World Development Vol. 93, pp.

108124, 2017
0305-750X/ 2017 Elsevier Ltd. All rights reserved.

www.elsevier.com/locate/worlddev
http://dx.doi.org/10.1016/j.worlddev.2016.12.028

Does Local Development Inuence Outmigration Decisions?

Evidence from Indonesia
SMRITI TIWARI*
Skidmore College, Saratoga Springs, USA
Summary. This paper takes advantage of the exogenous phasing of the district-level elections in Indonesia to establish an unbiased
causal impact of local economic development conditions on outmigration decisions. Does the change in service delivery due to decen-
tralization aect the internal migration, a phenomenon not new to Indonesia and driven by the historic uneven development across the
archipelago? Furthermore, because the services are no longer centered in Java and Bali, will the migration to these historically popular
provinces change? If yes, to what extent? This discussion is of central importance not only in the Indonesian context where rising pop-
ulation pressures in Java and Bali provinces are the governments ultimate challenge but also in the development literature, where
empirical research on such questions is made impossible by the lack of appropriate data and context. Two dierent datasets, the Indone-
sian Family Life Survey and the Indonesian Census, are used to conduct the analyses. They provide consistent results. A household in a
district that went through election is 19% less likely to be a migrant-sending household. This is true particularly for districts that are in
Java and Bali. A household in a district in Java and Bali that went through election is 27% less likely to be a migrant-sending household.
2017 Elsevier Ltd. All rights reserved.

Key words decentralization, local development, outmigration, Indonesia

1. INTRODUCTION exogenous phasing of direct elections in districts to extract the

Many developing countries employ rural development
strategies that focus on employment generation, infrastructure
development, access to credit for small farmers, delivery of
health-care services, educational improvement, land reform,
etc. Even though the primary motivation for such interven-
tions is improving the quality of life in rural areas, an unin-
tended consequence of such programs may be reducing rural
to urban migration. Todaro (1997) and Lipton (1982) claim
that rural development programs are an essential solution to
deter excessive migration from rural to urban areas. Policy
makers often suggest improving living conditions and diversi-
fying economic opportunities in rural areas as a remedy to
decrease migration to cities, and thus reducing population
pressures in urban centers (Rhoda, 1983).
However, some analyses show that such programs have not
been successful in doing so because rural development activities
also increase access to cities, strengthen ruralurban integra-
tion, and raise education and skill levels. This could lead to
an increase in ruralurban migration instead (Adepoju, 1983;
Rhoda, 1983; Campbell, 1988; Becker, Hamer, & Morrison,
1994; Lucas, 1997; ILO, 1998). Beauchemin and Schoumaker
(2005) nd mixed results in the context of Burkina Faso.
Depending on the level of analysis, public facilities and infras-
tructure have either no eect or are more likely to increase out-
migration. However, they nd that having markets in villages
and large companies in secondary towns reduce the risk of
moving to big urban cities by about 50%. The discussion on
the potential impact of local development on migration behav-
iors does not provide clear evidence in the literature. Most of
all, this question is dicult to explore empirically due to the
lack of relevant context and appropriate data to study it.
The primary challenge while studying the migration-
development interactions is that they are not independent of
each other. This endogenous relationship makes establishing
causal impact in both directions challenging. In order to over-
come the endogeneity issue, this paper takes advantage of the
108
causal impact. In doing so, this work contributes to current
migration literature by using a natural experiment to establish
causal validity in the determination of migration decisions
(Palloni, Massey, & Ceballos, 2001; Munshi, 2003).
After more than three decades of highly centralized govern-
ment, Indonesia went through a phased decentralization pro-
cess 1 (both scal and political) during which wide-ranging
responsibilities over basic service delivery for education,
health, infrastructure, and general public administration were
devolved to the district governments. The decentralization
laws, which were passed in 1999 and took eect in 2001,
devolved approximately 25% of the national budget to the dis-
tricts in the form of block grants and dramatically increased
their authority over almost all sectors of government
(Skouas, Ambar, Dasgupta, & Kaiser, 2011). Additionally,
the district heads were selected through direct elections starting
in 2005. The unique and the most important feature of this pro-
cess is that timing of the direct election in a district depended on
the end of the tenure of the previous headwhich was exoge-
nous. Does this change in service delivery due to decentraliza-
tion aect the internal migration, a phenomenon not new to
Indonesia and driven by the historic uneven development across
the archipelago? Furthermore, because the increase in service
delivery is observed particularly in districts outside of Java
and Bali, will the migration from these districts change? If yes,
in what direction and to what extent?
The literature that explores the intersection between devel-
opment and migration provides ample of empirical evidence
on how migration may aect dierent development indicators,
* I would like to thank Dr. Emmanuel Skouas (Lead Economist, Poverty
Global Practice, World Bank Group) for his time and suggestions in
August 2013. I am also extremely grateful to him for sharing the pilkada
dataset, without which this study would not have been possible. I also
thank Dr. Paul Winters (Professor, Department of Economics, American
University) for his guidance. Final revision accepted: December 18, 2016.
 DOES LOCAL DEVELOPMENT INFLUENCE OUTMIGRATION DECISIONS?EVIDENCE FROM INDONESIA
such as migrants own income and health, education, and
health outcomes of children of a migrant sending household,
and the welfare of the migrant sending community as a whole
(Hildebrandt & McKenzie, 2005; Adams, 2006; Acosta,
Fajnzylber, & Lopez, 2007; Yang, 2008; Gibson, Mckenzie,
& Stillman, 2011; Antman, 2011 2). However, empirical
research on the role of economic development in the migrant
sending community in determining outmigration has been
slow to emerge.
This has been particularly true of decentralization, which is
less an investment in a particular rural development strategy
but more a change in the management of rural development.
Decentralization has become more widespread in developing
countries 3 as economists and policy makers have increasingly
advocated it as a way to reduce the role of the state in general
by fragmenting central authority and introducing more inter-
governmental competition. In countries with many ethnicities,
decentralization is also regarded as a way of avoiding social
and political tensions and ensuring local cultural and political
autonomy (Oates, 1993; Bardhan, 2002). Local governments,
with an informational advantage over the national govern-
ment, are in a better position to deliver public services. Empir-
ical literature examining the link between decentralization and
local economic development provides evidence of a strong
positive relationship (Santos, 1998; Lin & Liu, 2000; Faguet,
2004; Stansel, 2005; Iimi, 2005; Hammond & Tosun, 2011).
However, decentralization usually takes place in one shot for
the entire country, leaving no possibility for evaluation except
perhaps a before and after analysis, which is most likely to
provide biased estimates. The fact that decentralization in
Indonesia was a phased process provides a unique opportunity
to undertake a successful evaluation.
The empirical analyses are done using two rich and unique
datasetsIndonesian Family Life Survey (IFLS) and the
Indonesian Census. These datasets are compiled together with
the Pilkada dataset, assembled by Skouas, Ambar,
Dasgupta, and Kaiser (2014). The Pilkada dataset contains elec-
toral information from 1999 to 2009 and information about
district-level characteristics. Most importantly, it provides dates
of direct elections for all the districts, which helps to identify the
treated districts. The main outcome variablemigration status,
comes from IFLS and the census. IFLS is well suited for this
analysis, as it collects detailed current and retrospective infor-
mation about the migration histories of individuals. However,
the sample is restricted to 13 of the 34 provinces in Indonesia
and is representative of 83% of the Indonesian population.
The census samples generally cover the entire country but are
limited in terms of migration modules, and only provide infor-
mation about whether an individual is in a dierent district
from ve years prior. Since both datasets have pros and cons
in terms of the information they oer, the results deduced from
analyzing the two separately are complementary.
The results indicate that households in districts that went
through direct elections were signicantly less likely to have
outmigration than those that did not. This decrease was
mostly driven by the fall in the outmigration from districts
in Java and Bali. Migration from districts of other provinces
were unaected by the direct elections.
Mechanisms through which local development, spurred by the
democratization and decentralization, inuences households
migration decisions (especially in the context of Indonesia) are
discussed in the following section. Background on the Indone-
sian decentralization process is discussed in Section 3. The rele-
vance of the datasets is discussed in Section 4. In Section 5, the
econometric strategies used to analyze the data are described fol-
lowed by the results in Section 6 and conclusion in section 7.
109
2. CONCEPTUAL FRAMEWORK IN THE INDONE-
SIAN CONTEXT
(a) Internal migration in Indonesia
According to the literature, internal migration in both colo-
nial and post-colonial Indonesia points to three factors that
may have played a role in inuencing internal migration. They
are: (i) income dierential generated due to high colonial
pressures in Java, (ii) transmigration policy undertaken by
the Dutch administration, later inherited by the post-
independence Indonesian government, and (iii) the persistence
of uneven development across the region due to path depen-
dency of the institutions established by the Dutch rulers.
(i) Income dierential generated due to high colonial pressures in
Java 4
The rst reason is the nature of the land in the pre-colonial
era that resulted in the historically uneven development of the
Indonesian archipelago. The Inner Indonesia (provinces 2,
4, 8, 1012, 34 in Figure 1), with volcanic topography and rich
agricultural lands, became the center for Dutch rulers as they
maintained larger colonial pressures on Java than other parts
of Indonesia. The concentration of colonial activity in Java led
to an increased demand for labor, which in turn increased the
income dierential between Java and other regions of Indone-
sia (Hugo, 1980). As a result, Javas population exploded in
the nineteenth century. During 18021900, Javas population
grew, on average, 2.1% annually (Van Lottum & Daan, 2012).
(ii) Transmigration policy
The second reason is the transmigration policy, established
by the Dutch in 1905 (then known as kolonisatie or coloniza-
tion) as a means of relieving the population pressure in the
Inner Indonesia by giving people incentives to move to
Outer Islands. 5 Despite being abandoned due to the high
costs associated with the transmigration policy, and its
insignicant 6 role in reducing the population pressures in
inner islands, the policy was reinstated every time for the same
purpose. As a result, Indonesia not only inherited the policy
post-independence but also made it a centerpiece of their
development program. Suharto, Indonesias president for
almost three decades continued and accelerated the transmi-
gration program. By 1989, a cumulative total of approxi-
mately one million families, or ve million people, had been
shipped to the outer islands as part of the ocial program
(Fearnside, 1997).
(iii) Path dependency of the institutions set up by the Dutch
rulers
The concentration of colonial activity in Java led to a num-
ber of growing urban centers, such as present-day Jakarta,
Surabaya, and Semarang (Hugo, 1980). These cities attracted
people from the crowded interior because of the employment
opportunities they oered (Pelzer, 1945). After the indepen-
dence from the Dutch rulers, the Indonesian government
remained heavily central to Java, thus maintaining the path
of uneven development across Indonesia. Interprovincial
migration was on the rise throughout twentieth century with
7% in 1970, 8.2% in 1990, and 10.1% by 2000 (Hill,
Resosudarmo, & Vidyattama, 2008; Tirtosudarmo, 2009).
The inner islands, that constitute 7% of the total land in
Indonesia, have consistently been home to around 60% of
the Indonesian population for the past few decades.
The literature provides some evidence on the relative
strengths of these three factors. Due to the abandonment of
110 WORLD DEVELOPMENT
Figure 1. Administrative map of Indonesia.
the transmigration program after the Asian nancial crisis of
1997 and the subsequent transition to democracy,
Tirtosudarmo (2009) speculates that migration has become
predominantly a function of the labor market economy and
was constantly increasing in recent years, driven by the wage
dierentials between provinces and the attractiveness of the
Jakarta area. Van Lottum and Daan (2012) test the relative
importance of these three factors for the period during
19302000 and nd that the presence of an urban primate
such as Jakarta was the most important determinant of migra-
tion. They nd that migrants tend to base their decisions on
the more tangible prospect of economic betterment, either in
the shape of improvements of ones expected income or the
attractiveness of moving to a primate city.
(b) Decentralization and local development 7
After more than three decades of highly centralized govern-
ment, Indonesia went through a phased decentralization pro-
cessscal decentralization in 2001, and political
decentralization starting in 2005. Economists and policy mak-
ers have increasingly advocated decentralization as a way to
reduce the role of the state in general by fragmenting central
authority and introducing more intergovernmental competi-
tion. Especially in countries with many ethnicities, decentral-
ization is also regarded as a way of avoiding social and
political tensions and ensuring local cultural and political
autonomy (Oates, 1993; Bardhan, 2002; Martinez-Vazquez
& McNab, 2003). Local governments, with informational
advantage over the national government, are more ecient
as they are in a better position to deliver public services. As
a result, decentralization has become more widespread in
developing countries. 8
The empirical evidence on the role of decentralization on
economic growth initially did not provide consistent results.
In a cross-country study, Davoodi and Zou (1998) nd a sig-
nicantly negative relationship between decentralization and
growth in developing countries. Zhang and Zou (1998) and
Xie, Zou, and Davoodi (1999) also nd a negative relationship
in the case of China and the United States respectively. On the
contrary, some empirical work in the same contexts using
more recent data nd opposite results. In the cross-country
context, Hunther and Shah (1998) provide correlation between
scal decentralization and the quality of governance, which
bolsters economic growth. Lin and Liu (2000) and Akai and
Sakata (2002) nd that decentralization contributes to local
economic growth in the case of China and United States 9
respectively. Iimi (2005) highlights the fact that these studies
fail to account for the endogeneity issues and oers an instru-
mental variable approach. Focusing on the cross-county data
of 51 countries, Iimi (2005) concludes that decentralization is
instrumental in promoting economic growth.
A few studies have explored the role of decentralization in
public service deliverywhich is crucial to local economic
development. As earlier mentioned, because decentralization
processes take place in one shot in most countries, studies that
evaluate them are unable to capture its unbiased impact.
Santos (1998) and Faguet (2004) respectively provide before
and after evidence from Porto Alegre in Brazil and post-
1994 decentralization initiative in Bolivia. During 1989 and
1996, access to sanitation and enrollment in elementary and
secondary schools more than doubled in Porto Alegre.
Faguet (2004) reports that public investment in education,
water, and sanitation rose signicantly in three-quarters of
all municipalities, and service delivery responded to local
needs with public education expenditure being larger in areas
with lower literacy rate or with fewer private schools.
Although these results may be biased, as they capture both
the impact of decentralization and the general trend without
decentralization, it seems likely that there has been substantial
impact on the pattern of resource allocation across localities,
especially in poorer ones. Galasso and Ravallion (2005)
employ careful econometrics to show that decentralization
improved pro-poor program benets in Bangladesh.
Apart from the econometric issues, there are a few concep-
tual reasons why we may be seeing dierent results in dierent
contexts. Bardhan (2002) and Martinez-Vazquez and McNab
(2003) argue that decentralization is able to achieve the
intended goals if and only if the structure of local accountabil-
ity is in place and the local governments are not being cap-
tured by the usual local elites. In other words, for
decentralization to work, it has to be complemented with a
 DOES LOCAL DEVELOPMENT INFLUENCE OUTMIGRATION DECISIONS?EVIDENCE FROM INDONESIA
change in existing structures of power within communities
such that there is an improvement in the opportunities for par-
ticipation and engagement of the usually disadvantaged or dis-
enfranchised in the political process (Bardhan, 2002; Iimi,
2005). There is some empirical evidence on the impact of man-
dated representations of historically disadvantaged groups like
women (Chattopadhyay & Duo, 2004) and landless house-
holds (Foster & Rosenzweig, 2004).
In the case of Indonesia, Skouas et al. (2011) nd that the
directly elected district heads become more responsive to local
needs at least in the area of health. Skouas et al. (2014) nd
that the political decentralization increased the district govern-
ments expenditure in general, but was most pronounced
specically in the sectors of education, housing, and agricul-
ture. They conclude that the success of political decentraliza-
tion that started in 2005 in increasing expenditures is in part
because the scal decentralization was already in place starting
in 2001. They argue that political decentralization is capable of
strengthening the incentives of local governments to better uti-
lize the opportunities provided by the scal decentralization.
The results from Kis-Katos and Sjahrir (2014) also conrm
the importance of scal decentralization of 2001 in increasing
expenditure in the sectors of education, health and physical
infrastructure. Soejoto, Subroto, and Suyanto (2015) nd that
decentralization in Indonesia had a positive impact on public
expenditure, economic growth, and human development index.
According to Sen (1989), development is the process of
expanding the substantive freedoms that people enjoy. Expen-
diture in areas such as education, health, food, medicine and
housing, as well as community projects in education, health,
and recreational facilities, should be seen as development as
long as they enhance peoples wellbeing and capabilities.
Another aspect that is important to discuss here is the time
frame within which decentralization could inuence local
development conditions. Faguets (2004) evaluation of Boli-
vias historic decentralization in 1994, based on a dataset from
1987 to 1996, nds the above-mentioned positive impacts after
just two years. The success in education and sanitation is seen
within seven years in Porto Alegre, Brazil (Santos, 1998). The
positive changes due to the political reservation policy for
women in India come eight years after the policy was put in
place in 1992. In the case of Indonesia, positive eects are vis-
ible as quickly as four years after the rst direct elections in
2005 (Skouas et al., 2014) and eight years after the scal
decentralization in 2001 (Kis-Katos & Sjahrir, 2014). In that
way, it can be said that as the districts increase spending in
the sector of health, education, housing, agriculture, and
physical infrastructure as a result of scal and political decen-
tralization, local economic conditions at the district level
improve.
(c) Local development and outmigration
An improvement in local development conditions through
employment generation, infrastructure development, access
to credit for small farmers, delivery of health-care services,
educational improvement, land reform, etc., could reduce
rural to urban migration. This seems intuitive because people
migrate in expectation of nding a better or more stable liveli-
hood, and of improving their social and economic status
(Lipton, 1982; Stark, 1991; Todaro, 1997). As the economic
conditions improve, there is less incentive to migrate to a dif-
ferent place. However, it can also be argued that local develop-
ment increases access to cities, strengthen ruralurban
integration, and raise education and skill levels. This could
lead to an increase in outmigration instead (Adepoju, 1983;
111
Rhoda, 1983; Campbell, 1988; Becker et al., 1994; Lucas,
1997; ILO, 1998).
Given that the role of local development in determining out-
migration is theoretically ambiguous, it becomes crucial to
empirically test it in the Indonesian context. Would the
improvement in service delivery and local development condi-
tions at the district level, as a result of decentralization and
democratization, alter the existing migration pattern? If yes,
in what direction?
Due to the historical reasons discussed earlier, the develop-
ment pattern in Indonesia is very uneven. Districts outside of
Java and Bali have been much less developed with fewer and
lower quality public services and amenities than those in the
islands of Java and Bali. There are dierences in the levels
of development even within the islands of Java and Bali.
For example, the districts in Jakarta are much more developed
than those in other parts of Java. It is also clear from the his-
torical migration patterns of Indonesia that migrants tend to
base their decisions to migrate on the prospect of economic
betterment (Van Lottum & Daan, 2012). Not surprisingly,
the direction of migration in Indonesia has been from outer
islands to inner islands of Java and Bali and migration ow
within the islands of Java and Bali is toward Jakarta.
In addition to assessing the general impact of direct elections
on public spending, Skouas et al. (2014) also investigated if
there was any heterogeneity of impacts between the inner
and the outer islands. They nd that the overall positive eect
of electoral reforms on public spending was primarily due to
increases in the districts outside Java and Bali. If the new dis-
trict heads outside of Java and Bali are responding to the
needs of their respective districts by investing more in educa-
tion, health, housing, and agriculture, what can be expected
to happen to the outmigration from these districts? Would
that be dierent from the outmigration patterns from the dis-
tricts in islands of Java and Bali?
It is important to note that the relative economic benet for
a move from outer islands to inner islands is much higher than
that from a district in Java and Bali to Jakarta. As a result, the
economic conditions in districts of outer islands would have to
improve a lot more than that in districts of inner islands in
order for potential migrants to be indierent to migrate.
And, it will take much more time for districts of outer islands
to get there than for those of inner islands. Since the data used
for the analyses are only three (IFLS) and ve (census) years
after the direct elections, the results might be constrained by
time, especially for assessing outmigration from districts of
outer islands. At the same time, the local development condi-
tions and migration are theoretically jointly determined. In
other words, they are not independent of each other or they
are simultaneously occurring at the same time (De Haas,
2010). Hence, local development spurred by the decentraliza-
tion process could inuence outmigration decisions within a
short period that data is available for.
To summarize, the goal of this paper is to empirically
explore the extent to which local development, spurred by
the democratization and decentralization at the district level,
aects outmigration. Separate analyses are undertaken to
assess whether there is a dierence in the outmigration pattern
from districts in outer islands and those in inner islands.
10
3. DECENTRALIZATION IN INDONESIA
Figure 2 shows a timeline of the Indonesian decentralization
process. For over three decades (19651998), Indonesia was
under a highly centralized and autocratic government (New
112 WORLD DEVELOPMENT
Figure 2. Phased in decentralization.
Order) under President Suharto. Despite that, the regime did
allow local political representation and local government.
Local governments in Indonesia were comprised of a regional
head, executive agencies, and the local assembly (DPRD).
During Suhartos regime, the local government representatives
were chosen from a closed list of candidates (mostly military)
by the Ministry of Home aairs. As a result, majority of basic
service delivery at the local level was still centrally controlled
and these local governments depended on earmarks and dis-
cretionary transfers as well as their own sources of limited rev-
enues (Malley, 2003).
As the Indonesian economy grew, there were growing pres-
sures for democratization and decentralization, especially after
the Asian Financial Crisis. This ultimately led to the downfall
of New Order regime in 1998 and subsequently, in June 1999,
Indonesia held its rst relatively free and fair elections in
34 years. The process created a new batch of more assertive
local legislatures (DPRD) and transferred the role of basic ser-
vice delivery to the district governments as opposed to pro-
vinces (World Bank, 2003). The local legislatures had the
general oversight of a larger consolidated local budget
(APBD) only in 2001. This round of elections is referred to
as indirect because the election appointed the local legislatures
who in turn selected local heads. Unlike Suhartos era, the new
legislatures could choose from more open lists of candidates,
and faced less interference from the Ministry of Home Aairs
(Decree No. 22/1999). The 1999 DPRD also had the right to
impeach local heads, which meant a premature end of their
terms. The ve-year tenure cycle of the local heads could only
be interrupted by death, illness, a no-condence vote, or the
creation of a new district. However, because the 1999 DPRD
also had the right to impeach local heads and end their terms,
the timing of the ve-year tenure cycle could dier across local
governments. The political powers of the local legislatures to
select, control, and even dismiss local heads raised concerns
on two fronts. First, DPRD seemed to overreach their powers
that blurred the roles of the legislative and executive agencies.
Second, DPRDs politics gave way to corruption among the
legislatures and local heads. Regional heads were found to
be accepting and distributing bribes to secure and maintain
oce of the head of local governments 11 (Mietzner, 2007).
The concerns about indirect political accountability led to
the second wave of local government electoral reform toward
direct elections (Pilkada Langsung) under Law No. 32/2004
(Erb & Sulistiyanto, 2009; Schiller, 2009). This reform made
the local head more directly accountable to the people by stip-
ulating that (s)he would be directly elected by citizens, and
provided a clearer denition of the heads political functioning
(Skouas et al., 2014). Based on the new Pilkada amendment,
the government decided to conduct the rst batch of direct
elections in June 2005 in the districts where the tenure of the
indirectly elected (by DPRD) district heads were ending. The
rst batch of direct elections concerned all the regional head
positions that had opened between December 2004 and April
2005. If the tenure of any DPRD head was due to end in the
rst few months of 2005, the government extended their terms
up to the direct elections. In some districts, a caretaker govern-
ment had been appointed prior to the December 2004 law,
who were allowed to serve till the elections. By June 2005,
155 districts had directly elected heads.
During 2001 and 2007, the number of districts increased from
336 to 434, and new district head positions opened up in the
newly created child districts. Given that it is dicult to compare
the migration levels of districts aected by splits, this analysis
covers only those districts that are unaected by splits.
Table A2 provides the number of districts that held direct elec-
tions during 200507. By 2007, 304 out of a total of 434 dis-
trictsincluding undivided and split districtshad held
direct elections. Among the 238 districts that remained undivided
during 200509, 163 districts had held direct elections for district
heads by 2007, including 101 districts that did so in 2005 12.
The most important feature of this process for the purpose
of this analysis is that the timing of the direct elections for
undivided local governments is random. The key reason
behind the phasing of the direct elections is the fact that the
regional governments in Indonesia were established at dier-
ent times since its independence. Hence, the ve-year tenure
cycle of the district heads was dierent for dierent districts
(Pratikno, 2009).
4. DATA
Several datasets are compiled together to conduct this anal-
ysis. First, the Pilkada dataset, assembled by Skouas et al.
(2014), oers electoral information for the period 19992009.
It provides the most relevant information for this analysis,
i.e., whether or not a district had undergone direct elections
(treatment) by 2005 or not. It also provides detailed informa-
tion on district characteristics, some of which are used as con-
trol variables. A district is considered treated if it went
through direct elections between June and December of 2005
and control if it did not. Among the 238 districts that did
not split, 101 districts went through direct elections between
June and December of 2005. The 62 districts that held it dur-
ing 200607 are omitted from the analyses for two reasons.
First, enough time needed to have passed between the treat-
ment period (JuneDecember 2005) and the post-treatment
survey (2008 for IFLS). Secondly, omitting these districts from
the analyses is justied because they can possibly belong to
either the treatment or control groups and including them is
likely to bias the impacts of direct elections. In summary,
migration behavior of households in 101 districts that had
elections in the treatment period is being compared with that
of households in 75 districts that did not have elections until
after post-treatment survey.
To establish causality, the fact that the direct elections were
in fact exogenous and can be used as a natural experiment
needs to be established (Meyer, 1995; Robinson, McNulty,
& Krasno, 2009). The exogeneity of the treatment relies on
two features: (a) direct elections in 2005 and later were held
in the districts where the terms of the previous district heads
 DOES LOCAL DEVELOPMENT INFLUENCE OUTMIGRATION DECISIONS?EVIDENCE FROM INDONESIA 113

were coming to full term, and (b) the cross-district variation in
when the previous district heads term ended was due to the
historical dierences in the timing of election (indirectly by
DRPD) or appointment (under the New order system) of that
individual.
In order to prove the exogeneity of the direct elections, two
probit models are analyzed. Table 1a presents the marginal
eects from the rst probit model on whether a district has
had direct elections during 200507 (as per Law 34/2004).
The explanatory variables include: characteristics related to
the district economy in 2005; share of urban population in
the district in 2005; whether a district is classied administra-
tively as a rural or an urban district; share of various sectors
such as agriculture, service, mining, manufacturing, and
nance to total district GDP; distances to province capital
and Jakarta; and regional xed eects.
Table 1b presents the marginal eects from the second
probit model of whether a district had an indirectly
appointed district head in 2000, as this was the year with
the highest proportion (36%) of district heads appointed
indirectly by the local district parliaments (DPRD) after
passing of the law (Law No. 22/1999). The explanatory vari-
ables include: characteristics related to the district economy
in 2000; share of urban population in the district in 2000;
whether a district is classied administratively as a rural or
an urban district; share of various sectors such as agricul-
ture, service, mining, manufacturing, and nance to total dis-
trict GDP; distances to province capital and Jakarta; and
regional xed eects.
The rst column in Tables 1a and 1b reports results for all
the districts, while the second column reports results for the
238 districts that did not split. For both models, almost

Table 1a. Marginal eects of the determinants of direct elections in districts in Indonesia
Dependent Variable: Treatment = 1 if District had a direct election between 2005 and 2007; Treatment = 0 otherwise
All districts Undivided districts
Unemployment rate in 2004 0.0044 0.0112
(0.792) (0.574)
Ln per capita Real GDP, 2005
0.0122
0.0532
(0.880) (0.568)
Share of Electricity to total GRDP, 2005
1.116
0.734
(0.746) (0.845)
Share of Mining to total GRDP, 2005 0.336 0.311
(0.376) (0.472)
Share of Agriculture to total GRDP, 2005 0.0044 0.275
(0.992) (0.570)
Share of Manufacturing to total GRDP, 2005 0.242 0.197
(0.540) (0.644)
Share of Construction to total GRDP, 2005 1.389 1.951*
(0.144) (0.093)
Share of Financial Services to total GRDP, 2005 0.576 0.248
(0.698) (0.873)
Share of Services to total GRDP, 2005 0.323
0.412
(0.611) (0.553)
Distance from district to Provinces capital 0.000 0.000
(0.257) (0.938)
Distance from district to Jakarta 0.000 0.0002*
(0.210) (0.075)
To Provinces capital, by Ocean or Land 0.0187
0.0029
(0.854) (0.982)
To Jakarta, by Ocean or Land 0.0077
0.0075
(0.950) (0.954
(1 = Kota; 0 = Kabupaten) 0.248 0.257
(0.116) (0.143)
Sumatera Island 0.002 0.0771
(0.988) (0.602)
Kalimantan Island
0.114
0.120
(0.460) (0.484)
Sulawesi Island
0.269
0.386**
(0.103) (0.041)
Nusa Tenggara & Maluku Island
0.351**
0.360*
(0.047) (0.086)
Papua Island
0.700**
0.640
(0.021) (0.127)
Urbanization (share of population that is urban), Susenas 2005
0.378
0.371*
(0.114) (0.178)
Share of villages in the district with asphalt road in 2005
0.0149 0.186
(0.923) (0.306)
Observations 315 222
*** ** *
Note: Robust standard errors adjusted for clustering at the province level; p < 0.001, p < 0.05, p < 0.1. p-Values are in parentheses.
114 WORLD DEVELOPMENT

Table 1b. Marginal Eects of the determinants of indirect elections in districts in Indonesia
Dependent Variable: Treatment = 1 if District had a head who was indirectly elected in 2000; Treatment = 0 otherwise
All districts Undivided districts
Unemployment rate in 1999
0.0395
0.0304
(0.198) (0.340)
Ln per capita Real GDP, 2000 0.106 0.0778
(0.295) (0.455)
Share of Electricity to total GRDP, 2000 0.997 1.887
(0.846) (0.743)
Share of Mining to total GRDP, 2000 0.053 0.026
(0.922) (0.967)
Share of Agriculture to total GRDP, 2000 0.428 0.283
(0.441) (0.626)
Share of Manufacturing to total GRDP, 2000 0.330
0.0625
(0.527) (0.907)
Share of Construction to total GRDP, 2000 0.633 2.494
(0.606) (0.108)
Share of Financial Services to total GRDP, 2000 2.164 1.518
(0.212) (0.408)
Share of Services to total GRDP, 2000 0.149
0.628
(0.860) (0.507)
Distance from district to Provinces capital 0.000
0.000
(0.901) (0.525)
Distance from district to Jakarta 0.000 0.0004***
(0.127) (0.005)
To Provinces capital, by Ocean or Land 0.119 0.219
(0.412) (0.280)
To Jakarta, by Ocean or Land 0.0955
0.118
(0.514) (0.549)
(1 = Kota; 0 = Kabupaten)
0.118
0.207
(0.586) (0.337)
Sumatera Island
0.116 0.039
(0.473) (0.842)
Kalimantan Island
0.299
0.372
(0.113) (0.112)
Sulawesi Island
0.411**
0.591***
(0.032) (0.008)
Nusa Tenggara & Maluku Island
0.364**
0.623***
(0.089) (0.011)
Papua Island

Urbanization (share of population that is urban), Susenas 2005 0.259 0.0139

(0.430) (0.970)
Share of villages in the district with asphalt road in 2005
0.0348 0.282
(0.862) (0.270)
Observations 226 164
*** **
Note: Robust standard errors adjusted for clustering at the province level; p < 0.001, p < 0.05, *p < 0.1. p-values are in parentheses.

all district-level characteristics are insignicant for both
specications, conrming that whether a district had an
indirect election in 2000 and whether a district had a direct
election in 2005 are independent of the district-level character-
istics that can potentially be inuential in determining migra-
tion. These results support the exogeneity of the direct
elections, and hence, allow for causal inferences in the analyses
performed using their variation in the districts as a natural
experiment.
The main outcome variablemigration statuscomes from
two sets of datasets: Indonesian Family life Survey (IFLS) and
the Indonesian Census. Both datasets have pros and cons in
terms of the information they oer. Thus, the results deduced
from separately analyzing the two dierent datasets are com-
plementary. The IFLS, conducted by RAND Corporation in
collaboration with several universities, is an ongoing survey.
The longitudinal survey collects information on individuals
living in 13 of the 27 provinces in the country and the sample
is representative of 83% of the Indonesian population. Cur-
rently, four IFLS waves have been conducted, respectively in
1993, 1997, 2000, and 2007/2008. IFLS dataset has had above
90% re-contact rates in every follow-up wave, making it rich
and informative. The most important IFLS feature for the
purposes of this analysis is that it collects both current and ret-
rospective information about the lives of the respondents. The
dependent variablehousehold 13 migration statusis identi-
ed mostly from the retrospective information collected as a
part of migration history survey. Detailed information on
household and household heads characteristics is also avail-
able. The 2000 and 200708 waves are used for the analysis.
Because IFLS constitutes of only 13 out of the 34 provinces
in Indonesia, the analysis based upon it may not present the
entire picture. To incorporate all of the 34 provinces, the
Indonesian census is used to do a separate analysis.
 DOES LOCAL DEVELOPMENT INFLUENCE OUTMIGRATION DECISIONS?EVIDENCE FROM INDONESIA
The Central Bureau of Statistics (Indonesia) has been collect-
ing population census every ten years and intercensal popula-
tion survey every ve years since 1971 14. The 10% sample of
2010 census and the 0.51% sample of 2005 intercensal popula-
tion survey are used for the analysis. 15 The census collects
information on all people, Indonesian and foreign residing
in the territorial area of Indonesia, regardless of the residence
status while the intercensal survey collects information on per-
manent residents only. The surveys have individuals demo-
graphic information, their social and economic activities,
and their household characteristics. However, the migration
module in the census does not provide a detailed account of
individuals migration history. The questionnaire only asks
where the individual was ve years ago. The migration vari-
able is constructed by checking whether one is in a dierent
district now than they are 5 years ago. This could be problem-
atic in the identication of people who migrated within the
5 years period. For example, if an individual had migrated
during 200609, it would not be recorded using the census.
Hence, the census will be useful in understanding a longer-
term migration pattern while the IFLS will be crucial in under-
standing temporary migration. Each dataset is discussed in
detail below.
(a) Indonesian family life survey
Figure 3 oers a chronology of the collection of datasets and
the identication strategy for migrant sending households for
IFLS. As discussed earlier, the treatment period is June to
December of 2005. The fourth wave of IFLS was conducted
between November 2007 and May 2008. The post treatment
period (t = 1) is considered to be the period of 29 months
between January 2006 and May 2008 and the pre-treatment
Figure 3. Timeline for Indonesian family life survey.
Figure 4. Timeline for census.
115
period (t = 0) is the same duration of time between January
2002 and May 2005. A household is considered a migrant-
sending household in a period t if at least one member of
the household has migrated for at least six months.
The IFLS dataset also contains retrospective information on
migrants who had returned by the time of the survey as well.
This feature is useful not only because it helps include migrants
who would otherwise not be identied, but also because it
makes the construction of a panel over two periods possible.
A total of 12,692 households living in 263 districts 16 were sur-
veyed in the fourth wave of IFLS. Out of the 263 districts rep-
resented in IFLS, 137 districts did not split and did not go
through direct elections between January 2006 and December
2007. A balanced panel over two periods with 5,086 households
in each period is constructed for the analysis. The sample attri-
tion is not systematic, implying that there are no unique char-
acteristics among those who do not appear in the fourth wave.
Around 57% of the households lived in districts that went
through direct elections between June and December of 2005.
(b) Census
Figure 4 oers a chronology of the collection of datasets and
the identication strategy for migrant sending households for
the census sample. The treatment period is June to December
of 2005. The 2010 census was conducted in May of 2010 and
the 2005 census was conducted in June 2005. 17 The period
between the 2000 and 2005 Census is considered the pre-
treatment period (t = 0) and the period between the 2005
and 2010 Census is considered the post-treatment period
(t = 1). A household is considered a migrant-sending house-
hold in a period t if at least one member of the household
has migrated for at least six months.
116 WORLD DEVELOPMENT

Among the 238 districts that did not split, 101 districts went
through direct elections between June and December of 2005.
The 62 districts that held elections during 200607 are omitted
from the analyses to avoid any contamination. This leaves a
total of 176 18 districts, all of which are represented in the cen-
sus. A total of 114,482 households from these 174 districts
appear in the 2005 census sample while a total of 3,360,798
households from these 174 districts appear in 2010 census sam-
ple 19. Hence, a repeated cross-sectional data over two periods
is constructed. In 2005, around 58% of the households lived in
districts that went through direct elections between June and
December of 2005.
In addition to showing that the treatment was random, the
descriptive statistics and the results of the test of dierences in
means of household, household heads characteristics, district-
level characteristics, and the household migration behavior in
IFLS baseline and census baseline are respectively presented in
Tables 2 and 3. Most of the variables, such as household
heads age, education, gender, and marital status are self-
explanatory. Some that need explanations are discussed
below. As Islam is the most practiced religion in Indonesia,
the religion variable takes on a value of 1 if the household
head is a Muslim and 0 if they practice any other religion.
The oor is considered concrete if they are made up of

Table 2. Summary statistics by treatment status for IFLS

Control Treatment Test of dierence Bias
Mean Standard Deviation Mean Standard Deviation
A. Controls
Household heads Characteristics
Age 45.96 14.49 45.59 14.87 0.378 15.66
Gender 0.785 0.411 0.814 0.383 0.001 27.50
Marital Status 0.813 0.389 0.832 0.373 0.076 24.33
No Education 0.165 0.371 0.116 0.320 0.000 36.91
Some Education 0.753 0.43 0.813 0.39 0.000 46.85
Undergraduate and above 0.080 0.272 0.071 0.257 0.200 14.28
Religion is Islam 0.93 0.254 0.872 0.334 0.000 37.81
Household Characteristics
Household size 5.50 2.69 5.28 2.506 0.002 31.82
Concrete Floor 0.765 0.424 0.701 0.458 0.000 52.27
Toilet 0.663 0.473 0.629 0.483 0.015 49.18
Electricity 0.918 0.273 0.893 0.308 0.002 24.79
District Characteristics
Unemployment rate in 2002 0.0519 0.0179 0.049 0.0178 0.4970 217.06
Ln per capita Real GDP, 2003 15.43 0.695 15.35 0.602 0.447 32.57
Share of Electricity to total GRDP, 2003 0.012 0.008 0.011 0.013 0.755 13.80
Share of Mining to total GRDP, 2003 0.0459 0.137 0.0473 0.130 0.955 4.579
Share of Agriculture to total GRDP, 2003 0.269 0.203 0.284 0.191 0.662 30.85
Share of Manufacturing to total GRDP, 2003 0.195 0.184 0.155 0.137 0.128 46.05
Share of Construction to total GRDP, 2003 0.0449 0.0246 0.0526 0.0314 0.108 55.80
Share of Financial Services to total GRDP, 2003 0.051 0.0279 0.054 0.0343 0.555 21.26
Share of Services to total GRDP, 2003 0.114 0.060 0.128 0.0638 0.173 91.28
Distance from district to Provinces capital 83.97 52.86 84.45 61.15 0.960 2.208
Distance from district to Jakarta 657.34 480.49 764.4 421.41 0.234 65.59
To Provinces capital, by Ocean(1) or Land(0) 0.0323 0.178 0.024 0.152 0.750 12.67
To Jakarta, by Ocean(1) or Land(0) 0.452 0.502 0.553 0.500 0.228 319.07
(1 = Kota; 0 = Kabupaten) 0.290 0.457 0.305 0.463 0.840 28.55
Urbanization (share of population that is urban), Susenas 2003 0.493 0.328 0.468 0.353 0.673 27.09
Share of villages in the district with asphalt road in 2003 0.7094 0.246 0.722 0.231 0.760 21.06

B. Overall Migration
Pre-treatment Migration 0.137 0.343 0.1586 0.365 0.0282
Post-treatment Migration 0.1405 0.347 0.138 0.345 0.831

C. Migration from Java and Bali

Pre-treatment Migration 0.123 0.328 0.155 0.361 0.009
Post-treatment Migration 0.125 0.331 0.124 0.329 0.925
D. Migration from other regions
Pre-treatment Migration 0.167 0.373 0.165 0.371 0.912
Post-treatment Migration 0.176 0.381 0.161 0.367 0.397
Number of Observations by treatment 2,205 2,881
Total Number of Observations (per period)a 5,086
a
IFLS consisted of 147 districts that did not split and did not go through direct elections in 2006 and 2007. Out of the 147, 85 had direct elections between
June and December 2005. Out of the 87 districts in Java and Bali, 47 districts had gone through direct elections between June and December of 2005. And
out of the 60 districts in other regions, 38 districts went through direct elections between June and December of 2005.
 DOES LOCAL DEVELOPMENT INFLUENCE OUTMIGRATION DECISIONS?EVIDENCE FROM INDONESIA 117

Table 3. Summary statistics by treatment status for census

Control Treatment Test of dierence Bias
Mean Standard Deviation Mean Standard Deviation
A. Controls
Household heads Characteristics
Age 46.00 13.85 46.19 13.87 0.024 36.08
Gender 0.861 0.346 0.867 0.339 0.008 12.25
Marital Status 0.832 0.374 0.838 0.368 0.004 12.72
No Education 0.122 0.327 0.119 0.324 0.245 9.60
Some Education 0.758 0.429 0.769 0.422 0.000 20.15
Undergraduate and above 0.121 0.326 0.112 0.315 0.000 15.15
Household Characteristics
Household size 4.04 1.76 3.98 1.73 0.000 26.22
Concrete Floor 0.887 0.317 0.875 0.330 0.000 18.5
Toilet 0.789 0.407 0.792 0.405 0.286 10.52
Rural 0.523 0.499 0.547 0.498 0.000 107.49
District Characteristics
Unemployment rate in 1999 0.0302 0.0169 0.0286 0.0172 0.5625 70.74
Ln per capita Real GDP, 2000 15.23 0.663 15.21 0.589 0.859 9.29
Share of Electricity to total GRDP, 2000 0.010 0.008 0.009 0.011 0.778 18.73
Share of Mining to total GRDP, 2000 0.0567 0.172 0.0505 0.134 0.798 8.13
Share of Agriculture to total GRDP, 2000 0.283 0.212 0.295 0.194 0.677 19.85
Share of Manufacturing to total GRDP, 2000 0.173 0.179 0.151 0.137 0.344 27.00
Share of Construction to total GRDP, 2000 0.0475 0.0282 0.0532 0.0334 0.236 45.04
Share of Financial Services to total GRDP, 2000 0.0484 0.029 0.0504 0.0327 0.680 18.71
Share of Services to total GRDP, 2000 0.119 0.060 0.129 0.067 0.289 47.43
Distance from district to Provinces capital 99.38 89.48 92.85 78.01 0.609 21.06
Distance from district to Jakarta 838.75 636.38 875.08 518.37 0.679 13.91
To Provinces capital, by Ocean(1) or Land(0) 0.067 0.251 0.0505 0.220 0.652 19.31
To Jakarta, by Ocean(1) or Land(0) 0.547 0.501 0.616 0.489 0.359 89.52
(1 = Kota; 0 = Kabupaten) 0.307 0.464 0.313 0.466 0.928 19.67
Urbanization (share of population that is urban), Susenas 2000 0.501 0.332 0.445 0.352 0.304 67.71
Share of villages in the district with asphalt road in 2000 0.698 0.241 0.683 0.251 0.683 30.24

B. Overall Migration
Pre-treatment Migration 0.086 0.2804 0.0909 0.288 0.004
Post-treatment Migration 0.084 0.277 0.0749 0.263 0.000

C. Migration from Java and Bali

Pre-treatment Migration 0.0729 0.259 0.077 0.266 0.051
Post-treatment Migration 0.0805 0.272 0.066 0.249 0.000

D. Migration from other regions

Pre-treatment Migration 0.104 0.306 0.105 0.307 0.726
Post-treatment Migration 0.095 0.294 0.096 0.295 0.417
Number of Observations by treatment 48,510 65,972
Total Number of Observations (per period)a 114,482
a
Census consisted of all the 174 districts that did not split and did not go through direct elections in 2006 and 2007. Out of the 176, 101 had direct elections
between June and December 2005.

ceramic, marble, granite, stone, tiles, terrazzo, cement, or ent from the districts that did not go through direct elections.
bricks. The district-level controls include: characteristics Rosenbaum and Rubin (1985) have suggested an alternative
related to the district economy (per capita GDP and unem- for assessing whether a comparable counterfactual has been
ployment rate); share of urban population; whether the district found. The process entails the estimation of bias by examining
is classied administratively as a rural or an urban district; the standardized dierence in means and is dened as below:
share of various sectors such as agriculture, service, mining,  
 
manufacturing, and nance to total district GDP; distances X
X 
 T C 
to province capital and Jakarta; and regional xed eects. bias 100   q  1
 s2T
s2C 
Based on the test of dierences in means of the household  2

and household heads characteristics, the treatment and con-
trol households look signicantly dierent in both datasets. where X T and sT are the mean value and standard deviation
However, the districts that went through direct elections for the exogenous factor for the treatment group being consid-
between June and December of 2005 are not statistically dier- ered and X T and sT are similarly dened for the control group.
118 WORLD DEVELOPMENT
Bias is simply the normalized dierence in mean between two
groups in absolute value. Unlike the raw dierence in means, it
accounts for the dierences in sample sizes and standard devi-
ations between control and treatment groups. As evident from
Tables 2 and 3, there is a variation in sample size and standard
deviations between the two groups. Rubin (2001) suggests that
a bias less than 25% provides evidence that the counterfactual
is reliable. We nd that the bias is higher than 25% for more
than half of the characteristics in IFLS and for a third of
the characteristics in the census. In addition, the two methods
do not provide consistent results. For example, the test of dif-
ferences in means shows that based on strictly all district-level
characteristics, the treatment and control districts are not sta-
tistically dierent. However, bias calculation shows a bias of
more than 25% for many of these characteristics. In order to
get an unbiased estimate of the impact, dierence in dierence
with conditioning variables approach is employed. The above-
described biases are accounted for by controlling for the dier-
ences in observables between treated and control groups.
Part B of Tables 2 and 3 consists of the pre-treatment and
post-treatment migration outcomes for the entire sample of
IFLS and the census, respectively. For both of these samples,
the migration patterns are statistically dierent in the pre-
treatment period with a slightly higher average migration from
the treatment districts. In the IFLS sample, the migration rate
barely increased for the control districts while it decreased for
the treatment in the post-treatment period. As a result, the
migration patterns are not statistically dierent in the post
treatment period. For the census, however, migration rates
slightly decreased in both treatment and control districts and
remain statistically dierent in the post-treatment period.
Part C reports the same statistics for districts in Java and
Bali. The averages for these regions in the IFLS sample
matched that of the entire sample in IFLS. On the other hand,
in the census sample, migration rate increased from the control
districts while it decreased from the treatment districts.
Although the changes are very small, the migration rate of
the treated districts is signicantly dierent from that of the
control districts in both periods. Lastly, Part D reports the
pre-treatment and post-treatment migration outcomes for dis-
tricts outside of Java and Bali. The migration rates from the
regions outside of Java and Bali were higher than those from
Java and Bali, which is consistent with the historical internal
migration pattern in Indonesia. In the IFLS sample, migration
rate increased for the control districts while it decreased
slightly for the treated districts. And, in the census, migration
rate decreased for both groups. Additionally, migration pat-
terns for the two groups are not statistically dierent for both
periods in both samples.
(c) Econometric specication
The goal of this paper is to study the relationship between
local development and households outmigration behavior.
For this purpose, the exogenous direct elections held at the
district level are used as a natural experiment. In order to
assess the households migration behavior as a response to
the increase in the expenditure at the district level after the
direct election, migration behavior in households that
belonged to the district that went through the direct elections
should be compared to what migration behavior would have
been had the districts not gone through direct elections. The
identication of this unobserved counterfactual is the well-
known challenge in impact evaluation literature. Measuring
impact as the dierence in mean outcomes between treatment
and control households may give a biased estimate of the
impact.
The bias arises if there are unobserved characteristics that
aect the probability of participation in the program that
are also correlated with the outcome of interest (migration,
in this case). In order to get an unbiased estimate of the impact
of the program, three sources of bias need to be accounted for:
observable dierences, time-invariant unobservable dier-
ences, and time-variant unobservable dierences between the
treatment and control groups. Dierence in dierence with
conditioning variables 20 is used because it accounts for the
rst two sources. By denition, dierence-in-dierence method
controls for the time-invariant unobservable dierences
between control and treatment because the time-invariant
unobservable dierences are canceled out by taking the dier-
ence in outcomes over time (Meyer, 1995; Wooldridge, 2010;
Gertler, Martinez, Premand, Rawlings, & Vermeerch, 2011).
Inclusion of these variables not only controls for any remain-
ing dierences between households prior to the direct elections
but also increases precision of the estimates (Angrist &
Pischke, 2008).
The last source of bias could arise from the violation of the
key identifying assumption (parallel trend assumption) of the
dierence in dierence method. This assumption postulates
that there are not any time-variant observable dierences
between treatment and control group. In other words, in the
absence of the treatment, the trend in treatment and control
would have been the same. The parallel trend assumption is
tested in two ways. First, average migration rates from 1990
to 2008 for the treatment and control groups are plotted as
shown in Figure A1 in the Appendix. Migration rate is steady
for both groups until June of 2005 (treatment). After the treat-
ment, however, there is a fall in the migration rate for the
treatment group but a slight increase among the control
group. But because the average migration rates between these
two groups are not very dierent before the treatment, the
graph does not show a clear parallel trend. Thus, additionally,
a placebo test is performed as a second test by redoing the dif-
ference in dierence method with pre-treatment period as
time = 1 and a period prior to the pre-treatment as time = 0.
The placebo dierence-in-dierence estimate is not dierent
from zero. This implies that the parallel trend assumption
holds and time-variant unobservable dierences do not exist
in this case (Khandker, Koolwal, & Samad, 2010) (Results
provided in Table A1).
The following model captures the impact of the direct elec-
tion on the households decisions to migrate:
M it b0 b1 T ij b2 t b3 T ij  t cK it dLjt
it 2
where Mit is a binary variable that takes a value of 1 if any
member from household, i, migrated in period t and 0 if
not. The period, t is equal to 0 for the pre-treatment period
and 1 for the post-treatment period. Hence, this model design
entails two observations for each household i: one for the
baseline period and one for the post treatment period. The
variable, Tij, is also a binary variable that takes a value of 1
if the household i belongs to a district j that went through
direct elections between June and December of 2005 (treat-
ment) and 0 if it does not. The variable, t * Tij, is the interac-
tion of time and treatment, and Kit is a vector of control
variables such as household and household heads characteris-
tics in time t. Ljt is a vector of district characteristics, and 2it is
the random error.
 DOES LOCAL DEVELOPMENT INFLUENCE OUTMIGRATION DECISIONS?EVIDENCE FROM INDONESIA
Figure A1. Test of parallel trend assumption.
This equation is estimated using a probit regression. Robust
standard errors are estimated and intra-cluster correlations are
accounted for by clustering the standard errors at the province
level in all the regressions. In this equation, b1 captures the
time-invariant unobservable dierences between treatment
and control households, and b2 captures any general trends
over time not caused by the treatment status. The coecient
of interest, b3, is an estimate of impact as it captures the dier-
ence in the change over time in treatment versus control house-
holds. This estimate is simply the average treatment eect of
the direct election on migration decisions (Heckman,
Lalonde, & Smith, 1999). The DD estimate of the impact of
direct election on migration is the test of the null hypothesis
that b3 = 0.
5. RESULTS
The marginal eects of the probit model for the above-
discussed specication using IFLS and the census 21 are pre-
sented in Tables 4 and 5 respectively. Equation (1) is estimated
stepwise: column (1) provides estimates without any controls;
column (2) with only household and household heads charac-
teristics; column (3) with household and household heads
characteristics and district-level characteristics and regional
xed eects. Column (4) provides marginal eects for just Java
and Bali region and column (5) provides marginal eects for
regions outside Java and Bali.
(a) Entire samples
The relationship between the direct elections and migration
behavior can be understood by analyzing three marginal
eects associated with treatment (b1 ), time (b2 ), and the inter-
action between time and treatment (b3 ). These marginal eects
help us understand the dierences in the migration behavior
between households in districts that went through direct elec-
tion (treatment) and those that are in districts that did not
(control) from the pre-direct election to post-direct election
119
period. The marginal eect associated with the time variable
is negative and statistically signicant only when household
and household head characteristics are controlled for in col-
umn (2). This implies that levels of migration have decreased
in both treatment and control groups over this period but
the signicance of the trend disappears once district and regio-
nal controls are introduced in column (3). The marginal eects
associated with the treatment variable are positive suggesting
slightly higher migration levels for households in districts that
have gone through direct election. However, this estimate is no
longer signicantly dierent from zero when district- and
regional-level characteristics are controlled for, indicating that
the dierences are negligible once the dierences in district-
level are accounted for.
The marginal eects associated with the interaction between
the treatment and time variable indicate that direct elections
had a negative impact on migration. For the overall IFLS
sample, it is evident from column (3) that a household in a dis-
trict that went through direct elections between June and
December of 2005 is 2.89 percentage points less likely to be
a migrant-sending household. This is a large eect, equal to
19.26%, 22 given the average migration rate of around 0.15
at baseline. The analysis done with the census sample shows
similar resultsa household in a district that went through
direct election between June and December of 2005 is 0.72 per-
centage points less likely to be a migrant-sending household.
This is an 8% decrease given an average migration rate of
around 0.09. Additionally, the marginal eects of the interac-
tion term on migration are almost exactly same across the col-
umns and are statistically signicant, especially in the IFLS
sample. This indicates that there is little reason to believe that
any bias is being introduced by heterogeneity in the eect of
the domestic economic conditions across households in the
Indonesian districts, or by any other unobserved heterogeneity
correlated with control variables.
Based on the results from column (3), it is evident that
households with older and married heads are signicantly less
likely to send migrants. In comparison to household heads
without any education, those with some or higher (undergrad-
120 WORLD DEVELOPMENT

Table 4. Dierence in dierence model using IFLS dataset

(1) (2) (3) (4) (5)
Entire Sample Entire Sample Entire Sample Java and Bali Islands Not Java and Bali Islands
Time 0.0042
0.013**
0.006
0.002
0.013
(0.591) (0.029) (0.351) (0.573) (0.438)
Treatment 0.022* 0.0263** 0.0177 0.046***
0.007
(0.101) (0.022) (0.288) (0.000) (0.801)
Time * Treatment
0.0239**
0.0232**
0.0289**
0.039***
0.012
(0.024) (0.030) (0.023) (0.003) (0.662)
Household characteristics NO YES YES YES YES
Household heads age
0.001***
0.001***
0.001***
0.001***
Household head is male
0.012**
0.009
0.000
0.027
Household head is married
0.453***
0.051***
0.056***
0.034
Household head has some education 0.059*** 0.059*** 0.039* 0.09***
HHs education > undergrad or higher 0.121*** 0.121*** 0.097*** 0.149***
Household size 0.009* 0.012*** 0.011*** 0.016***
Household size squared 0.0003 0.0001 0.000 0.000
Household has concrete oors 0.019** 0.022*** 0.018* 0.029*
Household has a toilet 0.019* 0.017* 0.032***
0.01
Household has electricity 0.0013
0.002
0.006
0.007
Religion is Islam 0.009 0.029* 0.047
0.0001
District characteristics & Regional controls NO NO YES YES YES
Unemployment rate in 2002
0.058
1.05** 1.61***
Ln per capita Real GDP, 2003
0.004 0.013** 0.0002
Share of Electricity to total GRDP, 2003
1.91***
1.61***
1.378
Share of Mining to total GRDP, 2003
0.126
0.188* 0.051
Share of Agriculture to total GRDP, 2003
0.082
0.044 0.071
Share of Manufacturing to total GRDP, 2003
0.101
0.072
0.154
Share of Construction to total GRDP, 2003
0.126
0.137* 0.04
Share of Financial Services to total GRDP, 2003 0.296 0.332
0.028
Share of Services to total GRDP, 2003
0.179 0.069
0.289
Distance from district to Provinces capital 0.0002* 0.001***
0.0001
Distance from district to Jakarta 0.000
0.000** 0.000
To Provinces capital, by Ocean(1) or Land(0)
0.088***
0.039
0.063***
To Jakarta, by Ocean(1) or Land(0) 0.048*** 0.064***

(1 = Kota; 0 = Kabupaten)
0.046
0.085*** 0.131
Share of population that is urban, Susenas 2003 0.081 0.164***
0.18**
Share of villages with asphalt road in 2003
0.039*
0.028*** 0.12**
Total Number of Observations 10,172 10,172 10,172 6,550 3,622
*** **
Note: Robust standard errors adjusted for clustering at the province level; p < 0.001, p < 0.05, *p < 0.1. p-values are in parentheses.

Table 5. Dierence in dierence model using census

(1) (2) (3) (4) (5)
Entire Sample Entire Sample Entire Sample Java and Bali Islands Not Java and Bali Islands
Time
0.002
0.013***
0.006** 0.000
0.010**
(0.807) (0.000) (0.042) (0.900) (0.045)
Treatment 0.0045 0.0085** 0.0108*** 0.0123*** 0.0103
(0.457) (0.038) (0.007) (0.007) (0.120)
Time *
Treatment
0.0135*
0.0080*
0.0072*
0.012*** 0.001
(0.096) (0.057) (0.053) (0.000) (0.873)
Household heads characteristics NO YES YES YES YES
District characteristics & Regional controls NO NO YES YES YES
Number of Observation 3,475,277 3,475,277 3,475,277 2,545,276 930,001
*** **
Note: Robust standard errors adjusted for clustering at the province level; p < 0.001, p < 0.05, *p < 0.1. p-Values are in parentheses

uate and above) education are more likely to send migrants.
Also, households with concrete oors and toilets are more
likely to send migrants than those without. As the poorest
are less likely to have education, concrete oors, and toilets,
the results indicate that the poorest households are less likely
to migrate. This is in line with the literature that conceptual-
izes the non-linearity or the inverted U-curve in the relation-
ship between migration and development process (Martin &
Taylor, 1996).
Among the district-level characteristics, households in dis-
tricts with a higher share of energy and electricity sector in
the real GDP are less likely to migrate. The results suggest that
 DOES LOCAL DEVELOPMENT INFLUENCE OUTMIGRATION DECISIONS?EVIDENCE FROM INDONESIA
migration is lower out of districts from where province capi-
tals can be reached by ocean than by land and higher out of
those from where Jakarta can be reached by ocean rather than
by land. Lastly, likelihood of migration decreases as the share
of villages with asphalt roads increase in districts.
(b) For Java and Bali
The results are similar when the sample is restricted to the
districts in Java and Bali only. On average, before direct elec-
tions, the households in districts that went through direct elec-
tions participated in migration more than those in districts
that did not. The general trend in migration is on the decline
for the period between pre-treatment and post-treatment.
According to the results from column (4) of Table 4, a house-
hold in a district in Java and Bali that went through direct
elections between June and December of 2005 is 3.9 percent-
age points less likely to be a migrant-sending household. This
is a large eect, equal to 27.85%, given the average migration
rate of around 0.14. The analysis done with the census sample
shows the same resulta household in a district in Java and
Bali that went through direct election between June and
December of 2005 is 1.2 percentage points less likely to be a
migrant-sending household. This is a 16% decrease given an
average migration rate of around 0.075. The concern that
these results could be driven by the lack of variability of treat-
ment in Java and Bali is invalid in this case because around
53% of the districts in Java and Bali went through elections
in 2005.
(c) For other provinces
For both IFLS and census samples, the direct elections had
no impact on the migration pattern of districts in other pro-
vinces.
For districts in Java and Bali as well as those outside, house-
hold heads with some years of education or undergraduate
(and higher) degrees are more likely to send migrants. House-
holds with concrete oors are more likely to send migrants than
those without. For districts outside of Java and Bali, the likeli-
hood of migration increases with increases in the unemploy-
ment rate of the previous year, decreases for districts from
where the province capital can be reached by ocean, decreases
as the share of urban population increases, and increases when
the share of villages with asphalt roads increases.
However, some interesting distinctions emerge for districts
in Java and Bali. The unemployment rate for the year prior
to the baseline signicantly decreases the probability of outmi-
gration in districts in Java and Bali. Migration is lower from
districts that are in a kota (city) versus from districts in a kabu-
paten (regency 23). As people are already in areas with rela-
tively more abundant economic opportunities, instead of
looking elsewhere, people could be persistent in looking for
jobs where they are. It could also be the case that as more peo-
ple lose their jobs, they are unable to aord to migrate. A
higher GDP is associated with a higher probability of outmi-
gration but as the share of energy and electricity, mining
and construction in the GDP of the districts increases, it
decreases. The higher the distance from the district to Jakarta,
the less likely are households to have migrants. The likelihood
of migration also increases if the travel to Jakarta is by the
ocean. A higher share of urban population increases the like-
lihood of outmigration in districts in Java and Bali. For exam-
ple, as the population in Jakarta has increased over time,
people have moved out of Jakarta city to its suburban areas.
This is not surprising for islands that make up only 7% of
121
the total land area of Indonesia and host 60% of its total pop-
ulation. As the share of villages with asphalt road increases,
the likelihood of outmigration decreases for districts in Java
and Bali.
In conclusion, analysis performed separately using two dif-
ferent datasets provides consistent results. The results indicate
that districts that went through direct elections were signi-
cantly less likely to have outmigration. The results for the
entire sample give us a general picture of how the movements
across the archipelago are aected by decentralization and
democratization. For a country like Indonesia, where internal
migration has been historically abundant and driven by the
disparities in economic opportunities and amenities, it is inter-
esting to nd that overall migration is decreasing. But this
decrease was mostly driven by the fall in the outmigration
from districts in Java and Bali.
The dierence in results between inner and outer islands can
be attributed to the dierence in relative economic benet
from migration. The incentive to migrate to Jakarta decreases
with very little improvement in economic conditions in the dis-
tricts in Java and Bali. In contrast, despite the higher increase
in service delivery and a relatively big improvement in eco-
nomic conditions, the incentive to migrate to inner islands
remains unaected for districts in the outer islands. Although
the outmigration from districts of outer islands was unaf-
fected, it is noted that it did not continue with the increasing
trend as shown in Figure A1. This is contrary to
Tirtosudarmos (2009) speculation that migration to Java
and Bali would respond to the labor market economy and
thus, would increase.
6. CONCLUSION
Most of the literature studying the relationship between
migration and local development is on the role of migration/
remittances on development outcomes of the migrant sending
households and communities (Antman, 2011; Gibson et al.,
2011; Adams, 2006; Hildebrandt & McKenzie, 2005; Acosta
et al., 2007; Yang, 2008). The literature, especially empirical,
studying the extent to which local development facilitates or
deters outmigration is less abundant because the causality in
this direction is theoretically ambiguous and empirically di-
cult to establish. On one hand, improving living conditions
and diversifying economic opportunities in rural areas is
expected to decrease migration to cities (Lipton, 1982;
Rhoda, 1983; Todaro, 1997). On the other hand, rural devel-
opment activities also increase access to cities, strengthen
ruralurban integration, and raise education and skill levels.
This could lead to an increase in ruralurban migration
instead (Adepoju, 1983; Becker et al., 1994; Campbell, 1988;
Lucas, 1997; Rhoda, 1983; ILO, 1998).
The endogenous relationship between development and
migration makes it dicult to empirically establish causality.
In order to overcome this issue, the analyses presented in this
paper take advantage of the exogenous phasing of district-
level elections in Indonesia. With historically established
uneven development paths and internal migration patterns,
Indonesia makes an ideal case study. The empirical analyses
conducted using two dierent datasets (IFLS and the
Indonesian census) provide coherent results.
Policy makers believe that it will aid in eciently delivering
services at the local level, thus improving living conditions and
diversifying economic opportunities in rural areas. Even
though the initial empirical evidence on the role of democrati-
zation and decentralization were mixed, more rigorous studies
122 WORLD DEVELOPMENT
recently have shown that decentralization could be key in
enhancing local economic development by more ecient ser-
vice delivery (Lin & Liu, 2000; Faguet, 2004; Iimi, 2005;
Stansel, 2005; Hammond & Tosun, 2011).
This discussion is of central importance not only in the
Indonesian contextwhere rising population pressures in Java
and Bali provinces are its governments ultimate challenge
but also in the development literature, where empirical
research on such questions is made impossible by the lack of
appropriate data and context. Indonesia, with historically
uneven development paths and high internal migration rates,
is an ideal case study. Its journey from decades of highly cen-
tralized government into democratization and decentralization
in the twenty rst century, coupled with the availability of rel-
atively good datasets, make this study possible.
The results provide empirical evidence that a household in a
district that went through direct election between June and
December of 2005 is 19.26% (IFLS) and 8% (census) less likely
to be a migrant-sending household. The results for the entire
sample give us a general idea of how the internal migration
across the archipelago is aected by democratization and
decentralization at the district level. For a country like
Indonesia, where internal migration has been historically
abundant and driven by the disparities in economic opportu-
nities and amenities, a decline in overall migration as a result
of democratization and decentralization is a promising result.
The decline in migration is particularly strong for districts
that are in Java and Bali - a household in a district in Java
and Bali that went through direct election between June and
December of 2005 is 27.85% (IFLS) and 16% (Census) less
NOTES
1. Discussed in detail later.
2. To name a few.
3. Some countries that have undergone decentralization in the past few
decades are Bolivia, Brazil, India, China, Indonesia, South Africa, Uganda,
etc.
4. Bali is another province that is a popular migrant destination. Even
though agriculture is still the biggest employer in Bali, in the last three
decades, tourism has taken o and is now the biggest contributor to its GDP.
5. Due to Javas importance as a core region, other parts of the
archipelago are usually referred to collectively as the Outer Islands.
6. During 190531, 27,338 people were moved, an average of only 1013
per year (Jones, 1979).
7. According to Tiebout (1956) model, as dierent local governments
oer dierent public tax-expenditure bundles, individuals are expected to
move based on their preferences. According to this model,
decentralization could directly lead to migration. However, Bardhan
(2002) points out that this assumption fails in the context of poor
countries even in mobile societies. It is unlikely that relatively poor people
take on an expensive process, such as migration, in search of higher
welfare benets.
8. Some countries that have undergone decentralization in the past few
decades are Bolivia, Brazil, India, China, Indonesia, South Africa,
Uganda, etc.
likely to be a migrant-sending household. The results indicate
that even though the internal migration rate is on the decline,
it is driven by the reduction in the likelihood of outmigration
from the districts in Java and Bali. In general, the decision to
migrate is driven by the need for economic betterment. The
relative economic benet for a move from outer islands to
inner islands is much higher than that from a district in Java
and Bali to Jakarta. As a result, just a small improvement in
the economic conditions in districts of Java and Bali leads to
a decrease in outmigration. In contrast, outmigration from
districts outside of Java and Bali are unaected despite a big-
ger improvement in their economic conditions.
In comparison to the districts in Java and Bali, it may take
longer for local development to inuence outmigration in dis-
tricts outside of Java and Bali because of a bigger gap in the
levels of development between the inner and outer islands.
The 2010 census is only ve years after the election but it does
not provide detailed migration information on the years
between the censuses. The fourth wave of IFLS provides retro-
spective migration information, but it was conducted barely
two and half years after the elections.
This paper takes advantage of a unique natural experiment
to study the migration-development interaction that has not
been empirically explored before and is theoretically ambigu-
ous. The results support the theory that improvement in eco-
nomic conditions reduces outmigration from districts in
Indonesia. This is true especially for districts in Java and Bali,
where outmigration fell despite a smaller improvement in eco-
nomic conditions in comparison to the districts outside of Java
and Bali.
9. Stansel (2005) and Hammond and Tosun (2011) conrm the results for
the United States.
10. Based entirely on the description provided by Skouas et al. (2014).
11. Malley (2003) reports that a bribe, equivalent to the price of a car,
could virtually guarantee staying in oce or getting an election vote.
12. All districts had direct elections by 2010.
13. Even though the treatment is at the district level, all the analyses are
done at the household level following the New Economics of Labor
Migration model that the decision to migrate is a joint household decision,
with the household sharing the costs and benets of migration with the
migrant through an explicit or implicit sharing rule.
14. The censuses were in 1971 1980, 1990, 2000 and 2010. The rst
intercensal survey was in 1976, 1985, 1995 and 2005.
15. These samples are made available by Integrated Public Use Micro-
data Series, International (IPUMSI).
16. There were 434 districts in Indonesia in 2007.
17. The identication of migrant sending households in the post
treatment period is unclear in the post treatment period. The 2010 census,
conducted in May 2010, asks the individuals location from ve years ago
(May 2005). This means that individuals who migrated in May of 2005
and the treatment period (June to December of 2005) are included in the
migrant identication in the post treatment period.
 DOES LOCAL DEVELOPMENT INFLUENCE OUTMIGRATION DECISIONS?EVIDENCE FROM INDONESIA
18. Two districts held direct elections before June of 2005 and are not
included in the analyses.
19. IPUMSI provides a 0.51% sample of the 2005 census and a 10%
sample of the 2010 census, which is why, 2010 census sample consists of
more households than the 2005 census. Results remain consistent when a
smaller sample of the 2010 census is used as well.
20. Another strategy to address the observable dierences would be to
use dierence in dierence with inverse propensity weightings. Results
using this method are consistent with those using dierence in dierence
with conditioning variables.
REFERENCES
Acosta, P., Fajnzylber, P., & Lopez, J. H. (2007). The impact of
remittances on poverty and human capital: Evidence from Latin
American household surveys, Policy Research Working Paper No.
4247. Washington, DC: The World Bank.
Adams, R. H. (2006). International remittances and the household:
Analysis and review of global evidence. Journal of African Economies,
15, 396425.
Adepoju, A. (1983). Issues in the study of migration and urbanization in
Africa south of Sahara. In P. A. Morrison (Ed.). Population
movements: Their forms and functions in unbanization and development
(pp. 115149). Liege: Ordina IUSSP.
Akai, N., & Sakata, M. (2002). Fiscal decentralization contributes to
economic growth: evidence from state-level cross-section data for the
United States. Journal of Urban Economies, 52(1), 93108.
Angrist, J. D., & Pischke, J. S. (2008). Mostly harmless econometrics: An
empiricists companion. Princeton University Press.
Antman, F. (2011). The intergenerational eects of paternal migration on
schooling and work: What can we learn from childrens time
allocations?. Journal of Development Economics, 96(2), 200208.
Bardhan, P. (2002). Decentralization of governance and development.
Journal of Economic Perspective, 16(4), 185205.
Beauchemin, C., & Schoumaker, B. (2005). Migration to cities in Burkina
Faso: Does the level of development in sending areas matter?. World
Development, 33(7), 11291152.
Becker, C. M., Hamer, A. M., & Morrison, A. R. (1994). Beyond urban
bias in Africa: Urbanization in an era of structural adjustment.
Portsmouth London: Heine mann J. Currey.
Campbell, E. K. (1988). Rural urban migration and rural development
interrelations in West Africa: Some obstacles to goal achievement.
African Urban Quarterly, 3(12), 3449.
Chattopadhyay, R., & Duo, E. (2004). Women as policy makers:
Evidence from a randomized policy experiment in India. Econometrica,
72(5), 14091443.
Davoodi, H., & Zou, H. F. (1998). Fiscal decentralization and economic
growth: A cross-country study. Journal of Urban Economies, 43(2),
244257.
De Haas, H. (2010). Migration and development: A theoretical perspec-
tive. International Migration Review, 44(1), 227264.
Erb, M., & Sulistiyanto, P. (2009). Deepening democracy in Indonesia?:
Direct elections for local leaders (Pilkada). Jakarta: Institute of
Southeast Asian Studies.
Faguet, J. P. (2004). Does decentralization increase government respon-
siveness to local needs?: Evidence from Bolivia. Journal of Public
Economics, 88(3), 867893.
Fearnside, P. M. (1997). Transmigration in Indonesia: Lessons from its
environmental and social impacts. Environmental Management, 21(4),
553570.
Foster, A. D., & Rosenzweig, M. R. (2004). Democratization and the
distribution of local public goods in a poor rural economy. Economics
Department, Brown University.
Galasso, E., & Ravallion, M. (2005). Decentralized targeting of an
antipoverty program. Journal of Public Economics, 89(4), 705727.
Gertler, P. J., Martinez, S., Premand, P., Rawlings, L. B., & Vermeerch, C.
M. J. (2011). Impact evaluation in practice. Washington, DC: The
World Bank.
Gibson, J., Mckenzie, D., & Stillman, S. (2011). The impacts of
international migration on remaining household members: Omnibus
123
21. The results are consistent even when the census is restricted to only
those districts that are represented in IFLS.
22. The average migration rate among the households in treatment
districts at baseline is 0.155 (Table 2). A fall of 2.89 percentage points is
19.26% (2.89/0.155) of the baseline average migration rate.
23. Indonesia is divided into provinces, which constitute regencies and
cities. Although considered at the same level, regencies are larger in size
and cities have non-agriculture economic activities. Both regencies and
cities are further divided into districts.
results from a migration lottery program. The Review of Economics and
Statistics, 93(4), 12971318.
Hammond, G. W., & Tosun, M. S. (2011). The impact of local
decentralization on economic growth: Evidence from US counties.
Journal of Regional Science, 51(1), 4764.
Heckman, J. J., Lalonde, R. J., & Smith, J. A. (1999). The economics and
econometrics of active labor market programs. In D. Card, & O.
Ashenfelter (Eds.), Handbook of labor economics (pp. 18652097).
Amsterdam: North Holland.
Hildebrandt, N., & McKenzie, D. J. (2005). The eects of migration on
child health in Mexico. Economia, 6, 257289.
Hill, H., Resosudarmo, B., & Vidyattama, Y. (2008). Indonesias changing
economic geography. Bulletin of Indonesian Economic Studies, 44(3),
407435.
Hugo, G. (1980). Population movements in Indonesia during the colonial
period. Research School of Pacic Studies and Asian Studies,
Australian National University.
Hunther, J., & Shah, A. (1998). Applying a simple measure of good
governance to the debate on scal decentralization (Vol. 1894). World
Bank Publications.
Iimi, A. (2005). Decentralization and economic growth revisited: An
empirical note. Journal of Urban Economics, 57(3), 449461.
ILO (1998). Migration and population distribution in developing coun-
tries: Problems and policies. In United Nations (Ed.), Population
distribution and migration. Proceedings of the United Nations expert
group meeting on population distribution and migration, Santa Cruz,
Bolivia, January 1822, 1993. New York: United Nations.
Jones, G. W. (1979). Indonesia: The transmigration programme and
development planning. In R. J. Pryor (Ed.), Migration and development
in South-East Asia: A demographic perspective (pp. 212221). Kuala
Lumpur, Malaysia: Oxford University Press.
Khandker, S. R., Koolwal, G. B., & Samad, H. A. (2010). Handbook on
impact evaluation: Quantitative methods and practices. Washington,
DC: The World Bank.
Kis-Katos, K., & Sjahrir, B. S. (2014). Does local governments
responsiveness increase with decentralization and democratization?
Evidence from sub-national budget allocation in Indonesia. IZA
Discussion Paper No. 7884.
Lin, J. Y., & Liu, Z. (2000). Fiscal decentralization and economic growth
in China. Economic Development and Cultural Change, 49(1), 121.
Lipton, M. (1982). Rural development and the retention of the rural
population in the countryside of developing countries. Canadian
Journal of Development Studies, 3, 1137.
Lucas, R. (1997). Internal migration in developing countries. In M. R.
Rosenzweig, & O. Stark (Eds.), Handbook of population and family
economics (pp. 721798). Amsterdam: Elsevier Science Publishers.
Malley, M. (2003). New rules, old structures and the limits of democratic
decentralization. In E. Aspinall, & G. Fealy (Eds.), Local power and
politics in Indonesia: Decentralisation & democratisation (pp. 183218).
Jakarta: Institute of Southeast Asian Studies.
Martin, P., & Taylor, J. E. (1996). The anatomy of a migration hump. In
J. E. Taylor (Ed.), Development strategy, employment and migration:
Insights from models (pp. 4362). Paris: OECD.
Martinez-Vazquez, J., & McNab, R. M. (2003). Fiscal decentralization
and economic growth. World Development, 31(9), 15971616.
Meyer, B. D. (1995). Natural and quasi-experiments in economics. Journal
of Business and Economic Statistics, 13(2), 151161.
124 WORLD DEVELOPMENT

Mietzner, M. (2007). Party nancing in post-soeharto Indonesia: Between World Bank (2003). Decentralizing Indonesia: A regional public expendi-
state subsidies and political corruption. Contemporary Southeast Asia: ture overview report. Washington, DC: The World Bank.
A Journal of International and Strategic Aairs, 29(2), 238263. Xie, D., Zou, H. F., & Davoodi, H. (1999). Fiscal decentralization and
Munshi, K. (2003). Networks in the modern economy: Mexican migrants economic growth in the United States. Journal of Urban Economics, 45
in the US labor market. Quarterly Journal of Economics, 118, 549599. (2), 228239.
Oates, W. E. (1993). Fiscal decentralization and economic development. Yang, D. (2008). International migration, remittances and household
National Tax Journal, 46(2), 237243. investment: Evidence from Philippine migrants exchange rate shocks.
Palloni, A., Massey, D. S., & Ceballos, M. (2001). Social capital and The Economic Journal, 118(528), 591630.
international migration: A test using information on family networks. Zhang, T., & Zou, H. F. (1998). Fiscal decentralization, public spending
American Journal of Sociology, 106, 12621298. and economic growth in China. Journal of Public Economies, 67(2),
Pelzer, K. J. (1945). Pioneer settlement in the Asiatic tropics: Studies in land 221240.
utilisation and agricultural colonization in Southeastern Asia. New
Haven: Yale University Press.
Pratikno (2009). Political parties in Pilkada: Some problems for demo-
cratic consolidation. In M. Erb, & P. Sulistiyanto (Eds.), Deepening
democracy in Indonesia? Direct elections for local leaders (Pilkada)
(pp. 5373). Singapore, Jakarta: Institute for South East Asian Studies. APPENDIX
Rhoda, R. (1983). Rural development and urban migration. Can we keep
them down on the farm?. International Migration Review, 17(1), 3464.
Robinson, G., McNulty, J. E., & Krasno, J. S. (2009). Observing the
counterfactual? The search for political experiments in nature. Political
Analysis, 17(4), 341357.
Rosenbaum, P. R., & Rubin, D. B. (1985). Constructing a control group
using multivariate matched sampling methods that incorporate the Table A1. Placebo test for parallel trend assumption
propensity score. The American Statistician, 39, 3338.
Rubin, D. B. (2001). Using propensity scores to help design observational (1) (2) (3)
studies: Application to the tobacco litigation. Health Service & Entire Entire Entire
Outcomes Research Methodology, 2(34), 169188. sample sample sample
Santos, B. S. (1998). Participatory budgeting in Porto Alegre: Toward a
Time 0.0174*
0.017*
0.0172**
redistributive democracy. Politics and Society, 26(4), 461510.
Schiller, J. (2009). Electing district heads in Indonesia: Democratic (0.043) (0.055) (0.050)
deepening or elite entrenchment?. In M. Erb, & P. Sulistiyanto (Eds.), Treatment 0.006 0.0103 0.0015
Deepening democracy in Indonesia? Direct elections for local leaders (0.590) (0.415) (0.913)
(Pilkada) (pp. 147173). Singapore: Institute for South East Asian Time * Treatment 0.01061 0.0157 0.158
Studies. (0.258) (0.282) (0.278)
Sen, A. (1989). Development as capability expansion. Journal of Devel-
opment Planning, 19, 1458. Household heads characteristics NO YES YES
Skouas, E., Ambar, N., Dasgupta, B., & Kaiser, K. (2011). Electoral District characteristics & Regional NO NO YES
accountability, scal decentralization and service delivery in Indonesia, controls
Policy Research Working Paper No. WPS5614. Washington, DC: The Total Number of Observations 10,172 10,172 10,172
World Bank.
Skouas, E., Ambar, N., Dasgupta, B., & Kaiser, K. (2014). Electoral Note: Robust standard errors adjusted for clustering at the province level;
***
accountability, scal decentralization and service delivery in Indonesia, p < 0.001, **p < 0.05, *p < 0.1. p-Values are in parentheses.
Policy Research Working Paper No. WPS6782. Washington, DC: The
World Bank.
Soejoto, A., Subroto, W. T., & Suyanto, Y. (2015). Fiscal decentralization
policy in promoting Indonesia human development. International Table A2. Timing of direct elections of district heads during 200507
Journal of Economics and Financial Issues, 5(3), 763771.
All districts (304) Undivided districts (163)
Stansel, D. (2005). Local decentralization and local economic growth: A
cross-sectional examination of US metropolitan areas. Journal of 2005Q1 0 0
Urban Economics, 57(1), 5572. 2005Q2 155 67
Stark, O. (1991). The migration of labor. Cambridge, Oxford: B. Blackwell. 2005Q3 36 25
Tiebout, C. M. (1956). A pure theory of local expenditure. Journal of 2005Q4 12 9
Political Economy, 64(5), 416424.
2006Q1 16 13
Tirtosudarmo, R. (2009). Mobility and human development in Indonesia,
UNDP Human Development Research Paper No. 2009/19. New York: 2006Q2 22 20
United Nations. 2006Q3 9 5
Todaro, M. (1997). Urbanization, unemployment and migration in Africa: 2006Q4 27 7
Theory and policy, Policy Research Division Working Paper No. 104. 2007Q1 2 2
New York: Population Council. 2007Q2 4 4
Van Lottum, J., & Daan, M. (2012). The determinants of internal 2007Q3 7 4
migration in a developing country: Quantitative evidence for Indone- 2007Q4 14 7
sia, 19302000. Applied Economics, 44(34), 44854494.
Wooldridge, J. M. (2010). Econometric analysis of cross section and panel
data. MIT Press.

Available online at www.sciencedirect.com

ScienceDirect