. Thus, this article investigates the extent to which decentraliza-
tion affects education quality directly and through the improved provision
of educational assistance. It raises the question: “What works to improve
education quality?” More precisely, the question is: “What is the mecha-
nism through which educational assistance and a decentralized education
system work together toward changing students’ learning outcomes within
a fragmented society and educational system?”
V I R G I A. S A R I 125
Indonesia provides a relevant case to study educational assistance and

changes in the quality of education within the context of a decentralized
educational system. Within the last two decades, the country has success-
fully recovered from the Asian financial crisis, joined the ranks of lower
middle income countries, and halved its poverty rate. Nonetheless, inequal-
ities in the quality and quantity of education remain acute, particularly in
remote and rural areas (World Bank 2013, 13), despite the tripling of gov-
ernment expenditures devoted to the education sector since 2000 (World
Bank 2017). Educational assistance is central to addressing this problem;
more important, it has historically been designed as a part of key govern-
ment antipoverty policies. Educational assistance accounts for the largest
share of total government social assistance expenditures (Yusuf and Sum-
ner 2015, 344–345). In addition, the Indonesian government undertook
major reforms by carrying out decentralization in 2001. The rapid nation-
wide implementation of the system transformed the previously highly cen-
tralized government structure and has resulted in the transfer of education
management from the central government to independent district educa-
tion offices. The extent to which educational assistance leads to changes in
education quality thus depends on the capacity of this changing institutional
framework.
Most importantly, Indonesia is unique in terms of how the educational
system has developed historically, particularly with respect to the role of
private education. The educational system has developed under the dom-
inant influence of religious and local principles and of political interests
(Suratno 2014). In contrast to the developed world, private education plays
an important role in closing the education gap by providing education to
low-income and marginalized populations (Bangay 2005, 170–171). Private
education provides close to 40 percent of lower and upper secondary school-
ing and continues to be in demand because it is provided mainly by com-
munities themselves and thus is influenced by their particular ethnic or
religious orientation (Stern and Smith 2016).
The study described here set as its objective to evaluate the effective-
ness of both structural and pragmatic policies in achieving inclusive progress
in education quality. To achieve this objective, I examined the evolution
of educational assistance in Indonesia, assessing it as an indirect mech-
anism through which a decentralized education system affects the qual-
ity of education. I argue that it affects education quality by influencing
the provision of educational assistance to public and private schools, and
that the institutional feature of education provision, which is subject to lo-
cal norms, is an important mechanism behind it. To test this, I used in-
formation on education outcomes and characteristics of schools and their
localities from a longitudinal study of the Indonesian Family Life Survey
(IFLS) and analyzed the impact of decentralization on education quality
in Indonesia, as measured by average student performance. The natural
experiment of the implementation of government decentralization was ex-

ploited to compute difference-in-difference (DID) estimators using standard
DID and semi-parametric DID.
Research context
Decentralization, educational assistance,
and education quality
In a decentralized educational system, certain authorities for certain tasks

(e.g., teaching, hiring, curriculum design, and school-based management)
are delegated from the central government to the district education office.
The debate on the impact of decentralization on education quality remains
open. Conceptually, decentralization can positively affect education qual-
ity through increased government spending (Falch and Fisher 2012, 278)
and better allocation of education spending as a result of local institutions’
increased autonomy. Advocates of decentralization argue that local insti-
tutions, such as the district education office and the schools themselves,
have better knowledge about local demands and needs (De Grauwe 2005;
Di Gropello and Marshall 2011). However, decentralization can also nega-
tively affect learning outcomes by increasing inefficiency in the delivery of
public services. The main critique of decentralization and public service de-
livery is that it is prone to corruption and rent-seeking behaviors (Mueller
2004; Cheikbossian 2008), and the magnitude of this impact depends on
the capacity of local institutions (Crook and Sverrisson 1999).
Because empirical evidence is scarce, the link between decentralized
education and learning outcomes is inconclusive. Using panel data on in-
ternational test scores for 25 OECD countries, Falch and Fischer (2012,
277–278) found that increased spending following decentralization had a
positive impact on learning outcomes. In contrast, several country-specific
studies have highlighted the negative impact of decentralization on educa-
tion outcomes (Merrouche 2007; Leer 2016). Thus, further in-depth inves-
tigation into whether and how decentralization affects education quality in
developing countries is called for.
Educational assistance
In addition to decentralization, educational assistance has recently become

an important tool for improving various education outcomes, including
school attendance and learning outcomes. However, international evidence
of the impact of educational assistance in improving education quality is
still scarce and remains inconclusive, with a large bias toward assessment of
participation rather than of learning outcomes. A study evaluating the im-
pact of a program of scholarships for the poor in Cambodia using a regres-
sion discontinuity design (RDD) found that the program increased school
attainment but did not improve students’ test scores in the medium term
(Filmer and Schady 2014, 683–684). In Colombia, a CCT program con-
ditioned on children’s school attendance (Familias en Accion) was found
to increase school attainment, yet beneficiary and nonbeneficiary children
turned out to have scored similarly overall and in language and mathemat-
ics (Baez and Camacho 2011, 24–25). Interestingly, if combined with the
right incentives, the same program was found to raise children’s learning
outcomes in the long term, following eight to 12 years of program expo-
sure (Barrera-Osorio, Linden, and Saavedra 2017).
Similarly, Ganimian and Murnane (2016) reviewed more than 200
impact evaluation studies on different sets of education initiatives and con-
cluded that simply reducing the cost of education for students did not lead to
higher levels of achievement. Translating such initiatives into better student
performance required components such as providing incentives to teachers,
parents, and communities and putting a monitoring system in place (Gan-
imian and Murnane 2016, 739). In sum, the research stresses the weak
link between educational assistance and learning outcomes and highlights
the importance of institutional features in determining the success of such
programs.
The study described here aimed to provide further clarity on the mech-
anism by which decentralized education affects education outcomes. The
research focused on assessing the transmission mechanism and investigat-
ing the role of institutions—the district education office and the schools,
if combined with the right incentives—in determining the size and magni-
tude of the effect of decentralized education and educational assistance on
learning outcomes. I examined three important mechanisms and tested the
following:
(1) The role of educational assistance: The study tested whether decentraliza-
tion affected education quality through increased access to educational
assistance.
(2) School management: The study tested (i) whether decentralization af-
fected education quality through improved effort on the part of teachers
and (ii) the heterogeneity of the impact of decentralization according
to how the school principal was elected.
(3) Localities and social norms: The study tested whether decentralization af-
fected education quality more among public schools if the school was
located in an active community.
Education quality and policies in Indonesia
Education remains important for Indonesia as a means to progress out of

poverty. Lack of education plays a key role in explaining the large share
of working poor in Indonesia’s population (Priebe, Howell, and Sari 2014,
23). Evening out the quality of education at all levels is important in solving
the skills shortage and to achieving the country’s aspiration of avoiding the
“middle-income trap” (World Bank 2014).
Nevertheless, there has been only slow progress in improving edu-
cation quality, despite the country’s achievement in expanding access to
education. The secular decline in returns to schooling between 1993 and
2007 reflects the fact that quality of education does not respond well to
market demand (Purnastuti, Miller, and Salim 2013) and that the quality
of education in Indonesia is comparatively lower than in other Asian na-
tions and in other less-developed countries (Purnastuti, Salim, and Joarder
2015, 191). Learning outcomes have also been on a downward trajectory
compared with global outcomes. Of 70 countries surveyed, Vietnam ranked
eighth in PISA science scores, while Indonesia ranked 62nd (OECD 2016).
Moreover, unequal access to good-quality education and unequal distribu-
tion of schools and high-quality teachers remain evident, to the particular
disadvantage of rural and remote areas (World Bank 2013, 13), where pri-
vate schools must fill the gaps by providing education in poorer, rural, and
remote areas (Bangay 2005, 170–171).
Since independence in 1945, the government of Indonesia has been
aware of the importance of providing equal access and educational ex-
perience across the regions and has continued to address these problems
through educational reforms and increased financial resources. It is impor-
tant to note the shift in strategy from providing hard infrastructure during
the New Order era (1996–1998), such as building schools in villages, to di-
rectly targeting schools, teachers, and students—aiming to not only increase
access to schooling, but also improve its quality. The share of education-
related spending among total government expenditures tripled between
2000 and 2013, with education becoming the second largest area of spend-
ing after government administration (see Figure 1).
Educational assistance is a key element in improving access to and
quality of education and is currently at the heart of social assistance pro-
grams in Indonesia. The allocated budget for educational assistance has con-
stituted the largest share of total government expenditure for targeted social
programs for a little over two decades, surpassing that of health, village, and
community programs (Yusuf and Sumner 2015, 344–345). This assistance
is mainly provided in the form of direct cash transfers to schools and stu-
dents, such as school subsidies (JPS), School Operational Assistance (BOS),
scholarships for the poor (BSM), and cash transfers for children, as part of
the Family Cash Transfer Program (PKH).
Despite the expansion of spending in the last two decades, however,
evidence of the impact of educational assistance in improving education
quality in Indonesia is like global evidence discussed above—scarce and
inconclusive. There is a large bias toward assessment of participation in
schooling (attendance or attainment) rather than of learning outcomes.
A study using household information from survey data on 100 villages in
FIGURE 1 Trends in government expenditures, by sector (2001–2013)
SOURCE: Author’s illustration, based on the Indonesia Consolidated Fiscal Dataset (COFIS) (World Bank 2017).
Indonesia found that the public scholarship program was effective in sup-
pressing the drop-out rate among junior secondary school students by 3
percent during the financial crisis period (Cameron 2009, 314). Regression-
based estimates from World Bank Jakarta (2014, 54–55) showed evidence
that following the School Operational Assistance, schooling participation at
the junior secondary school level increased. The impact was pronounced
among students in the lowest income distribution, although no strong link
was found in terms of improvement in school transition. On the other hand,
a quasi-experimental study based on household survey data found that a
school assistance program during the oil crisis between 2003 and 2005 was
not effective in reducing the drop-out rate (Kharisma 2016, 10–12).
Decentralization of the educational system in Indonesia began in 2001,
as a part of the “big bang” decentralization across government sectors, and
it led to major changes in education management throughout the coun-
try. What makes Indonesia a unique case study is that the decentralization
occurred on a large scale and in all parts of the country simultaneously,
transforming the previous constituencies into 34 provinces and 413 dis-
tricts (Nasution 2016, 4). Currently, the government consists of five levels:
the central government, the provincial government, the kabupaten (district),
the kecamatan (subdistrict), the kota (municipality). The main change as a
result of decentralization is the transfer of authority and fiscal responsibility
from the central government to districts and municipalities, which are now
responsible for managing their own budget expenditures and for providing
public goods, including education (Nasution 2016, 7).1
Prior to decentralization, the education sector was highly centralized,
with the Ministry of Education (MoEC) and the Ministry of Religious Af-
fairs (MoRA) being responsible for both the planning and management of
the national education system, as well as for its financing. Public school
teachers were thus central government employees, rather than being cho-
sen by subnational governments or local communities. In contrast, follow-
ing decentralization, the district offices now have the authority to allocate
and design local education policies, including managing education finance,
designing local regulatory frameworks, developing curricula at the school
level, and providing early childhood, basic, and secondary education (World
Bank 2013). Thus, an assessment of changes in education quality in gen-
eral needs to take into account the heterogeneity in the capacity of district
education offices in managing these additional resources.
This is particularly so in the case of the transfer of responsibility for
educational assistance. Agustina et al. (2009, 25) illustrated the role of in-
stitutions in managing public scholarship programs, whereby districts and
head teachers play a significant role in targeting and allocating beneficiaries
at the school level. While the quota for each province is determined at the
central level (by the MoEC and MoRA), the school and the district decide
who will receive the resources. Often, the school principal is in charge of
disbursing funds when students or parents are not available at the right
time to receive the educational assistance directly. Therefore, the role of
institutions—schools and the district education office—along with the
community should not be overlooked when assessing overall changes in
education quality and the role of educational assistance, particularly within
the context of a decentralized educational system. This is particularly rele-
vant in a heavily populated and geographically challenging country such as
Indonesia, in which local institutions—district education offices and local
schools—are often responsible for administering government programs.
Earlier studies have investigated the heterogeneous effect of decen-
tralization in Indonesia, such as by urban-rural area and engagement of
school committee (e.g., Leer 2016). However, they shed limited light on the
role of institutions (i.e., district offices and school officials) and local norms
in allocating the increased education financing and deciding on the type of
school that receives them. The research described here aims to fill the gap
by providing greater clarity and a more nuanced assessment concerning
the conditions through which decentralization and educational assistance
can affect educational outcomes in public and private schools. I have done
this by examining the role of educational assistance, school efforts, and
communities as the mediating factor through which decentralization affects
educational quality.
TABLE 1 Sample selection from IFLS 2000 and 2007

No. of schools Full sample DID sample SDID sample
Public 2,366 2,062 619
Private 2,404 2,036 145
Total 4,944 4,272 764
SOURCE: Author’s calculations, based on IFLS 2000 and 2007 rounds.
Data and estimation strategy

Data
The analysis presented here is based on the Indonesian Family Life Survey
(IFLS), a multipurpose, ongoing longitudinal survey that provides in-depth
information on the economic and noneconomic well-being of individuals,
households, and communities.2 The IFLS collects repeated cross-sectional
information on community facilities such as health and education facilities.
Its school module draws information on three levels of schooling—primary,
junior secondary, and higher secondary. Originally, the IFLS collected data
on nine schools in each of the 321 enumeration (community) areas. The
schools were randomly selected from a sample drawn on households’
knowledge about school facilities. Within each school, a further random
sample was drawn to obtain information on students’ test scores. The
present study used information on school and community characteristics.
This article aims to test which mechanism—direct or indirect—
explains the impact of decentralization on education quality. For indirect
impact channels, of particular interest is the possible misallocation of ed-
ucation financing and reduced school and community efforts after decen-
tralization. Thus, the analysis focuses on the role of educational assistance,
school localities, and social norms in affecting the direction and magnitude
of decentralization in determining improvement in education quality. This
approach required information not only about the schools, but also about
the characteristics of the localities (villages and communities). This provides
the justification for the use of IFLS data, as that survey collects in-depth in-
formation on school, village, and community characteristics, which allows
for testing the hypothesis. IFLS school modules collect data on students’
test scores, school facilities, teachers’ education and experience, and school
decision-making in financing, as well as—most important—in-depth infor-
mation on the type and amount of educational assistance received by the
school.
To estimate impact, I compared two groups: public schools (the treat-
ment group) and private schools (the control group).3 Table 1 shows the
number of observations included in the analyses for each of the estima-
tion strategies applied. The unit of analysis in this article is the school.
The observations included all public and private schools. We excluded
observations with extreme values in terms of educational outcomes. We
followed standard measurements of educational outcomes—i.e., students’

performance in test scores. The IFLS collected information on students’ test
scores in national examinations for mathematics and language subjects.
The information was collected from a random sample of students for each
of the schools surveyed. Our key dependent variables for educational
outcomes are thus the average student score in mathematics and language
tests in the national examination at the school level.
The aim here is to understand whether decentralization affected the
quality of education, particularly among the public schools affected by the
policy. To estimate the effect, information was needed not only on educa-
tional outcomes when the public schools received the treatment, but also
on outcomes in the absence of decentralization. Essentially, the effect of de-
centralization would be the difference in the average of the two. However,
the challenge is that the latter information is not observed. Thus, estimat-
ing the effect would involve a measurement of information that one never
observes. To counter this, I have employed two methods popular in the lit-
erature of policy evaluation studies.
I exploited the “natural experiment” of decentralization, which oc-
curred in 2001, by sampling a repeated cross-section of data on schooling
and education outcomes and applying two estimation methods to estimate
the impact of decentralization on education quality in Indonesia: standard
DID, as illustrated in Athey and Imbens (2006), and semi-parametric DID
(SDID), as demonstrated in Abadie (2005).
Estimating students’ achievement
First, the direct impact of decentralization on education quality of schools

was estimated; later, the mechanisms behind the impact were studied. This
strategy for estimating the effect of decentralization on education quality in
Indonesia is similar to that illustrated in Leer (2016). Education quality is
measured in terms of the average test scores in language and mathematics.
Education quality is determined by the following function:

Yspt = α0 + α1 Publicspt + α2 Publicspt × PostDecent + Xspt + Zspt + νspt (1)
where Y is the average outcome variable of school s in province p at year

t, Public is a binary variable coded 1 for public school as the treatment
group and Public × PostDecent is the variable of interest, with α 2 capturing
the impact of decentralization on the education quality of public schools
relative to private schools. X is a vector of covariates influencing the out-
come variables. Geographic variations—urban-rural, and school and village
characteristics—were controlled for. Appendix Table A1 lists and describes
the variables. Standard errors are clustered at the unit of analysis—the
school. Using DID and SDID, I also estimate the indirect mechanism, in-
vestigating the heterogeneity effect of decentralization in different settings
(active vs. passive communities) and in schools using different ways of se-
lecting the principal.
It is important to note that the function of education quality was mod-
ified and extended to fit the main purpose of the study. First, different def-
initions of public versus private schools were applied: I did not incorporate
schools under the MoRA within the control group. The reason is that MoRA
schools have never been decentralized to date. After decentralization, ed-
ucation financing has increased among public schools (under both MoEC
and MoRA administration). Thus, categorizing public religious schools as
private schools (as in Leer 2016) can potentially contaminate the control
group, leading to biased DID estimators. Second, as briefly discussed ear-
lier, school localities and local norms matter in determining the direction
and magnitude of decentralization’s impact on education quality. Hence, I
incorporated the two vectors of covariates (X and Z ) into equation (1), as
they have been overlooked in earlier studies.
Difference-in-difference estimation
DID estimation is a popular method in the applied economic literature for

measuring the impact of a policy change (Heckman, Lalonde, and Smith
1999, 30). It measures the effect of a policy or an intervention by comparing
the changes in the outcome of interest in the treatment and control group
before and after the intervention. This method is feasible for use with the
data, as the education outcome variables were observed over time.
Using DID, I estimated the average treatment effect—in this case, the
impact of decentralization on average education outcomes of students in
public schools relative to private schools (the control group). The parameter
of interest is the interaction between the treatment group (a dummy; 1 is
equal to public schools) and the treatment period variable (a dummy; 1 is
equal to the postdecentralization period). To ensure the causality of the DID
estimate, certain assumptions need to be satisfied, namely that the parallel
trend and the treatment effect are constant across groups and across period.
I followed the set-up of linear DID for a continuous outcome variable as
illustrated in Athey and Imbens (2006) and Puhani (2012, 86).
The equation below captures the treatment effect:
ϕ = E [Y1 |PD = 1, D = 1, X ] − E[Y0 |PD = 1, D = 1, X ] (2)
where Y1 represents the potential outcomes if treated and Y0
the potential outcomes if not treated, PD is a binary variable
indicating coded 1 if the observation is post-decentralization and 0 if not,
D is a binary variable indicating treatment status coded 1 if the observation
receives treatment and 0 otherwise, and X is a vector of covariates.
Enrollment in the treatment is then specified as:
I = 1 [PD = 1, D = 1] = PD × D (3)
where D = 1 indicates that DID in the linear model captures the outcome
of interest (Y), as demonstrated in this equation:
Y = I × Y1 + (1 − I) × Y0 (4)
The potential outcome Y0 in the linear model is specified as:
E [Y |PD, D, X ] = αT + βD + θX (5)
Equation (5) implies an important assumption in the DID linear model, that
the effect of time period α is constant across control and treatment groups.
Also, the difference between the two groups β is constant across period.
Following the specifications in equations (2) and (5), the equation can
be rewritten as:
E [Y1 |PD = 1, D = 1, X ] = α + β + ϕ + X θ (6)
If assumptions as implied in equations (3), (4), and (5) are satisfied, then
the treatment effect can be obtained by the following:4
E[Y |PD, D, X ] = T D × [ϕ + αT + βD + X θ]
+ (1 − PDD) × [αPD + βPD + X θ ]

= αPD + βPD + ϕPDD + X θ (7)
The treatment effect ϕ is then captured by the interaction term between the
treatment status variable D and the time period PD.
Semi-parametric DID
The challenge in this study was to ensure that it addressed the imbalances
in the characteristics of public and private schools prior to decentralization.
The DID estimator relies on strong assumptions, and one of the key assump-
tions is the parallel trend—that the average outcome for both treatment and
control groups moved in a parallel direction before and after the treatment.
This will require the average characteristics before the treatment to be bal-
anced across the groups, and violation of this assumption would result in
biased DID estimators (Abadie 2005, 1–2).
I tested this by plotting the average language and math test scores,
pre- and postdecentralization, of both public and private schools. The trend
of average education outcomes of both types of schools showed that they
moved in parallel paths, although it was evident that students of public
schools on average scored higher in both language and math than pri-
vate school students (Appendix Figure A1). Further, I performed balance
checks on the mean difference of pretreatment characteristics between pub-
lic and private schools and observed imbalances in some of the covariates
(Appendix Table A2).
To address this issue, I estimated the impact of decentralization on ed-

ucation quality using the semi-parametric DID (SDID) approach in order
to strengthen the reliability of the estimates. SDID can be used to examine
the impact of an intervention and is particularly useful when longitudinal
or repeated cross-sectional data are available. It addresses the imbalances
in the characteristics of control and treatment groups using a reweighting
technique (Houngbedji 2015, 1–2). The inference takes into account the
fact that the propensity score (the probability to be treated) is estimated.
The average treatment effect is obtained through the following equation:5
Yt D − ∅ (X0 )
E × (8)
P (D = 1) 1 − ∅ (X0 )
The implementation of SDID to investigate the impact of decentralization

on the average of education outcomes of public schools in comparison with
private schools is crucial. This is because (as discussed earlier) I had observed
some differences in the characteristics of the two groups that may have
affected the average test scores.
In practice, SDID estimation involved two steps. First, the propensity
score was calculated using the set of covariates used in the identification
model, as captured in equation (1). Next, the sample was reweighted using
the estimated propensity score to construct the SDID sample. Observations
with propensity scores of less than 0.01 or greater than 0.99 were dropped,
by using the absdid syntax in Stata.
Results
Descriptive statistics and balance checks
This section briefly discusses the descriptive statistics of the sample and
explains the results of balance checks on pretreatment differences between
private and public schools. The study consisted of information on observ-
able characteristics of schooling and localities (i.e., village and community
characteristics) in the baseline year 2000 and the follow-up year 2007.
Table 2 provides descriptive statistics of the situation before the reform.
Decentralization had taken place after the year 2000, as indicated by
the observed jump in the share of public schools managed by district educa-
tion offices. The MoEC centrally managed public schools prior to the reform.
Decentralization also affected the revenue and expenditure structures of vil-
lages and schools. The governmental contribution to total village revenues
rose by approximately 20 percent in 2007, and school revenue continued
to double during the period 2007–2014 (Appendix Figure A2). This trend
was even more pronounced among private schools (Appendix Figure A3).
Despite the reform, the government still retained a significant role
in providing education financing, from the village to the school level.
TABLE 2 Descriptive statistics for covariates prior to decentralization (2000)

Variables Private Public Difference t-coeff. Sig. level
Language test 5.219 6.279 −1.060 −20.128 ***
Math test 4.333 5.767 −1.434 −19.593 ***
Sanitation 0.609 0.545 0.064 2.706
Village revenue 103.976 119.976 −16.000 −1.346
Poor household (%) 24.794 28.721 −3.927 −2.906
Urban 0.672 0.537 0.134 5.792 ***
Distance to district 19.401 19.135 0.267 0.219
office
Community groups (n) 4.658 4.731 −0.073 −0.918
Teacher with primary 0.002 0.000 0.002 1.488
education
Teacher with lower 0.067 0.050 0.017 1.514
secondary education
Teacher with upper 0.147 0.169 −0.022 −1.213
secondary education
Teacher with tertiary 0.903 0.842 0.061 3.583
education
Teacher years of 12.405 16.587 −4.182 −12.955 ***
experience
Teacher working hours 22.911 29.862 −6.951 −15.437 ***
*
p < 0.10 **p < 0.05 ***p < 0.01
NOTE: Survey weights applied.
Government funding continued to account for 50–60 percent of village and

school revenue and contributed the largest share of financing for educa-
tional assistance in both public and private schools (Appendix Figure A4,
Panel A). The average value of government educational assistance tripled
in real terms since the reform, with a pronounced rise among private schools
(Appendix Figure A4, Panel B). However, whether the increased education
financing led to changes in educational outcomes depended on institutional
capacity and on the effective allocation of spending.
The data also show that public schools provided a higher quality of
education relative to private schools. On average, students in public schools
consistently performed better in both language and math tests, by 1.1
and 1.4 points, respectively. This parallels findings from other studies and
confirms that private schools in Indonesia provide low-cost education to
the poorer regions and have less access to education finance and resources,
limiting the quality of the education provided (Bangay 2005, 171–172;
Kristiansen and Pratikno 2006, 515). Differences in outcomes were driven
by teachers’ effort, with private schools on average having 6 percent more
teachers with tertiary education but with teachers there spending fewer
working hours than those at public schools (Table 2).
To arrive at nonbiased estimates, it is important to ensure that there
were no significant observed differences between the two types of schools.
TABLE 3 Impact of decentralization on education quality: DID and SDID

estimation results
Language test Math test
Estimation Basic Extended Basic Extended
DID −0.532*** −0.703*** −0.764*** −1.045***
SE [−8.14] [−7.03] [−11.59] [−10.10]
Observations 3,904 1,806 3,880 1,794
SDID −0.442*** −0.350** −0.429*** −0.370**
SE [−4.48] [−3.07] [−3.76] [−2.61]
Observations 764 509 762 510
*
p < 0.10 **p < 0.05 ***p < 0.01
NOTE: DID shows the difference-in-difference estimators, obtained using the diff syntax developed by Villa
(2016), and SDID shows the semi-parametric difference-in-difference results, estimated using absdid Stata
syntax with the following specifications: logistic option to estimate the propensity score for calculating weight to
reweight the original sample; drop observations with propensity score of less than 0.01 and greater than 0.99.
Basic model controls no covariates, and extended model estimates the treatment effect controlling for other
factors−school, village, and geographic variations. Standard errors are reported in parentheses and are clustered
at the school level. In this and subsequent tables, SE means standard error and t-statistics are shown in brackets.
Notice that t-test results showed the observed mean difference of some of
the covariates are related to average test scores. This presented difficulty
in the estimation, as education outcomes among public schools could re-
sult from preexisting characteristics rather than being an exogenous effect
of decentralization. To respond to this issue, I implemented two estimation
strategies, as was explained above. As a result, statistically significant differ-
ences in the means of any pretreatment characteristics between public and
private schools were no longer observed (Appendix Table A2).
The impact of decentralization on education quality in

Indonesia
The analysis begins by measuring the direct impact of decentralization

on education quality in Indonesia, focusing on two outcome variables—
average student score in language in and math tests. Table 3 displays the
results for the calculated difference estimators. As there was a significant
difference in education outcomes in the baseline period, results are re-
ported for both DID and SDID estimators using the original and reweighted
samples.
Decentralization had a significant impact on average learning out-
comes; there was also a counterintuitive trend whereby private schools
seemed to benefit more from decentralization. The estimated effects were
statistically significant and robust to different econometric specifications
and various sets of covariates.6 Following decentralization, the estimated
differences in the average scores in language and math tests were negative
0.53–0.70 and 0.76–1.1 points, respectively. The differences were more
evident for average math test scores. The DID estimators tended to over-
estimate the changes in the average test scores. This was expected, as a
significant different in average education outcomes was observed between

public and private schools. For instance, Leer (2016) estimated the impact
of decentralization on education quality in Indonesia using the DID method
and found negative coefficients, although they were not statistically signif-
icant. In this article, both DID and SDID estimators were statistically signif-
icant and larger in magnitude. One reason is that more time-variant school
characteristics (e.g., teachers’ average working hours, teachers’ school,
principal’s education) and school localities (e.g., community groups, village
characteristics) were controlled for. Furthermore, the analyses accounted
for all levels of schools (i.e., primary and secondary), hence providing a
larger sample size for estimating the impact.
Nevertheless, the sign of the estimated coefficients needs to be inter-
preted carefully. The coefficients show the estimated difference between the
average performance of public schools and that of private schools, given
the treatment of decentralization. Thus, these findings do not necessarily
suggest that decentralization was detrimental to education quality. On the
contrary, they counterintuitively demonstrate the “catching up” of private
schools. In other words, while an increase in average test scores was ob-
served in both public and private schools (Appendix Table A2), the im-
provements at public schools have been less sizable than at private schools.
Private schools seem to have benefited more from decentralization than
public schools.
It is also noteworthy that the estimated DID coefficients are potentially
lower-bound estimates of the impact. Aside from the 2001 decentraliza-
tion, the Indonesian government implemented two key educational policy
changes—school operational assistance grants and the 20 percent rule of
education spending. However, the impact should be at minimum, as bud-
get increases occurred incrementally at one percent annually during the
period of analysis. According to the Indonesia Consolidated Fiscal Database
(World Bank 2017), the share of total education spending (national and
subnational) to the gross domestic product grew by 4.3 percent from 2001
to 2007 vis-à-vis the approximately 6 percent increase from 2008–2009 and
reached 19.6 percent in 2016.
The mechanism
The interesting question is why public schools did not benefit as much
as private schools, which theoretically were not affected by the reform.
This article goes a step further by investigating the mechanisms that
could explain the conditions and channels through which decentralization
facilitated improved education outcomes. As highlighted at the outset, I am
particularly interested in examining the role of educational assistance as an
indirect mechanism. In doing so, three possible channels were explored,
focusing on the roles of key development actors—local institutions, schools,
and communities. Decentralization may have facilitated collusion between

district and village officials, resulting from a shift in preferences when
allocating education finance toward private schools—particularly in the
form of increased provision of educational assistance to minority schools.
Increased provision of educational assistance programs
Studies have examined the importance of educational assistance pro-

grams in enhancing education quality, including increased enrollment rates
(Sparrow 2007), school participation (Afridi 2011), and improved learning
outcomes (Anand, Mizala, and Repetto 2009). My hypothesis was that de-
centralization helps to improve education outcomes through increased pro-
vision of educational assistance. Two possible mechanisms were examined:
(1) whether the decentralized education system increased access to edu-
cational assistance (measured by the share of students who received such
assistance), and (2) the amount of educational assistance (measured by the
average amount received per student). Both were examined at the school
level. Naturally, one would expect decentralization to lead to an increase
in government-provided educational assistance. To investigate the link, the
following equation used DID and SDID estimations in which the outcome
variable was each of the measures of access to and amount of educational
assistance.
transferspt = ∝0 + ∝1 Publicspt + ∝2 Publicspt x PostDecent

+ ∝3 Sspt + ∝4 Xspt

+ ∈spt (9)
where transfersp was each of the measures of access to (ratio of beneficia-

ries of educational assistance) and amount of (average amount of trans-
fer received per student) educational assistance in school s in province p
at time t; Public was a dummy variable indicating the treatment group;
Public x PostDecent was the key variable of interest, with α 2 capturing the
impact of decentralization on the provision of educational assistance re-
ceived by the public schools relative to their private school counterparts; S
was a vector of variables controlling the school’s characteristics; and X vp
controlled for time-variant village characteristics.
Besides assessing the impact of decentralization on cumulative educa-
tional assistance, it is also important to see how decentralization affected
different types of educational assistance. To take into account the complex
structure of educational assistance programs in Indonesia, the estimations
segregated the type of educational assistance by provider and by type of
benefits. “Providers” were classified as government, school committee, or
community scholarship, while benefits were classified as cash- and in-kind
educational assistance. As captured in the data, community and school com-
mittee involvement in providing educational assistance has been increasing,
TABLE 4 Effect of decentralization on provision of educational assistance

Estimation Type of educational assistance
Panel A: Ratio of School
beneficiaries Government committee Community In kind Cash
DID 0.855 −1.733* −1.599 3.49 1.167
SE [−1.18] [−2.13] [−1.28] [−0.53] [−1.21]
Observations 1,624 1,314 1,298 71 1,620
SDID 0.245 5.696 1.477*** 3.49 1.167
SE [−0.31] [−1.54] [−8.71] [−0.53] [−1.21]
Observations 358 33 6 71 1,620
Panel B: Size of School

transfer Government committee Community In kind Cash
DID 0.656 7.09 −26.91 0.153 −14.45
SE [−0.07] [−0.56] [−0.61] [−0.01] [−0.42]
Observations 1,617 1,307 1,288 71 1,616
SDID −6.758 −2.893 25.04 NA 0.928
SE [−0.82] [−0.37] [−1.51] NA [−0.05]
Observations 345 32 6 NA 356
*
p < 0.10 **p < 0.05 ***p < 0.01
NOTE: NA = not applicable.
although the government is still the predominant contributor to the total

share (Appendix Figure A4).
First, I investigated whether decentralization increased access to edu-
cational assistance at the school level. It was measured as the ratio of bene-
ficiary students to the total number of students in the school; the results are
presented in Panel A of Table 4.7 Decentralization did not have an impact on
the provision of government educational assistance. On the other hand, de-
centralization affected the provision of educational assistance by the school
committee and the community. The DID estimator shows that following de-
centralization, the average share of beneficiaries for school committee edu-
cational assistance programs in public schools decreased by 1.7 percent, al-
though the impact disappeared when SDID estimators were estimated. The
ratio of beneficiaries of community educational assistance in public school
increased by 1.5 percent compared with private schools.
I also tested whether decentralization increased the average amount
of educational assistance received per student. As demonstrated in Panel B
of Table 4, decentralization did not affect the size of educational assistance
across any type of provider (i.e., government, school committee, or commu-
nity programs) and type of educational assistance (i.e., in-kind versus cash).
Overall, the results suggest that one of the mechanisms through which
decentralization affects education quality is through increased access to
educational assistance. It increases the allocation of educational assistance
provided by the school committee more toward private school, relative to
public school. This explains the “catch-up” of education quality in private

schools after decentralization. However, it is also important to note that
the results did not show a statistically significant impact on the increase of
government educational assistance after decentralization, which I had hy-
pothesized. This may explain the earlier finding that public schools seemed
to benefit less from decentralization, indicating ineffective allocation and
use of the increased educational financing to public schools, which echoes
the finding from earlier studies. For instance, the World Bank (2013,
12–13) argued that the current trend in education spending in Indonesia
was unlikely to translate into better student and teacher outcomes due to
inefficiency in education spending, significant inequality in the distribution
of educational access and inputs, and lack of incentives for district officials
to optimize spending utilization.
School efforts and management
The other possible explanation for the catch-up in education quality among
private schools is that the schools increased their efforts to improve stu-
dents’ performance. Advocates of a decentralized education system argue
that it provides schools with more financial resources coming from the
community. Also, decentralization provides greater freedom for schools to
tailor their expenditure based on the school’s need. To test this, I proxied
school effort with (1) teachers’ behavior and (2) the number of active school
days and estimated DID and SDID estimators for decentralization’s effect on
schools’ efforts. Three indicators were used to measure teachers’ effort: the
average number of hours spent teaching in class, the average number of
hours spent in other jobs, and teachers’ average monthly salary. The speci-
fication was:
effortspt = ∝0 + ∝1 Publicspt + ∝2 Publicspt xPostDecent
+ ∝3 Sspt + ∝4 Xspt

+ ∈spt , (10)
where effortsp was each of the measures of effort in school s in province p;

Public was the parameter of interest indicating the treatment group; S was
a vector of variables controlling schools’ characteristics; and Xv p controlled
for time-variant village characteristics.
Table 5 shows that, while in the presence of decentralization teachers’
hours increased in both public and private schools, the impact in public
schools was five hours per week smaller compared with private schools.
This is an important finding with a crucial policy implication. Following the
increase in the average monthly wage of public school teachers, their effort
is now less in the presence of decentralization. This was confirmed by the
positive effect observed in the average number of working hours spent in
another job.
TABLE 5 The impact of decentralization on schools’ efforts

Teaching hours Other working hours School days
Estimation Basic Extended Basic Extended Basic Extended
DID −7.095*** −7.581*** 6.881*** 6.883*** −0.0116 0.0199
SE [−11.59] [−7.92] [−11.45] [−7.5] [−0.68] [−1.31]
Observations 3,934 1,849 3,925 1,844 4,085 1,853
SDID −5.817*** −6.602*** 3.512** 1.734 0.0569 0.0573
SE [−5.21] [−4.72] [−2.84] [−1.1] [−1.86] [−1.21]
Observations 801 544 798 542 820 545
*
p < 0.10 **p < 0.05 ***p < 0.01
This finding may indicate the lack of a strong performance monitor-

ing system in place between the district education office and public schools.
Following decentralization, the district education office was responsible for
hiring teachers. Instead of improving the performance of current teachers,
the extra resources were used to hire more teachers—despite evidence as-
serting that an increased number of teachers is often not correlated with
improved student performance (World Bank 2013, 12). There was also lim-
ited evidence that decentralization facilitated an increase in the number of
active school days (see Table 5).
If decentralization led public schools to improve less in terms of edu-
cation quality than the average private school, it is of interest to measure
whether collusion between the village/community and the school drove
this trend. This relates to the literature on potential rent-seeking behavior
among school officials: Schools that depend less on the state or district gov-
ernment would show improved education quality because they have the
capacity to allocate resources based on the needs of the school, rather than
having the allocation tied up with universal guidelines from the education
ministry. To test this hypothesis, the schools were classified based on how
the school head was elected: “nondemocratic” and “democratic” schools.
Nondemocratic schools were those whose school head was chosen by the
government, the state, or the district education office; democratic schools
were those whose school head was elected by the community (i.e., through
a foundation or school committee).
Table 6 displays the estimated effect of decentralization on average test
scores by different categories of schools. The key finding is that decentral-
ization facilitated increased education outcomes among public schools only
when there was a complete transfer of authority up to the level of district
education office—in this case, where the district education office selected
the school head. This is indicated by the positive DID estimators on both
language and math test scores for the “district” schools subsample. Among
nondemocratic schools whose principal was selected by the district educa-
tion office, decentralization was found to increase the average scores by
TABLE 6 The heterogeneous impact of decentralization, by type of election

of school head
Type of schools
Estimation (I) (II) (III) (IV)
Panel A: Language test All schools Central District Community
DID −0.703*** −0.710*** 0.715*** −0.0812
SE [−6.78] [−9.32] [−9.73] [−0.95]
Observations 1806 1774 1774 1774
SDID −0.350** 0.121 −0.816* 0.151
SE [−3.07] [−0.41] [−2.52] [−0.52]
Panel B: Math test All schools Central District Community

DID −1.045*** −0.638*** 0.658*** −0.376***
SE [−10.30] [−7.87] [−8.42] [−4.80]
Observations 1794 1762 1762 1762
SDID −0.370** 0.0029 −1.104* −0.065
SE [−2.61] [−0.01] [−2.11] [−0.20]
*
p < 0.10 **p < 0.05 ***p < 0.01
0.70 and 0.65 points for language and math, respectively, in public schools
compared with private schools. On the other hand, results among non-
democratic schools, whose principal was selected by the central govern-
ment, were consistently negative across both average test scores.
To summarize, school efforts and management provide an alternative
explanation for the conditions under which decentralization worked to im-
prove education quality. First, decentralization increased the average work-
ing hours of teachers more among private schools that in public schools.
This explains the “catch-up” story of private schools benefiting more after
decentralization. Second, decentralization improved the average outcome
of public schools more, relative to their counterparts, only when a certain
degree of authority was transferred from the central government to the dis-
trict education office.
Community engagement
Another mechanism through which decentralization may have affected
education outcomes is community efforts; I hypothesized that decentral-
ization had a greater impact on education outcomes where the community
was active. This relates to a broader literature on the community role
in development and in the delivery of public services (Stiglitz 2002;
Bovaird 2007), and in the improvement of education quality in particular
(Kendall 2007; Pradhan et al. 2014). To test this possibility, the heterogene-
ity of decentralization’s impact on education outcomes by the schools’
TABLE 7 Heterogeneous impact of decentralization: active and passive

communities
Language test Math test
Estimation Passive Active Passive Active
community community community community
DID −0.724*** −0.609*** −1.127*** −0.783***
SE [−5.41] [−3.45] [−8.30] [−4.22]
SDID −0.546** −0.212 −0.463* −0.286
SE [−2.76] [−1.54] [−2.44] [−1.35]
*
p < 0.10 **p < 0.05 ***p < 0.01
locality—whether the school is located in an active or a passive

community—was examined. Localities where the number of active so-
cial groups was above average were defined as active communities, while
those where the number of active social groups was below average were
classified as passive communities.8 The results are shown in Table 7.
Negative and statistically significant DID estimators were observed for
average test scores in both passive and active communities. This confirms
the important role of communities, which facilitated improvement in edu-
cation outcomes, particularly among private schools. However, the impacts
were no longer statistically significant under SDID estimators, which indi-
cated that the impact of decentralization on average education outcomes
was only being observed among schools where the community was less ac-
tive. This counterintuitive result can be explained by the fact that school
committees in Indonesia have limited influence over decision-making on
school budgets and planning. A study using a randomized controlled trial
found that school committees in Indonesia had a strong impact on learn-
ing outcomes when they had linkages with village councils; thus, increas-
ing community participation alone does not necessarily suffice (Pradhan
et al. 2014, 123). This finding is important, as it underlines the importance
of school localities when examining the direction and magnitude through
which decentralization affects education quality.
Discussion
Despite the expansion of participation in schooling, progress in improving
the quality of education in developing countries remains slow. Educational
assistance, among other initiatives, has been central in addressing the
challenge, yet the success of such programs depends on the capacity of
the institutions administering them. With a popular shift toward a decen-
tralized education system, this study attempted to answer the following
questions: How does decentralized education play a role in the provision
of educational assistance, and thus explain changes in education quality?
Under what conditions does a decentralized education system lead to the

fair distribution of educational assistance to public and private schools?
Do local and social norms enhance the institutional features of education
provision in the process? In the first step, panel data on school information
were used to estimate the DID estimators on the effect of decentralized
education on education quality, measured in terms of average student test
scores in language and math. The findings highlighted that decentralization
in general facilitated improved student outcomes in both control and treat-
ment schools. Counterintuitively, the impact was stronger among private
than public schools.
The study then took a step further by looking at the mechanism, ques-
tioning under what conditions decentralization works to improve education
outcomes, particularly among private schools. It was of particular interest
to study the role of educational assistance. Increased provision of educa-
tional assistance by government and the community, particularly among
marginalized private schools, was one of the key mechanisms. Decentral-
ization increased the ratio of beneficiaries of government and community
scholarships in private schools. Decentralization had no statistically signif-
icant impact on increases in cash or in-kind educational assistance. This
means that while decentralization may improve access to educational assis-
tance, it does not trigger innovations in terms of the forms that educational
assistance takes.
Other mechanisms were also examined in an attempt to find explana-
tions for the increased resources going toward private schools. Decentral-
ization appears to have affected education quality through increasing teach-
ers’ efforts and was heterogeneous in terms of its relationship with type of
school management. The increase in teachers’ workings hours was more
pronounced among private schools and was heterogeneous depending on
the characteristics of the locality—active versus passive communities.
These results lead me to argue that there are two aspects to de-
centralization in improving education quality in a developing economy.
First, decentralization facilitates collusion between village authorities and
marginalized private schools, with substantial increases in government
and community educational assistance and financial resources, especially
to private minority schools. Interestingly, despite dominant rent-seeking
behavior and motives of self-interest, the increased allocation of public
resources to private schools had a positive impact on student achieve-
ment outcomes. These results also emphasize the role of social norms in
undermining the efficient allocation of public goods after decentralization.
In contrast, it may be that the partial nature of decentralization ex-
plains the moderate progress of education quality among public schools
relative to private schools in the postdecentralization period. The results
presented here suggest that decentralization only succeeded in facilitating
an increase in education quality in public schools if there was a certain
transfer of authority toward the district and school levels (i.e., when the
district education office had the authority to select the principal of public
schools). Public schools with a school head selected by the district educa-
tion office (rather than by the central government) experienced greater
increases in average student achievement.
These study findings focusing on education quality relate to the
broader literature of decentralization and public service delivery and to the
political economy behind them (Faguet and Sanchez 2008; Di Gropello and
Marshall 2011; Leer 2016). By looking at how decentralization affected the
provision of educational assistance to public and private schools, this study
also contributes to the literature on decentralization and social assistance
(Bardhan and Mookherjee 2005). To unravel the key mechanism that
may indicate collusion between school authorities and village officials, this
study is positioned in the literature on the effects of social and ethnic frag-
mentation on public service delivery (Alesina, Gennaioli, and Lovo 2014;
Tajima, Samphantharak, and Ostwald 2017), particularly on the concept of
rent-seeking behavior (Mueller 2004; Cheikbossian 2008). The study also
explored the role of community, which related to the literature on commu-
nity engagement in development and public service delivery (Stiglitz 2002;
Bovaird 2007) and in improving education quality in particular (Kendall
2007).
In conclusion, the unintended consequences of decentralizing educa-
tion management demonstrate a positive and promising story in providing
inclusive education in Indonesia. Private schools play an important role in
bridging the gap, providing access to education in rural and remote areas.
While the private schools are, in general, inferior in terms of quality and
facilities, their catching up with public schools would mean increased pro-
vision of a better quality of education to children from low-income families.
The results also yield important policy implications. First, as school efforts
are one of the key direct mechanisms, government interventions to im-
prove education quality should focus on reforming the management quality
of public schools. This includes initiatives such as increasing incentives for
teachers (to increase the quality, rather than the quantity, of teaching), pro-
viding teacher training, and strengthening the role of the school committee
in the school decision-making process.
Notes
The views expressed in this paper are those of to the provincial level (Nasution 2016, 135).
the author and not those of The World Bank. Due to this, we restrict the analysis to the pe-
1 It is important to note that in 2014, riod 2000–2007.
the Indonesian government enacted Law No. 2 The survey currently has five waves,
23, which led to a shift in authority for se- covering a 21-year span from 1993 to
nior secondary schools from the district level 2014; it collects information from 13 out
of 26 provinces in the nation. The sam- ization impacts persisted in sign and mag-
pling scheme is stratified at the province nitude and did not change across different
level and in urban/rural locations, and the models. Results are available upon request.
13 provinces represent up to 80 percent of 7 In this article, educational assistance
the population (Frankenberg and Thomas refers to scholarship programs that take
2000, 4) the forms of in-kind assistance (e.g., school
3 The control group refers to a group books and uniforms) and cash assistance
of schools that were not affected by (e.g., school fee waivers, registration fee
decentralization—i.e., private schools. The waivers). Assistance includes need-based,
treatment group refers to schools that were talent and merit, retrieval and transition
affected by the decentralization decree. It in- scholarships that are provided by such
cludes all government schools under MoEC sources as government, community groups,
administration. parent organizations, and school committees.
4 The diff Stata syntax developed by 8 The social groups refer to commu-
Villa (2016) was used. nity programs or activities such as village
5 The absdid Stata syntax developed by cooperative youth groups, village improve-
Houngbedji (2015) was used to carry out the ment programs, community public works,
estimation. and village mobile libraries. “Active” social
groups refer to the mentioned activities that
6 For robustness, the same DD and SDID
presently have been conducted or are being
estimations were run with a different set of
carried out on a routine basis.
controls added, and the estimated decentral-
References
Abadie, Alberto. 2005. “Semiparametric difference-in-differences estimators,” The Review of Eco-
nomic Studies 72(1): 1–19.
Afridi, Farzana. 2011. “The impact of school meals on school participation: Evidence from rural
India,” Journal of Development Studies 47(11): 1636–1656.
Agustina, Cut Dian, Dandan Chen, Andrew Ragatz, and Imam Setiawan. 2009. “Scholarships pro-
grams in Indonesia: Past, present and future,” Washington, DC: World Bank East Asia and
Pacific Regional Office.Alesina, Alberto, Caterina Gennaioli, and Stefania Lovo. 2014. “Public
goods and ethnic diversity: Evidence from deforestation in Indonesia,” Working Paper 20504.
Cambridge, MA: National Bureau of Economic Research.
World Bank Jakarta. 2014. Assessing the Role of the School Operational Grant Program (BOS) in Improving
Education Outcomes in Indonesia. Jakarta, Indonesia: World Bank Jakarta.
Anand, Priyanka, Alejandra Mizala, and Andrea Repetto. 2009. “Using school scholarships to esti-
mate the effect of private education on the academic achievement of low-income students in
Chile,” Economics of Education Review 28(3): 370–381.
Athey, Susan, and Guido W. Imbens. 2006. “Identification and inference in nonlinear difference-
in-differences models,” Econometrica 74(2): 431–497.
Autor, David. 2014. “Skills, education, and the rise of earnings inequality among the ‘other 99
Percent’,” Science [Online], 344(6186): 843–851.
Baez, Javier E., and Andriana Camacho. 2011. “Assessing the long-term effects of conditional
cash transfers on human capital: Evidence from Colombia,” Policy Research Working Paper.
Washington, DC: World Bank.
Bangay, Colin. 2005. “Private education: Relevant or redundant? Private education, decentralisa-
tion and national provision in Indonesia,” Compare: A Journal of Comparative and International
Education 35(2): 167–179.
Bardhan, Pranab, and Dilip Mookherjee. 2005. “Decentralizing antipoverty program delivery in
developing countries,” Journal of Public Economics 89(4): 675–704.
Barrera-Osorio, Felipe, Leigh L. Linden, and Juan Saavedra. 2017. “Medium- and long-term educa-
tional consequences of alternative conditional cash transfer designs: Experimental evidence
from Colombia,” Working Paper 23275. Cambridge, MA: National Bureau of Economic Re-
search.
Becker, Gary S. 1962. “Investment in human capital: A theoretical analysis,” Journal of Political
Economy 70(5): 9–49.
Bovaird, Tony. 2007. “Beyond engagement and participation: User and community coproduction
of public services,” Public Administration Review 67(5): 846–860.
Cameron, Lisa. 2009. “Can a public scholarship program successfully reduce school drop-outs in a
time of economic crisis? Evidence from Indonesia,” Economics of Education Review 28(3): 308–
317.
Cheikbossian, Guillaume. 2008. “Rent-seeking, spillovers and the benefits of decentralization,”
Journal of Urban Economics 63(1): 217–228.
Crook, Richard C., and Alan Sturla Sverrisson. 1999. “To what extent can decentralized forms of
government enhance the development of pro-poor policies and improve poverty-alleviation
outcomes?” Working paper. Accessed March 8, 2018, at: http://citeseerx.ist.psu.edu/viewdoc/
download?doi=10.1.1.197.1906&rep=rep1&type=pdf.
De Grauwe, Anton. 2005. “Improving the quality of education through school-based manage-
ment: Learning from international experiences,” International Review of Education 51(4): 269–
287.
Dimova, Ralitza, and Virgi A. Sari. (forthcoming). “Declining and levelling off of earnings inequality:
Boon or bane for Indonesia?” GPID Working Paper. London: GPID Network, Kings College
London.
Di Gropello, Emanuela, and Jeffery H. Marshall. 2011. “Decentralization and educational perfor-
mance: Evidence from the PROHECO Community School Program in rural Honduras,” Edu-
cation Economics 19(2): 161–180.
Faguet, Jean-Paul, and Fabio Sánchez. 2008. “Decentralization’s effects on educational outcomes
in Bolivia and Colombia,” World Development 36(7): 1294–1316.
Falch, Torberg, and Justina AV Fischer. 2012. “Public sector decentralization and school perfor-
mance: International evidence,” Economics Letters 114(3): 276–279.
Filmer, Deon, and Norbert Schady. 2014. “The medium-term effects of scholarships in a low-income
country,” Journal of Human Resources 49(3): 663–694.
Frankenberg, Elizabeth, and Duncan Thomas. 2000. “The Indonesia Family Life Survey (IFLS):
Study design and results from waves 1 and 2,” RAND unrestricted draft paper. Santa Mon-
ica, CA: RAND Corporation. Accessed March 8, 2018, at: http://www.vanneman.umd.edu/
socy699j/ifls2design.pdf.
Ganimian, Alejandro J., and Richard J. Murnane. 2016. “Improving education in developing coun-
tries: Lessons from rigorous impact evaluations,” Review of Educational Research 86(3): 719–
755.
García, Sandra, and Juan E. Saavedra. 2017. “Educational impacts and cost-effectiveness of condi-
tional cash transfer programs in developing countries: A meta-analysis,” Review of Educational
Research 87(5): 921–965.
Heckman, James J., Robert J. Lalonde, and Jeffrey A. Smith. 1999. “The economics and economet-
rics of active labor market programs,” in Orley C. Ashenfelter and David Card (eds.), Handbook
of Labor Economics, Volume III. Amsterdam: Elsevier.
Houngbedji, Kenneth. 2015. “Abadie’s semiparametric difference-in-differences estimator,” Stata
Journal, StataCorp LP, vol. 16(2), pp 482–490, June.
Kendall, Nancy. 2007. “Parental and community participation in improving educational quality
in Africa: Current practices and future possibilities,” International Review of Education 53(5/6):
701–708.
Kharisma, Bayu. 2016. “Can a school operational assistance fund program (BOS) reduce school
drop-outs during the post-rising fuel prices in Indonesia? Evidence from Indonesia,” MPRA
Paper. Bandung: University of Padjadjaran.
Kristiansen, Stein, and Pratikno. 2006. “Decentralising education in Indonesia,” International Jour-
nal of Educational Development 26(5): 513–531.
Leer, Jane. 2016. “After the Big Bang: Estimating the effects of decentralization on educational
outcomes in Indonesia through a difference-in-differences analysis,” International Journal of
Educational Development 49(C): 80–90.
Masino, Serena, and Miguel Niño-Zarazúa. 2016. “What works to improve the quality of stu-
dent learning in developing countries?” International Journal of Educational Development 48:
53–65.
Mueller, Dennis C. 2004. “Public choice: An introduction,” in Charles K. Rowley and Friedrich
Schneider (eds.), The Encyclopedia of Public Choice. Dordrecht: Kluwer Academic Publishers.
Nasution, Anwar. 2016. “Government decentralization program in Indonesia,” ADB Working Paper
Series. Tokyo: Asian Development Bank Institute.
OECD. 2016. “PISA 2015 Results in Focus,” Paris: Organization for Economic Co-operation and
Development.
Pradhan, Menno, Daniel Suryadarma, Amanda Beatty, Maisy Wong, Arya Gaduh, Armida Alis-
jahbana, and Rima Prama Artha. 2014. “Improving educational quality through enhancing
community participation: Results from a randomized field experiment in Indonesia,” Ameri-
can Economic Journal. Applied Economics 6(2): 105–126.
Priebe, Jan, Fiona Howell, and Virgi A. Sari. 2014. “Poverty and the labour market in Indonesia:
Employment trends across the wealth distribution,” TNP2K Working Paper. Jakarta: TNP2K.
Puhani, Patrick A. 2012. “The treatment effect, the cross difference, and the interaction term in
nonlinear ‘difference-in-differences’ Models,” Economics Letters 115(1): 85–87.
Purnastuti, Losina, Paul W. Miller, and Ruhul Salim. 2013. “Declining rates of return to education:
Evidence for Indonesia,” Bulletin of Indonesian Economic Studies 49(2): 213–236.
Purnastuti, Losina, Ruhul Salim, and Mohammad Abdul Munim Joarder. 2015. “The returns to
education in Indonesia: Post reform estimates,” The Journal of Developing Areas 49(3): 183–
204.
Romer, Paul M. 1990. “Human capital and growth: Theory and evidence,” Carnegie-Rochester Con-
ference Series on Public Policy 32: 251–286.
Sparrow, Robert. 2007. “Protecting education for the poor in times of crisis: An evaluation of
a scholarship programme in Indonesia,” Oxford Bulletin of Economics and Statistics 69(1): 99–
122.
Stern, Jonathan M. B., and Thomas M. Smith. 2016. “Private secondary schools in Indonesia: What
is driving the demand?” International Journal of Educational Development 46(C): 1–11.
Stiglitz, Joseph E. 2002. “Participation and development: Perspectives from the comprehensive de-
velopment paradigm,” Review of Development Economics 6(2): 163–182.
Suratno, Tatang. 2014. “The education system in Indonesia at a time of significant
changes,” Revue internationale d’éducation de Sèvres, online. Accessed March 8, 2018, at:
https://journals.openedition.org/ries/3814.
Tajima, Yuhki, Krislert Samphantharak, and Kai Ostwald. 2017. “Ethnic segregation and public
goods: Evidence from Indonesia,” SSRN Electronic Journal, online. Accessed March 7, 2018, at:
https://doi.org/10.2139/ssn.3001672.
Villa, Juan Miguel. 2016. “diff: Simplifying the estimation of difference-in-differences treatment
effects,” Stata Journal 16(1): 52–71.
World Bank. 2013. Spending More or Spending Better: Improving Education Financing in Indonesia.
Jakarta: World Bank Office Jakarta.
World Bank. 2014. “Indonesia: Avoiding the trap,” Development Policy Review. Jakarta: World
Bank Office Jakarta.
World Bank. 2017. “Open Budgets Portal: Indonesia. The Indonesia Consolidated Fiscal Database
(COFIS),” Available at: http://boost.worldbank.org/country/indonesia (accessed 8 March
2018).
World Bank. 2018. World Development Report 2018: Learning to Realize Education’s Promise. Washington,
DC: World Bank.
Yusuf, Arief Anshory, and Andy Sumner. 2015. “Growth, poverty, and inequality under Jokowi,”
Bulletin of Indonesian Economic Studies 51(3): 323–348.
Appendix
TABLE A1 List and description of variables

Variable Description
Math score Average students’ score in national math examination
Bahasa score Average students’ score in national language
examination
Postdecentralization Dummy; 1 = observations in 2007 (postdecentralization)
Public Dummy; 1 = public schools (as the treatment group)
Sanitation Dummy; 1 = sanitation facility available in the village
Village revenue Total village revenue
Poor households (%) Average share of poor households in the village
Urban Dummy; 1 = village located in urban area
Distance to district Distance (in km) to the nearest district office from the
office village head’s office
Community groups (n) No. of active community groups in the village
Teacher—primary No. of teachers with primary education in the school
Teacher—lower No. of teachers with lower secondary education in the
secondary school
Teacher—upper No. of teachers with upper secondary education in the
secondary school
Teacher—tertiary No. of teachers with tertiary education in the school
Teacher—hours of work Average no. of hours teacher spent in the school
Teacher—other job Average no. of hours teacher spent in other job
School head—primary Dummy; 1 = school head’s highest level of education is
primary
School head—lower Dummy; 1 = school head’s highest level of education is
secondary lower secondary
School head—upper Dummy; 1 = school head’s highest level of education is
secondary upper secondary
School head—tertiary Dummy; 1 = school head’s highest level of education is
tertiary
TABLE A2 Balance of covariates across public and private schools (2000)
Original sample Reweighted sample
Variables Private Public Difference t-coeff Sig. level Private Public Difference t-coeff Sig. level
*** ***
Language test 5.219 6.279 −1.060 −20.128 5.330 6.134 −0.803 −8.075
*** ***
Math test 4.333 5.767 −1.434 −19.593 4.669 5.683 −1.014 −7.725
Sanitation 0.609 0.545 0.064 2.706 0.590 0.494 0.097 2.095
Village revenue 103.976 119.976 −16.000 −1.346 90.782 100.498 −9.715 −0.541
Poor household (%) 24.794 28.721 −3.927 −2.906 27.798 28.736 −0.939 −0.353
***
Urban 0.672 0.537 0.134 5.792 0.593 0.491 0.102 2.216
Distance to district office 19.401 19.135 0.267 0.219 19.152 21.960 −2.808 −1.189
Community groups (n) 4.658 4.731 −0.073 −0.918 4.535 4.672 −0.137 −0.885
Teacher with primary education 0.002 0.000 0.002 1.488 0.007 0.000 0.007 2.068
Teacher with lower secondary education 0.067 0.050 0.017 1.514 0.070 0.057 0.012 0.565
Teacher with upper secondary education 0.147 0.169 −0.022 −1.213 0.217 0.153 0.064 1.860
Teacher with tertiary education 0.903 0.842 0.061 3.583 0.895 0.865 0.030 0.955
***
Teacher’s years of experience 12.405 16.587 −4.182 −12.955 12.053 16.085 −4.032 −6.907
***
Teacher’s working hours 22.911 29.862 −6.951 −15.437 23.398 29.678 −6.280 −7.822
*
p < 0.10 **p < 0.05 ***p < 0.01
NOTE: The original sample is the subsample used for standard DID estimation and estimating using diff syntax in Stata developed by Villa (2016). The reweighted sample is used for
estimating Abadie (2005) semi-parametric DID, and the results are obtained by using absdid Stata syntax developed by Houngbedji (2015). Survey weights were applied.
TABLE A3 Estimates of differences in education quality

Year 2000 Year 2007
Language test Math test Language test Math test
Private ( = control) 5.327 4.525 7.035 6.71
SE [0.040] [0.057] [0.029] [0.048]
Public ( = treatment) 6.211 5.703 7.36 6.926
SE [0.027] [0.037] [0.019] [0.028]
Observations 1630 1628 1653 1658
diff. −0.884***
DD estimate
*
p < 0.10 **p < 0.05 ***p < 0.01
NOTE: t-statistics are in brackets.
FIGURE A1 Trends in the average language and math test scores prior to
matching (original sample, 1993–2007)
SOURCE: Author’s calculations, based on IFLS rounds.

FIGURE A2 Trends in village revenue and expenditure (2000–2014)
FIGURE A3 Trends in school revenue (2007–2014)

FIGURE A4 Trends in education assistance, by provider (2000–2014)
NOTE: The amounts shown in Panel B are in real values.


Educational Assistance and Education Quality in in

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Educational Assistance and Education Quality in in

Uploaded by

Copyright:

Available Formats

Educational Assistance and

THE EQUITABLE PROVISION of quality education is important for growth and

political actors may jeopardize an education system, eventually contributing

Indonesia provides a relevant case to study educational assistance and

experiment of the implementation of government decentralization was ex-

In a decentralized educational system, certain authorities for certain tasks

In addition to decentralization, educational assistance has recently become

Education quality and policies in Indonesia

Education remains important for Indonesia as a means to progress out of

FIGURE 1 Trends in government expenditures, by sector (2001–2013)

TABLE 1 Sample selection from IFLS 2000 and 2007

Data and estimation strategy

followed standard measurements of educational outcomes—i.e., students’

Estimating students’ achievement

First, the direct impact of decentralization on education quality of schools

where Y is the average outcome variable of school s in province p at year

DID estimation is a popular method in the applied economic literature for

The potential outcome Y0 in the linear model is specified as:

E [Y1 |PD = 1, D = 1, X ] = α + β + ϕ + X θ (6)

+ (1 − PDD) × [αPD + βPD + X θ ]

To address this issue, I estimated the impact of decentralization on ed-

The implementation of SDID to investigate the impact of decentralization

TABLE 2 Descriptive statistics for covariates prior to decentralization (2000)

Government funding continued to account for 50–60 percent of village and

TABLE 3 Impact of decentralization on education quality: DID and SDID

The impact of decentralization on education quality in

The analysis begins by measuring the direct impact of decentralization

significant different in average education outcomes was observed between

and communities. Decentralization may have facilitated collusion between

Increased provision of educational assistance programs

Studies have examined the importance of educational assistance pro-

transferspt = ∝0 + ∝1 Publicspt + ∝2 Publicspt x PostDecent

where transfersp was each of the measures of access to (ratio of beneficia-

TABLE 4 Effect of decentralization on provision of educational assistance

Panel B: Size of School

although the government is still the predominant contributor to the total

public school. This explains the “catch-up” of education quality in private

School efforts and management

effortspt = ∝0 + ∝1 Publicspt + ∝2 Publicspt xPostDecent

where effortsp was each of the measures of effort in school s in province p;

TABLE 5 The impact of decentralization on schools’ efforts

This finding may indicate the lack of a strong performance monitor-

TABLE 6 The heterogeneous impact of decentralization, by type of election

Panel B: Math test All schools Central District Community

TABLE 7 Heterogeneous impact of decentralization: active and passive

locality—whether the school is located in an active or a passive

Under what conditions does a decentralized education system lead to the

TABLE A1 List and description of variables

TABLE A3 Estimates of differences in education quality

SOURCE: Author’s calculations, based on IFLS rounds.

FIGURE A2 Trends in village revenue and expenditure (2000–2014)

SOURCE: Author’s calculations, based on IFLS rounds.

FIGURE A3 Trends in school revenue (2007–2014)

SOURCE: Author’s calculations, based on IFLS rounds.

FIGURE A4 Trends in education assistance, by provider (2000–2014)

NOTE: The amounts shown in Panel B are in real values.

You might also like