Unraveling the Welfare Effects of Administrative and Political Decentralization in Indonesia

Unraveling the Welfare Effects of Administrative and Political Decentralization in Indonesia Deepak Singhania November 29, 2015 Abstract Decentralization of power has brought sweeping changes to governance systems of many developing countries over the past few decades. During 1997-2007 the World Bank has given $22 billion to its client countries to facilitate this process. However, the empirical evidence on the effects of decentralization on welfare outcomes has been both limited and ambiguous. Additionally, empirical comparison of different types of decentralization namely administrative, political and fiscal is missing. In this paper, I analyze and compare welfare effects under district-level administrative and political decentralization in Indonesia. My identification strategy exploits an interesting natural experiment where the two types of decentralization were temporally spread over two sub-periods between 2000 and 2008. A differences-in-differences estimation strategy with individual panel fixed effects implies that focusing only on political or administrative decentralization could be misleading and confounding. Specifically, there is a joint effect from the combination of political and administrative decentralization that has been neglected in the existing studies. Households/villages facing both types of decentralization display significantly stronger welfare effects than those facing only political or administrative or no decentralization at all. Keywords: Decentralization, Elections, Administrative, Political, Fiscal, Institutions JEL Codes: D72, D73, O43 I am grateful to my chair Steven Helfand for his continuous support and advice on this topic. This paper has improved greatly from timely guidance by my committee members Aman Ullah, Anil Deolalikar and Joseph Cummins. I have benefitted from discussions with participants at the 6th Annual Giannini Conference and at the Economics Brown Bag Seminar in University of California, Riverside. Contact details: dsing007@ucr.edu. Department of Economics, University of California, Riverside. Phone: +1 951-295-3128. 1

1 Introduction There are two opposing schools of thought on the consequences of decentralization 1. One is of the view that decentralization increases welfare by improving the flow of information between citizens and policymakers, as well as by increasing accountability on the part of policymakers. Bardhan (2002) notes a range of benefits of decentralization. This includes fragmentation of central authority to more diverse local governments, more responsive and efficient government, efficiency in local provision of public services with the help of modern technologies, and reduced social and political tensions. The other school of thought argues that decentralization can have negative effects on welfare by encouraging rent-seeking behavior and local government capture by elites. There is no consensus on which school of thought is correct. It is a matter of empirical enquiry and the effects of decentralization will depend on the inter-linkages between local and federal government, and on the institutions that define principal-agent relations (Mookherjee, 2015). While there can be different interpretations for the term decentralization, experts have largely identified three different types which apply to local government administrative, political and fiscal decentralization (Rondinelli, 1981). Administrative decentralization refers to the de-concentration of administrative responsibilities; fiscal decentralization involves delegation of public expenditure/revenue related responsibilities; and political decentralization relates to enhancing local accountability. These different types of decentralization differ in terms of power transferred to local governments; political decentralization transfers the most power, followed by fiscal and then administrative (Ribot, 2002). In this paper, I address two crucial gaps in the empirical literature on the decentralization of governance. The first gap relates to an empirical comparison of different types of decentralization. Studies that estimate the effects of different types of decentralization separately tend to overlook the omitted variable bias that could arise from their joint effects (for instance, see Skoufias et al. (2011); Martinez-Bravo (2014); Martinez-Bravo and Mukherjee (2015)). In my study I show that their joint effects can be significantly different from their individual effects for the case of district-level administrative and political decentralization in Indonesia. The second gap is that most empirical studies consider macroeconomic outcomes as measures of welfare which often provide an insufficient level of disaggregation for measuring changes in welfare at a microeconomic level. That is, within specific treatment groups, such a methodology is unable to discern the heterogeneous effects of the treatment under study. I address this limitation by considering outcomes 1 These two schools of thought emerge from the second generations of federalism which is different from the first generation (popularly known as fiscal federalism). The second generation assumes away from the assumption of perfect mobility of a voter in a developed country context and takes into account the ground realities in developing country context which involves issues like corruption and local elites. 2

for the units below the level of treatment, i.e. for households and villages, which will allow for analyzing heterogeneities in treatment. Moreover, with the help of these units of analysis, I will also be able to obtain standard errors that are robust to clustering on household or village groupings. The specific questions addressed in this study are: 1. Do different forms of decentralization affect household welfare outcomes, such as income and consumption, differently? 2. What are the mechanisms public resources or business environment through which decentralization affects household welfare outcomes? My identification strategy exploits an interesting natural experiment from Indonesia in the post-suharto regime that involved a series of decentralization reforms at various levels of government. I focus on administrative and political decentralization at district level. Administrative decentralization refers to the gradual splitting of districts between 2001 and 2009. Political decentralization relates to the first ever direct elections of district heads, which occurred as the terms of the existing, legislatively appointed, district heads ended. Owing to these developments, there were four different types of districts by 2007-08: those that split; those that had elections; those that split and had an election; and those that had none of the above. The results suggest two sets of findings. The first set answers a common question raised in the existing literature, that is, whether decentralization affects welfare. Significant effects for household level outcomes are not found when a split or an election serves as an individual treatment. At the village level, however, the effects of a split are generally stronger than that of an election. These findings contradict to the hypothesized effects of administrative vis-a-vis political decentralization in the theoretical literature. The second set of findings, which is a unique contribution of this paper, relates to the joint effects of the two types of decentralization. Specifically, all the welfare measures show significant improvement at the household and village level in districts that faced both a split and an election. This result has strong implications for the existing studies, which focus on either kind of decentralization separately and tend to ignore their joint effects. Using timing as a proxy for the intensity of treatment, I expect that the districts which experienced more years of decentralization should display larger welfare effects. However, I don t find the evidence of these larger welfare effects. In the near future, I plan to analyze some other district-level heterogeneity in treatment such as differences in districts population density after splitting. In addition, I plan to study 3

heterogeneity in treatment for villages within districts, for which, I will compare the change in distance from villages to district headquarters after splitting. This paper is organized as follows. The next section discusses the existing decentralization literature. In the third section I describe the background for my empirical study by discussing the treatment, the data, the empirical model and the parallel trends. I present results in the fourth section, and the fifth section concludes. 2 Understanding Decentralization: A Literature Review Two excellent studies summarizing much of the decentralization literature are provided by Bardhan (2002) and Mookherjee (2015). While emphasizing the importance of the issue of decentralization, both point out that empirically these issues are understudied in the developing country context. For understanding decentralization in developed countries, the theories relating to the first generation of fiscal federalism by Tiebout (1956) and Oates (1972) are well-suited. These theories focus on efficiencies and inefficiencies of decentralization with an assumption about voters ability to reveal their preferences for local public good by voting with their feet. But this and other such assumptions are not valid in developing countries. The second generation theories move away from the first generation by focusing on political economy and corruption issues. According to the World Bank decentralization is defined as the transfer of authority and responsibility for public functions from the central government to intermediate and local governments or quasiindependent government organizations and/or the private sector. A large literature identifies three kinds of decentralization administrative, political and fiscal (Ribot, 2002). Below, I discuss these three types and the related empirical research 2. Administrative decentralization Administrative decentralization refers to the transfer of administrative responsibilities to local authorities for implementing programs and policies designed at central level. It is considered to be of the weakest form of decentralization because the bodies at local government authorities are accountable to the central government only. Countries have found it naturally convenient to first adopt administrative decentralization before further delegation and devolution of centrally controlled roles. For instance, the Indonesian government had identified districts as their administrative regions in 1950s, but these districts were not given fiscal powers until 2001. 2 Since different types of decentralization are not mutually exclusive, segmentation of literature into these types is not easy. So, I divide it in the way researchers have described the prevalence of one over the other. 4

Empirical studies that focus primarily on the provision of public services like health, sanitation and education have shown positive effects due to efficiency of service delivery (Alderman (1998); Coady (2001)), better informed local officials (Carneiro et al. (2015); Azfar et al. (2001)) and community participation (Galasso et al. (2001); Wade (1997)). In Indonesian case, Burgess et al. (2011) have shown that splitting of Indonesian districts led to increased deforestation and lower timber prices. Political decentralization Political decentralization is the strongest form of decentralization because the agents local authorities are directly accountable to their principals local voters. An interesting case is that of Indian villages, where during 1990s elections were constitutionally formalized and made compulsory for village level governments, known as gram panchayat (Foster and Rosenzweig (2001); Anderson et al. (2012); Chattopadhyay and Duflo (2004)). All these studies have shown significant effects of decentralization. In Indonesia two kinds of studies exist. Those that focus on immediate governance outcomes like public expenditure, and those that focus on final public resource outcomes like availability of teachers and doctors. Martinez-Bravo (2014) has shown that inherited appointed officials lead to greater electoral fraud and clientelistic spending. Sjahrir et al. (2014), while analyzing the determinants of excessive administrative spending, show that elections or the proliferation of districts don t lead to increased administrative spending. Skoufias et al. (2011) and Mukherjee (2014) use district level data to study the effects of only election on various outcomes which include education and health related indicators. The former finds no effect of election on health and educational outcomes while the latter finds an increase in the number of public doctors and health workers and number of public teachers. Fiscal decentralization Fiscal decentralization is a mix of the other two, whereby public revenue/expenditure related powers are given to local authorities. Most of the decentralization literature has focused on estimating the impact of fiscal decentralization at macro level. The findings on the impact of fiscal decentralization on economic growth are quite mixed. In case of Indonesia, Pal and Roy (2015) conduct a before and after comparison to show the effects of universal district level fiscal decentralization on grassroots politics and local development. They show that communities with homogenous socio-cultural characteristics experience change in leader selection methods and a positive impact of fiscal decentralization on local development. 5

3 The Decentralization Experiment: Background, Data, Empirical Strategy and Parallel Trends 3.1 Background: Decentralization in Indonesia The legacy of the Indonesian government system, which involves formation of districts (kabupatens) and municipalities (kotas), dates back to the Dutch colonial rule during the early 20th century. Post colonial rule during 1950s, the country alternated between a centralized and a decentralized governance system, until finally settling down with the guided democracy of the then President Sukarno. The topdown authoritarian rule continued under the new order regime of President Suharto in 1967. After the fall of Suharto s authoritarian regime in May 1998 in the aftermath of the Asian Financial Crisis of 1997, Indonesia began a process of decentralization under the leadership of its successive presidents (Figure 1). The transfer of power to local governments involved political, fiscal and administrative decentralization which I discuss below. Political decentralization (Direct election of district heads) The first democratic parliamentary elections 3 were held in 1999 out of which two prominent levels of governments the central government and the districts/municipalities 4 emerged to carry out most of the governance activities (Figure 2) 5. Multiple parties were allowed to compete in democratic elections for district legislatures. These elected district legislatures in turn appointed district heads when the terms of the existing, centrally appointed, district heads ended 6. Since these district heads were appointed by local legislatures, they didn t generally have enough power to make independent decisions. Furthermore, there were possibilities of collusion between the two bodies for advancing their self-interest. To address these issues, in 2004 a new law was introduced. As per this law, beginning in the year 2005, district heads would also be chosen through public elections once the term of the existing, legislatively appointed, district heads ended. This had two implications for governance at district level. One is that the district heads became accountable to their citizens, and their term no more depended on legislatures. The second is that they were vested with substantial fiscal powers through fiscal decentralization. 3 These legislative elections happen every five years. 4 Henceforth I refer this as districts. 5 Although provinces (equivalent to a state in many countries) are another hierarchical level between central government and districts/municipalities, they are mainly responsible for coordinative role between districts/municipalities and central government. Real autonomy to govern, as in decision making powers in various public policy issues, rests with districts/municipalities. Please refer Figure 2. 6 Usually, a district head s term is of five years. Before 1999 elections, district heads were appointed by the Ministry of Home Affairs. 6

Fiscal decentralization (Autonomous expenditure making power to districts) With the benchmark laws of 1999 and 2004 (Law 22/1999 and 32/2004) the central government became responsible for five functions foreign relations, defense, legislations, macroeconomic policies and religious affairs while districts were given full autonomy to govern and administer the interest of the local people. District responsibilities included health, education, local infrastructure, public order and peace, etc 7. Effective from 2001, these fiscal powers, mostly expenditure related, were given to all the districts simultaneously with some modifications from 2005 onwards. Administrative decentralization (splitting of districts) During 2001-2004 and 2007-2009 some of the districts were split up. There was a moratorium on splitting during 2005-06 which coincided with the beginning of direct elections for district heads. The number of districts increased from 292 to 434 during 2001-2007 (Figure 3) 8. There was no specific rule for splitting. Some of the reasons for splitting as argued by Fitrani et al. (2005) included geographic dispersion, political and ethnic diversity, natural resources and scope for bureaucratic rent seeking. However, Burgess et al. (2011) have shown that the timing of splitting was random. In conclusion, between 2000 and 2007/2008 there were four sets of districts based on governance structure: those that split; those that had elections; those that split and had an election; and those that had none of the above 9. In addition, both split and elections occurred over time. 3.2 Data Data on elections and splitting Data on the splitting of districts come from Central Bureau of Statistics (BPS) in Indonesia. I obtained the data for election dates from Burgess et al. (2011) 10. This data contains information on the timing of end of legislatively appointed district heads term, as well as on the timing of first direct elections. The end of existing district heads terms don t coincide perfectly with the timing of elections, hence, I proxy for regime change with the former to ensure that the treatment of political decentralization is fairly random. Also, intuitively one would expect to see changes once the term of the existing district head ends because the opportunities for rent seeking behavior would cease. 7 Please note that districts largely had expenditure making powers. Revenue generation was still a primary function of the central government. 8 Some of the districts were split during 1999-2000 as an experiment, however, they were not given autonomous power until 2001 and hence I consider them splitting in 2001. Also, it wouldnt be until a period of one year that the newly created districts could operate by themselves; hence, I dont consider the districts that split in 2007. 9 Note that fiscal decentralization was a universal event and so it is not considered as one of the treatments. 10 I acknowledge the help of Benjamin Olken at MIT for generously sharing this data. 7

Household level data For household level outcomes I have used the Indonesian Family Life Survey (IFLS). It is a panel survey that spans over 14 years, from 1993 to 2007. The panel is available across four rounds (1993, 1997, 2000 and 2007) and about 90% of the households surveyed in 1993 could be tracked up to 2007. IFLS is representative of about 83% of the Indonesian population living in 13 of the 26 provinces. This dataset contains information at individual, household and community levels. For the main results I use the 2000 and 2007 waves of IFLS, and for showing parallel trends I compare 1993 and 2000 rounds. The outcome variables at the household level are log of real per capita income and consumption. Village level data I obtain village level outcomes from the 2000 and 2008 waves of the Indonesian Village Potential Statistics (PODES) 11. It is a village census survey of socio-economic characteristics with information from more than 60,000 villages on schools, health centers, electricity, roads, markets, industries, etc. This survey is conducted every three years. A limitation of this survey is that it doesn t have village identifiers to make a panel of villages. I use names of villages, sub-districts, districts and codes of districts and provinces to create a panel of villages. I successfully create a panel of about 94% of villages in the baseline 12. The outcomes that I consider at the village level fall broadly under two categories. The first category represents public resources for which district governments are directly responsible such as education, health, infrastructure and public order and peace. For educational resources I use the number of government owned junior high schools and distance to private or government owned junior high school 13. Indicators for health related resources include the availability of and distance to community health centers (puskesmas) which is exclusively financed by district governments, as well as the availability of doctors 14. Infrastructure is measured by the availability of paved and wide roads, the availability of street lights and the share of households with electricity. For public order and peace, I use the availability of and distance to police stations, as well as a dummy for the occurrence of crimes such as murder or robbery. 11 For parallel trends analysis at the village level, due to inaccessibility to PODES data from 1993 or 1996 round I rely on INDO-DAPOER data from the World Bank. This is a district level data with information only on six relevant characteristics for 1996 and 2000 in the pre-treatment period. Moreover, these characteristics are available for aggregated districts that were split in the post period. Since the treatment was at district level, the parallel trends analysis at district level should be sufficient for evidence of districts following similar trend in the pre-treatment period with respect to the outcomes. 12 I don t include Jakarta in my analysis because districts in Jakarta are not autonomous. Also, due to high frequency conflicts and unstable government regime in Papua and Papua Barat provinces I don t include them in the analysis. 13 I focus only on junior high school because primary education was almost universal in Indonesia before decentralization, and senior secondary education was usually funded by provinces. 14 There is no information if doctors are public or private, but a major portion of local government expenditure was directed towards hiring personnel which included teachers and doctors (Mukherjee (2014)). 8

The second category of outcomes is representative of economic activities and business environment in a village. This includes distance to a shopping complex and marketplace, number of food and leather industries, the availability of different types of cooperatives, the proportion of agricultural land that is technically irrigated and the education level of the head of a village. 3.3 Empirical Strategy Binary Treatment In order to estimate the effect of decentralization when the treatment is binary I use the following specification, which is differences-in-differences estimation strategy with individual fixed effects. Y idt = α i + β 1 P ost t + β 2 (Election P ost) dt + β 3 (Split P ost) dt + β 4 (Split Election P ost) dt + β 5 X idt + ɛ idt (1) In this specification, treated group effects are absorbed by individual fixed effects making this specification stronger than the usual differences-in-differences. For example, the time invariant unobservables (or observables) for individual units i, that are not necessarily common across all individual units within a treated group could include characteristics related to geographical or political boundaries. In equation (1), Y idt represents outcomes for households (or for villages) in district d at time t. The variables Post, Election and Split are dummies for year, election and split respectively. β 1 captures time trend and β 2, β 3 and β 4 capture the effect of having only an election, only a split and both respectively. Time-varying controls are captured through β 15 5. I assume that the error terms are uncorrelated with the treatment, and I provide evidence for this by estimating specification (1) assuming that the pre-treatment and post-treatment periods are 1993 and 2000 instead of 2000 and 2007 (see section 3.4). Heterogeneity in intensity of treatment (timing of treatment) Hypothesizing that there will be heterogeneity in treatment due to different years of decentralization faced by districts, I estimate the following specification. Y idt = α i + β 1 P ost t + Σ p β p+1 (Election p P ost) dt + Σ q β q+1 (Split q P ost) dt + Σ p β p+7 (Split Election p P ost) dt + β 10 X idt + ɛ idt (2) 15 Because of the nature of the topic under study, there are substantive restrictions in the number of controls that I can use. For one obvious reason, all possible time invariant controls are captured by individual fixed effects. In case of time varying controls it is hard to think of household or village level characteristics that would not be endogenous to decentralization because decentralization can induce large structural changes across various spheres. For instance, education of household members would itself be a function of the treatment, or rural/urban status of a village would also depend on the newly appointed district government. 9

The only difference between specifications (1) and (2) is that the former had dummies for treatment, whereas the latter includes dummies for different years of treatment. Here the superscripts p and q represent a bin for number of years since the election and split respectively 16. For household level regressions p takes values 1 and 2 (because elections happened in 2005 and 2006, and we observe households in 2007) while j takes values 3 through 6 (because splitting happened during 2001-2004). For the village sample these values would correspondingly add up by one, because village outcomes are observed in 2008 17. 3.4 Parallel Trends In order to show that the treatment and the control groups were following similar trends in the pretreatment period I estimate equation (1) for households and for villages in this pre-treatment period. Table 1 shows regression results for household level outcomes, income and consumption, assuming that the treatment happened between 1993 and 2000. The coefficients of interest are on the three types of treatments; that is, whether a households district faced a split or an election or faced both a split and an election between 1993 and 2000. All of the coefficients are insignificant except for income of the households who belonged to districts that had faced only an election. Even for these households the significance is below the commonly acceptable level. Moreover, consumption and income were falling before the split or elections and income was falling in the case of both, implying that any positive impacts found for these cases would be underestimated. Table 2 shows parallel trends in village level outcomes for six variables. All of the coefficients are insignificant except for the presence of a community health center in districts that faced a split, for which, it is at 10% level of significance. We can thus establish from these results that the treatment groups were following similar trends for various outcomes before they got treated. Hence, any effect discussed in the next section is likely to be due to the treatment under study. 4 Results Binary Treatment The main results from equation (1) are presented in Table 3 for household outcomes, and Tables 4a, 4b and 4c for village level outcomes. There are three types of regression specifications for every outcome one with split as the only treatment, one with election as the only treatment and one with both the treatments and their interaction. 16 Subscripts on coefficients, such as (p+1), represent different coefficients. 17 It is important to note here that there is no superscript on Split in the interaction variable Split*Election p *Post because timing of both type of decentralization is conditional on timing of elections which happened in the second sub-period, and for the same reason it has the same subscript and superscript as Election p *Post. 10

According to Table 3, columns 1-2 and 4-5 imply that neither a split nor an election had any significant effect on household income or consumption, except that splitting resulted in a significant 12% increase in household consumption at about 5% significance level. When we include the treatments and their interaction (columns 3 and 6), we see a positive and significant effect of about 23-24% on both income and consumption, due to the joint treatment. Interestingly, the sign of the coefficient of split flips between columns 1 and 3 and both the significance level and magnitude drops between columns 4 and 6. This implies that focusing on either kind of decentralization separately (as is done in the existing studies) might lead to a misspecified model. This might confound the differential effect of an election conditional on splitting and vice-versa. The village level regressions further confirm these findings. Overall, there is a significantly positive change for most indicators of public resources and business environment in the specifications with splitting as the only treatment, as well as in the ones with election as the only treatment (see Tables 4a, 4b and 4c). However, most of these effects are significant for the former than the latter. There is a significant positive effect of splitting on various measures of public resources. These measures are as follows: health resources, where the availability of community health centers increased, the distance to such centers was reduced and the availability of doctors increased; education, where the numbers of schools per 1000 people increased and distance to the school was reduced; infrastructure, where there were wider roads and the share of households with electricity increased; and public order and peace, where distances to police stations were reduced as was the proportion of villages reporting crimes like robbery and murder. Some similar effects are found in the specification where an election is the only treatment (see Table 4b). There are indications of a significant improvement in the business environment, although more for districts that split. Distance to shopping complexes and the marketplaces were reduced; the availability of types of cooperatives increased and more village heads were educated at least up to junior high school. One exception is in the number of food and leather industries which were reduced somewhat. When we include both the treatments and their interaction in a single specification these results change substantially. The effects of only a split for most of the outcomes, and the effects of only an election for all the outcomes become insignificant or the magnitude drops as soon as the joint effects of a split and an election are taken into account (see Table 4c). These joint effects are significantly positive, and even the magnitudes are larger compared to specifications with only split or only elections. These findings have two implications. One is that higher decentralization has stronger welfare effects. Possibly, the forces at play with each kind of decentralization include efficiency in the case of administrative decentralization and accountability in the case of political decentralization. Second implication is 11

that the studies that focus on any one type of decentralization tend to neglect their joint effects 18. Heterogeneity in intensity of treatment (timing of treatment) It would be reasonable to hypothesize that given the positive effects of decentralization there will be higher welfare effects in districts with more years of decentralization. Tables 5 and 6 present regression results using equation (2) for individual and village level outcomes. 19. For household level regressions, we don t see any trend in the number of years of split or number of years of election (Table 5). Same is true for village level outcomes (Table 6). However, considering the joint effect of election and split there are higher effects on income and consumption (for consumption the significance level at 5.3%) if the treatment was there for two years compared to one year. But for village level outcomes only in some cases the districts where both treatments were for highest period of three years the effects were highest. So, there is no strong support for the hypotheses that longer duration of treatment would mean higher effects. One reason could be that the post-treatment period is too short for the heterogeneous effects of timing to start reflecting in the data. 5 Conclusion This paper contributes to the limited pool of empirical literature on decentralization, particularly in a developing country context. The process of decentralization involves various stages that can be segregated into administrative, fiscal and political types of decentralization. Each type, having its own merits and drawbacks, has different implications for development, as is true for their combinations. Decentralization in Indonesia provides an interesting setting for comparing the individual and joint effects of these different types of decentralization. My findings both empirical and methodological add uniquely to our understanding of decentralization of governance. As an empirical inquiry, by using both household level and village level data, I have shown that there is an additional effect due to increased decentralization. The effects of administrative decentralization combined with political decentralization lead to significantly positive effects on various welfare outcomes. Methodologically, I have argued that studies which neglect this joint effect might be affected by omitted variable bias. Considering only one kind of decentralization can give biased estimates that would confound the effects of the other kind of decentralization. 18 It would have been interesting to compare the findings in this paper with those in the existing literature. However, in most of the existing studies the outcomes and data sources are different. The only comparable studies at hand are by Skoufias et al. (2011) and Mukherjee (2014). Both use district level data and study the effects of only election on various outcomes which include education and health related indicators. The former finds no effect of election on health and educational outcomes while the latter finds an increase in the number of public doctors and health workers (at 1% significance) and number of public teachers (at 10% significance). 19 Henceforth, all the results rely on the specification that includes all the possible treatments. 12

Next steps for this research project will involve testing for heterogeneities in treatment. One source of heterogeneity is at the level of districts; specifically, differences in districts population density after splitting. The other type of heterogeneity is at the level of villages. Here I will compare the changes in the distance from villages to district headquarters after splitting. 13

References Alderman, H. (1998). Social assistance in Albania : decentralization and targeted transfers. Technical Report LSM134, The World Bank. Anderson, S., Francois, P., and Kotwal, A. (2012). One Kind of Democracy. The Centre for Market and Public Organisation 12/292, Department of Economics, University of Bristol, UK. Azfar, O., Satu, K., and Patrick, M. (2001). Conditions for Effective Decentralized Governance: A Synthesis of Research Findings. University of Maryland, Center for Institutional Reform and the Informal Sector, College Park. Bardhan, P. (2002). Decentralization of Governance and Development. Journal of Economic Perspectives, 16(4):185 205. Burgess, R., Hansen, M., Olken, B. A., Potapov, P., and Sieber, S. (2011). The Political Economy of Deforestation in the Tropics. Working Paper 17417, National Bureau of Economic Research. Carneiro, P., Koussihoud, O., Lahire, N., Meghir, C., and Mommaerts, C. (2015). Decentralizing Education Resources: School Grants in Senegal. Working Paper 21063, National Bureau of Economic Research. Chattopadhyay, R. and Duflo, E. (2004). Women as Policy Makers: Evidence from a Randomized Policy Experiment in India. Econometrica, 72(5):1409 1443. Coady, D. (2001). An Evaluation of the Distributional Power of Progressa s Cash Transfers in Mexico. IFPRI Working Paper. Fitrani, F., Hofman, B., and Kaiser, K. (2005). Unity in diversity? The creation of new local governments in a decentralising Indonesia. Bulletin of Indonesian Economic Studies, 41(1):57 79. Foster, A. D. and Rosenzweig, M. R. (2001). Democratization, Decentralization and the Distribution of Local Public Goods in a Poor Rural Economy. SSRN Scholarly Paper ID 300421, Social Science Research Network, Rochester, NY. Galasso, E., Ravallion, M., Nizamuddin, S., Sen, B., Tirole, J., Walle, D. V. D., and Walton, M. (2001). Decentralized Targeting of an Anti-Poverty Program Emanuela Galasso and Martin Ravallion. Journal of Public Economics. Martinez-Bravo, M. (2014). The Role of Local Officials in New Democracies: Evidence from Indonesia. American Economic Review, 104(4):1244 87. 14

Martinez-Bravo, M. and Mukherjee, P. (2015). An Empirical Investigation of the Legacies of Non- Democratic Regimes: The Case of Soehartos Mayors in Indonesia. Unpublished. Mookherjee, D. (2015). Political Decentralization. SSRN Scholarly Paper ID 2640076, Social Science Research Network, Rochester, NY. Mukherjee, P. (2014). Accountability of Public Officials under Appointments and Elections - Evidence from Indonesia. Unpublished. Oates, W. E. (1972). Fiscal Federalism. Harcourt Brace Jovanovich. Pal, S. and Roy, J. (2015). When Political Change Signals Community Resolve? Fiscal Decentralization, Grassroots Politics and Local Development. SSRN Scholarly Paper ID 2433972, Social Science Research Network, Rochester, NY. Ribot, J. (2002). African Decentralization: Local Actors, Powers and Accountability. Technical Report 8, UNRISD Programme on Democracy, Governance and Human Rights. Rondinelli, D. A. (1981). Government Decentralization in Comparative Perspective Theory and Practice in Developing Countries. International Review of Administrative Sciences, 47(2):133 145. Sjahrir, B. S., Kis-Katos, K., and Schulze, G. (2014). Administrative Overspending in Indonesian Districts: The Role of Local Politics. World Development, 59(C):166 183. Skoufias, E., Narayan, A., Dasgupta, B., and Kaiser, K. (2011). Electoral accountability, fiscal decentralization and service delivery in Indonesia. Policy Research Working Papers. The World Bank. Tiebout, C. M. (1956). A Pure Theory of Local Expenditures. Journal of Political Economy, 64. Wade, R. (1997). How Infrastructure Agencies Motivate Staff: Canal Irrigation in India and the Republic of Korea. In Infrastructure Strategies in East Asia. EDI, World Bank, a. mody, ed. edition. 15

APPENDIX Background Figure 1: Timeline of decentralization process in Indonesia Figure 2: The administrative set-up of Indonesian government in 2007 16

Figure 3: Evolution of number of districts/municipalities during 2001-07 Parallel Trends Table 1: Regression for household outcomes assuming that the treatment happened in 2000 and the baseline is 1993 (1) (2) Income Consumption Split*Post -0.22-0.07 (0.12) (0.08) Election*Post -0.06-0.04 (0.11) (0.05) Election*Split*Post -0.16 0.19 (0.15) (0.12) Controls Yes Yes N 12130 12130 r2 0.08 0.04 *0.10 **0.05 ***0.01 Clustered standard errors at district level are in parentheses Controls include household size and household head age 17

Table 2: Regression for village outcomes assuming that the treatment happened in 2000 and the baseline is 1993 Community Jr. Sec. Electricity Water Sanitation health School Roads center (nos.) Split*Post -0.14 0.61 0.01 0.04-0.00-1.42 (3.10) (2.35) (2.95) (0.02) (0.01) (1.88) Election*Post -0.96 0.50-0.61-0.00-0.01-0.07 (1.46) (2.88) (2.33) (0.01) (0.01) (1.55) Election*Split*Post 5.56 5.38-2.08 0.04 0.00 2.49 (5.94) (4.45) (2.90) (0.03) (0.01) (2.94) Controls N 526 526 526 526 526 526 r2 0.57 0.07 0.26 0.23 0.10 0.03 *0.10 **0.05 ***0.01 Clustered standard errors at province level are in parentheses Healthcenters and schools are in number per 000 population Electricity, Water and Sanitation are as a proportion of total households Road is in terms of proportion of villages with asphalt roads All regressions are weighted by population share in the province 18

Main Results Table 3: Decentralization and household outcomes Income Consumption 1 2 3 4 5 6 Split*Post 0.02-0.11 0.12 0.08 (0.08) (0.11) (0.06) (0.07) Election*Post 0.10 0.04 0.07 0.05 (0.06) (0.07) (0.06) (0.06) Election*Split*Post 0.24 0.23 (0.09) (0.11) Controls Yes Yes Yes Yes Yes Yes N 13316 13316 13316 13316 13316 13316 r2 0.03 0.03 0.03 0.13 0.13 0.13 *0.10 **0.05 ***0.01 Clustered standard errors at district level are in parentheses Controls include household size and household head age 19

Table 4a: Decentralization and village outcomes (Specification- Only Split) Shopping Industries Financial Technical Village Head Marketplace Complex (Food/Leather) Cooperatives Irrigation Education Distance Nos. Dummy Proportion of ag. land Dummy Split*Post -6.01-1.15-0.57 0.03 0.03 0.07 (2.01) (0.49) (0.31) (0.02) (0.01) (0.01) N 110533 107770 118917 118918 118918 117716 r2 0.01 0.00 0.02 0.00 0.07 0.10 Health Education Infrastructure Public order/peace Community Jr. High Paved Wide Electricity Street Police Doctors Center School Roads Roads Households Lights Station Dummy Distance Dummy Nos. Distance Dummy Proportion Dummy Dummy Distance Dummy Split*Post 0.02-2.34 0.02 0.04-1.83 0.00 0.05 0.05 0.02 0.00-2.03-0.07 (0.01) (0.37) (0.01) (0.01) (0.25) (0.01) (0.01) (0.01) (0.02) (0.01) (0.82) (0.02) N 118918 118918 118918 118917 118918 118029 118029 118917 118918 118918 111586 118918 r2 0.01 0.03 0.05 0.02 0.05 0.03 0.03 0.31 0.16 0.03 0.06 0.40 Crime *0.10 **0.05 ***0.01. Clustered standard errors at district level are in parentheses. 20

Table 4b: Decentralization and village outcomes (Specification- Only Election) Shopping Industries Financial Technical Village Head Marketplace Complex (Food/Leather) Cooperatives Irrigation Education Distance Nos. Dummy Proportion of ag. land Dummy Election*Post -3.51-0.12 0.22 0.05 0.01 0.01 (1.90) (0.48) (0.32) (0.02) (0.01) (0.01) N 110533 107770 118917 118918 118918 117716 r2 0.01 0.00 0.02 0.00 0.06 0.09 Health Education Infrastructure Public order/peace Community Jr. High Paved Wide Electricity Street Police Doctors Center School Roads Roads Households Lights Station Dummy Distance Dummy Nos. Distance Dummy Proportion Dummy Dummy Distance Dummy Election*Post 0.02-0.50 0.03 0.01-0.31 0.01 0.00 0.04 0.01 0.02-0.65-0.04 (0.01) (0.35) (0.01) (0.01) (0.25) (0.01) (0.01) (0.01) (0.02) (0.01) (0.77) (0.02) N 118918 118918 118918 118917 118918 118029 118029 118917 118918 118918 111586 118918 r2 0.01 0.02 0.05 0.02 0.04 0.03 0.02 0.30 0.16 0.03 0.05 0.40 Crime *0.10 **0.05 ***0.01. Clustered standard errors at district level are in parentheses. 21

Table 4c: Decentralization and village outcomes (Specification- Both) Shopping Industries Financial Technical Village Head Marketplace Complex (Food/Leather) Cooperatives Irrigation Education Distance Nos. Dummy Proportion of ag. land Dummy Split*Post -4.62-1.08-0.25 0.03 0.01 0.06 (2.83) (0.76) (0.40) (0.02) (0.01) (0.02) Election*Post -1.69 0.18 0.63 0.05-0.01-0.02 (1.35) (0.49) (0.50) (0.02) (0.01) (0.01) Election*Split*Post -7.94-1.07-0.29 0.07 0.03 0.07 (2.80) (0.69) (0.41) (0.02) (0.01) (0.01) N 110533 107770 118917 118918 118918 117716 r2 0.01 0.00 0.02 0.00 0.07 0.10 Health Education Infrastructure Public order/peace Community Jr. High Paved Wide Electricity Street Police Doctors Center School Roads Roads Households Lights Station Dummy Distance Dummy Nos. Distance Dummy Proportion Dummy Dummy Distance Dummy Split*Post 0.01-1.72 0.01 0.02-1.35 0.01 0.05 0.02 0.01-0.02-0.71-0.07 (0.01) (0.63) (0.01) (0.01) (0.45) (0.02) (0.02) (0.02) (0.03) (0.02) (1.40) (0.03) Election*Post 0.01 0.38 0.02-0.01 0.39 0.01-0.01 0.01-0.00 0.01 0.64-0.04 (0.01) (0.22) (0.01) (0.01) (0.21) (0.02) (0.01) (0.01) (0.03) (0.01) (0.71) (0.02) Election*Split*Post 0.04-2.39 0.04 0.04-1.80 0.01 0.04 0.07 0.02 0.02-2.29-0.09 (0.01) (0.46) (0.02) (0.01) (0.31) (0.02) (0.01) (0.01) (0.02) (0.02) (1.08) (0.02) N 118918 118918 118918 118917 118918 118029 118029 118917 118918 118918 111586 118918 r2 0.01 0.03 0.05 0.02 0.05 0.03 0.03 0.31 0.16 0.03 0.06 0.40 Crime *0.10 **0.05 ***0.01. Clustered standard errors at district level are in parentheses. 22

Table 5: Decentralization and household outcomes (Specification- Only Split) Income Consumption 1 2 4 yrs Split*Post -0.05 0.01 (0.24) (0.11) 5 yrs Split*Post 0.30 0.20 (0.37) (0.23) 6 yrs Split*Post -0.21 0.08 (0.11) (0.09) 1 yrs Election*Post 0.06 0.04 (0.08) (0.09) 2 yrs Election*Post 0.01 0.06 (0.08) (0.07) 1 yrs 0.07 0.10 Election*Split*Post (0.08) (0.07) 2 yrs 0.39 0.36 Election*Split*Post (0.11) (0.19) Controls Yes Yes N 13316 13316 r2 0.03 0.13 *0.10 **0.05 ***0.01 Clustered standard errors at district level are in parentheses Controls include household size and household head age 23

Table 6: Decentralization and village outcomes (Intensity of treatment: Year) Shopping Industries Financial Technical Village Head Marketplace Complex (Food/Leather) Cooperatives Irrigation Education Distance Nos. Dummy Proportion of ag. land Dummy 4 yrs Split*Post 0.67 0.46-1.84-0.03 0.01 0.06 (2.02) (1.32) (0.49) (0.07) (0.02) (0.05) 5 yrs Split*Post -3.37 0.35 1.08-0.03 0.01 0.03 (2.24) (0.75) (0.73) (0.05) (0.02) (0.03) 6 yrs Split*Post -5.14-1.72 0.05 0.02 0.03 0.08 (6.03) (1.67) (0.43) (0.03) (0.02) (0.03) 7 yrs Split*Post -5.88-1.50-0.80 0.07-0.01 0.05 (3.15) (0.72) (0.52) (0.03) (0.03) (0.03) 1 yrs Election*Post -0.51-0.15 3.48 0.08-0.03 0.00 (1.37) (0.52) (1.69) (0.04) (0.03) (0.03) 2 yrs Election*Post -0.62 0.22 0.04 0.04-0.01-0.01 (1.44) (0.48) (0.63) (0.03) (0.01) (0.02) 3 yrs Election*Post -2.75 0.20 0.67 0.06-0.00-0.03 (1.52) (0.65) (0.62) (0.03) (0.01) (0.01) 1 yr Elec*Split*Post -7.26-2.94-1.00 0.06 0.02 0.05 (5.24) (1.78) (0.89) (0.05) (0.03) (0.03) 2 yr Elec*Split*Post -8.60-2.22-0.67 0.09 0.01 0.08 (6.33) (1.28) (0.38) (0.02) (0.02) (0.02) 3 yr Elec*Split*Post -7.76-0.14 0.04 0.06 0.04 0.06 (3.09) (0.72) (0.49) (0.03) (0.01) (0.02) N 110533 107770 118917 118918 118918 117716 r2 0.01 0.00 0.02 0.00 0.07 0.10 24

Table 6 (contd.) Health Education Infrastructure Public order/peace Community Jr. High Paved Wide Electricity Street Police Doctors Center School Roads Roads Households Lights Station Dummy Distance Dummy Nos. Distance Dummy Proportion Dummy Dummy Distance Dummy 4 yrs Split*Post -0.01 1.12-0.00 0.02-0.99-0.04 0.03-0.01 0.03-0.01 1.83-0.07 (0.02) (0.65) (0.02) (0.01) (0.78) (0.04) (0.03) (0.02) (0.10) (0.03) (1.40) (0.06) 5 yrs Split*Post -0.00-0.80 0.00 0.05-0.55-0.01 0.01-0.00 0.06-0.02 1.50-0.13 (0.01) (0.90) (0.03) (0.02) (0.61) (0.04) (0.01) (0.02) (0.05) (0.02) (1.01) (0.07) 6 yrs Split*Post 0.02-3.62 0.02 0.05-2.87-0.00 0.12 0.01-0.04-0.01-4.39-0.10 (0.01) (1.40) (0.02) (0.01) (0.96) (0.04) (0.03) (0.02) (0.04) (0.02) (3.38) (0.04) 7 yrs Split*Post -0.00-0.95 0.01-0.01-0.38 0.04 0.01 0.03 0.04-0.02 1.12-0.02 (0.02) (0.59) (0.02) (0.01) (0.42) (0.03) (0.01) (0.02) (0.04) (0.02) (0.77) (0.04) 1 yrs Election*Post -0.00 0.68 0.02-0.02 0.32 0.07-0.01 0.12 0.09-0.03 1.75-0.03 (0.01) (0.19) (0.02) (0.01) (0.42) (0.04) (0.01) (0.05) (0.05) (0.02) (0.76) (0.06) 2 yrs Election*Post 0.02 0.25 0.02-0.01 0.46 0.01-0.01 0.02 0.02 0.02 0.34-0.03 (0.01) (0.29) (0.02) (0.01) (0.25) (0.02) (0.01) (0.02) (0.03) (0.02) (0.83) (0.03) 3 yrs Election*Post -0.00 0.45 0.01-0.01 0.34 0.00-0.00-0.01-0.03 0.00 0.70-0.05 (0.01) (0.27) (0.01) (0.01) (0.24) (0.02) (0.01) (0.02) (0.03) (0.02) (0.81) (0.03) 1 yr Elec*Split*Post 0.02-2.10 0.01 0.05-1.27 0.06 0.09 0.10-0.02 0.02-2.70-0.10 (0.02) (0.67) (0.03) (0.02) (0.42) (0.04) (0.03) (0.03) (0.06) (0.03) (1.75) (0.05) 2 yr Elec*Split*Post 0.01-2.08 0.05 0.05-1.80-0.04 0.05 0.09-0.02 0.03-1.68-0.11 (0.02) (0.73) (0.02) (0.01) (0.59) (0.03) (0.02) (0.02) (0.03) (0.02) (2.20) (0.04) 3 yr Elec*Split*Post 0.05-2.60 0.04 0.04-1.91 0.03 0.03 0.06 0.05 0.02-2.50-0.08 (0.02) (0.65) (0.02) (0.01) (0.36) (0.02) (0.01) (0.02) (0.03) (0.03) (1.21) (0.03) N 118918 118918 118918 118917 118918 118029 118029 118917 118918 118918 111586 118918 r2 0.01 0.03 0.05 0.02 0.06 0.03 0.03 0.31 0.17 0.03 0.06 0.40 Crime *0.10 **0.05 ***0.01. Clustered standard errors at district level are in parentheses. 25