Rumayya. UWA Business School. Economics

Transcription

1 Three essays in political economy of decentralization in Indonesia Rumayya This thesis is presented for the degree of Doctor of Philosophy of The University of Western Australia UWA Business School Economics 2017

2

3 Abstract During the late 1990 s Indonesia embarked upon major political, governmental and fiscal reforms. As part of this reform the central government transferred almost all of its functions to district governments. A new system of intergovernmental transfers was established to finance the functions transferred to the districts. The new transfers consist of Revenue Sharing (DBH), General Allocation Funds (DAU) and Special Allocation Fund (DAK). Since 2004 another part of the reform has been the change in the electoral system of the president, from parliamentary election to direct election by the people. Also, starting in 2005, district government heads were selected through direct elections among citizens. Within this context, three issues are examined within the thesis. First, the thesis examines whether intergovernmental transfers in Indonesia were manipulated in order to pursue the political interests of the president. I employ the regression discontinuity design (RDD) on data set of transfer to district government during the period , and presidential election results at district level in the electoral years 2004 and I find that district governments - where the current president barely won in the most recent election - receive a larger Special Allocation Grant (DAK). I also find that the incidence and magnitude of this political manipulation is greater in the last two years of the presidential term. Looking at the degree of manipulation, it seems that transfers with smaller monetary value and more complex allocation formula are more prone to manipulation. Second, the thesis tests whether economic growth and unemployment rates matter in the re-election of incumbent district leaders in Indonesia. Applying the Probit and Hekcprobit model on Indonesia s local direct elections during , I find that both unemployment and GDP per capita growth has an impact on election outcomes in the election year. Yet, it is only the average annual GDP per capita growth, during the incumbent district leaders whole tenure that matters for his/her re-election. However, when I separate luck (district s performance due to regional or national economy) from competence (district economy own-performance), I find that in the election year only competence matters for re-election, with no evidence of influence of luck; whereas, in the average annual performance of the incumbents tenure, luck matters for re-election. This suggests that voters put more attention and vigilance on the incumbents performances in the last year, rather than on their whole tenure. i

4 Third, this thesis investigates the effect of different types of transfers on voter turnout in Indonesia s local election. I apply the OLS and Beta regression on a panel data set of variety of intergovernmental transfers received by Indonesia s district government, and also on the district leader direct election data, within the period I also employ the Two Stage Least Squares (2SLS) to address the risk of endogeneity bias. I use the share of oil sectors in the district s GDP as an instrument, since the district s oil endowment is more likely to be determined by exogenous geological factors. I find that the increase in share of Total Revenue Sharing (DBH) and Revenue Sharing in oil (DBH oil), the most unrestricted type of transfer, to total district government expenditure increases voter turnout. I argue that an increase in central government transfer could stimulate voter turnout in local elections if it is designed to give local government higher fiscal flexibility in its spending allocation. In addition, I also find evidence that an increase of the local government s fiscal flexibility in spending diminishes corruption practice at district level, which implies improvement in accountability. ii

5 iii

6 Table of Contents Abstract... i Table of Contents... iv List of Tables... vi List of Figures... viii List of Abbreviations... ix Acknowledgements... xi Statement of Candidate Contribution... xiii Chapter 1: Introduction Background The Political Economy of Intergovernmental Transfer Economic Performance and District s Leader Re-election Intergovernmental Transfer and Turnout in Local Election... 7 References Chapter 2: The President and Intergovernmental Transfers Allocation in Indonesia Introduction Literature Review Intergovernmental transfers in Indonesia Presidential Election in Indonesia Data and Methodology Empirical results Validity & Robustness Checks Discussion Conclusion Appendix References Chapter 3: Re-election of Incumbent District Leaders in Indonesia - Does Economy Matter? Introduction Literature Review Local Election in Indonesia: Institutional Setting Model Specification Data Empirical Results Robustness Analysis Conclusion Appendix References iv

7 Chapter 4: Intergovernmental Transfer and Voter Turnout in Indonesia s Local Election Introduction Literature Review Fiscal Decentralization and Intergovernmental Transfers in Indonesia Political Decentralisation in Indonesia: Direct Local Election of District Heads Empirical Strategy Empirical results Robustness analysis Discussion Conclusion References Chapter 5: Conclusion Summary of Research Findings Policy Implication Limitation of study and suggestion for further research References v

8 List of Tables Table 2.1 Specific Allocation Funds (Dana Alokasi Khusus/DAK) sectors Table 2.2 The 2004 presidential election results 33 Table 2.3 The 2009 presidential election results 34 Table 2.4 Descriptive statistics of central government transfer 35 Table 2.5 Descriptive statistics of subcomponents of DAK.. 36 Table 2.6 Pre-direct election mean-differences winning vs. losing districts.. 43 Table 2.7 The presidential election impact on central government transfer allocation.. 47 Table 2.8 The presidential election result impact on sub-components of DAK Table 2.9 The presidential election impact on transfer allocation, last two-years in office Table 2.10 The election impact on allocation of sub-dak, last two-years in office. 52 Table 2.11 The transfer s covariates discontinuity test, all samples Table 2.12 Effect of current presidential election outcome on lag five years of transfer Table 2.13 Effect of current presidential election on lag five years of sub-dak Table 2.A1 Districts characteristics in pre-direct election presidency (2004) Table 2.A2 The presidential election impact on transfer, first two-years in office. 70 Table 2.A3 The presidential election impact on transfer, first two-years in office. 70 Table 2.A4 Descriptive statistics of districts characteristics, all samples Table 2.A5 Effect of presidential election on lag five years of transfer, last two years. 71 Table 2.A6 Effect of presidential election on lag five years of sub-dak, last two years.. 73 Table 2.A7 Effect of presidential election on lag five years of sub-dak, first two years..73 Table 3.1 Description of the variables Table 3.2 Descriptive statistics Table 3.3 The impact of GDP per capita growth on probability of re-election: Probit Table 3.4 The impact of unemployment change on probability of re-election: Probit Table 3.5 The impact of GDP per capita growth on probability of re-election: Heckprobit..102 Table 3.6 The impact of unemployment on probability of re-election: Heckprobit Table 3.7 Regional luck & competence effect of GDP per capita growth: Probit Table 3.8 Regional luck & competence effect of unemployment change: Probit Table 3.9 National luck & competence effect of GDP per capita growth: Probit Table 3.10 National luck & competence effect of unemployment change: Probit Table 3.11 Regional luck & competence effect of GDP per capita growth: Heckprobit Table 3.12 Regional luck & competence effect of unemployment change: Heckprobit Table 3.13 National luck & competence effect of GDP per capita growth: Heckprobit Table 3.14 National luck & competence effect of unemployment change: Heckprobit vi

9 Table 3.A1 IV-Probit base regression income per capita growth Table 3.A2 IV-Probit base regression change in unemployment Table 3.A3 IV-Probit relative performance to growth of regional districts Table 3.A4 IV-Probit relative performance to regional districts in unemployment Table 3.A5 IV-Probit relative performance to national income per capita growth Table 3.A6 IV-Probit relative performance to national unemployment change Table 4.1 Descriptive statistics Table 4.2 The impact of share of revenue sharing to total district s expenditure to turnout Table 4.3 The impact of share of general allocation funds to to turnout 163 Table 4.4 The impact of share of specific allocation fund to turnout. 164 Table 4.5 The impact of district s expenditure flexibility to turnout. 168 Table 4.6 The impact of share revenue sharing from oil to turnout Table 4.7 The impact of share revenue sharing from oil to turnout: 2SLS 174 Table 4.8 The impact of share revenue sharing from oil to districts corruption: 2SLS. 180 vii

10 List of Figures Figure 2.1 Compositions of Indonesia Fiscal Equalisation Funds Figure 2.2 Electoral margin of the current president in power Figure 2.3 Central Government Transfer to District Governments, Figure 2.4 Central Government Transfer The last 2 Years of Tenure Figure 2.5 Selected Central Government Transfer Covariates Figure 2.6. Lagged 5 Years of Central Government Transfer Figure 2.7 Treatment effects, size and formula complexity of DAK subcomponents Figure 2.A1 Central Transfer The First 2 Years of Tenure Figure 2.A2 Lagged 5 Years of Central Transfer, Last 2 Years of President Tenure..72 Figure 4.1 Composition of Local Government s Revenue in Indonesia 151 Figure 4.2 Voter Turnout in the Local Leader Direct Election ( ) 155 viii

11 List of Abbreviations 2SLS APBN BPS DAK DAU DBH DPR GDP INDODAPOER IV Two-Stage Least Squares Anggaran Penerimaan dan Belanja Negara (State Budget Revenue and Expenditure) Biro Pusat Statistik (Central Bureau of Statistics) Dana Alokasi Khusus (Special Allocation Fund) Dana Alokasi Umum (General Allocation Funds) Dana Bagi Hasil (Revenue Sharing) Dewan Perwakilan Rakyat (House of Representatives) Gross Domestic Product Indonesia Database for Policy and Economic Research Instrumental Variables estimator KPU Komisi Pemilihan Umum (Indonesia s General Election Commission) KPUD MoHA OLS PILKADA RDD Komisi Pemilihan Umum Daerah (Regional General Election Commission) Ministry of Home Affairs Ordinary Least Squares Pemilihan Kepala Daerah (Local Election) Regression Discontinuity Design ix

12 x

13 Acknowledgements I am especially indebted to my coordinating supervisor Professor Anu Rammohan, who has provided me with support, patience, and encouragement throughout my research journey. I would also thank my co-supervisor Associate Professor Sam Tang for all his input during my study. I greatly acknowledge the financial support from the Australian Award Scholarship (AAS)-AusAid, which enable me to pursue PhD study at the University of Western Australia (UWA). I would also like to acknowledge the AusAID liaison officers at the UWA, Debra Basanovic and Deborah Pyatt for kindly assistance on various issues during my study. I also would like to extend my gratitude to the Business School research team, especially Adam Hearman and Mei Han for providing me with administrative support. Also, I am thankful to Ms. Julia Lightfoot for kindly proofreading my thesis. My thanks are extended to the other PhD students at Economics Discipline, UWA for friendly working environment. Especially for my office neighbors: Prayudhi Azwar, Riznaldi Akbar, Noor Syaifudin, Yashar Tarverdimamaghani, Rini Priyati, Longfeng Ye, Judiana Manihuruk, and Hay An. My special thanks to Marco and Ravhani, for fruitful discussion and warm friendship during my stay in Perth. Also, I am grateful to my parents and parents-in-law, for their endless pray. Most importantly, I would like thank my wife, Onish Akhsani, my son, Musa and my little girl, Maryam for their love, support, patience and inspiration during completion of this thesis. xi

14 xii

15 Statement of Candidate Contribution This thesis contains published work and/or work prepared for publication, some of which has been co-authored. The bibliographical details of the work and where it appears in the thesis are outlined below. Chapter 2: Rumayya & Rammohan, A. 2017, The President and Intergovernmental Transfers Allocation in Indonesia. Submitted to Bulletin of Indonesian Economic Studies (BIES). Contribution: Rumayya : 80 % Anu Rammohan : 20% Rumayya 03 May 2017 xiii

16 xiv

17 Chapter 1: Introduction 1.1 Background There are three layers of government in Indonesia: central government, provincial government, and district government. Since 1966, under the Suharto s authoritarian New Order regimes, Indonesia was highly centralized. Most decisions were made by the central government, whereas the local government acted merely as a cashier for the national government (Henderson and Kuncoro 2004). In 1998, Suharto resigned, following the Asian financial crisis that hit the country in 1997, and his vice president Bacharuddin Jusuf Habibie succeeded him as President. The Habibie presidency came under pressure from regions rich in natural resources, which voiced demands for greater transfer and autonomy(tadjoeddin et al. 2001). In fact, the threat of territorial disintegration escalated in those regions. Responding to this threat of territorial disintegration, the national parliament in May 1999 enacted two decentralization laws (Law No. 22 and No. 25 of 1999) that mandated the transfer of much authority from the central government to district governments. The central government s authority was restricted to defence, religion, justice, foreign affairs, and macroeconomic policy. The law also decreed that those decentralised functions must be accompanied by intergovernmental transfers to finance those functions. There are three forms of such transfers: Revenue Sharing (Dana Bagi Hasil/DBH), General Allocation Fund (Dana Alokasi Umum/DAU) and Special Allocation Fund (Dana Alokasi Khusus/DAK). These transfers are all allocated by central government to district government using certain mathematical formula (formula-based). Nevertheless, the laws were only implemented in 2001 (World Bank 2008). 1

18 Habibie also liberalised Indonesia's press and political party laws, and held an early democratic election in 1999, which resulted in the end of his presidency. His presidency was the third, and the shortest, after independence. In June 1999, the legislative election was held. The people of Indonesia directly elected their national parliament members, and the national legislative body constituted from the election comprised a total of 500 seats. In October 1999, this national parliament conducted a presidential election, in which Abdurrahman Wahid and Megawati Sukarnoputri elected as president and vice-president respectively. They were due to rule for the next five-year term ( ). However, in July 2001, after a series of conflicts with the national parliament, President Abdurrahman Wahid was impeached. Following the impeachment, the parliament raised Megawati to the presidency to complete the five-year term. The impeachment of President Abdurrahman Wahid motivated the parliament, in 2002, to amend the Constitution of Indonesia and include new articles on the presidential election system (Ufen 2006). Under the new amendments the president would be elected not by the parliament, but directly by the people. Also, impeachment of the president would require the consent of the judiciary, to prevent unwarranted influence on the president from partisan interests in parliament, and to stabilise the position of the president (Ufen 2006). As a result, starting from 2004, the president has been elected directly by the people for a term of five years. Indonesia took another important step of political reform in 2004 by revising the decentralisation law on regional autonomy (Law 32/2004). This new law ordained that the heads of local governments be directly elected by the local populace in the local elections (Pemilihan Kepala Daerah/PILKADA). Hence, starting in 2005, district heads were selected through direct elections among the citizens. Previously, district heads were 2

19 indirectly elected by the local parliaments. It was believed that this democratic reform would make the district heads more accountable to their electorates (Kaiser et al. 2006). Given this background, this thesis explores three research questions: (1) Has the central government s transfer allocation to the district governments been politically manipulated by the president? (2) Does economic performance influence incumbent district heads re-election? (3) How does differences in the design of intergovernmental transfer impact upon voter turnout in the local elections? I address each question in three separate and self-contained chapters in the thesis. The empirical analysis used to address the questions utilized election data from Indonesia s General Election Commission (KPU). A substantial work of this thesis comprised of collecting the election dataset, especially in the local elections, as there has not been a single published master database on local collections across Indonesia. The local elections are implemented and recorded by the local KPU in each region, with the election outcome subsequently reported to the central KPU in Jakarta. However, at the time of this study there has not been any effort from the central KPU to compile the local election data to a single national master dataset. Hence, the construction of a national master dataset on Indonesia s local elections is one of the distinctive contributions of this thesis. 3

20 1.2 The Political Economy of Intergovernmental Transfer The electoral reform of the presidential election, from indirect elections by the parliament to direct elections by the people, is considered a major progress in Indonesia s democratization. The new electoral system shifts the President s accountability directly to the people, rather than to the parliament. However, the combination of direct elections and the immunity of the President from impeachment, without consent of the judiciary, arguably make the President stronger in relation to the parliament (Ufen 2006). Within this context, in Chapter 2 of this thesis, I examine whether intergovernmental transfers in Indonesia were manipulated in order to pursue the political interests of the President. This is an important and interesting question to investigate, given that the transfer allocation is drafted on a yearly basis, for the following budget year, by the President and his/her cabinet. The central government s fiscal transfer is part of the national budget (Anggaran Penerimaan dan Belanja Negara/APBN/State Budget Revenue and Expenditure) that is determined yearly. It commences with a budget draft assembled by the executive (president and his/her cabinet) for the following budget year. The president leads the ministry in the budget formulation process. The draft includes an estimation of the future revenue and sectoral spending allocation. Once completed, the draft is then sent to parliament (Dewan Perwakilan Rakyat/DPR/House of Representatives) for review and consultation. Due to the ministry s influence, it is possible for the president to influence the regional distribution of central government transfers, especially in the DAU and DAK allocations, where the allocation mechanism is entirely a top-down process. 4

21 Although all intergovernmental transfers in Indonesia are formula based, it is interesting to evaluate whether political consideration still matters in the allocation. Indeed, most of the studies in the literature of tactical redistribution have focused almost exclusively on discretionary government spending, implicitly assuming that rules-based programs are implemented without regard to special interests (Litschig 2012). In that respect, Chapter 2 of this thesis contributes to the literature and seeks to examine whether rule-based transfers might be subject to political manipulations (Banful 2011; Caldeira 2012; Litschig 2012). This is particularly in regards to the smaller branch of literature that examines the presidential pork barrel (Larcinese et al. 2006; Taylor 2008; Berry et al. 2010). The empirical analysis in Chapter 2 applies the Regression Discontinuity Design (RDD), the emerging quasi-experimental technique in the estimation. I adapt the RDD in close electoral races to causally identify transfer manipulations in relation to election outcome, by comparing districts where the current president won by a small margin, with districts where the current president lost by a small margin, in the last presidential election. I employ a newly developed method proposed by Calonico et al. (2014) that is arguably more credible than the regression discontinuity design commonly used in the literature (Migueis 2013; Brollo and Nannicini 2012; Bracco et al. 2015). The empirical results of Chapter 2 demonstrate that for the overall presidency of the directly elected president ( ), only in the Special Allocation Grant (DAK) is there evidence of political manipulation (districts where the president won marginally receive larger DAK). However, no evidence of political distortion was found in the other types of transfers within that period. Nonetheless, the study in Chapter 2 finds evidence of political manipulation in all types of transfers, except for Revenue Sharing from tax (DBH Tax), during last two years of the presidency. The results of this Chapter 5

22 demonstrate that a formula-based/rules-based system of allocating transfers anchored in the Law, and based on apparently technocratic inputs, is not immune to politically motivated targeting. 1.3 Economic Performance and District s Leader Re-election It was believed that the local government electoral reform toward local direct elections would improve district government accountability in their electorates. Yet, there are indications that the democratic accountability mechanisms might have been compromised. For instance, Sjahrir et al. (2014) discovers that the introduction of local district government heads does not reduce the degree of district government administrative expenditure overspending, which they claimed indicates the failure of local elections in establishing accountability. Moreover, Mietzner (2005, 2010) reports that the political process on the local level has been partly characterised by both money politics and powerful local elites. On the other hand, Skoufias et al. (2014) finds that local direct election improves the local government s accountability to the poor. Hence, whether the democratic accountability of local election has been - in practice - compromised, remains an important empirical question. In Chapter 3 I attempt to answer the question by investigating whether incumbent district heads were held accountable to his/her performance in the first two cycles of local direct elections in Indonesia. Specifically, I investigate whether voters punished incumbents when the economy was doing poorly, and rewarded them when the economy was doing well. In the empirical analysis, I apply insight from the literature on attribution errors in the context of retrospective economic voting (Wolfers 2007; Leigh and McLeish 2009; Hayes et al. 2015) to separate the effect of re-election of incumbents due to luck and competence. Attribution error is defined as, a tendency to systematically fail to take 6

23 sufficient account of environmental factors when aiming to assess competence (Patty and Weber 2005). However, I differ from previous studies by utilising regional districts economy performances as reference points for the evaluation of competence, and not just the national economy. The estimates of the chapter suggest that the extent of such attribution error is only evident when district performance is benchmarked using regional performance within the overall period of the incumbents tenure; not, however, in the election years. It appears that voters vigilance in distinguishing the incumbent district leaders competence from luck is more accurate for the performance in the election year rather than for the whole period of the incumbents tenure. It implies that performance in the last year of tenure is what matters most for the incumbents chances of re-election. 1.4 Intergovernmental Transfer and Turnout in Local Election Literature on decentralization warns that, although political decentralization in the form of local elections could improve the accountability of local governments to the citizens, it also puts democracy at risk of being captured by local elites who have special interests. This is because, on the local government level, collusion is easier to establish and maintain across different interest groups, since the transaction costs and information asymmetry are lower due to greater proximity (Bardhan 2002; Prud'homme 1995). In the context of Indonesia, Azis and Wihardja (2010) and Azis (2011) conducted field survey in 12 regions over the period , and found some evidence of local democracy being captured by the local elite. Azis (2011) contends that limited participation and poor quality of local leaders are among the factors that contribute to this local capture. 7

24 Voter turnout in elections is considered one of the important indicators of political participation (Geys et al. 2010). A high turnout level ensures that political leadership is represented by the majority of citizens, and not only by elites (Lijphart 1997). In the context of political decentralization, citizen participation in local elections is one of important condition that could ensure decentralisation success in establishing responsive and accountable local governments (Litvack et al. 1998). A higher turnout may give local politicians incentives to implement policies that benefit the electorate at large, at the expense of policies benefiting their own special interests. Hence, a high degree of democratic participation in terms of voter turnout in local elections may reduce the risk of local capture through more efficient monitoring of local politicians. In Chapter 4 I investigate how the existing designs of fiscal decentralization in Indonesia, which puts a higher emphasis on expenditure rather than revenue, influences voter turnout in local elections. Specifically, I examine the effect of different types of transfers on the level of voter turnout in Indonesia s local elections. This is an important issue to investigate given the current design of fiscal decentralization, where local governments are highly dependent on central government transfers to finance their expenditure. Approximately, more than 70% of local government revenue came from intergovernmental transfers during the study period. In the empirical analysis of Chapter 4 I employ the OLS and Beta regression techniques on a panel data-set of 497 district governments in Indonesia for the first two cycles of local government direct elections within the period I discovered and recorded that the types of intergovernmental transfers which gave fiscal flexibility to local governments in allocating their spending, increased voter turnout in local elections. To facilitate the interpretation of these findings I explored how fiscal flexibility in spending influences corruption at the district level. I found evidence that higher fiscal 8

25 flexibility in spending diminishes corruption practices, which indicates an improvement on governance and accountability at the local level. Chapter 4 contributes to the literature on institutional frameworks that determine the consequences of decentralisation, particularly in regard to the literature which argues that accountability is higher in local governments that rely less heavily on central government transfers (Geys et al. 2010). The reason is that local governments which depend on central government transfers to finance their functions diminish the interest of voters in holding the local politicians accountable for their decisions (Devarajan et al. 2009). This study, however, demonstrates that central government transfers could, on the contrary, increase voters interest in accountability, which can be observed by their turnout in elections; as long as the transfer is designed to give local government fiscal flexibility in its spending allocation (policy autonomy). 9

26 References Azis, I. J. (2011). Institutional Model of Decentralization in Action. Asian Development Bank Economics Working Paper Series, (288). Azis, I. J., & Wihardja, M. M. (2010). Theory of endogenous institutions and evidence from an in-depth field study in Indonesia. Economic & Finance Indonesia, 58(3), Banful, A. B. (2011). Do formula-based intergovernmental transfer mechanisms eliminate politically motivated targeting? Evidence from Ghana. Journal of Development Economics, 96(2), Bardhan, P. (2002). Decentralization of governance and development. The Journal of Economic Perspectives, 16(4), Berry, C. R., Burden, B. C., & Howell, W. G. (2010). The president and the distribution of federal spending. American Political Science Review, 104(04), Bracco, E., Lockwood, B., Porcelli, F., & Redoano, M. (2015). Intergovernmental grants as signals and the alignment effect: Theory and evidence. Journal of Public Economics, 123, Brollo, F., & Nannicini, T. (2012). Tying your enemy's hands in close races: the politics of federal transfers in Brazil. American Political Science Review, 106(04), Caldeira, E. (2012). Does the system of allocation of intergovernmental transfers in Senegal eliminate politically motivated targeting? Journal of African Economies, 21(2), Calonico, S., Cattaneo, M. D., & Titiunik, R. (2014). Robust Nonparametric Confidence Intervals for Regression Discontinuity Designs. Econometrica, 82(6), Devarajan, S., Khemani, S., & Shah, S. (2009). The politics of partial decentralization. In E. Ahmad & G. Brosio (Eds.), Does Decentralization Enhance Service Delivery and Poverty Reduction? (pp ). Cheltenham: Edward Elgar. Geys, B., Heinemann, F., & Kalb, A. (2010). Voter involvement, fiscal autonomy and public sector efficiency: evidence from German municipalities. European Journal of Political Economy, 26(2), Hayes, R. C., Imai, M., & Shelton, C. A. (2015). Attribution Error in Economic Voting: Evidence from Trade Shocks. Economic Inquiry, 53(1), Henderson, J. V., & Kuncoro, A. (2004). Corruption in Indonesia. NBER Working Paper Series, Larcinese, V., Rizzo, L., & Testa, C. (2006). Allocating the US federal budget to the states: The impact of the president. Journal of Politics, 68(2), Leigh, A. & M. McLeish. (2009). Are State Elections Affected by the National Economy? Evidence from Australia. Economic Record, 85 (269), Lijphart, A. (1997). Unequal participation: Democracy's unresolved dilemma presidential address, American Political Science Association, American Political Science Review, 91(01),

27 Litschig, S. (2012). Are rules-based government programs shielded from special-interest politics? Evidence from revenue-sharing transfers in Brazil. Journal of Public Economics, 96(11), Litvack, J., Ahmad, J. & Bird, R. (1998) Rethinking Decentralization in Developing Countries. Washington, DC: World Bank. Patty, J. W., & Weber, R. A. (2007). Letting the good times roll: A theory of voter inference and experimental evidence. Public Choice, 130(3-4), Prud homme, R. (1995). On the dangers of decentralization. World Bank Research Observer, 10, Tadjoeddin, M. Z., Suharyo, W. I. & Mishra, S. (2001). Regional disparity and vertical conflict in Indonesia. Journal of the Asia Pacific Economy, 6(3), Taylor, A. J. (2008). The presidential pork barrel and the conditioning effect of term. Presidential Studies Quarterly, 38(1), Ufen, A. (2006). Political Parties in Post-Suh arto Indonesia: Between politik aliran and Philippinisation (No. 37). GIGA German Institute of Global and Area Studies. Mietzner, M. (2005). Local democracy. Inside Indonesia, 85, Mietzner, M. (2010). Indonesia s Direct Elections: Empowering the Electorate or Entrenching the New Order Oligarchy? In Edward Aspinall and Greg Fealy, eds. Soeharto s New Order and its Legacy, eds. Canberra: ANU E Press Migueis, M. (2013). The effect of political alignment on transfers to Portuguese municipalities. Economics & Politics, 25(1), Sjahrir, B. S., Kis-Katos, K., & Schulze, G. G. (2014). Administrative overspending in Indonesian districts: The role of local politics. World Development, 59, Skoufias, E., Narayan, A., Dasgupta, B., & Kaiser, K. (2014). Electoral accountability and local government spending in Indonesia. World Bank Policy Research Working Paper, (6782). Wolfers, J. (2007). Are voters rational? Evidence from gubernatorial elections. Mimeo, University of Pennsylvania. World Bank. (2008). Spending for Development: Making the Most of Indonesia's New Opportunities. World Bank, Jakarta. 11

28 Chapter 2: The President and Intergovernmental Transfers Allocation in Indonesia 2.1 Introduction After more than thirty years of dictatorship, democracy was re-established in Indonesia following the fall of President Suharto s authoritarian regime in May As a part of the reform, Indonesia transformed herself from one of the most centralised into one of the most decentralised countries in the world. Under Law No. 22/1999 on Local Governments, the central government transfers all functions to district governments, except core functions commonly expected by central governments such as foreign policy, defence and monetary policy (World Bank 2008). The law also mandates that those decentralised functions must be accompanied by intergovernmental transfers to finance those functions. According to the law, intergovernmental transfers include Revenue Sharing (Dana Bagi Hasil/DBH) and grants. The grants include the General Allocation Fund (Dana Alokasi Umum/DAU) and the Special Allocation Fund (Dana Alokasi Khusus/DAK). These transfers (DBH, DAU and DAK) are all rules/formulas-based. Together these three transfers are called Fiscal Equalisation Funds (Dana Perimbangan) and account for the largest portion of local government total revenues, staying at average 70 percent within the period of Another governance reform that occurred in 2001 and 2002 related to the presidential election. The national parliament added a series of amendments to the Constitution of Indonesia so that the president should be elected directly by the people. Previously, the president was elected by the parliament. Additionally, the amendments also required the consent of the judiciary for the impeachment of the president by the 12

29 parliament. Since 2004, as a result of the amendments, the president has been elected by the people for a term of five years. This was a profound reform, with the President now accountable directly to the people rather than to parliament. However, the introduction of direct presidential elections and the amended laws of the Constitution whereby consent of the judiciary is required before a President can be impeached, arguably make the executive (that is, the president) become stronger in relation to the parliament (Ufen 2006). In this chapter I examine whether intergovernmental transfers in Indonesia were manipulated to pursue the political interests of the president. This is an important and interesting question, given that the transfer is part of the national budget (Anggaran Penerimaan dan Belanja Negara/APBN/State Budget Revenue and Expenditure) and drafted on a yearly basis by the president and his/her cabinet for the following budget year. The president, who leads the ministry in the budget formulation process, to certain extent, should have some degree of influence on the regional allocation of central government transfers. This study contributes to the literature on the political economy of intergovernmental transfers in several ways. First, it offers insights regarding the nature of the political economy of transfers in a post-authoritarian developing world. There is a wide range of countries which offer evidence of politically motivated targeting of transfers including developed nations such as Australia (Worthington and Dollery 1998), Sweden (Dahlberg and Johansson 2002), Portugal (Pereira 1996; Veiga and Pinho 2007) and the United States (Wright 1974; Wallis 1996, 1998; Anderson and Tollison 1991; Grossman 1994), and Albania (Case 2001), India (Khemani 2007; Cole 2009; Arulampalam et al. 2009) and Argentina (Porto and Sanguinetti 2001) from the developing world. Indonesia however offers a unique and interesting setting to evaluate 13

30 whether political manipulations of transfers take place in developing, post-authoritarian and a newly decentralised country. Moreover, to the best of my knowledge, this is the first study that quantitatively examines the political economy of all types of intergovernmental transfers in post-authoritarian Indonesia. The second contribution of this study to the political economy of intergovernmental transfer s literature is that it applies the emerging quasi-experimental technique in the estimation (Migueis 2013; Brollo and Nannicini 2012; Bracco et al. 2015). From an empirical perspective, every attempt to estimate the causal impact of an election outcome on the amount of transfers is clearly complicated by endogeneity issues. Without a credible source of exogenous variation in an election outcome, the empirical correlation between election and transfers can be completely driven by socio-economic factors influencing both dimensions. To causally identify transfer manipulations in relation to election outcome I adapt Regression Discontinuity Design (RDD) in close electoral races, as pioneered by Lee (2008). This is done by comparing districts where the current president has barely won, with districts where the current president has barely lost, in the last presidential election. Although I am not the first to estimate the impact of elections on the allocation of transfers using RDD, I employ a newly developed method proposed by Calonico et al. (2014b) in my RDD that is arguably more credible than the regression discontinuity design commonly used in the literature. RDD is chosen as method of analysis over the other quasi-experimental approaches (e.g. difference in differences, propensity score matching, instrumental variables) because of two considerations. First, it is a methodology that allows making causal conclusions that are nearly as reliable as a randomized experiment (Lee and Lemieux 2010). Second, the dataset characteristic of this study is fit with the RDD 14

31 condition: it can only be applied in cases where there is a specific threshold that determines who is eligible to receive the intervention being evaluated (Khandker et al. 2010). In the context of this study the intervention is the current president victory in the last presidential election at the district level, while the threshold defined as the vote margin of the president above zero. The third contribution to the literature is that it seeks to examine whether rulebased transfers might be subject to political manipulations (Banful 2011; Caldeira 2012; Litschig 2012). Formula-based resource allocation mechanisms are employed to prevent tactical redistribution over intergovernmental transfers. This formula-based resource allocation strategy has gained prominence in the developing world after the wave of decentralization in the last two decades (Banful 2011). The prevailing assumption is that resources which are distributed by a formula based on economic and welfare variables will suspend the arbitrariness that allows politically motivated targeting. Whether in practice rule-based transfers successfully eliminate tactical redistribution is an important empirical question. However, little is known about this issue. Most of the studies in the literature of tactical redistribution have focused almost exclusively on discretionary government spending, implicitly assuming that rules-based programs are implemented without regard to special interests (Litschig 2012). This study contributes to the literature by examining the effectiveness of formulas as a strategy for limiting political motivations behind central government transfers to local government in the Indonesian post-authoritarian period. This study also contributes to the small but growing literature that examines the presidential pork barrel (Larcinese et al. 2006; Taylor 2008; Berry et al. 2010). It focuses on presidential election effects on transfers received by state governments. Most tactical distribution literatures focus on the effect of party composition in the parliament and 15

32 many empirical tests also ignore the president (Berry et al. 2010). However, Indonesia s electoral changes from parliamentary election of the president to direct election by the people in 2004 make the tactical motives pottentialy compelling to examine in the Indonesian context. Finally, this study also contributes to the literature on political budget cycles in the context of developing countries. According to Brender and Drazen (2005), fiscal manipulation may work better in new rather than in established democracies because voters in the former may be inexperienced with electoral politics or may have less information available to evaluate fiscal manipulation. Using the whole dataset from two presidential term ( ), the empirical results show that in close elections, the districts where the president is winning the electoral race receive about 44.4 percentage points larger Special Allocation Grant (DAK). No evidence was found of political distortions in the other form of transfers (DBH and DAU) within that period. However, during last two years of the presidency I find evidence of political manipulation in all types of transfers. During the last two years of the presidential term, districts where the current president barely won in the last election had a 133 percentage point higher Special Allocation Grant (DAK), 87.5 percentage point higher Revenue Sharing (DBH), and a 73 percentage point higher General Allocation Fund (DAU). On the contrary, during the first two years of the presidential term, no evidence exists of political distortions in the allocation of any transfers. These results are consistent with the prediction of core-voters and political budget cycle in the literature. The results of this study also show that even a formula-based/rules-based system of allocating transfers anchored in the Law, and based on apparently technocratic inputs, is not always immune to politically motivated targeting. 16

33 I argue that variation of the magnitude of the electoral effect on types of transfer and their occurrence are influenced by the transfer allocation mechanism. The simpler the transfer allocation formula is and the bigger the size of the transfer, the less likely is the incidence and magnitude of politically manipulated transfers. The chapter is organized as follows: Section 2 discusses previous theoretical and empirical literature on distributive politics; Section 3 provides an overview of all types of Indonesian intergovernmental transfers, with particular emphasis on their rule of allocation; Section 4 presents the development of the presidential election system of postauthoritarian Indonesia. The data and the empirical strategy are described in Section 5. Section 6 presents the empirical analysis followed by a validity and robustness check in Section 7. In Section 8, I discuss possible reasons for differences in the magnitude and incidence of political influence across different types of transfers; and Section 9 presents the main conclusions. 17

34 2.2 Literature Review This study is related to three strands of literature. First, it relates to the literature on tactical redistribution of intergovernmental transfers. Within this strand of literature, there is a theoretical debate about the direction in which transfers will be affected based on the political characteristics of the recipient groups. Generally, there are two strands of modelling related to this debate: the swing voter model (Lindbeck and Weibull 1987; Dixit and Londregan 1998) which claims that politicians have preferences in sending funds to areas in which the voters are willing to compromise their votes in response to economic benefits; and the core voter model (Cox and McCubbins 1986) which argues that voters respond more strongly to economic incentives provided by the politicians they prefer. According to the core voter model, politicians just as risk-averse investors, will target more resources to areas in which their political support is concentrated in order to maximise their return in terms of votes. There is empirical support from various countries for both models. The evidence for the swing voter model can be found in studies from Sweden (Dahlberg and Johansson 2002), the U.S. (Ansolabehere and Snyder 2006; Levitt and Snyder 1995), Kenya (Barkan and Chege 1989), Ghana (Miguel and Zaidi 2003), Albania (Case 2001), and Peru (Schady 2000). Meanwhile, the evidence for the core voter model are found in India (Cole 2009), Sweden (Johansson 2003; Dahlberg and Johansson 2002), the U.S. (Wright 1974), Albania (Case 2001), and Peru (Schady 2000). However, this study extends the existing literature in two ways. First, I address Larcinese et al. s (2008) critique that election results are inherently endogenous to transfers received by local governments, and therefore, their impact cannot be accurately measured directly, by applying Regression Discontinuity Design (RDD) technique. Altough I am not the first researcher to use RDD in the tactical redistribution literature 18

35 (Migueis 2013; Brollo and Nannicini 2012; Bracco et al. 2015), I employ a newly developed method proposed by Calonico et al. (2014b) that is arguably more credible than the regression discontinuity design commonly used in the literature (see Section for details). The other extension focuses on the presidential impact on transfers. Most studies on distributive politics focus on the internal operations of parliaments, paying particular attention to party ideology and composition of parliaments. Nonetheless, there has been a small but growing literature that examines the presidential pork barrel (Larcinese et al. 2006; Taylor 2008; Berry et al. 2010). Larcinese et al. (2006) examined the U.S. federal budget allocation to the states during the period , finding that the states which heavily supported the incumbent president in past presidential elections tend to receive more funds, while marginal and swing states are not rewarded. They contend that their results show that presidents are engaged in tactical distribution of federal funds. Taylor (2008), also using U.S. data, examined cross-state distributions of procurement contracts from 1984 to 2004 to test hypotheses that a presidential electoral connection explains geographic patterns in federal spending. He finds that states that gave presidents disproportionately fewer of their popular votes in the re-election received more procurement dollars per capita. Taylor (2008) concludes that his results do not give evidence that presidents use pork barrel politics to get re-elected. Berry et al. (2010) analysed the geographic spending of nearly every domestic program in the U.S. over a 24-year period (1984 to 2007) using district and county fixed-effects estimation strategies, and finds that the federal government spends approximately 4 5% more in districts and counties when members of the president s party represent them. My contribution to this strand of literature is to provide empirical evidence from Indonesia, a developing country recently democratised and decentralised. The Indonesian 19

36 president arguably has extensive opportunities to influence the distribution of central government transfers, especially after the amendments of the constitution by the parliament in 2002 that mandate the president be directly elected by the people (see Sections 2.3 and 2.4 for discussion). Second, there is a small and growing empirical literature on political manipulation in rule-based transfers (Banful 2011; Caldeira 2012; Litschig 2012). Most of the studies in the literature of tactical redistribution have focused almost exclusively on discretionary government spending, implicitly assuming that rules-based programs are implemented without regard to special interests (Litschig 2012). In an attempt to limit politically motivated distribution of national resources, some governments, including Indonesia, have adopted rule-based/formulaic transfer mechanisms. However, an important unanswered question is whether formulas are sufficient to eliminate politically motivated targeting of transfers. Estimating panel data for 67 local governments of Senegal from 1997 to 2009, Caldeira (2012) finds evidence that, even though Senegal employs a formula-based resource allocation mechanism, the distribution of central resources follows a pattern of tactical redistribution more than patronage, where a larger transfer is received by areas with more swing voters. The allocation formula considers the cost of local public spending and the demographic importance of each jurisdiction. However, in practice equity concerns do not appear to affect the allocation of intergovernmental fiscal transfers. Empirical evidences of political manipulation of rule-based/formula-based transfer were also found by Banful (2011) and Litschig (2012). Banful (2011) finds evidence that per capita DACF (District Assemblies Common Fund), a formula-based grants in Ghana, were higher in districts where vote margins in the previous presidential 20

37 election were lower, suggesting that swing voter districts were targeted. Banful (2011) also finds that DACF formula indicators and their weighting were chosen and amended to produce politically desired patterns of transfers. A similar formula manipulation was also found by Litschig (2012) in Brazil, where a revenue-sharing program between the central and local governments that uses an allocation formula based on local population estimates was manipulated to suit political interests. The population estimates entering the formula were manipulated to target municipalities with roughly equal right-wing and non-right-wing vote shares. This manipulation is consistent with swing-voter literature. In this study I test whether the impact of political forces in the distribution of formula-based transfers is also happening in Indonesia. This is compelling, given the fact that Indonesia is a newly decentralised country and all types of central government transfers to local governments in Indonesia is formula-based. Third, this study also relates to the large empirical literature on opportunistic political business cycles (Nordhaus 1975; Lindbeck 1976; Rogoff and Sibert 1988; Rogoff 1990; Drazen and Eslava 2006; Persson and Tabellini 2004; Brender and Drazen 2005; Shi and Svensson 2006; Alt and Lassen 2006). In the seventies, two seminal contributions, Nordhaus (1975) and Lindbeck (1976), initiated the literature on opportunistic political business cycles. They presented models where opportunistic incumbents manipulate the economy before elections in order to maximise their probability of re-election. More recently, Drazen and Eslava (2006) developed a model where politicians use election-year fiscal policy to influence electoral results by targeting types of spending preferred by voters in more swing-voter regions at the expense of other expenditures they do not favour, or other voters. According to Brender and Drazen (2005), fiscal 21

38 manipulation may work better in new rather than in established democracies because voters in the former may be inexperienced with electoral politics or may have less information available to evaluate fiscal manipulation. In the Indonesian context, a recent study by Sjahrir et al. (2013) confirmed the existence of political budget cycles at the district level, especially when direct election of a local districts head was introduced. They found that prior to the local election, the incumbent districts heads spend more on discretionary funds directed toward religious or society/sport groups, which are budgeted as donations or social assistance, and subcategorised under others spending. In another study, Sjahrir et al. (2014) identified the contributing factors to administrative overspending. They found that the degree of political competition increases administrative spending (which is categorised under other spending), and the introduction of the direct head election failed to reduce the degree of overspending. However, to the best of my knowledge there has not been any study that attempts to investigate the political budget cycle in the context of central government transfers to district governments in Indonesia. This study contributes to the current literature by providing the first empirical examination of the impact of presidential election timingcycles on Indonesia intergovernmental grants distribution. The Indonesia case is relevant to the literature on opportunistic political business cycles because Indonesia is a relatively young democracy, and to date most of the research in this literature has focused on established democracies. 22

39 2.3 Intergovernmental transfers in Indonesia As explained in the previous section, the Indonesian decentralization since 2001 has changed substantially the fiscal relations between the central and local governments. This new system, provides three categories of transfer from the central government to regional governments: (i) Revenue Sharing (DBH), (ii) block grants, the General Allocation Fund (DAU) and (iii) Special Allocation Fund (DAK). These three are referred to collectively as fiscal equalisation funds, or balancing funds (Dana Perimbangan). The revenue-sharing arrangement (DBH) is sourced from central government income which is reallocated to the (producer) regions in shares as regulated by Law (UU) No. 33/2004. DBH is distributed as taxation DBH (DBH Pajak) and natural resources DBH (DBH Sumber Daya Alam). Taxation DBH consists of land and building tax (PBB), land and building transfer fees (BPHTB), and personal income tax (PPh). Natural resources DBH are comprised of forestry, mining, fisheries, oil, natural gas and geothermal revenues. 80% 70% 60% 50% 40% 30% DAU DBH DAK 20% 10% 0% Source: Author calculation based on data from Ministry of Finance Figure 2.1 Compositions of Indonesia Fiscal Equalisation Funds (Dana Perimbangan) 23

40 General Allocation Fund (DAU) is allocated on the basis of a jurisdiction s fiscal gap, which is the difference between fiscal needs and fiscal capacity. DAU is a block grant; hence, regional governments are free to make use of the funds without interference from the central government. However, approximately 80% of regionally managed DAU is used for routine expenditure, mainly the salary expenses for public employees at local government level. In contrast to DAU and DBH, over which local governments have discretion as to their use, DAK (Special Allocation Fund) must follow the guidelines determined by the central government. It is allocated on the basis of specified criteria and finance-targeted activities linked to central government priorities, ranging from education and health to rural facilities and the environment. Among the three, DAU has the largest portion. In fact, the General Allocation Fund (DAU) constituted the most important source of finance in the structure of local government revenue prior to decentralization and continues to do so now (see Figure 2.1), followed by the DBH and the DAK. I now turn to provide more details on each component of the fiscal equalisation funds transfers Revenue Sharing Funds (Dana Bagi Hasil/DBH) The allocation of DBH is based on a formula recorded in Law No.33/2004. According to the law, there are two subjects of DBH: taxes and natural resources. Additionally, Article 28 of Government Regulation No. 55/2005 mandates a reconciliation procedure between the central government and subnational governments to confirm the share of DBH. The sources of DBH Taxes are property taxes, land rent, and domestic personal income taxes. For property tax and land rent the local government has the major share of 64.8% and 64%, respectively. Provincial governments shares are 16.2% and 16%, while central governments are 10% and 20%, respectively. For property 24

41 tax 9% is allocated for cost of collection and the central government share of 10% is then redistributed: 65% to all local governments and 35% to local governments whose achieved tax collection in the previous year exceeded the estimated/targeted tax collection target, as an incentive. For land rent the 20% central government share is redistributed to all local governments. As for domestic personal income taxes, the central government s share has the majority with 80%, provincial governments have 8% and local governments have 12%. DBH natural resources consist of revenues from oil and gas, forestry, other general mining, geothermal and fisheries. For oil and gas revenues, the central government has the major share of 84.5% and 69.5%, respectively. Provincial governments shares are 3.1% and 6.1%, while local governments are 12.4% and 24.4%, respectively. For general mining, geothermal, and forestry, the central government s shares are less significant and account for 20% of revenue figures. Provincial governments share is 16% and the remaining 64% goes to local governments. Of the 64% received by local government, 32% is entitled to producing regions, while the remaining 32% is distributed equally to all cities/regencies in the same provinces. The central government keeps 20% of revenues from fisheries and distributes the remaining 80% equally to all local governments. DBH natural resources are administered by the central government and transferred to subnational governments on a quarterly basis. DBH has the same characteristics as unconditional grants (DAU), except for DBH natural resources from oil and gas. For revenues from oil and gas, provincial governments are required to allocate 0.1% of the amount for elementary education, while local governments are required to allocate 0.2% for the same sector. 25

42 In addition to Government Regulation No. 55/2005, the Indonesian government enacted Presidential Regulation 26/2010 specifically to cater to the DBH resource allocation arrangement. The regulation stipulated information transparency on national and local revenues for revenue streams from the extractive industry. The regulation also mandates the formation, structure and responsibilities of a Transparency Team, which is tasked to implement transparency in the management of state and subnational extractive revenues. To perform this responsibility, the Team may seek information, additional data, input, and/or consult with agencies of the central and local government as well as extractive companies for the purpose of reconciliation General Allocation Funds (Dana Alokasi Umum/DAU) The allocation of DAU is based on the formula stipulated in Law No. 33/2004. According to the law, at least 26% of net domestic revenues established in the national budget must be allocated for DAU, which uses weighted variables. While the law states the fixed variables to be used in the formula, the weight of each variable can differ subject to the central government s proposal and national parliament s (DPR) approval. DAU allocation for a given local government is calculated as the sum of its basic allocation and its fiscal gap. The fiscal gap is the difference between fiscal need and fiscal capacity. This formula can be stated as follows: DAU = Basic Allocation + Fiscal Gap = Basic Allocation + (Fiscal Needs Fiscal Capacity) The Basic Allocation is the amount needed to cover personnel spending of local government, which includes base salaries, family assistance, and other related allowances. Fiscal Needs is the financing requirements of the region in providing basic public services. It is formula-based and it is determined by calculating the average total expenditures of all governments and then modifying this figure by reference to five 26

43 different indices of need for budgetary resources. The five indices are population, land area, Human Development Index, Gross Regional Product, and a construction cost index. Fiscal Capacity is defined by financing sources of the local government derived from the variables of own-source revenues (Pendapatan Asli Daerah/PAD) and revenue sharing funds (Dana Bagi Hasil/DBH), which consist of revenue sharing from taxes and from natural resources Specific Allocation Funds (Dana Alokasi Khusus/DAK) DAK is conditional grants and formula-based, and it targets sectors considered as national priorities. DAK matches grants as well, requiring subnational government to allocate at least 10% of DAK nominal value as counterpart funds in a project. DAK allocation is conducted by the Ministry of Finance according to Law No. 33/2004. The DAK allocation process consists of 2 steps: (1) selection of eligible recipients, and (2) computation of the DAK amount to each eligible recipient. Three criteria are used for DAK screening process (Ministry of Finance 2010): 1. General criteria are based on consideration of a subnational government s fiscal capacity, with priority given to regions whose fiscal capacity is lower than the national average, indicated by net fiscal index. 2. Specific criteria are prepared with attention given to laws and regulation, such as regions with special autonomy (currently Papua Province and West Papua Province), and regions which meet certain regional characteristics. The regional characteristics are disadvantaged regions (coastal/island/border regions, natural disaster-prone regions, food security regions) and tourism regions 27

44 3. Technical criteria are based on considerations determined by related line ministries using indicators that illustrate infrastructural characteristics in each sector. To be eligible for DAK, a local government needs to pass one of the three criteria above (general, specific, or technical criteria). In the screening process, local governments are firstly screened through general criteria. If that is satisfied, the local governments are classified as eligible DAK recipients. Otherwise, they will be screened through specific criteria. If they cannot satisfy the specific criteria, they will be screened through the technical criteria based on technical indexes in the relevant sectors (Law No. 32/2004 and Law No. 33/2004). If they still fail to satisfy the technical criteria, they are classified as ineligible. Otherwise, they will be eligible for DAK. Technical data used in screening are provided by relevant government agencies (e.g. line ministries) and the Central Bureau of Statistics. Table 2.1 Specific Allocation Funds (Dana Alokasi Khusus/DAK) sectors Sector Reforestation X X X X X Education X X X X X X X X Health X X X X X X X X Roads X X X X X X X X Irrigation X X X X X X X X Governmental infrastructure X X X X X X X X Fisheries X X X X X X X Water X X X X X X Sanitation X Agriculture X X X X X X Environment X X X X X Demography X X X Forestry X X X Village infrastructure X X Trade X X Source: Directorate General of Fiscal Balance, Ministry of Finance. 28

45 However, the sectors may change every year, depending on central government policies and priorities (Ministry of Finance 2010). After eligible recipients are determined, the amount of DAK allocation for each local government in each sector are computed and determined by The Ministry of Finance. The computation is formulabased, using regional and technical weighting, indexes and weighted average (Ministry of Finance 2010). In recent years there has been a growing trend in the number of areas considered as national priorities, and as a result there has been a significant and steady increase in both the size and scope of DAK funding over the years since decentralization occurred. Initially DAK was used to fund only a single activity reforestation, but in 2003 five sectors were determined as national priorities with 354 local government recipients, and five new core DAK activities being added in In almost every subsequent year the list of these activities has been extended, so that by 2010 some 14 sectors were included as national priority, totalling 14 DAK activities with 490 local government recipients (see Table 2.1). It is worth noting that in 2010, the number of DAK recipient regions of 490 (districts and provinces), when compared with the total local governments in Indonesia (530) had reached 93%. So the magnitude of receiving DAK allocation, in any sense, does not even reflect the elements of 'specificness' mandated by the law. Meanwhile, DAK share in Fiscal Equalisation Funds (Dana Perimbangan) has been increasing from 1% in 2001 to 7% in Nominally the DAK in 2013 was more than 28 times larger than in Therefore, even though the total DAK allocation remains the smallest of the three fiscal balance components in percentage terms, its share has been increasing (as shown in Figure 2.1). 29

46 2.3.4 Role of the President in Allocating Transfers The central government fiscal transfer is part of the national budget (Anggaran Penerimaan dan Belanja Negara/APBN/State Budget Revenue and Expenditure). In order to understand the president s influence on intergovernmental transfers in Indonesia it is necessary to understand the central government budgeting process (APBN). The Government of Indonesia s (GoI) budget is determined yearly. It commences with a budget draft assembled by the executive (president and his/her cabinet) for the following budget year. The president leads the ministry in the budget formulation process (Blöndal et al. 2009). The draft includes an estimation of future revenue and sectoral spending allocation. Once completed, the draft is then sent to parliament (Dewan Perwakilan Rakyat/DPR/House of Representatives) for review and consultation. The consultations occur between executive and legislative, which mostly centres on sectoral allocation. If consensus is achieved between legislative and executive the budget draft is then enacted as a Law by the DPR. If consensus is not achieved, the Law mandates for the executive to use the previous year s budget (Blöndal et al. 2009). Due to the ministry s influence, the president is able to influence the regional distribution of central government transfers, especially in DAU and DAK allocation, where the allocation mechanism is entirely a top-down process (Savitri 2013). Local governments do not have the capacity to influence or challenge the DAU and DAK allocation. Their involvement at best is providing the updated data needed by central government for calculating the formula. While the law states the fixed variables to be used in the DAU formula, the weight of each variable can differ subject to the central government s (i.e. the president) proposal and national parliament s (DPR) approval. 30

47 Indeed, the DAU variables and weighting had been changed more than six times during (Shah et al. 2012). The DAK is allocated in the APBN according to the national priority in the Presidential Decree on government work plans for the current year. This means the sectors may change every year, depending on central government policies and priorities. In fact, the numbers of sectors and districts recipients of DAK have been growing each year (see Table 2.1), while the allocation formulas and variables in each sector also has changed almost every year. This is because each of DAK technical criteria and its allocation formula are determined by the relevant technical ministry in each specific sector, in coordination with the Ministry of Home Affairs, the Ministry of Finance, and the Ministry of National Development Planning (Usman et al. 2008). Meanwhile, in DBH, the central government has less influence because the revenue estimate is conducted together with local governments; also there is a reconciliation channel where local governments can dispute the amount of DBH allocation (Savitri 2013). 2.4 Presidential Election in Indonesia Ever since independence in 1945, except for the nine years of so called Parliamentary Democracy during the 1950s, the president was constitutionally deemed to be elected by the People s Consultative Assembly (Majelis Permusyawaratan Rakyat/ MPR), the legislative body heading the state structure. Suharto, who gained power in 1966, succeeded in consolidating the authoritarian regime originally established by Sukarno through his control, with military backing, of the ruling party, GOLKAR. In so doing, Suharto was elected as the president seven consecutive times and maintained his rule for 32 years (Kawamura 2003). 31

48 In October 1999, prior to the first ever free legislative election in 1999, the President and Vice President were elected by the MPR. Abdurrahman Wahid and Megawati Sukarnoputri as the president and the vice-president respectively, were supposed to rule for the next five-year period ( ). However, in July 2001 President Abdurrahman Wahid was impeached by the MPR because of conflicts between the President and the Parliament. Following the impeachment, the MPR elevated Megawati to the presidency to complete the five-year term (Kawamura 2013). The impeachment of President Abdurrahman Wahid motivated the MPR to include new articles on the presidential election system in the amendments of Constitution of Indonesia in Under the new amendment, the president should be elected not by the MPR but directly by the people, and that the impeachment of the president should require the consent of the judiciary so as to prevent undue influence on the position of the president from partisan interests in parliament and to stabilise the position of the president (Ufen 2006). In the new election system, the president s election is held in the same year as the general election for the House of People s Representatives (Dewan Perwakilan Rakyat/DPR) and the House of Local Representatives (Dewan Perwakilan Daerah/DPD), which is once every five years. Candidates for president and vice-president have to run as a pair and be backed by a political party (or a coalition of political parties) that has a certain share of parliamentary seats. In the 2004 presidential election, the Presidential Election Law provided that only parties (or coalitions of parties) with more than 20% of the vote in the parliamentary election or 15% of the parliamentary seats could put up a candidate. In the 2009 presidential election, the conditions for putting up candidates were raised to more than 25% of the vote or 20% of parliamentary seats. An independent candidate is not allowed 32

49 to run for election. The new amendment also regulates that a President-Vice President candidate pair is elected into office after receiving more than 50 percent of the vote nationally with at least 20 percent of the vote in more than half of the provinces of Indonesia. If no pair receives the amount of votes required, the election will continue into the second round with the pairs receiving the highest and second highest number of votes (Kawamura 2013). Table 2.2 The 2004 presidential election results Candidate First round Votes % Second round Votes % Susilo Bambang Yudhoyono (SBY) 39,838, ,266, Megawati Sukarnoputri 31,569, ,990, Wiranto 26,286, Amien Rais 17,392, Hamzah Haz 3,569, Total 118,656, ,257, Source: General Election Commission (KPU). In the 2004 presidential election, four out of the five pairs of candidates for president and vice-president were put up by coalitions of parties. Since so many parties win seats and there is no majority party in the parliament, parties have to build coalitions to win the presidential election. The winning GOLKAR party put up Wiranto paired with Salahuddin Wahid in a coalition with the PKB party and four minor parties. The PDIP, the second largest party, ran with Megawati and Hasyim Muzadi in PDS, a minor Christian party. The new PD party in cooperation with two minor parties ran Susilo Bambang Yudhoyono (SBY) and Yusuf Kalla as their presidential and vice-presidential candidates. The fourth coalition grouped the PKS party with the PAN party and seven other parties in support of Amien Rais and Siswono Yudo Husodo. Only the PPP party, which had failed in coalitional negotiations, ran its own candidate Hamzah Haz and Agum 33

50 Gumelar. The election runs in two rounds. The presidential election in 2004 was won by SBY after defeating Megawati by more than 60% in the second election round (see Table 2.2 for details of the election results) (Kawamura 2013). Table 2.3 The 2009 presidential election results Candidate Votes % Megawati Sukarnoputri 32,548, Susilo Bambang Yudhoyono (SBY) 73,874, Yusuf Kalla 15,081, Total 121,504, Source: General Election Commission (KPU). In the 2009 presidential election, the number of candidates decreased to three pairs because the conditions of putting up presidential candidates were tighter. Still all presidential candidates were put up by coalitions of political parties. The PD party, the winner of legislative election which became the ruling party in coalition with four Islamic parties ran by SBY, the incumbent president, paired with Budiono. The second coalition put the GOLKAR party with the new HANURA party, ran Wiranto with Yusuf Kalla, the incumbent vice-president, as their president and vice president candidate respectively. The third coalition was formed by the PDIP with the new GERINDRA party, of which the presidential and vice-presidential candidates were Megawati Sukarnoputri and Prabowo Subianto. This was the second election in which Indonesians elected their President and Vice President directly. The elections returned a president and vice president for the periods. In the election, President SBY won more than 60% of the vote in the first round, which enabled him to secure re-election without a run-off (see Table 2.3 for details of the results) (Kawamura 2013). 34

51 2.5 Data and Methodology Data The data set for central government transfers and socioeconomic characteristics for all Indonesian districts for the period is acquired from the Indonesia Database for Policy and Economic Research (INDODAPOER) of the World Bank. However, since the direct presidential elections only commenced in 2004, in the core analysis I only use the transfer data allocation for the period The transfer data is analysed in natural logarithm of per capita at constant price of The 2004 transfer is excluded from the analysis since it does not represent the effect of the direct election; rather it is allocated by the previous parliament-elected president. The number of districts in the sample increased from 336 in 2000 to 497 in 2013, due to district splitting within the period of observation. Meanwhile, the presidential election results of each district for the 2004 and 2009 elections are taken from General Election Commission (KPU). For completeness, I use dataset of all the districts available in the INDODAPOER, including the proliferated district. Table 2.4 Descriptive statistics of central government transfer (in thousands of IDR at 2000 price) Variable N Mean Std. Dev. DBH in Total 6,093 39,400,000 90,400,000 DBH from Tax 5,216 23,200,000 34,700,000 DBH from Natural Resources 5,090 23,500,000 82,200,000 DAU 5, ,000,000 97,800,000 DAK in Total 5,143 16,300,000 13,900,000 PAD 5,226 22,200,000 41,400,000 DBH per capita 5, DBH from Tax per capita 5, DBH from Resources per capita 5, DAU per capita 5, DAK per capita in Total 5, PAD per capita 5, Source: The Indonesia Database for Policy and Economic Research (INDODAPOER) of World Bank. 35

52 Table 2.4 presents the summary statistics (N = number of observation/districts; mean; and Std. Dev. = standard deviation) of the three types of transfers (DBH, DAU, and DAK) and also the districts own-source revenues (PAD) of the entire districts within the period of Clearly, DAU, the block grant, has the majority share of the resources transferred to municipalities, followed by DBH, the revenue sharing funds and DAK, the specific allocation funds. By comparing the transfers mean and its standard deviation, it is obvious that DAU has the least variance. To a certain extent, this illustrates that the transfer is relatively equally distributed across districts compared to the other forms of transfer. It is also interesting to notice how the districts own-source revenues (PAD) are larger than the specific allocation funds (DAK) in total, but in per capita terms PAD has the lowest figure. This illustrates how district governments are highly dependent on the central government in financing their activities. Table 2.5 Descriptive statistics of subcomponents of DAK (in thousands of IDR at 2000 price) Variable N Mean Std. Dev. Agriculture 1,827 3,139,924 1,408,345 Education 2,686 10,025,250 9,211,919 Environment 1, , ,004 Forest 200 1,000, ,300 Fishery 2,183 2,151,505 1,311,820 Government infrastructure 605 3,723,567 2,073,541 Health 2,659 5,623,787 4,094,492 Infrastructure 2,740 8,945,286 5,665,054 Water 2,003 2,072,061 1,077,875 Irrigation 2,122 2,181,423 1,230,792 Road 2,679 5,871,880 3,601,001 Demography , ,418 Trade , ,191 Village 110 1,727, ,904 Source: The Indonesia Database for Policy and Economic Research (INDODAPOER) of the World Bank. 36

53 Table 2.5 presents the summary statistics (N = number of observation/districts; mean; and Std. Dev. = standard deviation) of the DAK subcomponents of the entire districts within the period of I present 13 out of 14 components described in Table 1 because Sanitation DAK just exist in 2010, as a breakdown from Water DAK. In this study we regroup Sanitation DAK to Water DAK for the sake of analysis. It is obvious from Table 5 that the largest DAK per district is in the education sector, followed by Road and Health sectors, respectively Empirical strategy The aim of the empirical analysis is to identify the influence of presidential election results on government transfers received by the district governments. To do so, I compare the outcomes of the last presidential election in term of districts where the current President won, to districts where the current President lost. The main empirical challenge is that these two types of districts may be systematically different, since in districts where the incumbent president won may have more poverty, poorer education, fewer informed voters or better bureaucracies. This (potentially) unobserved heterogeneity can obscure the effect of presidential election result. To address these identification concerns, and claim a causal effect, in this study I use a sharp regression discontinuity design (RDD). This design exploits the variation in vote margin of the current president in power in the last election across districts. In particular, it compares the central government transfer received by the local government of districts in which the current president barely wins versus districts in which the current president barely loses the last presidential election. In this setup, the treatment is that the current president won the last election. 37

54 The assignment of the treatment depends on the electoral results in the last presidential election at district level. Let us define the assignment variable for district i as X i = v i,m max {v i,j m }, where v i,m is the vote share obtained by the current president in the last election in district i, and v i,j m is the largest vote share obtained by any other presidential candidate in that election in district i. X i represents the current president victory/loss margin in the previous presidential election. It measures by how much the current president won (or lost) in the previous election in district i. Defined in this way, a district is treated (i.e. the current president won the election) only if X i > 0. The RDD estimator is defined as: τ RD = lim x 0 E[Y i X i = 0] lim x 0 E[Y i X i = 0] (2.1) where Y i is the central government transfer received by district i. τ RD can also be interpreted as the average treatment effect at the discontinuity point τ RD = E[Y i (1) Y i (0) X i = 0], where Y i (1) and Y i (0) represent the potential central government transfer received by district i with and without treatment. The basic idea behind this approach is that districts near the cut-off (X i = 0) are most likely indistinguishable, so if there is a discontinuity at cut-off (τ RD ) it can be attributed solely to the treatment effect. Following Imbens and Lemieux (2008), we estimate τ RD using a local linear regression within a MSE-optimal bandwidth and robust inference methods as follow: Y i,j = α + τ RD D + β 1 (X i c) + β 2 D(X i c) + ε i,j (2.2) D = { 1 if X i > c 0 if X i < c c h X i c + h Where Y i,j is the outcome variable, which in the context of this study refers to natural logarithm of per capita transfer in district i at year j, whereas c is the treatment 38

55 cut-off (X i = 0), while D is a binary variable equal to one if X i > c, and h is the bandwidth of data used. This is non-parametric method that basically fits linear regression to observations within certain bandwidth on either side of the discontinuity point with different slopes and intercepts fit data on either side of the cut-off. The regression uses weights that are smaller for observations that are further from the discontinuity point. The observations are weighted using a triangle kernel. An important issue in the estimation of regression discontinuity design (RDD) is the sensitivity of local linear regressions to the choice of bandwidth (Imbens and Lemieux 2008; Calonico et al. 2014b). To address this concern, I use the estimator recently proposed by Calonico et al. (2014b) to choose the bandwidth (h), as this has better properties than the estimator proposed by Imbens and Kalyanaraman (2012). Calonico et al. (2014b) estimator is robust bias-corrected with data-driven bandwidth selectors. I implement this estimator using the STATA package rdrobust. More details are available at Calonico et al. (2014a). Another implementation of RDD is the parametric method, using cubic or higher order polynomials, often using statistical information criteria or cross-validation to choose the degree of the polynomial. However, I do not apply the parametric method in this work because as Gelman and Imbens (2014) point out, this method produces estimates that are highly sensitive to the degree of the polynomial. Also, the high order global polynomial regressions conventional inference for treatment effect in the regression discontinuity setting can be misleading, in the sense that confidence intervals are too narrow, which leads to over-rejection of the estimates. The over-rejection suggests that the global polynomial approximation is not accurate enough to allow the researcher to ignore that bias in the estimates of the treatment effects. The local linear and quadratic regression work substantially better in that the rejection rates are close to nominal levels, 39

56 and the standard errors are substantially smaller than those based on the high order global polynomial approximations. Regardless the estimation methods, the RDD estimates are invalid if districts can precisely manipulate close elections to their advantage, in which case observations close to the cut-off (X i = 0) may not in fact be comparable. The number of treated observations just above the cut-off should be approximately similar to the number of control observations below it. Also, if some districts were able to selectively manipulate the vote margin near the cut-off, it would not be possible to identify the causal effect of the treatment from the effect of those characteristics that allow districts to manipulate close elections. Where there is an unexplained abrupt change in the number of observations right at the cut-off, the RDD applications will tend to be less credible. This can be confirmed by observing in Figure 2.2, a histogram of the margin of victory/loss at district level of the current president in power on presidential elections held in 2004 and We observe that there is no significant jump in density between districts where the president in power narrowly loses and districts where the president in power narrowly wins. The number of observations above and below the cut-off (X i = 0) is arguably very similar. It appears there is no significant jump in the number of municipalities where the current president narrowly loses and narrowly wins. In a more rigorous statistical sense, the assumption is tested more formally by a density test to examine if, in a local neighborhood near the cut-off, the number of observations below the cut-off is considerably different from the number of observations above it. The density test was first introduced by McCrary (2008). However, Cattaneo, Jansson, and Ma (2015a) propose a local polynomial density estimator that does not require pre-binning of the data and leads to size and power improvements relative to McCrary (2008). I perform the density test with the command rddensity in Stata (Cattaneo, Jansson, and Ma 2015b) to evaluate whether there is discontinuity in the 40

57 distribution of the assignment variable (i.e. the current president vote margin) around the cut-off (X i = 0). The null hypothesis is that the density of the running variable is continuous at the cut-off, and its implementation requires the estimation of the density of observations near the cut-off, separately for observations above and below the cut-off. Such a discontinuity would be indicative of manipulation of the assignment variable (McCrary, 2008). The resulting estimated p-value is , meaning there is no evidence that the density is discontinuous at the cut-off. Therefore, a positive discontinuity relating transfers to the margin of victory/loss of the president at cut-off (X i = 0) must be driven by a causal effect of the president rewarding voters in those districts, given that voters or local districts government cannot precisely control the elections outcome. Table 2.6 reports the mean differences between districts where the president loses or wins, of several variables representing district characteristics in 2004, which is a year before the directly elected president running the administration and governing the assignment of equalisation funds. These variables represent district differences in fiscal, geographic, demographic, health, education and economy. In each variable, I present results for the full sample and for a window with margin 10% and 20% above and below the cut-off. The descriptive statistics of the variables appears in the Table 2.6 is in the Appendix, Table 2.A1. The columns on Table 2.6 show that even in full sample, districts in which the president is losing have some similarity with districts in which the president is winning, especially in the fiscal and health aspects. The difference is losing districts having larger rural population, less working age population, less unemployment, higher poverty and also less in education and economy characteristics than winning districts. However, when the sample is restricted to around the cut-off, differences between the two groups become smaller and mostly statistically insignificant. This suggests that districts near the cut-off 41

58 where the president barely loses (the control areas) have, on average, the same observable characteristics as districts near the cut-off where the president barely wins (the treatment areas). This evidence supports the strategy of using the RDD technique. Panel A. Vote margin in full sample (in %) Frequency Panel B. Vote margin 10% below and above cut-off (in %) Frequency Figure 2.2 Electoral margin of the current president in power (presidential election 2004 and 2009) 42

59 Table 2.6 Pre-direct election mean-differences between districts where the president losing and winning Fiscal per capita Geography Demography Health Education Economy Full sample Margin 10% from cut-off Margin 20% from cut-off PAD (in IDR) , , (10,396.85) (11,289.76) (16,261.63) DBH (in IDR) 33, , , (47,313.28) (61,216.21) (35,233.69) DAU (in IDR) -15, , , (52,228.68) (99,711.67) (69,774.26) DAK (in IDR) -4, , , (8,607.35) (20,239.25) (11,529.22) Area (in KM 2 ) 1, , , (1,272.99) (1,752.10) (1,570.76) % Rural population 12.71*** (3.47) (5.73) (4.31) Population (in number of people) -87, , , (66,576.37) (118,224.00) (93,341.59) % Population *** (0.47) (0.85) (0.59) Unemployment rate (%) -2.53*** (0.48) (0.87) (0.64) Poverty rate (%) 3.27** (1.29) (2.09) (1.69) % Birth attended by health worker -4.75* (2.54) (4.43) (3.53) Immunization coverage (%) (1.12) (2.45) (1.59) Morbidity (%) (1.03) (2.04) (1.50) Net Enrolment Rate SMP (%) -6.17*** ** (1.59) (2.88) (2.34) Net Enrolment Rate SMA (%) -5.26** (2.10) (4.41) (3.07) Literacy rate (%) -2.73** (1.10) (1.62) (1.85) GDP (in billion IDR 2000) -2,288.69** ,642.96* (911.69) (954.22) (958.69) Household per capita income (in IDR) -34,725.43*** 2, , (9,306.98) (12,271.83) (9,814.20) Access to electricity (%) *** ** (2.60) (4.94) (3.90) Notes: SMP=Junior High School, SMA=Senior High School. * denotes significant at 10%, **significant at 5% and *** significant at 1%. Standard errors in brackets. 43

60 2.6 Empirical results Graphical evidence The main aspects of the RD design can be summarised in an easy-tointerpret figure, which shows how an estimated regression function behaves for control and treated units relative to some summary of the actual data. This common RD plot gives an idea of overall fit while also exhibiting graphically the sharp RD estimate. In the context of this research control units refers to districts where the president loses the election, while treated units refer to districts where the president wins the election. The RD plots use two main ingredients. First, two smooth global polynomial regression curve estimates for control and treatment units separately to flexibly approximate the population conditional mean functions for control and treated units of the running variable. Second, sample means are constructed over non-overlapping regions of the support of the running variable for control and treatment units separately. These sample means provide an approximation of the population regression functions, but they also help visualise the dispersion of the data and to capture the behaviour of the cloud of points. I demonstrate this graphical approach in Figure 2.3 to clearly demonstrate the main point. Figure 2.3 explores the discontinuity at 0 percent when the president barely wins over the second strongest candidate. Panels A to C show the percentage points of margin of victory of the president in two elections (2004 and 2009) on the horizontal axis and the natural logarithm of per capita grants allocated to each district on the vertical axis. Each panel in Figure 2.3 presents a different type of transfer. Panel A presents total revenue sharing (DBH), Panel B presents general allocation grant (DAU) and Panel C presents special allocation grant (DAK). Each dot in the panel corresponds to the average transfer that follows the election, grouped by margin of victory intervals. 44

61 Panel A. Total Revenue Sharing (DBH) Per capita Log shared revenue funds per capita Electoral margin of SBY in the presidential election 2004 and 2009 Panel B. General Allocation Grant (DAU) Per capita Log general allocation funds per capita Electoral margin of SBY in the presidential election 2004 and 2009 Panel C. Special Allocation Grant (DAK) Per capita 10 Log special allocation funds per capita Electoral margin of SBY in the presidential election 2004 and 2009 Figure 2.3 Central Government Transfer to District Governments,

62 For observations with a positive vote margin those to the right of the vertical line in the middle of the graph the president won the election. For observations with a negative vote margin, the president loses the election. Observations closer to the vertical line indicate more narrowly won elections. To help organise the data visually, two fourthorder local polynomial best-fit lines are shown, fit to the data separately on each side of the cut-off. The Figure is constructed using the procedure developed by Calonico, Cattaneo, and Titiunik (2014a). For further details on these approaches, see Calonico, Cattaneo, and Titiunik (2014a). Panel A suggests that there is no discontinuity between districts where the president barely wins and barely loses in terms of total revenue sharing (DBH). Meanwhile in both Panel B and C the smooth-fitted polynomial regression line of DAU and DAK show a positive discontinuity at the cut-off, with a sharper increase in the latter rather than the former. These indicate that in close elections, districts where the president barely wins get considerably more DAU and DAK than very similar districts where the president barely loses. The following sections explore whether the graphical findings are confirmed by more formal RD estimates of the presidential election effect on the central government transfers to district governments. 46

63 2.6.2 Baseline results Table 2.7 presents the baseline regression results. Columns 1 to 5 in Table 2.7 show the results of a formal statistical test of the finding in Figure 2.3. I follow the now standard practice of using a nonparametric regression within a narrow window around the cut-off. In particular, a regression function is estimated separately above and below the cut-off by means of weighted linear regressions, with weights decreasing in the distance of each observation s vote margin from the cut-off. The nonparametric regressions require a window (also called the bandwidth ) to be specified, with observations outside of the bandwidth receiving a zero weight. I use the robust, bias-corrected, local linear regressions and optimal bandwidth selector proposed by Calonico et al. (2014b) that minimises the mean-squared error of the regression. Table 2.7 The presidential election result impact on central government transfer allocation DBH Total (1) DBH Tax (2) DBH Resources (3) DAU (4) DAK Total (5) Estimate ** (0.158) (0.137) (0.343) (0.152) (0.226) p-value Bandwidth N N below cut-off N above cut-off Notes: * denotes significant at 10%, **significant at 5% and *** significant at 1%. Robust standard errors in parenthesis. All estimation uses the robust, bias-corrected, estimator and bandwidth selector proposed by Calonico et al. (2014b). The coefficient estimate for the presidential election result of effect on transfers is consistently positive for all types of transfer. However, only DAK (Special Allocation Grant) is recorded to have a statistically significant impact. As the dependent variable is the log of the per capita grant, the coefficient estimate has the interpretation of the percentage change in per capita transfers due to the effect of the presidential election result. The result in column 5 of Table 2.7 indicates that districts just above the threshold 47

64 received around 44.4 percentage points more DAK than districts just below the threshold. Other than DAK the results suggest that the presidential election result has an insignificant effect on central government transfer allocations. Since there are 13 sub-categories of DAK, it is interesting to examine whether the influence of the presidential election may differ across the components of DAK because these components do not have the same characteristics. The 13 components are: Agriculture, Education, Environment, Forestry, Fishery, Government, Health, Infrastructure-Water, Infrastructure-Irrigation, Infrastructure-Road, Demography, Trade, and Village. Table 2.8 The presidential election result impact on sub-components of DAK Estimate Robust Standard Error Bandwidth N N below cut-off N above cut-off Agriculture 0.919*** Education 0.895*** Environment 1.140*** Forestry Fishery 1.052*** Government infra ** Health 0.933*** Infrastructure 0.982*** Water 0.995*** Irrigation 0.845*** Road 1.035*** Demography 0.559* Trade 1.748*** Village Notes: * denotes significant at 10%, **significant at 5% and *** significant at 1%. All estimation uses the robust, bias-corrected, estimator and bandwidth selector proposed by Calonico et al. (2014b). Table 2.8 presents estimates of the 13 sub-categories of DAK. The table provides evidence that the presidential election result has different effects on the components of per capital real total DAK. The common denominator to all estimations is that the 48

65 estimated effect of presidential election results on grants is always positive and generally very significant, with the exception for DAK-Forestry and DAK-Village, which are not significant. Numerically, the coefficients imply that the impact of presidential election results on transfers near the cut-off vary, from the lowest at 53.3% (DAK-Government) to the highest at 174.8% (DAK-Trade). It is worth noting that there are three other DAK components where the presidential election impact is more than 100%, other than DAK- Trade; these are DAK-Environment (114%), DAK-Fishery (105.2%), and DAK- Infrastructure-Road (103.5%) Does the electoral cycle matter? There are two strands of literature which explain budget manipulation: political business cycles literature and targeted distribution literature. The discussion in the previous section is motivated by targeted distribution literature, in which I have tested whether central government transfers have been manipulated to reward districts where the President won in the last election. In this section I will extend the analysis to ascertain whether the transfer manipulation is sensitive to election timing, as suggested by political business cycle literature. The motivation for this is because there has been a bulk of empirical evidence supporting the notion that political budget cycles (PBCs) may be found in young democracies, like Indonesia, where voters reward pre-election increases in spending (Brender and Drazen 2005; Shi and Svensson 2006). The hypothesis is that transfer manipulation is considerably greater the closer the election becomes, because at this time the central government is not only motivated to reward their voters for the past election, but is also attempting to ensure victory in the next election. To test this proposition I run RD design to central government transfer data and presidential election result data for the last two years of the president s term of office. 49

66 Panel A. Total Revenue Sharing (DBH) Per capita Log shared revenue funds per capita Electoral margin of SBY in the presidential election 2004 and 2009 Panel B. General Allocation Grant (DAU) Per capita Electoral margin of SBY in the presidential election 2004 and 2009 Panel C. Special Allocation Grant (DAK) Per capita Electoral margin of SBY in the presidential election 2004 and 2009 Figure 2.4 Central Government Transfer The last Two Years of Tenure ( and 2013) 50

67 I start the analysis using a graphical approach in Figure 2.4. The figure shows the relationship between the vote margin and the natural logarithm of per capita transfer for the last two years of the president s term of office (2008, 2009 and 2013). Unfortunately, the transfer data for 2014 was not available at the time the analysis was conducted. The presidential election effect on the central government per capita transfer is given by the discontinuity at the cut-off point. All panels in Figure 2.4 show that there is a substantial jump in central government per capita transfers for the narrow win districts. The formal statistical test of the finding in Figure 2.4 and the estimated size of the discontinuity are given in Table 2.9. Table 2.9 The presidential election impact on transfer allocation, last two-years in office DBH Total DBH Tax DBH Resources DAU DAK Total Estimate 0.875** ** 0.730** 1.330** (0.419) (0.335) (0.710) (0.350) (0.552) p-value Bandwidth N N below cut-off N above cut-off Notes: * denotes significant at 10%, **significant at 5% and *** significant at 1%. Robust standard errors in parenthesis. All estimation uses the robust, bias-corrected, estimator and bandwidth selector proposed by Calonico et al. (2014b). The coefficient estimate for the effect of the presidential election result is consistently positive and significant for almost all types of transfer, except DBH-Tax. This is opposite to the finding in the previous section where only Total DAK was significant. Transfers to narrow win districts in the last two years of the president were higher by 73% to 150% than the narrowly lost districts, depending on the types of transfer. The most significant impact goes to DBH resources (150%), followed by total DAK (133%), Total DBH (87.5%) and DAU (73%). As a benchmark, Figure 2.A1 and Table 2.A2 in the Appendix show graphical evidence and robust regression discontinuity 51

68 estimation for the first two years of the president s term in office. The figure and table clearly demonstrate that there is no evidence of discontinuity of central government transfer in the first two years of elected president tenure at the cut-off. Table 2.10 The election impact on allocation of DAK subcomponents, last two-years in office Robust N below N above Estimate Standard Bandwidth N cut-off cut-off Error Agriculture 0.929** Education 0.807** Environment 1.057** Forestry Fishery 1.206*** Government infra * Health 0.990** Infrastructure 1.080** Water 0.964** Irrigation 0.838** Road 1.073** Demography 0.559* Trade 1.748*** Village Notes: * denotes significant at 10%, **significant at 5% and *** significant at 1%. All estimation uses the robust, bias-corrected, estimator and bandwidth selector proposed by Calonico et al. (2014b). I explore this finding by estimating the effect of presidential elections on the subcomponent of DAK for the last two years of the presidential term in Table The estimation results suggest that indeed almost all DAK sub-components are positive and significant, except for DAK-Forestry and DAK-Village. The highest presidential effect is on DAK-Trade with transfer per capita 174.8% larger for narrowly won districts compared to narrowly lost ones. The lowest effect is on DAK-Demography in which narrowly won districts only got 55.9% higher than narrowly lost districts. Other DAKsectors where the presidential effect is larger than 100% are DAK-Fishery (120.6%), DAK-Infrastructure (108%), DAK-Road (107.3%) and DAK-environment (105.7%). 52

69 As a comparison, I re-produce Table 2.10 using the sub-component of DAK data for the first two years of the president s term. Table 2.A3 in the Appendix clearly demonstrates that there is some evidence of discontinuity of sub-component of DAK in the first two years of elected president tenure at the cut-off. These results are somewhat expected regarding the fact that DAK is the only central government transfer that is significantly influenced by president election results in the whole sample period ( ) as presented in Table 2.7. The treatment effect of election results on the DAKcomponents for the first two-year tenure of the president in narrowly won districts are as follows (from highest to lowest): DAK-Infrastructure-Water (104.5%), DAK- Infrastructure-Road (91.8%), DAK-Education (88%), DAK-Environment (86.9%), DAK-Health (80.4%), DAK-Fishery (77.2%) and DAK-Total Infrastructure (74.1%). 53

70 2.7 Validity & Robustness Checks This section presents the results of a series of RDD validity tests to check the robustness of findings in the previous sections. One important test for the validity of my RDD design is to examine whether the covariates of central government transfer do not exhibit any discontinuity in relation to the margin of victory. If there were such a discontinuity, it would be unclear whether any treatment effects were due to the presidential election results or due to some other underlying variable (Lee and Lemieux 2010). I choose 16 variables that, in principle, should be correlated with transfer and therefore should be considered as covariates of central government transfer to district governments. These variables are chosen to represent districts fiscal capacity, geographic-demographic condition, health-education status and level of economic development, based on data available in this study. They are: local governments own revenue per capita (PAD), area, population, share of population years old, share of rural population, unemployment rate, poverty rate, birth attended by health worker, immunisation coverage for children under 5 years old, morbidity, net enrolment rate at junior high-school (NER SMP), senior high-school net enrolment rate (NER SMA), literacy rate, GDP, household per capita income and access to electricity. I argue that these indicators can capture development disparity across districts which the central government tries to reduce by transfer mechanisms. I conduct the test by estimating local linear estimation within an MSE-optimal bandwidth and robust inference methods, proposed by Calonico et al. (2014b), to estimate the treatment effect. In this framework each covariate is a dependent variable, and the explanatory variables are vote margin of victory in presidential elections. 54

71 Table 2.11 The central government transfer s covariates discontinuity test, all samples Estimate Robust Standard Error Bandwidth N N below cut-off N above cut-off PAD per capita , Area (in KM 2 ) 1, Population -6, , , % Population , % Rural population , Unemployment rate (%) Poverty rate (%) , % Birth attended by health worker Immunization coverage for children under 5 years old (%) , , Morbidity (%) , NER SMP (%) , NER SMA (%) , Literacy rate (%) , GDP (in billion IDR 2000) Household per capita income (in IDR) -42,717 27, , Acces to electricity (%) , Notes: * denotes significant at 10%, **significant at 5% and *** significant at 1%. All estimation uses the robust, bias-corrected, estimator and bandwidth selector proposed by Calonico et al. (2014b). A detailed description of descriptive statistics of the variables appears in the Appendix, Table A4. As reported in Table 2.11, the null hypothesis of zero discontinuity in all covariates in relation to the margin of victory cannot be rejected. Therefore, the conclusion is that there is no statistical evidence of discontinuity in the covariates. Another inference from this result is that treated units are similar to control units and not affected by the treatment, because districts just above and just below the cut-off are similar in all those characteristics (i.e. covariates). I also show graphical evidence for six selected covariates in Figure 2.5, where the covariates do not change abruptly around the cut-off. These results give credibility to the design. 55

72 Panel A. Local Owned Revenue (PAD) Per capita Panel B. Household Per capita Income Electoral margin of SBY in the presidential election 2004 and 2009 Panel C. Poverty Rate Panel D. Household Access to Electricity Poverty rate (in %) Household Access to Electricity (in %) Log household per capita expenditure Electoral margin of SBY in the presidential election 2004 and Electoral margin of SBY in the presidential election 2004 and Electoral margin of SBY in the presidential election 2004 and 2009 Panel E. Literacy Rate Panel F. Immunization Coverage, Child<5yr Immunization Coverage for Children under 5 yr (in %) Electoral margin of SBY in the presidential election 2004 and Electoral margin of SBY in the presidential election 2004 and 2009 Figure 2.5 Selected Central Government Transfer Covariates

73 Another important falsification test involves examining validity assumption: whether the outcome variable (i.e. central government transfer) before the treatment (i.e. direct presidential election) was smooth across the cut off. If a smooth function is observed before the program takes place, it is plausible that the jump after the treatment is due to the direct presidential election results. Table 2.12 Effect of current presidential election outcome on lag five years of transfer DBH Total DBH Tax DBH Resources DAU DAK Total Estimate Standard error (0.178) (0.175) (0.441) (0.145) (0.343) p-value Bandwidth N N below cut-off N above cut-off Notes: * denotes significant at 10%, **significant at 5% and *** significant at 1%. Robust standard errors in parenthesis. All estimation uses the robust, bias-corrected, estimator and bandwidth selector proposed by Calonico et al. (2014b). Therefore, I re-estimate the treatment effect of presidential election results on pseudo-outcomes of central government transfer. I generate these pseudo-outcomes by creating five year lags of all transfer data to ensure that I relate the current presidential election victory margin of a district to the variables relating to transfers received by the districts in the last government period. Since the transfer could not have been affected by the treatment, the expectation is that the null hypothesis of no treatment effect will not be rejected. 57

74 Panel A. Total Revenue Sharing (DBH) Per capita Log shared revenue funds per capita Electoral margin of SBY in the presidential election 2004 and 2009 Panel B. General Allocation Grant (DAU) Per capita Electoral margin of SBY in the presidential election 2004 and 2009 Panel C. Special Allocation Grant (DAK) Per capita Electoral margin of SBY in the presidential election 2004 and 2009 Figure 2.6. Lagged Five Years of Central Government Transfer

75 Table 2.12 shows the treatment effects (i.e. the presidential election) on five year lags of DAU, DBH and DAK. Note that, consistent with the validity assumption, there is no significant effect of the treatment on any of these pseudo-outcomes. Figure 2.6 provides graphical evidence of these results. It visually demonstrates that there are no discontinuities around cut-off. Table 2.13 Effect of current presidential election outcome on lag five years of subcomponents of DAK Robust N below N above Estimate Standard Bandwidth N cut-off cut-off Error Agriculture ** Education Environment ** Forestry Fishery Government infra Health Infrastructure Water ** Irrigation Road Demography Notes: * denotes significant at 10%, **significant at 5% and *** significant at 1%. All estimation uses the robust, bias-corrected, estimator and bandwidth selector proposed by Calonico et al. (2014b). DAK- Demography and DAK-Village cannot be estimated because the cut-off lies outside the range of vote margin. Meanwhile, Table 2.13 demonstrates the treatment effects on the pseudooutcomes (lag five years) of DAK subcomponents. In contrast to Table 2.12, I do find a discontinuity in three of the sub-categories of special allocation grant (DAK): DAK- Agriculture, DAK-Environment, and DAK-Infrastructure-Water. This finding weakens credibility of treatment effects estimation of these three DAK sub-components in Table 2.8. However, in general, Table 2.12 and 2.13 results give support to the validity of the RDD applied to overall sample of central government transfer ( ) as estimated in Table 2.7 and

76 Likewise, I run the same test for pseudo-outcomes on the president s final two years in office to check the robustness of the effect of electoral cycles on central government transfer allocation estimated in Table 2.9 and Table 2.A5 and 2.A6 and Figure 2.A2 in the Appendix present the results of validity and robustness test of the political business cycle finding. I find no significant discontinuities in the lag five years of aggregate transfer variables (DAU, DAK and DBH) in the president s final two years in office (see Table 2.A5 and Figure 2.A2 in Appendix). However, I do find a discontinuity in one of sub-categories of special allocation grant: DAK-Demography (see Table 2.A6 in Appendix). Taken together, all these results, to certain extent, yield support to the validity of the RDD assumption in the estimation of political business cycle of central government transfers, since control (i.e. districts where the president barely loses) and treatment districts (i.e. districts where the president barely wins) exhibit, on average, the same levels of pre-direct-election transfer. 2.8 Discussion Tables 2.7 and 2.9 illustrate the main results of this article. The tables depict a certain pattern of incidence and magnitude of the political effect on transfers. In terms of incidence, total DAK (special allocation grant) is consistently found significant both in the overall period, and in the last two years of presidency. On the other hand, DBH-Tax (revenue sharing in tax) is consistently found to be insignificant regardless of time specification. This seems to indicate that DAK is prone to manipulation while DBH-Tax is fairly immune. I argue that differences in allocation design of the two types of transfers may be contributing to these findings. It is obvious when reviewing the transfer design in section 3 that DBH-Tax design has a contrasting feature compared to DAK. For instance, DBH-Tax s sectors and 60

77 allocation formulas are relatively simple and steady over the period of observation. Moreover, there is a reconciliation procedure between the central government and local governments to confirm the share of DBH. This procedure establishes a transparency between local and central government, and also provides a chance for local governments to dispute transfer allocations that are considered to be manipulated. On the contrary, the DAK process and formula are relatively complex and obscure. Under Law No. 33/2004 the amount of DAK will be determined annually in the APBN (State Budget Revenue and Expenditure). The absence of articles, as well as government regulations specifically regulating DAK, makes it a flexible component of balance funds. On the one hand, this flexibility can be used as an instrument for harmonizing interregional balance funds, and on the other hand, it causes uncertainty for regional governments as to the amount of DAK funding they will receive. The central government appoints DAK recipient sectors in accordance with their priorities as set out in the government work plan for the particular year. Changes to development priorities will be reflected in changes to the sectors and activities to which DAK is channeled. In addition, there is inconsistency in setting the criteria for the use of technical indicators per priority sector s activity. Technical criteria are formulated in the form of technical indices as formulated by the technical ministries of the relevant technical sectors. These indices are used to describe the condition of local governments in each activity/sector to be funded by DAK. Recipient local governments are required to meet a number of technical requirements to be eligible for DAK. However, these requirements are changing from year to year, as do the objectives set out for funded activities. Despite the fact that the Ministry of Finance has made the methodology for calculating DAK allocations open to the public, the technical data used in its formulation is available only to the responsible ministries. Thus, local governments are unable to predict expected DAK funding levels in the following year (Usman et al. 2008). 61

78 Furthermore, no mechanism exists to enable the local governments to complain when DAK allocations do not meet their specific needs or are out of balance with the allocations with other areas. The unpredictability of DAK - both the amount and the time of distribution - gives the impression that the political nuance is still dominant in determining the allocation of DAK. Compared to other types of transfer, DAK is the smallest component of balance funds (see Table 2.4). As a consequence, this transfer is arguably not transparent enough to be noticeable to the public. Yet, even though its aggregate volume may be small, payments to individual actors at district government level can be large. Given these characteristics of the DAK, it is highly reasonable that it is subject to persistent political manipulations. In short, my explanations for the vulnerability of DAK to political manipulation are: complexity and size. The more complex a transfer allocation formula is, the more it is susceptible to manipulation. The complexity can be roughly measured by the number of indicators needed to calculate the transfer allocation. As the number of indicators increase the more difficult it becomes for the public to understand and track the calculation, and so the more vulnerable is that transfer to manipulation. On the other hand, I argue that transfer size is also important in determining the degree of vulnerability of transfers to political manipulation, but in a different direction. The bigger a transfer is, the less it is susceptible to manipulation. This is because bigger transfers attract more public attention. The larger a transfer is, the more noticeable is its political salience, and the less likely it is to become the subject of political manipulation. I argue that complexity and size may also explain the magnitude of political distortion in central government transfer allocation in the last two years of a presidential term, as described in Table 2.9. In the table, the ranks of size of presidential election effect on transfers, from largest to smallest, are as follows (excluding the DBH-Total): DBH- 62

79 resources (150.6%), DAK-Total (133%) and DAU (73%). Whereas, from Table 2.4 and Figure 2.1, it is clear that DAU is the largest component of equalisation funds. Therefore, it is reasonable that the DAU allocations attract noticeable public attention more than any other types of transfer, which then explains why it has the smallest political manipulation next to election year. Meanwhile, the DBH-resource magnitude of manipulation near the election years, that is the largest among other types of transfer, might be explained by the complexity in the calculation of the transfer. Unlike the DAK, the DBH-resource manipulation is not subject to the formula, but it is on the measurement of resource revenue. I argue that the degree of asymmetric information in this transfer is quite severe. This is because both local and central government do not have adequate technical capacity in estimating the revenue generated in these sectors, and heavily rely on estimations provided by the resource extractive company (EITI 2015). The enactment of Presidential Regulation 26/2010, which stipulated information transparency on state and subnational revenues for revenue streams from the extractive industry, implicitly confirm the presence of information asymmetry within the transfer. The regulation also mandates the formation, structure and responsibilities of a Transparency Team. The Transparency Team is tasked to implement transparency in the management of state and local extractive revenues. To perform this responsibility, the Team may seek information, additional data, input, and/or consult with agencies of the central and local government as well as extractive companies. 63

80 Panel A. Treatment effect for overall period Panel B. Treatment effect for the last two years * Panel C. Size of DAK subcomponents Panel D. Complexity of DAK subcomponents Figure 2.7 Treatment effects, size and formula complexity of DAK subcomponents Note: *The blue bar are subcomponents of DAK which are not significant in overall period. 64

81 The complexity and size argument could also explain the differences of magnitude of presidential election effect in allocation of subcomponents of DAK, as shown in Tables 2.8 and I summarized the magnitude of presidential election effect of Tables 2.8 and 2.10, together with their criteria and size in Figure 2.7. For the record, Figure 2.7 only depicts the election (i.e. treatment) effect that is statistically significant and passes the validity test in previous section. In Panels A and B of Figure 2.7, it is clear that the largest magnitude of political distortion is on DAK-Trade. The districts where the current president barely wins receive 174.8% more DAK-Trade compared to districts where the current president barely loses. In Panels C and D, the DAK-Trade per capita ranks as the second lowest, having a per capita value less than IDR 5,000, while at the same time ranks as the second highest in terms of formula complexity by having more than 20 indicators to determine its allocation. This DAK-Trade magnitude of political effect is suitably fits with the complexity and size argument. DAK-Trade is recorded as having a small volume, which makes it less noticeable by the public. In addition, it also has a complex allocation formula which includes more than 20 indicators, making its allocation consistency to formula difficult for anyone to track. Using the same framework, we can also explain the magnitude of other DAK subcomponents in relation to each other s. These include explaining why DAK-Government infrastructure has the lowest magnitude in political manipulation: although it has a quite high complexity in its formula, it is sufficiently high in volume to attract public attention. 65

82 2.9 Conclusion In this chapter, I empirically examine whether intergovernmental transfers in Indonesia have been manipulated to pursue the political interests of the President. This is an interesting question given that in 2004 there was a shift in the electoral system from parliamentary election of the President to direct election of the President by the people. I also examine whether the manipulation (if indeed it exists) is sensitive towards the presidential election timing-schedule. I examine these questions using a data set on Indonesia intergovernmental transfer to districts government over the period of and direct presidential elections results at district level during the electoral years of 2004 and I employ regression discontinuity design (RDD) as my empirical strategy, exploiting the fact that districts where the current president barely wins or barely loses have almost identical characteristics. Therefore, any discontinuity of transfer received by district governments where the current president barely wins can be attributed as causal effect of presidential manipulation of transfer allocation. To briefly summarize my main results, I find that the president has an important impact on the allocation of the transfer to the districts. District governments where the current president has marginally won in the last election receive larger Special Allocation Grant (DAK) by 44.4 percentage points within the overall period of the sample. This finding lends support to the core voter model (Cox and McCubbins 1986). However, no evidence was found of political distortions in the other form of transfers (DBH and DAU) within that period. I also find that the incidence and the size of transfer manipulation becomes much more significant closer to the election. District government areas, where the current president barely won in the last election, receive larger DBH, DAU and DAK in the last two years of the president s term by the size of 133%, 87.5% and 73%, respectively. This 66

83 finding is consistent with opportunistic political business cycles literature (Brender and Drazen 2005). I offer two possible reasons to explain the differences in size and incidence of political manipulation across types of transfers: size and complexity. Large transfers should be less prone to manipulation because bigger transfers should attract more public attention, hence it should be more difficult to manipulate. On the other hand, transfers with a complex allocation formula should be more susceptible to manipulation, since it should be more difficult for the public to understand the formula and track the calculation. Finally, the results of this study adds to empirical evidence that has shown rulebased transfers are not immune to political manipulations (Banful 2011; Caldeira 2012; Litschig 2012). The flexibility to arbitrarily change the sectors/activities and their technical criteria in DAK (Special Allocation Grant) might be an important loop-hole through which the president can manipulate allocation. Nevertheless, I argue that it is possible that formulas can effectively limit the extent to which politics drives intergovernmental transfer allocation in Indonesia, with conditions of simplicity, transparency and accountability of the allocation mechanism, particularly for transfers with low volumes which are more prone to political manipulation. 67

84 Appendix Table 2.A1 Descriptive statistics of districts characteristics in pre-direct election presidency (2004) Variable N Mean Std. Dev. PAD (in million IDR) ,300 40,900 DBH (in million IDR) ,700 82,500 DAU (in million IDR) , ,000 DAK (in million IDR) 307 9,820 16,400 Area (in KM2) 369 5,273 9,862 % Rural population Population (in number of people) , ,785 % Population Unemployment rate (%) Poverty rate (%) % Birth attended by health worker Immunization coverage (%) Morbidity (%) Net Enrolment Rate SMP (%) Net Enrolment Rate SMA (%) Literacy rate (%) GDP (in billion IDR 2000) 455 3,475,074 7,520,231 Household per capita income (in IDR) ,006 72,383 Access to electricity (%)

85 Panel A. Total Revenue Sharing (DBH) Log shared revenue funds per capita Electoral margin of SBY in the presidential election 2004 and 2009 Panel B. General Allocation Grant (DAU) Log general allocation funds per capita Electoral margin of SBY in the presidential election 2004 and 2009 Panel C. Special Allocation Grant (DAK) Log specific allocation funds per capita Electoral margin of SBY in the presidential election 2004 and 2009 Figure 2.A1 Central Government Transfer The First Two Years of Tenure ( and ) 69

86 Table 2.A2 The presidential election outcome impact on transfer allocation, first two-years in office DBH Total DBH Tax DBH Resources DAU DAK Total Estimate (0.232) (0.175) (0.507) (0.182) (0.282) p-value Bandwidth N N below cut-off N above cut-off Notes: * denotes significant at 10%, **significant at 5% and *** significant at 1%. Robust standard errors in parenthesis. All estimation uses the robust, bias-corrected, estimator and bandwidth selector proposed by Calonico et al. (2014b). Table 2.A3 The election outcome impact on allocation of DAK subcomponents, first two-years in office Estimate Robust Standard Error Bandwidth N N below cut-off N above cut-off Agriculture Education 0.880*** Environment 0.869** Fishery 0.772** Government infra Health 0.804** Infrastructure 0.741** Water 1.045** Irrigation Road 0.918** Notes: * denotes significant at 10%, **significant at 5% and *** significant at 1%. All estimation uses the robust, bias-corrected, estimator and bandwidth selector proposed by Calonico et al. (2014b). DAK- Demography, DAK-Forestry and DAK-Trade could not be estimates because the cut-off is outside the range of vote margin. The coefficients in red are the DAK which not pass validity test as recorded in Table A7. 70

87 Table 2.A4 Descriptive statistics of districts characteristics, all samples Variable N Mean Std. Dev. PAD per capita (in million IDR) 5,992 46, ,000 Area (in KM2) 3,949 4,925 8,599 % Rural population 5, Population (in number of people) 6, , ,569 % Population , Unemployment rate (%) 3, Poverty rate (%) 5, % Birth attended by health worker 6, Immunization coverage (%) 4, Morbidity (%) 6, NER SMP (%) 6, NER SMA (%) 6, Literacy rate (%) 6, GDP (in billion IDR 2000) 6,094 3,846,467 8,783,680 Monthly Household per capita income (in IDR) 5, , ,761 Access to electricity (%) 5, Table 2.A5 Effect of current presidential election outcome on lag five years of transfer, last two years DBH Total DBH Tax DBH Resources DAU DAK Total Estimate Standard error (0.296) (0.233) (0.573) (0.189) (0.454) p- value Bandwidth N N below cutoff N above cutoff Notes: * denotes significant at 10%, **significant at 5% and *** significant at 1%. Robust standard errors in parenthesis. All estimation uses the robust, bias-corrected, estimator and bandwidth selector proposed by Calonico et al. (2014b). 71

88 Panel A. Total Revenue Sharing (DBH) Log shared revenue funds per capita Electoral margin of SBY in the presidential election 2004 and 2009 Panel B. General Allocation Grant (DAU) Electoral margin of SBY in the presidential election 2004 and 2009 Panel C. Special Allocation Grant (DAK) Electoral margin of SBY in the presidential election 2004 and 2009 Figure 2.A2. Lagged Five Years of Central Government Transfer, Last Two Years of President Tenure 72

89 Table 2.A6 Effect of presidential election on lag five years of subcomponents of DAK, last two years Estimate Robust Standard Error Bandwidth N N below cut-off N above cut-off Agriculture Education Environment Forestry Fishery Government infra Health Infrastructure Water Irrigation Road Demography ** Notes: * denotes significant at 10%, **significant at 5% and *** significant at 1%. All estimation uses the robust, bias-corrected, estimator and bandwidth selector proposed by Calonico et al. (2014b). DAK- Trade and DAK-Village cannot be estimated because the cut-off lies outside the range of vote margin. Table 2.A7. Effect of presidential election on lag five years of subcomponents of DAK, first two years Estimate Robust Standard Error Bandwidth N N below cut-off N above cut-off Agriculture Education * Environment Fishery Government infra Health * Infrastructure ** Water *** Irrigation * Road Notes: * denotes significant at 10%, **significant at 5% and *** significant at 1%. All estimation uses the robust, bias-corrected, estimator and bandwidth selector proposed by Calonico et al. (2014b). DAK- Forest, DAK-Demography, DAK-Trade and DAK-Village cannot be estimated because the cut-off lies outside the range of vote margin. 73

90 References Akhmedov, A., & Zhuravskaya, E. (2004). Opportunistic political cycles. Quarterly Journal of Economics, 119(4), Ames, B. (2001). The Deadlock of Democracy in Brazil. University of Michigan Press, Ann Arbor. Alt, J., & Lassen, D. (2006). Transparency, political polarization, and political budget cycles in OECD countries. American Journal of Political Science, 50(3), Anderson, G. M., & Tollison, R. D. (1991). Congressional influence and patterns of New Deal spending, Journal of Law and Economics, 34, 161. Ansolabehere, S., Snyder Jr., J.M. (2006). Party control of state government and the distribution of public expenditures. Scandinavian Journal of Economics, 108 (4), Arretche, M. and J. Rodden. (2004). Legislative bargaining and distributive politics in Brazil: an empirical approach, unpublished manuscript. Arulampalam, W., Dasgupta, S., Dhillon, A., & Dutta, B. (2009). Electoral goals and centerstate transfers: A theoretical model and empirical evidence from India. Journal of Development Economics, 88(1), Atlas, C. M., Gilligan, T. W., Hendershott, R. J., & Zupan, M. A. (1995). Slicing the Federal government pie: who wins, who loses and why. American Economic Review, 85, Banful, A. B. (2011). Do formula-based intergovernmental transfer mechanisms eliminate politically motivated targeting? Evidence from Ghana. Journal of Development Economics, 96(2), Barkan, J., Chege, M. (1989). Decentralising the state: district focus and the politics of reallocation in Kenya. Journal of Modern African Studies, 27 (3), Berry, C. R., Burden, B. C., & Howell, W. G. (2010). The president and the distribution of federal spending. American Political Science Review, 104(04), Blöndal, J. R., Hawkesworth, I., & Choi, H. D. (2009). Budgeting in Indonesia. OECD Journal on Budgeting, 9(2), 49. Bracco, E., Lockwood, B., Porcelli, F., & Redoano, M. (2015). Intergovernmental grants as signals and the alignment effect: Theory and evidence. Journal of Public Economics, 123, Brender, A., & Drazen, A. (2005). Political budget cycles in new versus established democracies. Journal of Monetary Economics, 52(7), Brollo, F., & Nannicini, T. (2012). Tying your enemy's hands in close races: the politics of federal transfers in Brazil. American Political Science Review, 106(04), Caldeira, E. (2012). Does the system of allocation of intergovernmental transfers in Senegal eliminate politically motivated targeting? Journal of African Economies, 21(2),

91 Calonico, S., Cattaneo, M. D., & Titiunik, R. (2014a). Robust data-driven inference in the regression-discontinuity design. Stata Journal, 14(4), Calonico, S., Cattaneo, M. D., & Titiunik, R. (2014b). Robust Nonparametric Confidence Intervals for Regression Discontinuity Designs. Econometrica, 82(6), Cattaneo, M. D., Jansson, M., & Ma, X. (2015). rddensity: Manipulation Testing based on Density Discontinuity. Technical report working paper, University of Michigan. URL: umich. edu/ cattaneo/software/rddensity/r/rddensitymanual. pdf. Camacho, A., Conover, E. (2011). Manipulation of social program eligibility; detection, explanations and consequences for empirical research. American Economic Journal: Economic Policy 3 (2), Case, A. (2001). Election goals and income redistribution: recent evidence from Albania. European Economic Review, 45 (3), Castells, A., & Solé-Ollé, A. (2005). The regional allocation of infrastructure investment: the role of equity, efficiency and political factors. European Economic Review, 49, Cole, S. (2009). Fixing market failures or fixing elections? Agricultural credit in India. American Economic Journal: Applied Economics, 1 (1), Cox, G. W., & McCubbins, M. D. (1986). Electoral politics as a redistributive game. The Journal of Politics, 48(02), Dahlberg, M., Johansson, E. (2002). On the vote-purchasing behavior of incumbent governments. American Political Science Review, 96 (1), Denemark, D. (2000). Partisan pork barrel in parliamentary systems: Australian constituencylevel grants. The Journal of Politics, 62(3), Dixit, A., & Londregan, J. (1998). Ideology, tactics, and efficiency in redistributive politics. Quarterly Journal of Economics, 113(2), Drazen, A., & Eslava, M. (2006). Pork barrel politics. NBER Working Paper Series, 12190, EITI (2015). Indonesia EITI Report : Contextual Report Volume Two. Extractive Industries Transparency Initiatives (EITI). Coordinating Ministry for Economic Affairs Republic of Indonesia. Finan, F. S. (2003). Political patronage and local development: a Brazilian case study. Unpublished manuscript, University of California-Berkeley. Ferejohn, J. (1974). Pork barrel politics. Palo Alto: Stanford University Press. Gelman, A., & Imbens, G. (2014). Why high-order polynomials should not be used in regression discontinuity designs (No. w20405). National Bureau of Economic Research. Grossman, P. J. (1994). A political theory of intergovernmental grants. Public Choice, 78(3-4), Hauk, W., & Wacziarg, R. (2007). Small States, big pork. Quarterly Journal of Political Science, 2,

92 Imbens, G. W., & Lemieux, T. (2008). Regression discontinuity designs: A guide to practice. Journal of Econometrics, 142(2), Imbens, G., & Kalyanaraman, K. (2012). Optimal Bandwidth Choice for the Regression Discontinuity Estimator. Review of Economic Studies, 79(3). Joanis, M. (2007). The road to power: partisan loyalty and the centralized provision of local infrastructures. Public Choice, 146(1), Johansson, E. (2003). Intergovernmental grants as a tactical instrument: Empirical evidence from Swedish municipalities. Journal of Public Economics, 87, Khandker, S. R., Koolwal, G. B., & Samad, H. A. (2010). Handbook on impact evaluation: quantitative methods and practices. World Bank Publications. Khemani, S. (2007). Does delegation of fiscal policy to an independent agency make a difference? Evidence from intergovernmental transfers in India. Journal of Development Economics, 82, Kawamura, K. (2003). Politics of the 1945 Constitution: Democratization and Its Impact on Political Institutions in Indonesia (No. 3). Institute of Developing Economies, Japan External Trade Organization (JETRO). Kawamura, K. (2013). Presidentialism and political parties in Indonesia: why are all parties not presidentialized? (No. 409). Institute of Developing Economies, Japan External Trade Organization (JETRO). Larcinese, V., Rizzo, L., & Testa, C. (2006). Allocating the US federal budget to the states: The impact of the president. Journal of Politics, 68(2), Lee, D. S. (2008). Randomized experiments from non-random selection in US House elections. Journal of Econometrics, 142(2), Lee, D. S., & Lemieux, T. (2010). Regression Discontinuity Designs in Economics. Journal of Economic Literature, 48, Levitt, S.D., Snyder, J.M. (1995). Political parties and the distribution of Federal outlays. American Journal of Political Science, 39 (4), Lindbeck, A. (1976). Stabilization policy in open economies with endogenous politicians. The American Economic Review, 66(2), Lindbeck, A., & Weibull, J. W. (1987). Balanced-budget redistribution as the outcome of political competition. Public choice, 52(3), Litschig, S. (2012). Are rules-based government programs shielded from special-interest politics? Evidence from revenue-sharing transfers in Brazil. Journal of Public Economics, 96(11), Ministry of Finance- Indonesia (2010). Handbook Supplement Directorate General of Fiscal Balance, Jakarta. McCrary, J. (2008). Manipulation of the running variable in the regression discontinuity design: A density test. Journal of Econometrics, 142(2), Migueis, M. (2013). The effect of political alignment on transfers to Portuguese municipalities. Economics & Politics, 25(1),

93 Miguel, E. and Zaidi, F. (2003). Do politicians reward their supporters? Regression Discontinuity Evidence from Ghana. Mimeo, University of California, Berkeley. Nichter, S. (2008). Vote buying or turnout buying? Machine politics and the secret ballot. American Political Science Review, 102 (1), Nordhaus, W. (1975). The political business cycle. Review of Economic Studies, 42, Pereira, P. T. (1996). A politico-economic approach to intergovernmental lump-sum grants. Public Choice, 88(1-2), Pereira, P. T. (1996). A politico-economic approach to intergovernmental lump-sum grants. Public Choice, 88, Persson, T., Tabellini, G. (2000). Political Economics: Explaining Economic Policy. MA, MIT Press, Cambridge. Persson, T., & Tabellini, G. (2004). Constitutional rules and fiscal policy outcomes. American Economic Review, 94(1), Porto, A., & Sanguinetti, P. (2001). Political determinants of intergovernmental grants: Evidence from Argentina. Economics & Politics,13(3), Rogoff, K., & Sibert, A. (1988). Elections and macroeconomic policy cycles. Review of Economics Studies, 55, Rogoff, K. (1990). Equilibrium political budget cycles. American Economic Review, 80, Samuels, D.J. (2002). Pork-barreling is not credit-claiming or advertising: campaign finance and the sources of the personal vote in Brazil. The Journal of Politics, 64 (3), Savitri, M.D. (2013). Performance-Based Grants in Indonesian Decentralization. Incorporating Incentives to Improve Public Service Delivery. Doctoral Dissertation. The Okuma School of Public Management Waseda University Schady, N.R. (2000). The political economy of expenditures by the Peruvian Social Fund (FONCODES), The American Political Science Review, 94 (2), Shi, M., & Svensson, J. (2006). Political budget cycles: Do they differ across countries and why? Journal of Public Economics, 90, Shah, A., Qibthiyyah, R., & Dita, A. (2012). General purpose central-provincial-local transfers (dau) in Indonesia: From gap filling to ensuring fair access to essential public services for all. World Bank Policy Research Working Paper, (6075). Sjahrir, B. S., Kis-Katos, K., & Schulze, G. G. (2013). Political budget cycles in Indonesia at the district level. Economics Letters, 120(2), Sjahrir, B. S., Kis-Katos, K., & Schulze, G. G. (2014). Administrative overspending in Indonesian districts: The role of local politics. World Development, 59, Snyder, J. M. (1989). Election goals and the allocation of campaign resources. Econometrica, 57,

94 Solé-Ollé, A., & Sorribas-Navarro, P. (2008). The effects of partisan alignment on the allocation of intergovernmental transfers. Difference-in-differences estimates for Spain. Journal of Public Economics, 92, Strömberg, D. (2008). How the Electoral College influences campaigns and policy: the probability of being Florida. The American Economic Review, 98(3), Taylor, A. J. (2008). The presidential pork barrel and the conditioning effect of term. Presidential Studies Quarterly, 38(1), Ufen, A. (2006). Political Parties in Post-Suharto Indonesia: Between politik aliran and Philippinisation (No. 37). GIGA German Institute of Global and Area Studies. Usman, S., Mawardi, M. S., Poesoro, A., Suryahadi, A., & Sampford, C. (2008). The Specific Allocation Fund (DAK): Mechanism and Uses (No ). Veiga, L. G., & Pinho, M. M. (2007). The political economy of intergovernmental grants: Evidence from a maturing democracy. Public Choice, 133(3-4), Wallis, J. J. (1996). What Determines the Allocation of National Government Grants to the States? NBER Working Paper on Historical Factors in Long Run Growth 90. Wallis, J. J. (1998). The political economy of New Deal spending revisited, again: With and without Nevada. Explorations in Economic History, 35(2), World Bank. (2008). Spending for Development: Making the Most of Indonesia's New Opportunities. World Bank, Jakarta. Worthington, A. C., & Dollery, B. E. (1998). The political determination of intergovernmental grants in Australia. Public Choice, 94(3-4), Wright, G. (1974). The political economy of new deal spending: an econometric analysis. Review of Economics and Statistics, 56 (1),

95 Chapter 3: Re-election of Incumbent District Leaders in Indonesia - Does Economy Matter? 3.1 Introduction After fifteen years of political change, Indonesia is now the third largest democracy in the world. Indonesia has successfully held four democratic general elections and three direct presidential elections, since 1999, after the authoritarian New Order regime collapsed. The Indonesian government has also implemented an extensive decentralization scheme since A wide range of administrative, fiscal and political power has been transferred to district governments. The country took another significant step towards political decentralization in 2005 in order to strengthen the democratic accountability of local governments. Starting in 2005, following a more presidential rather than parliamentary system, district heads were selected through direct elections among citizens. However, there are indications that the political process at the local level has been characterised partly by money politics and powerful local elites (Mietzner 2005, 2010), which suggest that the democratic accountability mechanisms might be compromised. The question of whether the democratic accountability has been compromised is an important empirical question. In this chapter I attempt to answer the question by investigating whether incumbent district heads were held accountable to his/her performance in the first two cycles of local direct election in Indonesia. Specifically, I investigate, whether voters punish incumbents when the economy is doing poorly and reward them when the economy is doing well. This study extends the literature on several aspects. First, this is the first study, on Indonesia s local elections that investigates the determinants of district government leader re-election. The majority of studies in this strand of literature are either using state/governor elections or national/cross-country elections (Hayes et al. 2015; 79

96 Leigh 2009; Berry and Howell 2007; Jérôme and Jérôme-Speziari 2005; Cutler 2002). Moreover, Indonesia is a newly democratised and decentralised developing country while most previous country-specific studies in the literature are established federal, democratic and developed countries. Second, departing from previous studies on economic voting that consider voter attribution errors, my study uses both national economy as well as regional districts economic performance as reference points for evaluation. This approach is based on insights from the yardstick competition literature (Besley and Smart 2007; Besley and Case 1995; Salmon 1987), which suggest that voters have asymmetric information about the incumbent candidate s competence. As a solution to the problem, voters may compare the performance in their jurisdiction with the neighboring ones and draw information about the relative competence of their administrator. The voters choose to re-elect or dismiss the incumbent based on the outcome of this comparison. Attribution error in this context is defined as, a voter tendency to systematically fail to take sufficient account of externalities when aiming to assess incumbent competence (Wolfers 2007). There are several studies which have documented the presence of attribution errors in the context of retrospective economic voting. In the United States, Wolfers (2007) shows that the incumbent governors of oil-producing U.S. states tend to enjoy a higher re-election probability when oil prices are rising. In Australia, Leigh and McLeish (2009) show that Australian voters reward state governments for both competence (unemployment in their state relative to the rest of Australia) and luck (unemployment common to all states). Hayes et al. (2015) have recently verified these findings using cross-country data. However, most of these studies only use performance of the higher economy (national or global) as a reference point to address voter attribution errors. 80

97 Third, I test and control for potential sample selection bias which mostly neglected in previous literature. The risk of selection bias exists, because data on re-election are only observed for an incumbent who is running for re-election. Furthermore, using Portuguese municipalities dataset, Castro and Martins (2013) find that the determinant of the mayors choice to run for an additional term in office is affected by the local economic performance. This implies that there are systematic differences between incumbent that running and not-running for re-election, which need to be addressed using a selection model. Fourth, this study contributes to the literature on decentralization in Indonesia, especially on the debate on local government accountability. Skoufias et al. s (2014) study shows that local governments become more accountable to their poor citizens, while Sjahrir et al. (2014) recently found that the introduction of direct election of district government heads does not reduce the degree of district government administrative expenditure overspending. They claim that this result implies failure of local election in establishing accountability. Previous literature also suggests the Indonesian postdecentralization accountability framework is weak (Lele 2012; Lewis 2010; Calavan et al. 2009; Lewis and Pattinasarany 2009) and some even argue it is the major deficit area in Indonesia decentralization (Lewis 2010; Calavan et al. 2009). I contribute to this debate by providing empirical evidence on whether local economy performance matters for incumbent district government head re-election. Hence, this study is giving a new perspective on the empirical evidence within the topic of local government electoral accountability in decentralized Indonesia. The estimates of this study suggest that the extent of such attribution error is only quantitatively significant when district performance is benchmarked using regional performance within overall period of incumbents. When regional districts GDP per capita 81

98 growth increase by 1 percentage point, the likelihood of an incumbent district head being re-elected, on average, increases by 3.2 percentage points. Whereas, a one percentage point increase in change of unemployment of regional districts increase the average likelihood of an incumbent district head being replaced by 13.1 percentage points. However, in election years voters do distinguish between district performance from regional districts and national performance. That is, incumbent district leaders seem to be rewarded or punished for economic performance that arises from competence. This is especially true in the case of unemployment reduction performance. The rest of this chapter is organised as follows. Section 2 reviews the literature on economic voting in local elections. Section 3 presents a short history of Indonesia s local election. Section 4 describes the model, and the dataset used is introduced in Section 5. The empirical results obtained are presented in Section 6, followed by robustness checks in Section7. Section 8 concludes. 3.2 Literature Review This study is a part of a larger literature on retrospective economic voting (for a literature review see Dassonneville and Lewis-Beck 2014; Duch and Stevenson 2008; Lewis-Beck and Stegmaier 2000, 2007; Van der Brug et al. 2007; Lewis-Beck and Paldam 2000; Anderson 2007). The classical interpretation of retrospective voting is a reward-punishment model. In this model voters will reward the incumbent with their vote when the economy is good, and will punish the incumbent by casting their vote for the challenger when the economy is bad. In the language of rational choice, voters (principals) are attempting to reduce moral hazard on the part of elected representatives (agents). By re-electing high-performing politicians and voting out poor performers, voters incentivize good behavior on the part of politicians that induces accountability. 82

99 This was Key s (1966) informal description of retrospective voting, later formalized by Barro (1973) and Ferejohn (1986). Another rational choice interpretation describes retrospective voters as selecting the leaders who will perform most competently after being elected (Fearon 1999). After learning about an incumbent s quality through his or her performance in office, voters can choose to re-elect a competent leader or take their chances on an unknown challenger (Duch and Stevenson 2008, Persson and Tabellini 2002). In the long run, elections serve the process of selecting good performers (Padro-i-Miguel and Snyder 2006, Ashworth 2005). Importantly, both the reward-punishment and selection models suggest that retrospective voting serves the normatively beneficial function of reducing moral hazard and adverse selection. The empirical literature on retrospective economic voting has shown that economic conditions regularly have significant effects on national-level elections across a wide range of countries (e.g., Canada: Happy 1992; Nadeau and Blais 1993, United Kingdom: Sanders 2005, United States: Fiorina 1981), levels of economic development (e.g., Gélineau 2007; Singer 2013), and institutional types (Anderson 2006; Powell and Whitten 1993; Whitten and Palmer 1999). Overall, the literature on economic voting predominantly focuses on national electoral contexts, whereas the local elections received relatively minor attention. However, some recent studies have started to investigate whether local incumbents, especially at subnational/district government, are held accountable for local economic conditions as well. Within these studies there are two streams of literature based on theoretical frameworks. First, studies which are inspired by the political business cycle literature, which empirically find that local incumbents strategically increase spending or reduce the tax burden when approaching the election year (Sakurai and Menezes-Filho 2011; Sakurai and Menezes-Filho 2008; Veiga and Veiga 2007; Vermeir and Heyndels 2006). 83

100 Second, studies based on the classic economic voting theorem, which assume that voters take into account the state of the local economy (i.e. economic outcomes), when casting their vote on local election. This is evident from recent research which shows that local incumbents are held accountable for how they performed while being in office, with their performances operationalised by means of indicators of local economic conditions (Boyne et al. 2009; Berry and Howell 2007; Oliver and Ha 2007). However, these studies failed to take into account the possibility of attribution error in their estimation. Literature from psychology has defined fundamental attribution error as a human tendency to systematically fail to take sufficient account of background or environmental factors when aiming to assess competence (Patty and Weber 2005). There is a growing empirical literature that investigates the impact of attribution error in economic voting. At a cross-country level, Leigh (2009) studies whether world leaders benefit from global macroeconomic swings relative to their own country s economic performance. He finds that voters do a poor job of distinguishing signal from noise; they are more sensitive to the global economy than to their own country s economic performance net of global swings. Recently, Hayes et al. (2015) also verify these findings. However, Hayes et al. (2015) also show that voter attribution errors are less likely in countries with a long tradition of democracy, educated voters, and free media. In the single-country context Wolfers (2007), using dataset from US gubernatorial election, finds that although voters evaluate their state s economic performance relative to the national economy, there is also evidence that voters also reward and punish incumbent state governors for economic fluctuations that are plausibly unrelated to gubernatorial actions (national or internationally driven). Whereas in Australia, Leigh and McLeish (2009) show that Australian voters reward state governors for both competence (unemployment in their state relative to the rest of Australia) and luck (unemployment common to all states). 84

101 3.3 Local Election in Indonesia: Institutional Setting Indonesia has implemented Big Bang governance reforms since 2001 with two of the most important pillars of this reform being decentralization and democratization. The fiscal decentralization reform, written into law in 1999 and implemented in 2001, devolved almost 40% of the overall budgetary responsibility to the regions (provinces and districts), and transferred a number of very important functions, such as primary and secondary education, health services, environmental protection, and infrastructure, predominantly to the districts (World Bank 2008). This devolution of authority to the local level has shifted substantial financial resources and decision-making power to units that previously were merely executing orders from the central government. The democratization reform took place through two waves. First, it happened in 1999 when Indonesia had its first democratic election after the authoritarian New Order regime collapsed. Elections for district and provincial legislatives were conducted along the same 5-year cycle as the national elections, implying that all local legislatures were elected at the same time (i.e., 1999, 2004, 2009 and most recently 2014). Before the Law No.32 was stipulated in 2004, district heads were elected by the democratically elected local parliaments whenever the terms of the effectively appointed district heads from the New Order came to an end. The concerns around indirect political accountability triggered the second wave of local government electoral reform toward local direct elections (Pemilihan Kepala Daerah/PILKADA) under Law No. 32/2004). This reform made the local government head (Regent/ Major or Bupati/Walikota) more directly accountable to the people by stipulating that she/he would be directly elected by the local citizens. The reform also provided a clearer definition of the government head s political functions. It was believed that this democratic reform would make the district heads more accountable to their 85

102 constituencies. The first batch of direct elections was conducted in June 2005 for all the regional head positions that had tenure come due between December 2004 and April A unique aspect of this reform was that the change in political selection occurred in a staggered manner, once the old terms subject to indirect elections had come to an end. The timing of the shift to direct elections in a district was determined by whether the district head selected by the previous system had served their full tenure, which resulted in direct elections being held in a little more than one-third of all districts in June The remaining districts continued to be under the existing regime until the tenure of their heads were over. The implication of varying local elections timing across districts is that evaluation of performance of each district must be adjusted with tenure of local heads, which mean even though the time interval of tenure is the same for all districts (5 years) the starting/end year of the tenure could be different across districts. On the negative side it complicates the calculation of district performance. On the positive side I argue that by having different election timing, performance of each district can be isolated from time specific shock. This means each district performance is less affected by other districts and facilitates performance assessment that is relatively more accurate in capturing individual district achievement. According to Ministry of Home Affairs (MoHA), in total there were 963 local elections within Since tenure of local district head is 5 years, elections between can be devided into two periods. The first period includes elections that occurred in , and the second period refers to elections that occurred between Most of the districts experienced their first PILKADA in the first 86

103 period, and the second PILKADA in the second period, except for several new districts that are having their first PILKADA in the second period. Overall this study managed to collect 88.06% local election data over This includes 362 out of 454 elections (80%) in the first period ( ) and 486 out of 509 elections (95%) in the second period ( ). For the record there was only one election in 2009 where there should have been no election in that year, since it was a national election year (the local election was originally scheduled in 2008). 3.4 Model Specification Retrospective economic voting model A Probit model will be used to explain how a district s economic performance (Eco) influences the probability of re-election of the head of district government, given certain control variables of political and demographic conditions of the districts (Cont). The model can be mathematically presented as follows: Prob(Reelect = 1 Eco, Cont) = Φ(Eco β + Cont γ) (3.1) Where β and γ are the vectors of the parameters to be estimated and Φ(.) is the normal cumulative distribution function. As the Probit model is estimated over a panel of 497 district governments (i) for the two cycles of local elections (E) that took place between 2005 and 2013, a panel data analysis is considered. The application of binary model to panel data analysis is straightforward. The structural model for the panel data to be estimated in this study can be written as follows: y i,e = α + Eco i,e β + Cont i,e γ + ε i,e (3.2) where, i = 1,.,497 and E = 1,2 87

104 Reelect i,e = { 1 if y i,e > 0 0 otherwise ε i,e ~ N[0,1] The y i,e variable shows the probability of the district head being re-elected, considering all explanatory variables, for districts i at election E. Nevertheless, y i,e is not observable. One can only observe Reelect i,e, which is a binary-choice variable to indicate whether or not the last district head is re-elected in the current election. Whereas α is a constant term, Eco i,e and Cont i,e are matrices of the set of observable independent variables that linearly determine y i,e, β and γ are vectors of coefficients associated with Eco i,e zero mean and unit variance (Greene 2003). and Cont i,e, and ε i,e is the error term, normally distributed with I conduct an LR test for random effects to test the suitability of panel regression estimation procedure (for details see Table 3.3 and 3.4). The LR test of rho=0 are resulting in Chibar 2 (01) that ranging from to 0.33 with estimated p-value ranging from to 0.282, meaning rho is statistically zero in all specifications. Thus, the panel-level variance component is unimportant, and the panel estimator is not different from the pooled estimator. Therefore, I decided to estimate a simple pooled Probit, where the presence of heteroscedasticity and autocorrelation is controlled for using robust standard error clustered by province. For the record I did not test for fixed effect since adding fixed effect on Probit models induces bias in the coefficients and standard errors, due to the incidental parameter problem (Wooldridge 2002). Risk of selection bias in equation (3.2) exists, however, because data on reelection are observed only for an incumbent running for re-election. If systematic differences exist between incumbents that are running and not-running for re-election, regressions based on a restricted, non-random sample of running incumbent only will be subject to specification error and generate biased results (Greene 2003; Van de Ven and 88

105 Van Praag 1981). To control for potential sample selection bias, a Heckprobit selection model is also applied in this study. This technique is based on Heckman s (1979) sample selection model which was designed for linear outcome equations. The Heckprobit has been adapted for discrete dependent variables where both the selection equation and the outcome equation are binary choices (Van de Venn and Van Praag 1981). Whether or not data is observed for incumbent re-election depends on the incumbent s eligibility and also the personal decision to run as candidate in the election. The selection model is a Probit model which in its application in this context can be expressed as: z i,e = θ + x i,e δ + u i,e (3.3) where, i = 1,.,497 and E = 1,2 Candidacy i,e = { 1 if z i,e > 0 0 otherwise u i,e ~ N[0,1] corr[ε i,e u i,e ] = ρ The z i,e is an unobservable variable representing the probability of incumbent district head running for re-election in terms of the explanatory variables contained in x i,e for districts i at election E. Candidacy i,e is a binary variable to indicate whether or not the last district head is running at election E. Whereas θ is a constant term, x i,e is the set of observable independent variables that linearly determine z i,e, δ is vectors of coefficients associated with x i,e, and u i,e is the error term of the selection equation, normally distributed with zero mean and unit variance, and ρ denotes the correlation between the error terms of the outcome and selection equations. 89

106 The value of ρ is used to evaluate the risk of selection bias. If it is statistically proven that ρ=0 then there is no evidence of selection bias, which mean the re-election and selection equation are independent, making estimation of the selection model unnecessary. In this case, it would be better to estimate only the re-election equation with the standard Probit model, since it will deliver more consistent and unbiased estimates. However, if ρ differs significantly from zero, standard Probit techniques applied to the re-election equation will produce biased results. The Heckprobit procedure instead provides consistent, asymptotically efficiently for all the parameters in such models (Pastore 2012; Van de Ven and Van Praag 1981). I consider five variables that determine the candidacy decision: district economic performance, number of candidates running in the election, district establishment period and electoral cycle. The better that the economic performance of a district is, the more likely is an incumbent to run for re-election, since the probability of winning would be greater. The more candidates running in the election the less likely is the incumbent to run for re-election since the number of candidates reflect the intensity of political competition; more candidates mean more competition and less probability for re-election. I also argue that district establishment period matters for candidacy. Newly established districts are more likely not to have an incumbent running in the election, especially those established within 1-2 years prior to the election. On the contrary, older districts are more likely to have an incumbent running in the election. The electoral cycle also matters in determining candidacy. It is reasonable that the first cycle of local elections has a higher candidacy than the second cycle. This is because the law mandates that a district head can only rule for two terms, so those incumbents that have been re-elected in the first cycle of local elections ( ) by law are not allowed to run for re-election in the second cycle ( ). 90

107 3.4.2 Filtering competence from luck The estimation of the effect of district economic performance on a district leader s probability of re-election in model (3.2) is ignoring the risk of attribution error discussed in the previous section. This is because the design of the model is unable to differentiate whether the incumbent was re-elected due to competence or luck. To distinguish between the effect of competence and the effect of luck, I follow the methodology proposed by Bertrand and Mullainathan (2001) which has recently been adopted by Wolfers (2007), Leigh (2009), Leigh and McLeish (2009) and Hayes et al. (2015). I restructure the model by separating the district economic performance into luck and competence in two ways. First, by using regional districts economic performance (Reg), as follows: y i,e = α + (Eco i,e Reg i,e )β + Reg i,e φ + Cont i,e γ + ε i,e (3.4) Secondly, by using national economic performance (Nat), as follows: y i,e = α + (Eco i,e Nat E )β + Nat E φ + Cont i,e γ + ε i,e (3.5) where, i = 1,.,497 and E = 1,2 Reelect i,e = { 1 if y i,e > 0 0 otherwise ε i,e ~ N[0,1] As in the previous equations the y i,e variable shows the probability of district head being re-elected, considering all explanatory variables, for districts i at election E, where y i,e is unobservable so that we have to rely on Reelect i,e, which is a binarychoice variable to indicate whether or not the last district head is re-elected in the 91

108 current election. Whereas α is a constant term, Eco i,e and Cont i,e are sets of observable independent variables that represent district s economics performance and politicaldemographic condition, respectively. The Reg i,e is an average economic performance of all districts within the same province for district i at election E. The Nat E is the national economic performance at local election E, which is constant across districts. Meanwhile γ is vector of coefficients associated with district political and demographic characteristics (Cont i,e ), and ε i,e is the error term, normally distributed with zero mean and unit variance. The β now represents vectors of coefficient characterized as the effect of the incumbent s competence, while φ is vectors of coefficient that measures the effect of luck. If the voters are rational and perfectly able to filter competence from luck then φ = 0. Conversely, the literature on attribution errors suggests that voters may fail to take sufficient account of background or environmental factors in assessments of competence, leading to φ > 0. However, like equation (3.2), equation (3.4) and (3.5) are also at risk of selection bias because data on re-election are still only observed for incumbents who run for reelection. Therefore, I also apply Heckprobit selection model as in equation (3.3) to control for potential sample selection bias in equation (3.4) and (3.5). 92

109 3.5 Data In order to analyze the Indonesia case, I collected data for the 497 districts over the period , covering the two electoral cycles of and The political data sets at district level are constructed based on information of local direct election results within from the Local General Election Commission (KPUD) and Mcculloch (2011) database of the district leaders from 2001 to The political data which is the main interest of this study is the re-election of the incumbent district leader. However, it is important to note that the Mcculloch's (2011) database is using 2001 as reference point to amalgamate back the data if districts subsequently split after Hence, the re-election of the incumbent district leader in local election 2005 for new district, which split from its mother district after 2001, is unobservable. To characterize the district economic performance, I use the per capita GDP growth and unemployment rate. These two are among three variables that have received the greatest empirical attention in the voting literature (Paldam 2004): the unemployment rate, inflation and GDP growth. Inflation is not included for two reasons: at the regional level there is no data on inflation and inflation is under the authority of the central bank, and not under the control of district head. The economic performance data are taken from the Central Bureau of Statistics (BPS) and World Bank Indonesia database (INDODAPOER). To control for variations in districts characteristics I include sets of demographic variables. These demographic variables are: share of population living in urban areas, share of population density, share of working-age population, share of retired-age population, total population, literacy rate, ethnic and religious diversity, and a dummy variable for districts located in Java (the most developed island in Indonesia). In addition to the demography, I also include variables to control for fiscal and political characteristics of the districts. 93

110 Table 3.1 Description of the variables Variables Description Re-elect Binary variable equal to 1 if the incumbent district leader is re-elected and 0 otherwise gyp gyp_regional gyp_national District per capita real GDP growth Average of district per capita real GDP growth within a province National per capita real GDP growth Economic Performance DgYp _regional DgYp_national Une UR UR_regional gyp - gyp_regional gyp - gyp_national District unemployment rate Percentage change in the districts unemployment rate Average percentage change of districts unemployment rate within a province UR_national D UR_regional D UR_national Percentage change in the national unemployment rate UR - UR_regional UR - UR_national Urban Percentage of district population living in urban areas Young Percentage of district population within the age 0-14 Old Percentage of district population age > 65 Demography LnPop Literacy Ethnic Natural logarithm of district total population Literacy rate at district level Ethnic fractionalisation at district level (probability that two randomly selected people are not from the same ethnic group, values range from 0 to 1) Religion Java Religion fractionalisation district level (probability that two randomly selected people are not from the same religious group, values range from 0 to 1) Binary variable equal to 1 if the district is located in Java Island Candidacy Vturnout Binary variable equal to 1 if the incumbent district leader is running for re-election and 0 otherwise Voter turnout in local election of district leader Political Ncandidate Independent Elect.Cyle Number of candidate running in district leader election Binary variable equal to 1 if there is independent candidate running in the election and 0 otherwise Binary variable equal to 1 for the second round of direct local election and 0 otherwise Newdistrict Binary variable equal to 1 if the district is located in established after 2001 and 0 otherwise Fiscal capacity Percentage of district s government own source revenue and revenue sharing to total revenue LGRevenue Natural logarithm of district s government total revenue percapita at constants price 2000 Sources: World Bank (INDODAPOER); Regional General Election Commission (KPUD); Statistics Indonesia (BPS). Fiscal 94

111 Table 3.2 Descriptive statistics All elections Elections with incumbent running Variables Obs. Mean Std.Dev Obs. Mean Std.Dev Dependent variable Re-elect Independent variables* gyp (term) gyp (election year) gyp_regional (term) gyp_regional (election year) gyp_national (tern) gyp_national (election year) Une (term) Une (election year) UR (term) UR (election year) UR_regional (term) UR_regional (election year) UR_national (term) UR_national (election year) Control Urban Young Old LnPop Literacy Ethnic Religion Candidacy Vturnout Ncandidate Independent Elect.Cyle Java Newdistrict Fiscal capacity LGRevenue Notes: * percentage change from the last year and average annual percentage change over term 95

112 A complete description of all variables employed in this study can be found in Table 3.1, while the descriptive statistics are reported in Table 3.2. For the record, the unemployment data at the district level is only available from 2007 onwards, and this explains why in Table 3.2 unemployment has the lowest number of observation compared to other variables. 3.6 Empirical Results Baseline results from retrospective model The baseline results of this study are presented in Tables 3.3 and 3.4. These are an estimation of the pooled Probit model of the equation (3.2). Following Veiga and Veiga (2010) and Castro and Martins (2013), in order to determine which time horizon of economic performance is important for the voters, I expressed the economic variables in two different ways: first, as percentage changes from the previous year at the election year; and second, as average percentage annual changes over the entire term of incumbents (started from the previous election year). Since the possibility of re-election is zero if no incumbent is running in the election, the baseline estimations drop observation of elections where the incumbent is not running for re-election. For each variable presented in Tables 3.3 and 3.4, the estimated coefficients and the average marginal effects are shown. The robust standard error for both the estimated coefficients and average marginal effects are presented in parentheses and the degree of statistical significance is indicated with asterisks. The number of observations, loglikelihood, and Pseudo-R 2, and likelihood-ratio test for random effect (Random Effect Test) are reported at the bottom of each table. A likelihood-ratio test formally compares the pooled estimator (Probit) with the panel estimator. The test in Table 3.3 and

113 statistically proves that rho is zero, which indicates the panel-level variance component is unimportant, and the panel estimator is not different from the pooled estimator. Table 3.3 reports the impact of GDP per capita growth (gyp) on probability of reelection of district leader incumbent, while Table 3.4 reports the impact of change in unemployment rate ( UR). Examining the estimation results of economic performance in Tables 3.3 and 3.4, I find that both the GDP per capita and the unemployment are to some degree relevant for district leader re-election. Based on the marginal effect of Probit regressions in Table 3.3, an incumbent district leader is 0.4 % (column 4) more likely to win a re-election for every extra percentage point of GDP per capita growth in the election year. To put this into context, the mean growth rate of GDP per capita in the election year in the sample is 3.6%, and incumbents are re-elected 66.8 % of the time (see Table 3.1). These results imply that at a growth rate of 4.6 %, incumbents would have a 67.2 % chance of re-election. Meanwhile, focusing on the marginal effect of change in the level of unemployment rate in Table 3.4, it shows that a 1 % decrease in the unemployment rate in the election year increases the probability of an incumbent district head winning the election by 3.9% (column 4). These baseline results suggest that unemployment can have a more substantial impact on re-election compared to GDP per capita growth. However, the fact that none of the economic performances are significant over the entire term seems to indicate that the voters are on average myopic rather than far-sighted (Helwig and Marinova 2015), in which case, they outweigh election-year performance and ignore overall performance under the incumbent government administration. 97

114 Table 3.3 The impact of GDP per capita growth on probability of re-election: Probit regression Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect (1) (2) (3) (4) gyp * 0.004* (0.010) (0.003) (0.007) (0.002) Urban 0.007*** 0.002*** 0.008*** 0.002*** (0.002) (0.001) (0.003) (0.001) Young 0.055*** 0.017*** 0.041** 0.013** (0.018) (0.005) (0.018) (0.005) Old 0.117** 0.037*** 0.090* 0.028* (0.046) (0.014) (0.053) (0.017) LnPop (0.094) (0.030) (0.101) (0.315) Literacy (0.007) (0.002) (0.008) (0.003) Ethnic (0.367) (0.117) (0.411) (0.129) Religion (0.404) (0.128) (0.420) (0.130) Vturnout (0.009) (0.003) (0.010) (0.003) Ncandidate *** *** *** *** (0.442) (0.013) (0.046) (0.013) Independent 0.577*** 0.184*** 0.609*** 0.191*** (0.182) (0.056) (0.188) (0.058) Java (0.210) (0.068) (0.206) (0.649) Constant (2.063) (2.206) No. Obs Log L Pseudo-R Random Effect Test (rho=0) Chibar 2 (01) Prob Chibar Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is re-elected for another term and 0 otherwise. 98

115 Table 3.4 The impact of unemployment change on probability of re-election: Probit regression Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect (1) (2) (3) (4) UR ** ** (0.096) (0.030) (0.051) (0.016) Urban 0.007* 0.002** 0.008** 0.002** (0.004) (0.001) (0.003) (0.001) Young 0.077*** 0.024*** 0.079*** 0.024*** (0.023) (0.007) (0.024) (0.007) Old 0.135** 0.042** 0.132** 0.041** (0.637) (0.019) (0.063) (0.019) LnPop (0.127) (0.003) (0.131) (0.041) Literacy (0.008) (0.003) (0.009) (0.003) Ethnic 0.903** 0.284** 0.865** 0.267** (0.407) (0.126) (0.414) (0.126) Religion ** *** ** ** (0.506) (0.154) (0.522) (0.158) Vturnout (0.012) (0.004) (0.013) (0.004) Ncandidate *** *** *** *** (0.056) (0.550) (0.016) Independent 0.569*** 0.179*** 0.618*** 0.191*** (0.205) (0.061) (0.204) (0.060) Java (0.227) (0.073) (0.245) (0.077) Constant (3.005) (3.063) No. Obs Log L Pseudo-R Random Effect Test (rho=0) Chibar 2 (01) Prob Chibar Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is re-elected for another term and 0 otherwise. 99

116 The demographic and political control variables are also found to be fairly relevant in explaining the re-election of the district head. In Tables 3.3 and 3.4 the regressions find some evidence that the likelihood of a district head being re-elected for another term is greater in more urbanized districts (Urban). This might be because the incumbent s political campaign is more effective to deliver in more urbanized districts due to better infrastructure. The share of young-working-age (Young) and old population (Old) also increases the likelihood of re-election of the incumbent district head. However, the share of old population marginal effect is more than twice that of the young population. This implies that either the votes of the aged in the society are easier to win, or the aged population is more loyal than the young. The district population size (LnPop), literacy rate (Literacy) and dummy for district located in Java Island (Java) are consistently insignificant in both Table 3.3 and 3.4. The other two demographic control variables: ethnic (Ethnic) and religious fractionalisation (Religion), are not significant for growth equation (Table 3.3). On the contrary, both variables are significant in the unemployment equation (Table 3.4) with an opposite sign. The ethnic fractionalisation is positive, which means the likelihood for reelection is higher for an incumbent running in more ethnically heteregenous districts. However, for religious heterogeneity, it is reversed. The incumbent running for reelection in a more religiously diverse district has a lower probability of being re-elected, compared to an incumbent running in a less religiously heterogenous district. It suggests that ethnic diversity contributes to political stability while religious diversity is destabilising. In both Table 3.3 and Table 3.4, the numbers of candidates running in the local election (Ncandidate) and dummy variable for the presence of independent candidates in 100

117 the election (Independent) have a significant and consistent impact on districts leader reelection probability, which implies that, the probability of being re-elected decreases when the number of contesting candidates increases. This makes sense, since we expect that an election with a large number of candidates will have a more intense political competition and weaken the probability of re-election. However, it could also because the higher is the probability of re-election of the incumbent (higher popularity), the more reluctant other potential candidate to challenge for the position. The marginal effect of the dummy variable for independent candidates is positive, which indicates that the incumbent running in the election where there are independent candidates has a better opportunity to be re-elected. The voter turnout, however, is consistently insignificant. To address the issue of selection bias, I use the Heckprobit model to re-estimate the growth and unemployment determinant of re-election. The results are presented in Tables 3.5 and 3.6. The tables also present the results of the first stage of the Heckprobit regression, the Candicady Selection Equation, in the second panel of the tables. The results of selection equation in Tables 3.5 and 3.6 indicate that number of candidates (Ncandidate) and dummy for the second electoral cycle (Elect.Cycle) have a highly significant and negative influence on the probability of re-election for the incumbent running as a candidate. These results are reasonable since more candidates mean more competition so less incentive for running as a candidate, whereas candidacy is lower in the second electoral cycle because the incumbents that have been re-elected in the first cycle are not allowed by law to run for re-election. 101

118 Table 3.5 The impact of GDP per capita growth on probability of re-election: Heckprobit Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect (1) (2) (3) (4) gyp 0.153* 0.004* 0.012** 0.003** (0.009) (0.002) (0.006) (0.001) Urban 0.006** 0.002** 0.006** 0.002** (0.002) (0.001) (0.003) (0.001) Young 0.039** 0.011** (0.015) (0.004) (0.163) (0.004) Old 0.104** 0.028** (0.042) (0.011) (0.051) (0.014) LnPop (0.087) (0.024) (0.097) (0.026) Literacy (0.006) (0.002) (0.007) (0.002) Ethnic (0.331) (0.091) (0.379) (0.102) Religion (0.367) (0.099) (0.383) (0.100) Vturnout (0.008) (0.002) (0.009) (0.002) Ncandidate *** ** *** ** (0.058) (0.018) (0.066) (0.020) Independent 0.598*** 0.162*** 0.663*** 0.174*** (0.172) (0.048) (0.180) (0.051) Java (0.190) (0.052) (0.191) (0.050) Constant (1.925) (2.042) Candidacy Selection Equation gyp (0.008) (0.009) Ncandidate *** *** (0.038) (0.045) Newdist (0.123) (0.127) Elect.Cycle *** *** (0.133) (0.140) Constant 1.129*** 1.112*** (0.201) (0.229) 102

119 Table 3.5 (Continued) Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect Model Criteria (1) (2) (3) (4) Total number of obs Censored obs Uncensored obs Log Likelihood Wald Chi 2 (12 df) Prob > Chi Selectivity Test (rho=0) Chi 2 (1 df) Prob > Chi Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is re-elected for another term and 0 otherwise. Tables 3.5 and 3.6 also present the significance test for the selection model (Wald test of independent equations) in the last panel. The significance test for rho in Table 3.5 statistically proves that ρ differs significantly from zero. This indicates a problem of selection bias which verifies the need to run a selection equation to include elections in districts with no incumbent running in the estimation sample in order to prevent biased and inconsistent coefficient estimates. This justifies the use of Heckprobit selection models for the growth model of re-election in Table 3.5. However, the selection test in Table 3.6 is statistically proven that ρ=0. This means there is no evidence of selection bias and the re-election and candidacy equation are independent, making estimation of the selection model unnecessary. In this case, it would be more appropriate to refer the unemployment effect on re-election to the estimation of equation with standard Probit model in Table 3.4 rather than the Heckprobit estimation in Table 3.6, since it will deliver more consistent and unbiased estimates. 103

120 Table 3.6 The impact of unemployment change on probability of re-election: Heckprobit Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect (1) (2) (3) (4) UR *** ** (0.112) (0.038) (0.047) (0.015) Urban 0.007* 0.002* 0.007** 0.002* (0.004) (0.001) (0.004) (0.001) Young 0.073*** 0.022** 0.076*** 0.023** (0.023) (0.009) (0.025) (0.009) Old 0.133** 0.040** 0.131** 0.039** (0.062) (0.020) (0.062) (0.019) LnPop (0.127) (0.040) (0.132) (0.042) Literacy (0.008) (0.003) (0.009) (0.003) Ethnic 0.884** 0.268** 0.853** 0.255** (0.401) (0.133) (0.408) (0.129) Religion ** ** ** ** (0.496) (0.170) (0.466) (0.139) Vturnout (0.012) (0.004) (0.012) (0.004) Ncandidate ** ** ** (0.085) (0.033) (0.075) (0.028) Independent 0.567*** 0.172** 0.617*** 0.184*** (0.207) (0.071) (0.204) (0.067) Java (0.224) (0.070) (0.240) (0.073) Constant (3.082) (3.221) Candidacy Selection Equation UR (0.058) (0.059) Ncandidate ** ** (0.049) (0.049) Newdistrict 0.322** 0.316** (0.142) (0.144) Elect.Cycle *** *** (0.104) (0.104) Constant 0.703*** 0.697*** (0.249) (0.248) 104

121 Table 3.6 (Continued) Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect Model Criteria (1) (2) (3) (4) Total number of obs Censored obs Uncensored obs Log Likelihood Wald Chi 2 (12 df) Prob > Chi Selectivity Test (rho=0) Chi 2 (1 df) Prob > Chi Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is re-elected for another term and 0 otherwise. The results of Table 3.5 indicate that after controlling for selection we observe a positive effect of income per capita growth in all time horizons. However, the effect is quite small: based on the marginal effect of Heckprobit regressions in Table 3.5, an incumbent district leader is 0.3 % (column 4) more likely to win re-election for every extra percentage point of GDP per capita growth in the election year, which is 0.1% lower than the baseline estimation in Table 3.3. Whereas, an extra percentage point of annual average GDP per capita growth within the incumbent s tenure increases the likelihood of incumbent re-election by 0.4%. These results suggest that in terms of GDP per capita growth, voters are evaluating incumbent performance within both the short run (election year) and long run (average within tenure) timeframe, with a slightly larger concern on the long-run performance. 105

122 3.6.2 Competence vs. Luck To test whether the impact of economic conditions on local elections is due to voters rewarding luck or competence, I separate the district economic performance into luck and competence in two ways. First, by using regional districts economic performance, as in equation (3.4) and secondly, by using national economic performance, as in equation (3.5). In the first specification luck is defined as performance due to common regional shocks, while competence is defined as the gap between district performance and average performance of other districts within the same provinces. The simple pooled Probit model estimation of this specification are presented in Tables 3.7 and 3.8. In the second specification luck is defined as national economic performance (performance due to national economy), while competence is defined as the gap between district and national performance. The pooled Probit estimation of the second specification is presented in Tables 3.9 and The likelihood-ratio tests for random effect (Random Effect Test) are reported at the bottom of each table of Table 3.7 to Table These tests formally compare the pooled estimator (Probit) with the panel estimator. The tests in Table 3.7 to Table 3.10 statistically prove that rho is zero, which indicates the panel-level variance component is unimportant, and the panel estimator is not different from the pooled estimator. The simple Probit model estimation of Table 3.7 to Table 3.10 only used election where there is an incumbent candidate running for re-election, since the possibility of re-election is zero if no incumbent is running in the election. 106

123 Table 3.7 Regional luck & competence effect of GDP per capita growth: Probit regression Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect (1) (2) (3) (4) DgYp_regional (0.010) (0.003) (0.006) (0.002) gyp_regional 0.102*** 0.032*** (0.032) (0.010) (0.027) (0.009) Urban 0.007*** 0.002*** 0.008*** 0.003*** (0.003) (0.001) (0.003) (0.001) Young 0.062*** 0.020*** 0.044*** 0.014*** (0.018) (0.005) (0.017) (0.005) Old 0.122*** 0.039*** 0.094* 0.029* (0.047) (0.015) (0.518) (0.016) LnPop (0.094) (0.030) (0.095) (0.030) Literacy (0.008) (0.002) (0.008) (0.003) Ethnic (0.369) (0.116) (0.405) (0.127) Religion (0.417) (0.132) (0.428) (0.132) Vturnout (0.009) (0.003) (0.010) (0.003) Ncandidate *** *** *** *** (0.043) (0.013) (0.047) (0.013) Independent 0.620*** 0.196*** 0.639*** 0.199*** (0.187) (0.057) (0.186) (0.057) Java (0.208) (0.067) (0.200) (0.062) Constant (2.043) (2.147) No. Obs Log L Pseudo-R Random Effect Test (rho=0) Chibar 2 (01) Prob Chibar Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is re-elected for another term and 0 otherwise. 107

124 Table 3.8 Regional luck & competence effect of unemployment change: Probit regression Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect (1) (2) (3) (4) D UR_regional ** ** (0.092) (0.029) (0.048) (0.015) UR_regional ** ** (0.191) (0.059) (0.158) (0.049) Urban 0.007* 0.002* 0.007** 0.002** (0.004) (0.001) (0.003) (0.001) Young 0.066*** 0.020*** 0.076*** 0.023*** (0.025) (0.008) (0.026) (0.008) Old 0.117* 0.037* 0.122* 0.038** (0.061) (0.019) (0.066) (0.020) LnPop (0.134) (0.042) (0.142) (0.044) Literacy (0.008) (0.002) (0.008) (0.003) Ethnic 0.887** 0.276** 0.856** 0.264** (0.397) (0.121) (0.407) (0.124) Religion ** ** ** ** (0.459) (0.139) (0.517) (0.156) Vturnout (0.012) (0.004) (0.013) (0.004) Ncandidate *** *** *** (0.057) (0.016) (0.060) (0.017) Independent *** 0.633*** 0.195*** (0.210) (0.062) (0.201) (0.060) Java (0.210) (0.067) (0.235) (0.074) Constant (3.057) (3.129) No. Obs Log L Pseudo-R Random Effect Test (rho=0) Chibar 2 (01) Prob Chibar Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is re-elected for another term and 0 otherwise. 108

125 Table 3.9 National luck & competence effect of GDP per capita growth: Probit regression Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect (1) (2) (3) (4) DgYp_national * 0.004* (0.010) (0.003) (0.007) (0.002) gyp_national (0.173) (0.055) (0.352) (0.110) Urban 0.007*** 0.002*** 0.008** 0.002*** (0.003) (0.001) (0.003) (0.001) Young 0.052*** 0.017*** 0.037* 0.012* (0.019) (0.006) (0.020) (0.006) Old 0.115** 0.037** (0.048) (0.015) (0.056) (0.018) LnPop (0.098) (0.031) (0.102) (0.031) Literacy (0.008) (0.003) (0.009) (0.003) Ethnic (0.365) (0.116) (0.412) (0.129) Religion (0.412) (0.131) (0.419) (0.130) Vturnout (0.009) (0.003) (0.010) (0.003) Ncandidate *** *** *** *** (0.045) (0.013) (0.046) (0.013) Independent 0.603*** 0.192*** 0.654*** 0.205*** (0.184) (0.056) (0.180) (0.055) Java (0.215) (0.069) (0.208) (0.065) Constant (2.299) (2.567) No. Obs Log L Pseudo-R Random Effect Test (rho=0) Chibar 2 (01) Prob Chibar Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is re-elected for another term and 0 otherwise. 109

126 Table 3.10 National luck & competence effect of unemployment change: Probit regression Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect (1) (2) (3) (4) D UR_national ** ** (0.096) (0.301) (0.500) (0.015) UR_national (0.324) (0.102) (0.242) (0.075) Urban 0.007* 0.002* 0.007** 0.002** (0.004) (0.001) (0.003) (0.001) Young 0.074*** 0.023*** 0.073*** 0.023*** (0.022) (0.007) (0.025) (0.007) Old 0.137** 0.043** 0.127** 0.039** (0.064) (0.019) (0.063) (0.019) LnPop (0.129) (0.041) (0.121) (0.037) Literacy (0.083) (0.003) (0.009) (0.003) Ethnic 0.927** 0.291** 0.829** 0.255** (0.413) (0.127) (0.412) (0.125) Religion ** ** ** ** (0.496) (0.151) (0.531) (0.161) Vturnout (0.012) (0.004) (0.013) (0.004) Ncandidate *** *** *** *** (0.056) (0.017) (0.055) (0.016) Independent 0.570*** 0.179*** 0.665*** 0.205*** (0.209) (0.062) (0.202) (0.060) Java (0.228) (0.073) (0.250) (0.078) Constant (3.055) (2.936) No. Obs Log L Pseudo-R Random Effect Test (rho=0) Chibar 2 (01) Prob Chibar Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is re-elected for another term and 0 otherwise. 110

127 Across these four specifications (Tables ), I find consistent evidence that luck only benefits incumbents when it is defined as regional district performance and within the timeframe of overall performance of the incumbents. I find that a one percentage point rise in annual average of per capita GDP growth of regional districts (gyp_regional) during an incumbent s tenure increases his/her probability of winning office by 3.2 percentage points (column 2 of Table 3.7), while a one percentage point decrease in unemployment of regional districts ( UR_regional) during the same period increases the incumbent s probability of re-election by 13.1 percentage points (column 2 of Table 3.8). However, I do not find any luck variables matter when competence is filtered using national performance (Tables 3.9 and 3.10), neither in term growth (gyp_national) nor unemployment ( UR_ national). Furthermore, I only find evidence of competence effect on re-election in terms of unemployment (D UR_regional) and no evidence of competence in terms of GDP per capita growth (DgYp_regional) when competence is filtered using regional performance (Tables 3.7 and 3.8). Based on the marginal effect of Probit regressions in Table 3.8, an incumbent district leader is 3.5 % (column 4) more likely to win re-election for every extra percentage point of unemployment reduction relative to his/her regional districts in the election year. However, when competence is filtered from luck using national performance I find evidence of competence effect on re-election for both growth and unemployment (Tables 3.9 and 3.10). I find that if district s growth has outpaced national growth (DgYp_national) by one percentage point over the election year, this only raises the incumbent s chances of re-election by 0.4 percentage points (column 4 of Table 3.9), while if district s unemployment decreases a percentage point faster than the decrease in 111

128 national unemployment (D UR_national) over the election year, this increases the incumbent s probability of re-election by 3.8 percentage points (column 4 of Table 3.10). Putting the results of Tables 3.7 to 3.10 together, it seems that voters are only successful in separating incumbent competence from national performance ( luck ) during the election years, yet fail to distinguish incumbent competence from regional performance ( luck ) within an incumbent s overall years of tenure. These results are somewhat reasonable, since voters are more likely to be more informed about the national economy than the regional economy, due to national media penetration, so it is easier for voters to differentiate the national performance from their own district s performance, rather than distinguishing it from regional performance. Moreover, a shorter period of performance (election year) is also more noticeable to voters to evaluate rather than a longer period of performance (average within tenure). However, risk of selection bias in estimation of Tables 3.7 to 3.10 exists because data on re-election are observed only for an incumbent who is running for re-election. To address the issue of selection bias, I use the Heckprobit model to re-estimate the luck & competence effects of growth and unemployment on re-election. The results are presented in Tables 3.11 to The second panel in these tables presents the results of the first stage of the Heckprobit regression (Candidacy Selection Equation). Meanwhile, the significance test for the selection model (Wald test of independent equations) is presented in the last panel of these tables. Overall, the selection test is only significant in the growth model (Tables 3.11 and 3.13) and none of the unemployment models (Tables 3.12 and 3.14). 112

129 Table 3.11 Regional luck & competence effect of GDP per capita growth: Heckprobit Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect (1) (2) (3) (4) DgYp _regional (0.010) (0.003) (0.006) (0.001) gyp_regional 0.087*** 0.025** 0.039* (0.032) (0.010) (0.023) (0.006) Urban 0.006** 0.002** 0.006** 0.002** (0.002) (0.007) (0.003) (0.001) Young 0.051*** 0.014*** 0.028* 0.007* (0.017) (0.005) (0.015) (0.004) Old 0.115** 0.032** 0.084* 0.022* (0.046) (0.013) (0.049) (0.013) LnPop (0.091) (0.026) (0.092) (0.024) Literacy (0.007) (0.002) (0.007) (0.002) Ethnic (0.353) (0.103) (0.372) (0.099) Religion (0.395) (0.112) (0.380) (0.099) Vturnout (0.008) (0.002) (0.009) (0.002) Ncandidate *** *** *** ** (0.054) (0.018) (0.066) (0.020) Independent 0.640*** 0.182*** 0.686*** 0.180*** (0.182) (0.053) (0.178) (0.050) Java (0.199) (0.058) (0.185) (0.049) Constant (2.047) (1.990) 113

130 Table 3.11 (Continued) Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect Candidacy Selection Equation (1) (2) (3) (4) DgYp _regional (0.008) (0.008) Ncandidate *** *** (0.039) (0.045) Newdistrict (0.127) (0.128) Elect.Cycle *** *** (0.136) (0.140) Constant 1.108*** 1.083*** Model Criteria (0.209) (0.238) Total number of obs Censored obs Uncensored obs Log Likelihood Wald Chi 2 (12 df) Prob > Chi Selectivity Test (rho=0) Chi 2 (1 df) Prob > Chi Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is re-elected for another term and 0 otherwise. 114

131 Table 3.12 Regional luck & competence effect of unemployment change: Heckprobit Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect (1) (2) (3) (4) D UR_regional ** ** (0.098) (0.034) (0.046) (0.015) UR_regional ** (0.217) (0.087) (0.160) (0.053) Urban 0.007* 0.002* 0.007** 0.002* (0.004) (0.001) (0.003) (0.001) Young 0.664*** 0.022*** 0.075*** 0.023*** (0.023) (0.008) (0.025) (0.009) Old 0.116* 0.038** 0.123* 0.038* (0.064) (0.019) (0.066) (0.020) LnPop (0.138) (0.044) (0.142) (0.045) Literacy (0.008) (0.002) (0.009) (0.003) Ethnic 0.888** 0.291** 0.855** 0.267** (0.396) (0.136) (0.404) (0.131) Religion ** ** ** ** (0.465) (0.171) (0.503) (0.169) Vturnout (0.012) (0.004) (0.013) (0.004) Ncandidate ** ** *** ** (0.074) (0.031) (0.072) (0.029) Independent 0.594*** 0.195*** 0.634*** 0.198*** (0.213) (0.067) (0.201) (0.065) Java (0.216) (0.076) (0.235) (0.076) Constant (3.004) (3.169) Candidacy Selection Equation D UR_regional (0.055) (0.055) Ncandidate ** ** (0.048) (0.048) Newdistrict 0.330** 0.319** (0.146) (0.149) Elect.Cycle *** *** (0.111) (0.113) Constant 0.789*** 0.789*** (0.254) (0.253) 115

132 Table 3.12 (Continued) Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect Model Criteria (1) (2) (3) (4) Total number of obs Censored obs Uncensored obs Log Likelihood Wald Chi 2 (12 df) Prob > Chi Selectivity Test (rho=0) Chi 2 (1 df) Prob > Chi Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is re-elected for another term and 0 otherwise. Table 3.13 National luck & competence effect of GDP per capita growth: Heckprobit Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect (1) (2) (3) (4) DgYp _national 0.015* 0.004* 0.013** 0.003** (0.009) (0.002) (0.006) (0.002) gyp_national (0.172) (0.046) (0.348) (0.094) Urban 0.006** 0.002** 0.006** 0.002** (0.002) (0.006) (0.003) (0.001) Young 0.040** 0.011** (0.016) (0.004) (0.017) (0.005) Old 0.104** 0.028*** (0.041) (0.010) (0.051) (0.013) LnPop (0.090) (0.024) (0.098) (0.026) Literacy (0.007) (0.002) (0.008) (0.002) Ethnic (0.319) (0.085) (0.380) (0.103) Religion (0.365) (0.097) (0.386) (0.103) Vturnout (0.008) (0.002) (0.009) (0.002) 116

133 Table 3.13 (Continued) Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect (1) (2) (3) (4) Ncandidate ** ** *** ** (0.063) (0.019) (0.067) (0.021) Independent 0.579*** 0.155*** 0.683*** 0.182*** (0.174) (0.049) (0.174) (0.049) Java (0.192) (0.053) (0.194) (0.052) Constant Candidacy Selection Equation (2.120) (2.475) DgYp _national (0.008) (0.009) Ncandidate *** *** (0.038) (0.045) Newdistrict (0.122) (0.127) Elect.Cycle *** *** (0.133) (0.139) Constant 1.114*** 1.090*** Model Criteria (0.205) (0.233) Total number of obs Censored obs Uncensored obs Log Likelihood Wald Chi 2 (12 df) Prob > Chi Selectivity Test (rho=0) Chi 2 (1 df) Prob > Chi Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is re-elected for another term and 0 otherwise. 117

134 Table 3.14 National luck & competence effect of unemployment change: Heckprobit Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect (1) (2) (3) (4) D UR_national *** ** (0.112) (0.049) (0.462) (0.015) UR_national (0.334) (0.108) (0.267) (0.093) Urban 0.007* 0.002* 0.007** 0.002* (0.004) (0.001) (0.003) (0.001) Young 0.074*** 0.024** 0.715*** 0.022** (0.023) (0.009) (0.025) (0.009) Old 0.137** 0.044** 0.127** 0.039** (0.064) (0.022) (0.062) (0.020) LnPop (0.129) (0.042) (0.121) (0.039) Literacy (0.008) (0.003) (0.009) (0.003) Ethnic 0.925** 0.296** 0.826** 0.253* (0.415) (0.147) (0.407) (0.130) Religion ** ** ** ** (0.494) (0.177) (0.474) (0.141) Vturnout (0.012) (0.004) (0.126) (0.004) Ncandidate ** * *** ** (0.084) (0.036) (0.076) (0.030) Independent 0.571*** 0.183** 0.664*** 0.203*** (0.207) (0.074) (0.203) (0.073) Java (0.223) (0.073) (0.248) (0.077) Constant (3.082) (3.090) Candidacy Selection Equation D UR_national (0.060) (0.060) Ncandidate ** ** (0.049) (0.049) Newdistrict 0.328** 0.318** (0.144) (0.146) Elect.Cycle *** *** (0.111) (0.110) Constant 0.718*** 0.717*** (0.249) (0.249) 118

135 Table 3.14 (Continued) Average within tenure At the election year Coef. Marginal effect Coef. Marginal effect Model Criteria (1) (2) (3) (4) Total number of obs Censored obs Uncensored obs Log Likelihood Wald Chi 2 (12 df) Prob > Chi Selectivity Test (rho=0) Chi 2 (1 df) Prob > Chi Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is re-elected for another term and 0 otherwise. All the selection tests in Tables 3.12 and 3.14 statistically prove that ρ=0. This means there is no evidence of selection bias and the re-election and candidacy equation are independent, making estimation of the selection model unnecessary. Thus, it would be appropriate to refer to the standard Probit model estimation in Tables 3.8 and 3.10 for the competence and luck effect of unemployment on re-election. The results of selection equation in Tables 3.11 and 3.13 indicate that, consistent with the baseline estimation, the number of candidates (Ncandidate) and dummy for the second electoral cycle (Elect.Cycle) have a highly significant and negative effect on the probability of an incumbent running for re-election. The significance test for rho in Table 3.11 statistically proves that ρ differs significantly from zero for the election year period (column 3 of Table 3.11). This indicates a problem of selection bias which verifies the need to run a Heckprobit selection model to estimate the competence and luck of the growth effect in the election year on re-election, as presented in Table However, the selection test appears to be insignificant for the average tenure estimation (column 2 of Table 3.11), whereas, in Table 3.13 the selection model verifies the need to run a 119

136 Heckprobit selection model in both time horizons: election year (column 2) and the average within an incumbent s tenure (column 4). The results of Table 3.11 indicate that after controlling for selection and filtering competence from luck by regional districts performance, we observe that neither competence and luck of district s GDP per capita growth at election year matters for re-election. However, the results of Table 3.13 indicate that after controlling for selection and national performance, only competence in district s GDP per capita growth matters for re-election, both in average within an incumbent s tenure and the election year. According to the marginal effect in Table 3.13, an extra 1 percentage point rise of district s annual average GDP per capita growth that outpaced annual average national growth (DgYp_national) increases the likelihood of incumbent re-election by 0.4% (column 2); whereas an extra percentage point of district s growth in the election year that outpaced national growth, increases the likelihood of incumbent re-election by 0.3% (column 4). 3.7 Robustness Analysis The assessment of the causal effect of district economic performance on reelection in the previous section is at risk of endogeneity since these performances are probably not exogenous, but will depend on the policies chosen by incumbents in response to anticipated votes (personal popularity). We can expect that incumbents who anticipate a close race (those with a lower probability of re-election) would devote more effort in improving their district s economic performance in order to win the election. In other words, the estimates in the previous section may suffer from endogeneity problems, since the probability of re-election could also affect economic performance. 120

137 As a robustness check to test for the presence of endogeneity, I estimate IV-Probit where I apply measures of districts revenue, fiscal capacity, regional and national performance as instruments for districts economic performance. These instruments are used because they have direct influence on district economic performance but not directly on re-election and their variations are largely determined by external, rather than internal, factors within districts. I will explain the reasons for using these instruments in the following passages. Most of districts government revenue is composed of central government transfers (nationally it is more than 90% of total districts revenue) which are allocated based on certain formula designed by the central government, which make it impossible for districts to influence. When district revenue is high, more resources are available for that district to improve its economic performance. The fiscal capacity on the other hand measures the proportion of revenue which district governments can flexibly use and allocate. These are composed of district governments own source revenue (PAD) and revenue sharing (DBH), where local government has the authority to allocate, unlike transfers that are largely designated by central government to fund specific spending. The higher the fiscal capacity, the greater is the capacity of local government to stimulate its economy. Meanwhile, we can expect that regional and national economies influence the district economy due to common regional or national shocks, but not the other way around. The internal factors within districts are arguably too small to influence regional and national economy. Furthermore, based on the estimation in the previous section we could, to a certain extent, be convinced that regional and national performance does not affect (or has minimum effect) on the voters decision to re-elect the incumbents. Tables 3.A1 to 3.A6 in the Appendix show the first- and second-stage estimations of IV-Probit for each of the specifications. The first-stage estimation is a linear regression 121

138 of the district economic performance against Fiscal capacity, LGRevenue, regional and national economic performance, controlling for demographic, political and geographic characteristics. The second-stage estimation (IV-Probit) is a Probit model of re-election against a fitted value of district economic performance from the first-stage, and a set-off control variables for districts demographic, political and geographic characteristics. By default, IV-Probit uses maximum likelihood estimation. As documented in Tables 3.A1 to 3.A6 in the Appendix, overall the first stage regression reveals that the instruments which are most frequently significant in influencing re-election is district government revenue (LGRevenue), followed by regional performance. Both instruments have a positive relationship with re-election. Meanwhile, the results of the second stage estimation (IV-Probit) in Tables 3.A1 to 3.A6, will be explained individually in the following passages. In each table I run a test of exogeneity (Wald test of exogeneity). Basically it tests whether correlation coefficient of the error of the first and second regression is statistically different from zero. If the test is statistically significant, we may reject the null hypothesis that the district economic performance variable is exogenous, which lends support to the IV-Probit estimation. If the test is not significant, then we cannot reject the null hypothesis and the multivariate Probit model estimates is more appropriate. The test results are presented in bottom panel of each table. For three of the six tables reported in the Appendix, the effect of district economic performance on re-election is statistically significant. These three tables are: Table 3.A1, 3.A4 and 3.A5. In these tables I also find that the exogeneity test of the instrumented variables are statistically significant, meaning that for these outcomes we can reject the null hypothesis that district economic performance is an exogenous variable. Therefore, the IV-Probit estimates is more appropriate in these three specifications, since it will produce consistent and more efficient estimates than the simple Probit model. 122

139 The IV-Probit estimation of Table 3.A1 finds that in baseline specification, GDP per capita growth (gyp) significantly influences re-election only for the overall period of the incumbent, but not for the election year. The IV-Probit coefficient of growth in this specification is significant at 1% and is recorded at 0.082, which is equal to a marginal effect of 0.2%. This is 50% lower from the marginal effect recorded in Table 3.5 for the same period, which is 0.4%. However, unlike Table 3.5, the IV-Probit estimation does not find statistical evidence of the impact of GDP per capita growth on the election year on re-election. Although the exogeneity test of average tenure specification in Table 3.A1 is significant, the test of the election year specification is not statistically significant, which means a regular Probit regression as in Table 3.3 is more appropriate. The IV-Probit estimation of Table 3.A4 finds that the coefficient of competence element of change in unemployment, filtered by regional performance (D UR_regional), is significant at 1% for the average tenure specification. The coefficient is recorded at 1.008, which is equal to a marginal effect of -0.37%. Unfortunately, it is incomparable since the previous estimation of competence for unemployment within average tenure is always statistically insignificant, but the fact that the marginal effect is negative supports the previous findings. Table 3.A4 also finds that competence effect of unemployment on re-election is statistically significant (at 5%) for election year specification. However, unlike the average tenure specification, the exogeneity test for election year specification is insignificant, meaning there is no sufficient evidence to consider district economic performance as an endogenous variable in this specification. Therefore, the Probit estimates might be consistent and more efficient. In Table 3.A5 the competence coefficients for GDP per capita growth filtered by national performance is positive and significant, for either the average tenure or election year specifications. However, the exogeneity test is significant only in average tenure specification. This means that endogeneity does not seem to be a critical issue for 123

140 election year specification, therefore the Probit estimates might be consistent and more efficient; however, the IV-Probit estimation is more appropriate for the average tenure specification. The coefficient of competence of the average growth within tenure is 0.082, which is statistically significant at 1% and equal to a marginal effect of 0.2%. Consistent with the finding in Table 3.A1, this is 50% slower than the previous estimation in Table Overall, even though the IV-Probit estimation of the district performance effect on re-election is lower than the estimation strategy in the previous section, the main finding is robust: district economic performances do seem to increase the probability of the incumbent s re-election. 3.8 Conclusion The main purpose of the paper is to investigate whether economic conditions affect re-election of incumbent districts leader across Indonesia. Since decentralisation policy was enacted in 2001, Indonesian local governments have a wider range of authority and resources for stimulating local economic activity. In this newly decentralised governance, the local governments are responsible for improving the well-being of the population that resides in their jurisdiction. Furthermore, to strengthen accountability of local governments, starting in 2005 district government heads were selected through direct elections among citizens. Considering this new governance structure, it is fair to expect that incumbent district head will be held accountable for local economic performance. I examine the question using dataset on the first two cycles of Indonesia s local election of district s government leader from 2005 to Overall it includes 848 local direct elections. I estimate a vote function that models the effects of the local economic environment, taking into account the effect of regional and national economy. I start by 124

141 testing whether the re-election probability of incumbent s district head is affected or not by local economic outcomes using simple Probit model and Heckprobit model. I then proceed to test for the importance of attribution error in a model that differentiates incumbent s district head re-election due to competence (district economy ownperformance) and luck (district s performance due to regional or national economy). I also employ IV-Probit model to check whether endogeneity affect re-election outcomes, and as robustness check of the Probit and Heckprobit results. I measure economic performance using GDP per capita growth and change in unemployment In brief, I find that when voters are successfully disentangling competence from luck is the election year. No evidence of attribution error occurs within this time horizon. This is especially true for reduction in unemployment; either it is benchmarked with regional districts or national. I find that one extra percentage point district s unemployment reduction will raise the incumbent s chances of re-election by 3.5 percentage points if it is benchmarked with regional districts and 3.8 percentage points if it is benchmarked against national. Meanwhile, in the case of GDP growth per capita in election year, voters can only successfully separate competence from luck if it is benchmarked against national performance. I find that one extra percentage point district s GDP per capita over the national economy will raise the incumbent s chances of re-election by 0.3 percentage points. However, voters overall failed to vote based on competence if performance is described as annual average change of the economy during incumbent tenure. The only case where competence matters is on GDP per capita growth benchmarked by national performance. An extra percentage point of GDP per capita annual average growth of district from national, will raise the incumbent s chances of re-election by 0.4 percentage points. The evidence of attribution error is also found in this time horizon, both for GDP 125

142 per capita growth and change in unemployment, but only when district performance is benchmarked against regional districts. Putting the results altogether it appears that voters ability to filter incumbent district leader competence from luck are more accurate for the performance in the election year rather than performance for the whole period of incumbent tenure. This might suggest that voters do not take into account the full extent of an incumbent s economic record. Instead, voters put attention on performance more in the last year rather than on the whole period of an incumbent (Achen and Bartels 2004). It implies that performance in the last year of tenure matters most for incumbent probability for re-election, which lends support on the recent empirical finding on the existence of political budget cycle in Indonesia in the local direct elections periods, especially if the incumbent runs for reelection (Sjahrir et al. 2013). 126

143 Appendix Table 3.A1 IV-Probit base regression income per capita growth Average within tenure At the election year First stage IV-Probit First stage IV-Probit (1) (2) (3) (4) gyp 0.082*** (0.029) (0.026) Urban ** * (0.010) (0.003) (0.017) (0.003) Young 0.173*** *** (0.061) (0.018) (0.090) (0.017) Old 0.246* ** (0.149) (0.045) (0.310) (0.048) LnPop 2.580*** *** (0.721) (0.096) (0.789) (0.099) Literacy (0.026) (0.009) (0.098) (0.009) Ethnic (1.605) (0.409) (2.559) (0.418) Religion (2.524) (0.509) (5.265) (0.488) Vturnout (0.019) (0.001) (0.035) (0.010) Ncandidate *** *** (0.225) (0.046) (0.318) (0.047) Independent 0.911* 0.685*** *** (0.551) (0.238) (1.060) (0.210) Java (0.723) (0.196) (1.595) (0.194) Elect.Cycle * *** (0.961) (0.148) (0.846) (0.155) Fiscal capacity (0.023) (0.024) LGRevenue 3.495*** 3.212*** (1.050) (1.105) gyp_regional 0.931*** 1.071*** (0.247) (0.232) gyp_national (0.982) (2.901) Constant *** *** (20.193) (2.085) (21.186) (2.100) 127

144 Table 3.A1 (Continued) Average within tenure At the election year First stage IV-Probit First stage IV-Probit Model Criteria (1) (2) (3) (4) Total number of obs Log Likelihood Wald Chi 2 (13 df) Prob > Chi Exogeneity Test (corr=0) Chi 2 (1 df) Prob > Chi Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is reelected for another term and 0 otherwise. Table 3.A2 IV-Probit base regression change in unemployment Average within tenure At the election year First stage IV-Probit First stage IV-Probit (1) (2) (3) (4) UR (0.235) (0.167) Urban *** * (0.005) (0.006) (0.008) (0.005) Young * * (0.012) (0.023) (0.026) (0.024) Old (0.056) (0.070) (0.075) (0.070) LnPop ** (0.324) (0.156) (0.403) (0.164) Literacy (0.009) (0.012) (0.009) (0.014) Ethnic ** ** (0.335) (0.420) (0.692) (0.450) Religion ** ** (0.385) (0.452) (0.727) (0.510) Vturnout (0.009) (0.013) (0.011) (0.014) Ncandidate *** * *** (0.037) (0.063) (0.049) (0.060) Independent ** *** (0.835) (0.236) (0.206) (0.229) 128

145 Table 3.A2 (Continued) Average within tenure At the election year First stage IV-Probit First stage IV-Probit (1) (2) (3) (4) Java (0.262) (0.366) (0.369) (0.360) Elect.Cycle (0.385) (0.231) (0.208) (0.215) Fiscal capacity (0.004) (0.006) LGRevenue ** (0.331) (0.399) UR_regional 1.076*** 0.980*** (0.123) (0.114) UR_national (0.748) (1.510) Constant ** (9.289) (3.513) (11.079) (3.744) Model Criteria Total number of obs Log Likelihood Wald Chi 2 (13 df) Prob > Chi Exogeneity Test (corr=0) Chi 2 (1 df) Prob > Chi Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is reelected for another term and 0 otherwise. 129

146 Table 3.A3 IV-Probit relative performance to income per capita growth of regional districts Average within tenure At the election year First stage IV-Probit First stage IV-Probit (1) (2) (3) (4) DgYp _regional (0.078) (0.057) gyp_regional * (0.251) (0.039) (0.224) (0.025) Urban ** (0.010) (0.003) (0.017) (0.004) Young *** *** (0.061) (0.031) (0.090) (0.025) Old 0.248* (0.147) (0.060) (0.312) (0. 058) LnPop 2.547*** *** (0.790) (0.095) (1.731) (0.095) Literacy * (0.258) (0.009) (0.098) (0.009) Ethnic (1.611) (0.405) (2.531) (0.425) Religion (2.630) (0.589) (5.217) (0.556) Vturnout (0.019) (0.009) (0.035) (0.009) Ncandidate *** *** (0.224) (0.051) (0.321) (0.075) Independent 0.898* 0.658** ** (0.532) (0.318) (1.070) (0.311) Java (0.729) (0.177) (1.598) (0.198) Elect.Cycle * *** (0.969) (2.880) (0.149) Fiscal capacity (0.024) (0.020) LGRevenue 3.443*** 3.273*** (1.152) (0.914) gyp_national (1.125) (3.240) Constant *** *** (21.655) (2.299) (23.676) (2.071) 130

147 Table 3.A3 (Continued) Average within tenure At the election year First stage IV-Probit First stage IV-Probit Model Criteria (1) (2) (3) (4) Total number of obs Log Likelihood Wald Chi 2 (14 df) Prob > Chi Exogeneity Test (corr=0) Chi 2 (1 df) Prob > Chi Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is reelected for another term and 0 otherwise. Table 3.A4 IV-Probit relative performance to regional districts in unemployment change Average within tenure At the election year First stage IV-Probit First stage IV-Probit (1) (2) (3) (4) D UR_regional *** ** (0.109) (0.214) UR_regional (0.121) (0.122) (0.113) (0.110) Urban *** 0.014*** ** (0.005) (0.004) (0.008) (0.005) Young (0.016) (0.016) (0.025) (0.032) Old (0.049) (0.048) (0.072) (0.076) LnPop ** (0.155) (0.151) (0.373) (0.159) Literacy (0.009) (0.009) (0.009) (0.008) Ethnic (0.009) (0.324) (0.642) (0.559) Religion * (0.436) (0.452) (0.757) (0.558) Vturnout (0.008) (0.008) (0.011) (0.013) Ncandidate * (0.035) (0.034) (0.046) (0.098) 131

148 Table 3.A4 (Continued) Average within tenure At the election year First stage IV-Probit First stage IV-Probit (1) (2) (3) (4) Independent (0.164) (0.160) (0.204) (0.304) Java (0.230) (0.240) (0.379) (0.287) Elect.Cycle ** 0.276** (0.121) (0.114) (0.194) (0.221) Fiscal capacity (0.000) (0.007) LGRevenue ** (0.009) (0.350) UR_national (0.022) (1.005) Constant ** (2.897) (2.883) (8.503) (2.868) Model Criteria Total number of obs Log Likelihood Wald Chi 2 (14 df) Prob > Chi Exogeneity Test (corr=0) Chi 2 (1 df) Prob > Chi Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is reelected for another term and 0 otherwise. 132

149 Table 3.A5 IV-Probit relative performance to national income per capita growth Average within tenure At the election year First stage IV-Probit First stage IV-Probit (1) (2) (3) (4) DgYp _national 0.082*** 0.046* (0.027) (0.025) gyp_national *** * (0.997) (0.233) (2.796) (0.393) Urban ** (0.010) (0.003) (0.017) (0.003) Young 0.172*** *** (0.060) (0.019) (0.090) (0.018) Old ** (0.147) (0.050) (0.311) (0.052) LnPop 2.568*** *** (0.718) (0.100) (0.787) (0.099) Literacy (0.026) (0.010) (0.098) (0.010) Ethnic (1.606) (0.411) (2.556) (0.419) Religion (2.523) (0.503) (5.248) (0.489) Vturnout (0.019) (0.009) (0.035) (0.010) Ncandidate *** *** (0.225) (0.047) (0.319) (0.047) Independent 0.945* 0.646*** *** (0.550) (0.236) (1.061) (0.206) Java (0.718) (0.199) (1.596) (0.195) Elect.Cycle ** ** (0.969) (0.165) (0.822) (0.168) Fiscal capacity (0.023) (0.024) LGRevenue 3.490*** 3.245*** (1.053) (1.095) gyp_regional 0.930*** 1.063*** (0.247) (0.332) Constant *** *** (20.256) (2.422) (21.715) (2.400) 133

150 Table 3.A5 (Continued) Average within tenure At the election year First stage IV-Probit First stage IV-Probit Model Criteria (1) (2) (3) (4) Total number of obs Log Likelihood Wald Chi 2 (14 df) Prob > Chi Exogeneity Test (corr=0) Chi 2 (1 df) Prob > Chi Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is reelected for another term and 0 otherwise. Table 3.A6 IV-Probit relative performance to national unemployment change Average within tenure At the election year First stage IV-Probit First stage IV-Probit (1) (2) (3) (4) D UR_national (0.236) (0.163) UR_national (0.701) (1.653) (1.455) (1.209) Urban *** * (0.005) (0.006) (0.008) (0.005) Young * (0.015) (0.024) (0.026) (0.026) Old (0.055) (0.070) (0.074) (0.071) LnPop ** (0.329) (0.158) (0.402) (0.167) Literacy (0.009) (0.012) (0.009) (0.014) Ethnic ** ** (0.330) (0.402) (0.688) (0.438) Religion ** ** (0.386) (0.468) (0.721) (0.514) Vturnout (0.009) (0.013) (0.011) (0.014) Ncandidate *** * *** (0.037) (0.065) (0.049) (0.062) 134

151 Table 3.A6 (Continued) Average within tenure At the election year First stage IV-Probit First stage IV-Probit (1) (2) (3) (4) Independent ** *** (0.183) (0.236) (0.207) (0.231) Java (0.263) (0.372) (0.370) (0.363) Elect.Cycle (0.362) (0.883) (0.204) (0.216) Fiscal capacity * (0.004) (0.006) LGRevenue ** (0.336) (0.393) UR_regional 1.078*** 0.978*** (0.123) (0.115) Constant ** (9.446) (3.647) (10.911) (3.967) Model Criteria Total number of obs Log Likelihood Wald Chi 2 (14 df) Prob > Chi Exogeneity Test (corr=0) Chi 2 (1 df) Prob > Chi Notes: Standard errors clustered by province are in parentheses. Significance level at which the null hypothesis is rejected: ***, 1%; **, 5%; and *, 10%. Dependent variable is Reelect, a dummy variable that takes value 1 when a district head is reelected for another term and 0 otherwise. 135

152 References Achen, C. H. and L. M. Bartels. (2004). Blind Retrospection: Electoral Responses to Drought, Flu, and Shark Attacks. Working Paper, Princeton University. Auberger, A., & Dubois, E. (2005). The influence of local and national economic conditions on French legislative elections. Public Choice, 125(3-4), Anderson, C. J. (2007). The end of economic voting? Contingency dilemmas and the limits of democratic accountability. Annu. Rev. Polit. Sci., 10, Anderson, C. (2006). Economic voting and multilevel governance: a comparative individual-level analysis. American Journal of Political Science, 50(2), Ashworth, S. (2005). Reputational dynamics and political careers. Journal of Law, Economics, and Organization, 21(2), Barro, R. J. (1973). The control of politicians: an economic model. Public choice, 14(1), Barro, R. J. (1999). Determinants of democracy. Journal of Political economy, 107(S6), S158- S183. Bertrand, M., & Mullainathan, S. (2001). Are CEOs rewarded for luck? The ones without principals are. Quarterly Journal of Economics, Boyne, G. A., James, O., John, P., & Petrovsky, N. (2009). Democracy and government performance: Holding incumbents accountable in English local governments. The Journal of Politics, 71(04), Berry, C. R., & Howell, W. G. (2007). Accountability and local elections: Rethinking retrospective voting. Journal of Politics, 69(3), Besley, T., & Smart, M. (2007). Fiscal restraints and voter welfare. Journal of Public Economics, 91(3), Besley, T., & Case, A. (1995). Incumbent behavior: vote seeking, tax setting and yardstick competition. American Economic Review, 85(1), Calavan, M., Zuhro, S., Setiyono, B., Connerley, E., & Ashari, L. (2009). Democratic Decentralization Strategic Assessment: Indonesia. United States Agency for International Development (USAID). Castro, V., & Martins, R. (2013). Running for office again: evidence from Portuguese municipal elections. Public Choice, 156(3-4), Cutler, F. (2002). Local economies, local policy impacts and federal electoral behaviour in Canada. Canadian Journal of Political Science/Revue canadienne de science politique, 35(02), Dassonneville, R., & Lewis-Beck, M. S. (2014). The economic voter and economic crisis. Acta Politica, 49(4), Duch, R. M., & Stevenson, R. T. (2008). The economic vote: How political and economic institutions condition election results. Cambridge University Press. 136

153 Duch, R. M., & Stevenson, R. (2010). The global economy, competency, and the economic vote. The Journal of Politics, 72(01), Fearon, J. D. (1999). Electoral accountability and the control of politicians: Selecting good types versus sanctioning poor performance. In Democracy, Accountability, and Representation, ed. Adam Przeworski, Susan C. Stokes and Bernard Manin. New York: Cambridge University Press pp Ferejohn, J. (1986). Incumbent performance and electoral control. Public choice, 50(1), Fiorina, M. P Retrospective voting in American national elections. New Haven: Yale University Press. Gélineau, Francois Presidents, Political Context, and Economic Accountability Evidence from Latin America. Political Research Quarterly 60 (3): Gélineau, F. (2007). Presidents, Political Context, and Economic Accountability Evidence from Latin America. Political Research Quarterly, 60(3), Greene, W. H. (2003). Econometric analysis. Upper Saddle River, NJ: Prentice Hall. Hellwig, T., & Marinova, D. M. (2015). More Misinformed than Myopic: Economic Retrospections and the Voter s Time Horizon. Political Behavior, 37(4), Happy, J. R. (1992). The effect of economic and fiscal performance on incumbency voting: The Canadian case. British Journal of Political Science, 22(01), Hayes, R. C., Imai, M., & Shelton, C. A. (2015). Attribution Error in Economic Voting: Evidence from Trade Shocks. Economic Inquiry, 53(1), Heckman, J. J. (1979), Sample Selection Bias as a Specification Error. Econometrica, 47(1), Jérôme, B., & Jérôme-Speziari, V. (2005). The 2004 French regional elections: politico-economic factors of a nationalized local ballot. French Politics, 3(2), Key, V. O The responsible electorate: rationality in presidential voting, Cambridge, MA: Belknap Press. Leigh, A. (2009). Does the World Economy Swing National Elections? Oxford Bulletin of Economics and statistics, 71(2), Leigh, A. and M. McLeish. (2009). Are State Elections Affected by the National Economy? Evidence from Australia. Economic Record, 85 (269), Leigh, A., & McLeish, M. (2009). Are State Elections Affected by the National Economy? Evidence from Australia. Economic Record, 85(269), Lele, G. (2012). The Paradox of Distance in Decentralized Indonesia. Jurnal Ilmu Sosial dan Ilmu Politik, 15(3), Lewis-Beck, M. S., & Paldam, M. (2000). Economic voting: an introduction. Electoral studies, 19(2), Lewis-Beck, M. S., & Stegmaier, M. (2000). Economic determinants of electoral outcomes. Annual Review of Political Science, 3(1),

154 Lewis-Beck, M. S., & Stegmaier, M. (2007). Economic models of voting. In The Oxford Handbook of Political Behavior, edited by Russell Dalton and Hans-Dieter Klingemann, Oxford: Oxford Univ. Press. Lewis, B. D. (2010). Indonesian decentralization: Accountability deferred. International Journal of Public Administration, 33(12-13), Lewis, B. D., & Pattinasarany, D. (2009). Determining citizen satisfaction with local public education in Indonesia: The significance of actual service quality and governance conditions. Growth and Change, 40(1), McCulloch, N. (2011). The Indonesian Sub-National Growth and Governance Dataset- Documentation. Available online at Mietzner, M. (2005). Local democracy. Inside Indonesia, 85, Mietzner, M. (2010). Indonesia s Direct Elections: Empowering the Electorate or Entrenching the New Order Oligarchy? In Edward Aspinall and Greg Fealy, eds. Soeharto s New Order and its Legacy, eds. Canberra: ANU E Press Nadeau, R., & Blais, A. (1993). Explaining election outcomes in Canada: economy and politics. Canadian Journal of Political Science, 26(04), Oliver, J. E., & Ha, S. E. (2007). Vote choice in suburban elections. American Political Science Review, 101(03), Padró i Miquel, G., & Snyder, J. M. (2006). Legislative effectiveness and legislative careers. Legislative Studies Quarterly, 31(3), Patty, J. W., & Weber, R. A. (2007). Letting the good times roll: A theory of voter inference and experimental evidence. Public Choice, 130(3-4), Persson, T., & Tabellini, G. E. (2000). Political economics: explaining economic policy. MIT press. Powell Jr, G. B., & Whitten, G. D. (1993). A cross-national analysis of economic voting: taking account of the political context. American Journal of Political Science, Pastore, F. (2012). To study or to work? Education and labor-market participation of young people in Poland. Eastern European Economics, 50(3), Sakurai, S. N., & Menezes-Filho, N. A. (2008). Fiscal policy and reelection in Brazilian municipalities. Public Choice, 137(1-2), Sakurai, S. N., & Menezes-Filho, N. (2011). Opportunistic and partisan election cycles in Brazil: new evidence at the municipal level. Public Choice, 148(1-2), Salmon, P. (1987). Decentralisation as an incentive scheme. Oxford review of economic policy, 3(2), Sanders, D. (2005). The political economy of UK party support, : forecasts for the 2005 general election. Journal of Elections, Public Opinion & Parties, 15(1), Singer, M. M. (2013). Economic voting in an era of non-crisis: the changing electoral agenda in Latin America, Comparative Politics, 45(2),

155 Sjahrir, B. S., Kis-Katos, K., & Schulze, G. G. (2013). Political budget cycles in Indonesia at the district level. Economics Letters, 120(2), Sjahrir, B. S., Kis-Katos, K., & Schulze, G. G. (2014). Administrative overspending in Indonesian districts: The role of local politics. World Development, 59, Skoufias, E., Narayan, A., Dasgupta, B., & Kaiser, K. (2014). Electoral accountability and local government spending in Indonesia. World Bank Policy Research Working Paper, (6782). Whitten, G. D., & Palmer, H. D. (1999). Cross-national analyses of economic voting. Electoral Studies, 18(1), Wolfers, J. (2007). Are voters rational? Evidence from gubernatorial elections. Mimeo, University of Pennsylvania. World Bank. (2008). Spending for Development: Making the Most of Indonesia's New Opportunities. World Bank, Jakarta. Wooldridge, J. (2002). Econometric analysis of cross section and panel data. Cambridge: MIT Press. Van der Brug, W., Van der Eijk, C., & Franklin, M. (2007). The economy and the vote: Economic conditions and elections in fifteen countries. New York: Cambridge University Press. Van de Ven, W. P., & Van Praag, B. M. (1981). The demand for deductibles in private health insurance: A probit model with sample selection. Journal of econometrics, 17(2), Veiga, F., & Veiga, L. (2007). Political business cycles at municipal level. Public Choice, 131(1 2), Vermeir, J., & Heyndels, B. (2006). Tax policy and yardstick voting in Flemish municipal elections. Applied Economics, 38(19),

156 Chapter 4: Intergovernmental Transfer and Voter Turnout in Indonesia s Local Election 4.1 Introduction Voter turnout has received substantial attention within political economy and political science literature. Turnout is generally thought to be an indicator of levels of civic engagement, social connectedness and trust in a society (Putnam 2000). By turning out to vote, people can send signals to government about their preferences and problems. Hence, high turnout levels ensure that political leadership is representative of the majority of citizens and not only of elites (Lijphart 1997). Moreover, since voting serves as a primary tool for citizens to control their governments and hold them accountable, a higher political involvement in the form of voter turnout is also argued to enhance public sector performance (Geys et al. 2010). Not surprisingly, there is a large amount of literature which studies the role of a host of factors on turnout (see Geys (2006b) for a review). In the context of decentralisation, voter turnout is considered as a critical element in ensuring its success (Litvack et al. 1998). Fiscal decentralisation transfers resources into the hands of local governments, which logically, are better informed about the preferences and circumstances of the residents than the central government. Thus, it is claimed that fiscal decentralisation potentially improves allocative efficiency by allowing greater differentiation in the provision of public goods and services (Oates 1972). Correspondingly, political decentralisation in the form of local elections is intended to increase the accountability of local government to its residents (Seabright 1996). It is expected that people tend to be more aware of the actions of local governments than they are of the actions of the central government because local governments are closer to their citizens (Shah 2006). 140

157 However, decentralisation also puts democracy at risk of being captured by local elites who have a special interest. This is because at the local level, collusion is easier to establish and maintain across different interest groups, since the transaction costs and information asymmetry are lower due to greater proximity (Bardhan 2002; Prud'homme 1995). In this context, obviously, citizen participation in local elections is an important condition that ensures decentralisation success in establishing responsive and accountable local governments (Litvack et al. 1998). A higher turnout may give local politicians incentives to implement policies that benefit the electorate at large, at the expense of policies benefiting their own special interests. Hence, a high degree of democratic participation in terms of voter turnout in local elections may reduce the risk of decentralisation through more efficient monitoring of local politicians. In this chapter I empirically analyse the types and design of intergovernmental transfers as one of the elements that influence political involvement of voters in local elections. Specifically, I investigate the effect of different types of transfers on the level of voter turnout in Indonesia s local elections. The reason for this focus is that there have been theoretical predictions (Devarajan et al. 2009) and empirical findings (Geys et al. 2010) in the literature that have documented citizen vigilance in monitoring how local governments spend public funds is influenced by how local governments finance their expenditures. Citizen is likely to be more vigilance on the careful use of public money when these public funds originate from own revenue sources rather than external transfers (Devarajan et al. 2009). Empirically, Geys et al. (2010) find that local government that are fiscally independent on external funds has better cost efficiency. Moreover, Indonesia represents an interesting case for studying local government elections because it underwent a major process of fiscal and political decentralisation reforms in 1999 with the enactment of Law 22/

158 Indonesia s fiscal decentralisation reform was effectively implemented in 2001, when transfer responsibilities for almost all governmental functions moved from central government directly to the level of district governments. Only the governmental functions of defence, security, justice, foreign affairs, fiscal policy and religion remained in the hands of the Central government. The law also mandates that those decentralised functions must be accompanied by intergovernmental transfers to finance those functions. Furthermore, the political decentralisation reform started in 2005, after the revised decentralisation law on regional autonomy (Law 32/2004) prescribed that the heads of local governments should be directly elected by the local population (PILKADA). Previously, district heads were indirectly elected by the local parliaments. It was believed that this democratic reform would make the district heads more accountable to their electorates (Kaiser et al. 2006). Using a panel data-set of 497 district governments in Indonesia for the first two cycles of local government direct elections over the period , I have documented the types of intergovernmental transfers which gave fiscal flexibility to local governments in allocating their spending, increased voter turnout in local elections. Quantitatively, my estimates suggest that a 10 percentage point increase in the share of revenue sharing (Dana Bagi Hasil/DBH) - the most unrestricted type of transfer - to total district government expenditure increases voter turnout with the range of approximation percentage points. I also re-estimate the model using a more complete fiscal flexibility measure and find similar results. To address for endogeneity bias, I extend the analysis by using revenue sharing in oil (DBH oil) which is a component of revenue sharing (DBH). I argue that using revenue sharing in oil (DBH oil) could reduce endogeneity bias since oil endowment is more likely to be determined by geological factors. I find that a 10 percentage point increase in share of oil revenue (DBH oil) to total district government expenditure increases voter turnout 142

159 with most credible estimates ranging from percentage points across different specifications and estimators. These estimates are higher than previous estimations which indicate that the endogeneity problem may have downwardly biased previous estimates. As a robustness check, I also estimate Two Stage Least Squares (2SLS) models, using variation in share of oil sectors in districts GDP as an instrument for share of revenue sharing in oil (DBH oil) to total district government expenditure. I argue that size and share of oil sectors in the district economy is exogenous to district characteristics that influence voters behaviours, since share of oil sectors in district economy is more likely determined by oil endowment, which is exogenously determined by geological factors. In the most credible results for the 2SLS estimations I find that a 10 percentage point increase in share of revenue sharing in oil (DBH oil) to total district government expenditure increases voter turnout by 1.05 percentage points. To facilitate the interpretation of these findings I explore how fiscal flexibility in spending, proxied by revenue sharing in oil (DBH oil), influences corruption at the district level. I contend that reduction of corruption practice is an indicator of better governance and accountability. Using an index derived from firm survey data to construct a proxy measure of corruption at the district level, I find evidence that higher fiscal flexibility in spending diminishes corruption practices. Hence, I argue that the increase of voter turnout stimulated by district governments fiscal flexibility in spending is most likely due to a stronger political accountability from a genuine increase in civic engagement of the electorates, rather than clientelistic politics using local budget spending flexibility to engage in pork-barrel targeted spending on poor electorates. The findings of this research contribute to the literature on voter turnout and decentralisation in several aspects. First, this study adds to the literature on voter turnout in developing economies. Previous research on electoral participation has paid little 143

160 attention to turnout in developing countries. Previous studies in the literature of voter turnout typically used data from mature and developed democratic countries (Blais 2006; Geys 2006a, 2006b), whereas in this study I utilise local election data from Indonesia which is a newly democratised and decentralised developing country. Second, this research contributes to the small and growing literature on the impact of decentralisation on voter turnout (Akramov et al. 2008, Blais et al and Michelsen et al. 2014). I depart from these studies by specifically examining the impact of different types and design of intergovernmental transfer on voter turnout, focusing on the issue of fiscal flexibility, rather than using the broad measure of decentralisation. Third, this work contributes to the literature on institutional frameworks that determine the consequences of decentralisation, particularly in literature which argues that accountability is higher in local governments that rely less heavily on central government transfers (Geys et al. 2010). Moreover, these studies contend that financing expenditures with local taxes leads to greater accountability (Shah 2004, Bahl and Martinez-Vazquez 2013, and Weingast 2009). The reason is that local governments which depend on central government transfers in financing their functions diminish the interest of voters in holding the local politicians accountable for their decisions (Devarajan et al. 2009). This study contributes to the literature by presenting evidence that central government transfers could, on the contrary, increase voters interest in accountability which is shown by their turnout in elections; as long as the transfer is designed to give local government fiscal flexibility in its spending allocation (policy autonomy). The remainder of the chapter is organised as follows. Section 2 reviews the theoretical considerations and related literature on voter turnout in the context of decentralisation reforms. Section 3 presents an overview of fiscal decentralisation in Indonesian and a brief explanation on different types of intergovernmental transfers 144

161 available. Section 4 presents an overview of Indonesia s local election. In Section 5, I introduce the empirical strategy and the data-set. Section 6 presents the main results and Section 7 explores the robustness of these results. Section 8 discusses the findings and their implications. Section 9 concludes. 4.2 Literature Review Theoretical consideration From a theoretical perspective, voter turnout in elections can be analysed with a classical rational voter theory (Downs 1957), where the net benefit of voting R is: R = ρb C (4.1) From this rational choice perspective, a voter would vote rather than abstain if, and only if, the net benefit of voting is positive (R = ρb C > 0). The net benefit of voting R, depends on B, the difference in benefit to the voter of one or the other candidate winning in the election of local governments multiplied by ρ, the probability of being pivotal and C, the cost associated with voting (Downs 1957; Riker and Ordeshook 1968). There are two possible interpretations of ρ in the literature. Downs (1957) interprets ρ as the probability of being pivotal in the election, whereas Matsusaka (1995) interprets ρ as the probability of being pivotal in the political decision-making process. In Downs s (1957) framework, this ρ is argued to be an inverse function of the electorate s size and a direct function of the election s closeness (Beck 1975, Hansen et al and Dhillon and Peralta 2002); whereas in Matsusaka s (1995) framework ρ is more likely to be affected by the institutional policy-making design (Michelsen et al. 2014). How does intergovernmental transfer contribute to voter turnout in local government elections? Intuitively, people are more prone to voting when the government to be elected has more power or authority (Reif and Schmitt 1980). In the context of intergovernmental transfers, the degree of power or authority is reflected by the extent to 145

162 which local governments are free to allocate the transfer on spending. The higher the proportion of unconditional transfer received by a local government to its total expenditure, the higher is the local government power and authority in policy making due to higher fiscal flexibility in spending. In the framework of equation (4.1) this insight is reflected in ρ, the probability of a vote being pivotal on public policy formulation (Matsusaka 1995). Therefore, a higher district government fiscal flexibility, either due to an increase in a share of unconditional transfers to its total expenditure or due to an increase in share of local tax (or other own-source revenue), increases the local election pivotalness (ρ) of the districts, hence an increase in turnout Relation to the literature This research is related to several strands of literature. First, it is connected to the empirical literature on the impact of decentralisation. Previous studies on decentralisation have predominantly focused on the effects of decentralisation on a variety of public sector outputs, such as infrastructure, education and health (Faguet 2014). Nevertheless, there are a growing number of studies that investigate decentralisation s effects on the quality of governance (see Martinez-Vazquez et al for comprehensive review). Within this literature, there is a small branch of studies that examines the impact of decentralisation on voter turnout at local elections (Akramov et al. 2008, Blais et al. 2011, Michelsen et al. 2014). Blais et al. (2011) studied the impact of decentralisation on turnout in Canada and Spain and found that decentralisation increases turnout rates in regional elections and reduces the differences in turnout between regional and national elections in both countries. Akramov et al. (2008) analysed voter turnout in local government elections in Pakistan using survey data of more than 3,500 voters after the elections in They found that the introduction of direct elections of the district nazims a key position in 146

163 local government might improve electoral participation and thus create a precondition for better local government accountability. Michelsen et al. (2014), using a dataset on local-level elections in Germany, exploited the institutional variation across municipalities to analyse the relation between the institutional design of local public goods provision and voter turnout. They found that decentralisation of local public good provision drives voter turnout. My study contributes to this branch of literature by specifically investigating the impact of different types of intergovernmental transfer design on voter turnout in local elections in Indonesia. This differs from previous literature which used broad measure of fiscal decentralization. Second, this work relates to the literature on the relationship between the degree of expenditure/revenue decentralisation and local government accountability. A majority of previous literature on decentralisation assume that both an expenditure and revenue collection is of the same degree. However, that is not always the case. In fact, in most cases decentralisation is partial, meaning that although local governments are given broad expenditure responsibility, revenue collection/taxation matters remain the responsibility of the Central government. This type of decentralisation is described as partial fiscal decentralisation (Brueckner 2009). Theoretically and conceptually, the impact of different degrees of revenue autonomy on governance and accountability has been investigated by Peralta (2006, 2011), Devarajan et al. (2009) and Weingast (2009). In her theoretical model, Peralta (2006, 2011) defines full decentralisation as a condition where the local public goods are funded with local taxes set by the local government, and partial decentralisation is defined as a condition where the local public goods are funded using transfers from the central government. Peralta (2011) demonstrates that one should observe more turnovers in a fully decentralised regime, hence improve accountability by voting out bad incumbents. Devarajan et al. (2009) conceptually predict that when a local government depends on central government 147

164 transfers in providing public goods and services, it is more likely that the distribution is biased towards serving narrow interest groups or extracting rents. In a partially decentralised setting, local governments could easily blame insufficient transfers from higher tiers of government for poor quality of public services. Weingast (2009) conceptually emphasises the critical importance of local government revenue generation in reducing corruption, ensuring local government responsiveness, and establishing political independence with the centre. Nevertheless, the empirical literature does not always support the theoretical prediction. Using a panel data-set of 987 German municipalities Geys et al. (2010) documented that the voters demand for efficient utilisation of public funds is higher in municipalities that rely less heavily on central government transfers. On the contrary, de Mello and Barenstein (2001), using cross-country data for up to 78 countries, showed that the relationship between decentralisation and governance depends on how subnational expenditures are financed. The higher the share in total local revenues of grants and transfers from higher levels of government, the stronger the positive association is between decentralisation and governance. Instead of examining the role of revenue or tax autonomy on accountability and governance, I contribute to this strand of literature by exploring the impact of local government fiscal flexibility in spending allocation on voter turnout in local elections, which I argue reflects voter interest in accountability at the local level. I further extend the analysis by investigating the direct impact of fiscal flexibility on corruption at the district government level, which I argue represents governance quality of local government. As a further contribution, this study utilises a data-set on Indonesia, a developing country which recently pursued democratisation and decentralisation reform at the beginning of the 21 st century. To the best of my knowledge this is the first study that examines voter turnout at district leader elections in Indonesia. 148

165 4.3 Fiscal Decentralization and Intergovernmental Transfers in Indonesia The Indonesian government structure is composed of several administrative levels: central government, provinces, districts, sub-districts and villages. Since 1966, under the Suharto autocratic regime, the country was highly centralised. Most of the power was retained by the centre. In 1998, following the Asian financial crisis that severely affected the country in 1997, the Suharto government fell. The new government was pressured by regions rich in natural resources, demanding greater transfers and autonomy. In fact, the threat toward territorial disintegration was escalating in these regions. Responding to this threat of territorial disintegration, in 2001, the government undertook a massive decentralisation reform and redistributed most of the power to district governments. The central government authority was reduced to cover only defence, religion, justice, foreign affairs and macroeconomic policy (Alm et al. 2001). Fiscal transfers from central to regional governments almost doubled from 14.9 percent of total government expenditure in 2000 to 23.7 percent in Relative to GDP, fiscal transfers actually doubled from 2.4 percent of GDP in 2000 to 4.8 percent of GDP in This occurred because laws mandated that the transferred functions should be accompanied by sufficient intergovernmental transfers to ensure that local governments had enough resources to deliver these new functions. According to the decentralisation laws (Laws 22/1999 and 25/1999) local government revenues come from three sources, which includes: Own-source revenues (Pendapatan Asli Daerah/PAD): locally-raised revenues collected, based on local government regulation in implementation of decentralisation (local taxes e.g. restaurant and hotel taxes, motor vehicle taxes, 149

166 entertainment taxes and retribution taxes e.g. parking retribution, traditional market retribution). Balancing funds: an intergovernmental transfer from the national budget aimed to equalise fiscal capacity across districts, subject to their needs and capacity, in order to provide basic public services. Other revenues: revenues other than own-source revenues and balancing funds, e.g. grants from third parties, donation and intergovernmental grants other than balancing funds (e.g. special autonomy funds for Aceh and Papua) There are three major components of balancing funds: the formula-based General Allocation Grants (Dana Alokasi Umum/DAU), the earmarked Special Allocation Grants (Dana Alokasi Khusus/DAK) and the revenue sharing (Dana Bagi Hasil/DBH) which consists of shared tax revenues (DBH tax), and shared (nontax) natural resources revenues (DBH resources). However, fiscal decentralisation in Indonesia is more focused on expenditure decentralisation, compared to revenue decentralisation. While local governments are given autonomy to generate local taxes and retribution, most taxes are still controlled by the central government. For example, regarding personal income taxes, 80% of the total is the central government s while the remaining 20% is shared between local governments. The central government also controls other major taxes such as corporate income taxes, sales taxes and value-added taxes. On the other hand, expenditure decentralisation is more obvious. Given limited independent revenue sources for most local governments, and the significant difference in fiscal capacity across districts, the central government financial support from balancing funds DAU, DBH, and DAK continue to be a major revenue source for most local governments. Figure 4.1 below summarises the composition of local governments revenues in Indonesia from 2001 to

167 100% 90% 80% 70% 60% 14% 20% 16% 17% 20% 16% 21% 18% 19% 15% 19% 19% 24% 16% 16% 21% 23% 24% 18% 20% 20% 22% 20% 17% 16% 14% 50% 40% 30% 56% 49% 49% 49% 44% 50% 51% 43% 51% 43% 41% 42% 42% 20% 10% 0% DAU DAK DBH PAD others Source: Author calculation based on data from Ministry of Finance Figure 4.1 Composition of Local Government s Revenue in Indonesia DAU (General Allocation Grants) has been the major source of revenue for local governments. On average, from , DAU accounted for 47% of total local governments revenues. The second largest contributor is DBH (revenue sharing) with 19%, and PAD (own-source revenues) is the third with 18%. Revenue from other sources, on average, accounts for 12% of total local governments revenues. Compared with the early years of decentralization in 2001, the DAU percentage in 2013 decreased from 56% to 42%, whereas PAD (own-source revenues) increased from 14% in 2007 to 22% in From this revenue structure it is obvious that the post-reform local governments financing in Indonesia is highly dependent on central government transfers. In the following passages I will provide more details on the design of each transfer in order to get an insight into how each transfer can influence the level of voter turnout in local elections. 151