When political change signals community resolve: Fiscal decentralization, grassroots politics and local development Sarmistha Paly
Jaideep Roy
z
June 7, 2015
Abstract This paper examines how the introduction of the Fiscal Decentralization (FD) at the turn of the Century that o¤ered greater freedom in local governance a¤ected grassroots politics and local development in Indonesia. We …nd that socio-culturally homogenous communities are more likely to experience a change in the way they select local leaders at the time of the introduction of FD and we identify multiple political transitions in this respect. We then compare local development outcomes of various politically treated communities (relative to those that did not experience any change) before and after FD to identify the causal e¤ect of FD on local development: FD positively a¤ected local development when it also induced community level political change (yielding new democracies and oligarchies). Moreover, this positive impact of change on local development was maximum when it gave rise to new grassroots democracy. We argue that change in grassroots politics at the critical juncture of the nation’s history was key in fostering local development as it signalled community resolve for development and forced local political leadership to be more pro-people. But as costs of bringing about such changes is smaller with sociocultural homogeneity, such communities are more likely to engineer political change to reap the bene…ts of FD. JEL Classi…cation: D7, H4, O1, R5. Keywords: Fiscal decentralization, Political transition, Community homogeneity, Community development; Multiple treatment e¤ects; Di¤erence-in-di¤erence method.
Address for correspondence: Sarmistha Pal, University of Surrey and IZA; Email:
[email protected]. We would like to thank Rand Corporation for access to the data and for support with the data processing and also Siddhartha Bandopadhyay, Cathleen Beegle, Mattias Cattaneo, Stephan Litschig, Dilip Mookherjee, David Newhouse, Menno Pradhan, Daniel Suryaderma and Zaki Wahhaj for their comments and feedback at earlier stages of the paper. Earlier draft versions of this work are available at SSRN and IZA DP No. 5286. The usual disclaimer applies. y Faculty of Business, Economics and Law, University of Surrey, Guildford GU2 7XH, UK. Email:
[email protected] z Department of Economics, Deakin University, Melbourne;
[email protected]
1
1
Introduction
It is now well documented that while democratization in the Western World has largely succeeded in enhancing the welfare of the majority, introduction of democracy in the Developing World has faced mixed fortuity. Arguments put forward by Lerner (1958), Lipset (1959), Parsons (1971) and recently by Acemoglu et al. (2008) suggest that this ambiguity in the performance of democracy in Less Developed Countries may stem from the fact that many of them lacked certain preconditions in their path to modernization for democracy to persevere and thrive. An important precondition in this regard is the degree of community homogeneity that …nds fervent mention in the writings of John Stuart Mill (1861: p.289): “among a people without fellow-feeling, the united public opinion necessary to the working of representative government cannot exist.” A similar view is also expressed in Dahl (1971) where he suggests that for democracy to prosper, communities could not have too many subcultural divisions or a high degree of pluralism. In this paper we investigate whether community homogeneity is a precondition for the initiation of the process to democratization itself, particularly if there is a regime change in the aggregate environment that ushers in a new era of freedom? In other words, are homogenous communities more likely to be our new local democracies in LDCs that experience structural changes in governance? Moreover, when the process of choosing institutions or rules for selecting economic policies is determined by the communities themselves, is homogeneity necessarily good for grassroots development? Existing literature on democracy and development provide little empirical evidence on both these questions and this paper addresses these important gaps in the literature by examining how the introduction of …scal decentralization in 2001 in Indonesia a¤ected changes in grassroots politics, whether and how community homogeneity could explain these changes, and how these various communities performed consequently on the development front. Fiscal decentralization (FD) is identi…ed by many as an indispensable elixir to good local governance (see World Bank (2004)). It devolves …scal powers to local governing bodies, making them more directly responsible for the development of the community. This in turn provides additional freedom to local communities to decide for themselves by exercising their prerogative in choosing local leaders.1 Our objective is then to study how the advent of this ‘additional freedom’(through the initiation of FD) is utilized by di¤erent types of local communities to enhance the prospects of their economic well-being. We …nd that at the time of the introduction of FD, a signi…cant fraction of Indonesian communities ‘initiated’changes in grassroots politics and these changes, largely explained by community 1
Delegation of this nature has several bene…ts stemming out of inter-jurisdictional competition a la Tiebout (1956), local informational advantages a la Hayek (1948) or preference matching a la Oates (1972). See also Montinola et al. (1995), Qian and Weingast (1996), Qian and Roland (1998), Maskin et al. (2000) that argue that decentralization is good for development. On the other hand, FD may also enhance the in‡uence of the local elites or increase economic and political corruption that distorts development mechanisms (see Bardhan (2002) and Bardhan and Mookherjee (2006)).). See also Prud’homme (1995), Tanzi (1996), Cai and Triesman (2004) that argue that decentralization can be harmful.
2
homogeneity, determined the di¤erential impact of FD on local development across them. We also …nd that change in grassroots politics was mostly seen amongst communities that were relatively homogenous but interestingly, some homogenous communities were also more likely to bring about change through the introduction of new oligarchies! Thus somewhat surprisingly we …nd that homogeneity was not necessarily a precursor to the spread of democratic practices around this critical juncture in the nation’s history even when such regime change did proliferate freedom of expression. In view of Mill’s postulation, what can explain this apparently paradoxical observation? We rationalize these …ndings through the following theoretical prism. When Indonesia initiated the process of empowering local governing bodies with …scal responsibility, local communities were more incentivized to revise the way their local governments were elected or functioned. But in spite of the increased incentives, not all communities were able to agree upon the changes they wanted, particularly when there were socio-cultural heterogeneities. While local politicians understood that FD transferred additional power to the local population, they could still continue to reap political rents in communities that the politicians felt were divided in their cause or were themselves not sure about their unity. Realizing this, community members found it necessary to signal to local politicians that they were united for a common cause. This could bring forth institutional changes in the way these communities chose their leaders.2 But …nding a common voice and organizing an ‘event’to exhibit that can get increasingly costly when the community is not homogenous (in their language of communication or if there were few existing sociocultural norms (or organizations) that help its members to mobilize politically).3 The work of Alesina and Ferrara (2000) insinuates this possibility by showing that culturally homogeneous communities are more keen in productive social participation. These possibilities transpired into a signaling environment between communities and a pool of potential political leaders where in a separating equilibrium, homogenous communities (with low cost of signaling) were able to initiate short term changes in local politics while heterogenous ones remained politically stable at the time FD was introduced. In equilibrium therefore, change in grassroots politics was expected to promote better development outcomes on average. We also observe that close to 86% of communities which initiated a change in local politics had a new leader in our dataset. Hence our prediction is more robust for this subset. Upon splitting it into new democracy and new oligarchy we …nd that new democracies with new leaders performed signi…cantly better in development than new oligarchies with new leaders. However there was no signi…cant di¤erence in the performance of new oligarchies and stable polities in this respect.4 If new oligarchies were not certain to perform better than existing polities, what can explain their appearance at a time of enhanced nationwide freedom? For a relatively homogenous community that had been democratic in the past, it was lucrative to bring about a change to signal intent. But 2
See Acemoglu and Robinson (2001) for a similar argument. See Banerjee et al. (2005), Banerjee and Somanathan (2007) and Munshi and Rosenzweig (2008) for other studies that look at the role of community homogeneity and other historically determined factors (like ex-colony status, caste structure in India) in political mobilazation and its impact on provision of public goods. 4 This is in line with some …ndings in the existing literature (see for example Bardhan and Mookherjee (2000), Sonin (2003) and Olken (2004)). 3
3
a change for them could only be a move away from their existing democratic foundations as after all, a change in leadership through existing democratic institutions was still an old practice in such communities. When they decided to move to a new oligarchy, in principle they chose a governance lottery where (a) with some probability the leadership under the new oligarchy would be good as the signal of intent will force the new leader to be more serious about the welfare of the community than what their old polity has been providing and would have continued to provide in the absence of the signal, (b) but with the remaining probability the new oligarchs would be successful in forming political and social networks mentioned above that hamper welfare. This inherent short term ‘risk’in initiating a new oligarchy may have partly restricted their performance. But communities could have still experimented with such a change for long term bene…ts. Nevertheless, if political change was used as an instrument to signal resolve and if at the same time Mill’s theory of positive relationship between democracy and community homogeneity were to be true, then one would expect homogenous communities that sacri…ced democracy around the time of FD to signal resolve would be more likely to return to democracy in subsequent opportunities once the costly signal was executed. Indeed we …nd evidence of this reversal in our data (see Section 4.1 for more on this). We look at the lowest level of administrative unit in Indonesia, which still has an independent political identity called a community, a rural village or an urban township. Using the Indonesian Family Life Survey (IFLS) data, we consider these local communities in 1997 and 2007, two years separated by the introduction of Law 22/99 and Law 25/99 in 2001, which can largely be treated as the FD exercise and exogenous for the communities under consideration.5 In this respect, we distinguish between communities that experienced some political transition after FD and those that did not. In particular, IFLS data allows us to categorize local polities as ‘Democratic’ if the community leader is elected by voting or consensus building among all citizens and ‘Nondemocratic’ (or oligarchic) if the leader is ‘chosen’ by few including local elite, local institutions and outside in‡uence. We also observe the tenure of the leader which we use to identify if there has been a change of leader over this period; 86% of the sample communities had a new leader in 2007.6 By comparing the local polities (democracy or oligarchy) and leader turnover in sample communities between 1997 and 2007, we …nd that about 87% of sample communities experienced a change in both local polity and local leader in the post-FD period. Accordingly, we classify the sample communities in terms of the observed political transition over this decade: (i) old polity, 5
The extensive FD process has its roots in Law 22/1999 on regional autonomy and Law 25/1999 on intergovernmental …scal relations enacted in January 2001. According to laws 22/1999 and 25/1999, the central government would retain the responsibility in just …ve areas: national security, foreign policy, monetary policy, …nance and development planning, justice and law enforcement. All other public services, including the health, education and infrastructure, would become the responsibility of district and local governments, with a more minor role being granted to provincial governments (Brodjonegoro, 2001; Bardhan and Mookherjee, 2006). 6 Given that the term of o¢ ce of a community leader has been 5 years since 2001 and given that all post FD community elections were concluded by 2003 (Rodriguez, 2010), it is most likely that community leaders in power in 2007 are the ones elected in the post-2001 year. We can thus rule out the possibility of any other elections in the post-FD years, at least for a large majority of our sample communities.
4
old leader; (ii) new polity (democratic/ nondemocratic) old leader; (iii) old polity (democratic/ non-democratic) new leader; (iv) new polity new leader which can further be classi…ed as new oligarchy (non-democratic) new leader and new democracy new leader. Our estimation strategy has therefore been to compare various types of politically transiting (treatment) communities with the politically stable control group that neither experienced a change in local polity nor a change in leader with a view to identify the causal impact of FD on the local development measure over 1997-2007. We use two measures of local development: (a) the …rst measure is the share of total community spending on local development, which includes spending on new social and physical infrastructure as well as their maintenance (see section 2). (b) Our second measure pertains to an ‘underdevelopment’ status of the community measured by whether the community has been selected for the government poverty alleviation program Inpres Desa Terttingal or IDT for short (see further discussion in section 4) during 1997-07. In view of multiple political treatments (ii)-(iv) of FD among the sample communities, we start with the multivalued treatment e¤ects model (a la, Cattaneo, 2010; 2013) to estimate the average treatment e¤ect on the treated. Given that the identi…cation in this model is based on observables only (Cattaneo, 2010), there is possibility of omitted variable bias in these estimates. Hence, we also run a variant of the di¤erence-in-di¤erence model to identify the causal e¤ect of FD on local development and we do so by comparing local development outcomes of various politically treated (relative to those untreated) communities before and after FD after controlling for various observable community characteristics. All our estimates also include district dummies and also (District*Year) dummies to account for unobserved district-level time-invariant and timevarying factors to limit omitted variable bias. As such our estimates highlight the treatment e¤ect of FD within districts, which became the centre of power after FD. Our most robust estimates come from a district …xed e¤ects di¤erence-in-di¤erence (or DID for short) model with instrumental variables (IVs) for heterogenous political transition measures. In view of our theoretical conjectures, identi…cation of the political transition in this paper relies on community homogeneity indices. We consider di¤erent dimensions of homogeneity including share of the largest population group by language spoken and also community’s adherence to Islam and age-old adat laws both of which would promote an ethic of mutual cooperation (see Pal and Wahhaj, 2015 and further discussion below). To address the fact that these homogeneity measures are unlikely to be random, that is they are likely to be correlated with other factors that can in‡uence …scal decentralization and subsequent local development, determination of political transition equation also controls for other possible correlates of homogeneity and interactions of all arguments including homogeneity measure with the FD dummy. Accordingly, this speci…cation would control for all other di¤erences across communities of di¤erent levels of homogeneity in the post-FD years, thus forcing identi…cation through a systematic variation in community homogeneity only (see further discussion in section 3.1). We then use the predicted values of di¤erent political transitions likelihoods obtained from this …rst stage to determine local development in our sample in the post-FD period. 5
The existing literature on FD predominantly analyzes the e¤ects of decentralization (in general) on public policy and development using cross country (see for example, Davoodi and Zou (1998), De Luca et al. (2002), De Mello and Barenstein (2001) and Fishman and Gatti (2002)) or panel (Enikolopov and Zhuravskaya (2007)) data. Using cross-section and panel data from about 75 developing and transition countries, Enikolopov and Zhuravskaya (2007) found that FD is more successful in those economies which are more politically decentralized though the nature and variation of political decentralization across the sample countries was not investigated in the paper. Second, a handful of studies use community level single cross-section data to assess the developmental impact of FD (see Besley et al. (2005) for a comprehensive literature review) to highlight the role of leader characteristics in this respect; as such, these studies are unable to examine the causal impact of FD. More recently, there are attempts to identify the e¤ect of local elections on local development and public goods provision in developing countries (e.g., Padro-i-Miguel et al. (2014); Martinez-Bravo (2014)). While the correlation between e¢ cacy of democracy and community homogeneity has been reported in the large, Padro i Miquel et al. (2014) validates it in rural China by providing evidence that one of the preconditions for exogenously introduced grassroots democracy to be e¤ective is the degree of community homogeneity in some vertical attribute (religion in their case) that allows better provision of public goods. Martinez-Bravo (2014) assesses the impact of the …rst post-Soeharto parliamentary election in Indonesia on a number of outcome variables including the changes in village expenses over 1999-2002. Our paper contributes to this existing and newly emerging literature in a number of ways. We identify how community homogeneity may induce political change chosen by the people, which in turn a¤ects the nature of local development in post-FD period, thus o¤ering an endogenous political mechanism through which FD can a¤ect local development. We identify …ner classi…cations of political transition after FD that not only distinguish between democratically elected or oligarchically selected heads (which closely correspond to elected and appointed heads in Martinez-Bravo, 2014) at a given point in time, but also the changes in local politics over 1997-2007. The rest of the paper is structured as follows. Section 2 describes the data while Section 3 explains the empirical methodology. Section 4 presents the …ndings and the paper concluded in Section 5. All tables and …gures are provided at the end.
2
Data
Our analysis is based on the community level data obtained from 1997 and 2007 Indonesian Family Life Survey (IFLS) from 312 rural and urban communities, drawn from 13 provinces including Jakarta, Bali, Java (central, east and south), Sumatra (north, west and south), Lampung, Wntenara and south Kalimantan.7 This is a particularly rich data set that provides community level information on a whole range of demographic characteristics and access to public goods, local 7
Although IFLS data are available for the years 1993, 1998 and 2000 as well, information on local politics could be found only in the 1997 and 2007 surveys.
6
governance and its public …nances, citizens’participation in planning and implementation of local development projects, as well as a range of public utilities, infrastructure and transport, health and education facilities (see Frankenberg and Thomas (2000) and Strauss et al. (2009) for study design and overview of the data set).
2.1
Background
With the introduction of the ‘1979 village law’, village a¤airs were brought under the supervision and close control of higher authorities. While the law stated that the village had “the right to manage its own a¤airs”, it immediately noted that this “does not mean autonomy”(General Clari…cation, section 7). The village was nothing more and nothing less than “the lowest level of the government structure directly under the sub-district chairman”. Since 1979, the head of villages classi…ed as ‘desa’ has been elected in village-level elections held every 8 years, while the heads of ‘kelurahan’villages (urban/city) were appointed by upper levels of administration. These have been the de jure selection rules of community leaders though in practice the process of selection of the community leader may vary (see Table 1). In short, Indonesia has been culturally and politically decentralized nation even though, and as discussed in the introduction, local leader selections may have been controlled by the central regime under Suharto; however the nation was under the unambiguously tight grips of central …scal control until 2001. Fiscal decentralization in post-Suharto Indonesia was largely an exogenous event for the communities, which has its roots in Law 22/99 and Law 25/99 enacted in January 2001. It gave local communities more autonomy in raising local revenues while enforcing strict budgetary cuts on the central leadership to supply development grants to these communities. It also granted administrative authority to local governments to hire sta¤ and conduct local government a¤airs with a minimum intervention of the central government; local community governments were made responsible to the district government who provided the bulk of the funds to local communities after FD; in other words the centre of power moved from the central government in Jakarta to the district governments located in district head quarters after FD. We study these local communities between 1997 and 2007, two years separated by the introduction of FD in 2001.
2.2
Description
The IFLS data available for the years 1998 and 2000 reveal that there were no local elections in the sample communities during those two years. This is expected as the country faced widespread economic and political turmoil during 1997-99 when there could hardly arise any opportunity for local communities to attend to their own politics. After months of struggle, the …rst elected president Wahhid took o¢ ce in October 1999. Things started to get back to normal from the turn of the century paving the way for Law 22/99 and Law 25/99 to be introduced o¢ cially in January 2001. All local community-level elections in the post-FD period did not take place at the same time as Rodriguez (2010) suggests that about 80% of post-FD local elections were completed by 7
2003 as it involved a fair amount of administrative changes as part of the new decentralization rules. Although we do not observe the timing of the local elections for the sample communities, we observe the tenure of the community leaders in 1997 and 2007. Given that the term of o¢ ce of a community leader has been 5 years since 2001, it is most likely that community leaders in power in 2007 are the ones elected in the post-2001 year; we can thus rule out the possibility of any other elections in the post-FD years, at least for a large majority of our sample communities. The 1997 and 2007 rounds of the IFLS asked community leaders about how a leader was selected, which we use to classify these communities. Answers to this question are coded as: (i) voting, (ii) all residents, (iii) local elites, (iv) local institutions and (v) others. Under voting (code (i) and (ii)), the standard notion of plurality was employed where the voters were local community residents. Otherwise, local leadership was determined by the choice of local elites (code (iii)) or by existing o¢ cials of local government bureaucracies (code (iv)). However it is not clear as to how others (code (v)) selected the local leader, and so we exclude these communities from our analysis. Accordingly, we classify local politics as follows: Democratic if a leader is selected by free and fair elections with voters being all community members (code (i) and (ii)), and Oligarchic if a leader is selected by community elites (codes (iii) and (iv)) who then remain uncontested.8 We assume that the incidence of oligarchies re‡ect the size of elite capture. Table 1 summarizes the methods of selection of community leaders in the sample community in 1997 and 2007. Measure of …scal decentralization: We proxy …scal decentralization by the binary variable Y r2007 that takes a value 1 for the year 2007 and 0 for 1997. Measures of political transition: Access to both 1997 and 2007 rounds of IFLS data allows us to classify political transition into the following types (where V stands for voting and O for oligarchy or elite capture): V O [or New Oligarchy], if the community was democratic in 1997 but oligarchic in 2007, OV [or New Democracy], if the community was oligarchic in 1997 but turned democratic in 2007, V V [or Stable Democracy], if the community remained democratic between 1997 and 2007, and OO [or Stable Oligarchy], if the community remained under elite capture between 1997 and 2007. Table 1 summarizes the methods of selecting a community leader in our sample in 1997 and also in 2007, which in turn highlights the fact that the de facto selection process may di¤er from de jure selection rules. In general, a slightly higher proportion of sample communities adhered to voting 8
This classi…cation closely follows the elected and appointed leaders in Martinez-Bravo (2014): in particular, democratically elected leaders are the ‘elected’ leaders while oligarchic ones correspond to the ‘appointed’ category in Martinez-Bravo (2014).
8
and consensus building in 2007 than in 1997. Second, a higher proportion of rural communities adhered to democracy while a higher proportion of urban communities adhered to oligarchy in both years. To a large extent this was the result of Village Law 1979 which retained the power of the government to select leaders for urban communities. About 71% of the sample communities tend to be politically stable. Among them, 62% continued to be democratic while the remaining 38% remained oligarchic. In other words, only 91 of 312 total communities (i.e., about 29% of total communities) experienced a change in local politics. Among these transient communities, 66% are new democracies while the rest are new oligarchs. Further, the data gives information on the tenure of the community leader. We use this information to construct a binary variable pertaining to leader turnover (leader to) that indicates if the community leader who was in power in 1997 was no longer there in 2007. Abour 83% of all sample communities had a new community leader in 2007 while about 86% of transient communities (new democracy or new oligarchy) had a new leader in 2007. Without any loss generality, we next de…ne a local polity to be democratic if the leader is selected by consensus building or voting and oligarchic if the leader is selected by few elites. Thus a stable polity refers to the cases where democratic (oligarchic) communities in 1997 remained democratic (oligarchic) in 2007. A second dimension of political transition is captured by whether there was a change in leader in communities with stable/new polity (democratic/oligarchic) over this decade. Combining these two dimensions of political transition, we now construct a new categorical variables STATUS that takes a value 0,1,2,3: STATUS = 0 if stable democracy/oligarchy with no change in leader = 1 if new polity (democracy/oligarchy) with no change in leader = 2 if old polity (democracy/oligarchy) with a new leader = 3 if new polity with a new leader In an attempt to distinguish between the role of new democracy and new oligarchy, we also construct a second categorical measure of political transition: STATUS1
= 0 if stable democracy/oligarchy with no change in leader = 1 if new polity (democracy/oligarchy) with no change in leader = 2 if old polity (democracy/oligarchy) with a new leader = 3 if new oligarchy with a new leader = 4 if new democracy with a new leader
To some extent this classi…cation has been guided by the relative small sample size as we attempt to ensure that there are reasonable number of observations in each category. A simple frequency distribution of the variable reveals the extent and nature of multiple political treatments of …scal decentralization in our sample (see table 2).9 Out of a total of 624 sample communities 9
We could have merged category 3 and 4 into one as ‘new polity new leader’, but did not because we wanted to distinguish the treatment e¤ects of new democracy from new oligarchy.
9
over two years 1997-2007 (i.e., 312 communities in each year), there were only 39 (about 12.5% of the total) communities with no change in local polity or leader (category 0) while only about 5% can categorized as new democracy/oligarchy with the old leader after FD (category 1).10 In contrast we have more cases where there was a change in polity, a change in leader or both: about 58% communities had stable democracy/oligarchy with new leader while the rest (about 24%) had new polity (democracy/oligarchy) with new leader. In particular, there were about 12.5% newly democratic/oligarchic communities each with a new leader. One can also merge Status=3,4 to create a combined category new polity with new leader though it is unable to distinguish the costs/bene…ts of new democracies from new oligarchies (we consider estimates of political transition using both Status and Status 1 to check the robustness of our results). Measures of local development: Fiscal decentralization o¤ered more autonomy in local spending (and not in revenue collection) and the delivery of public services including health, education and infrastructure (Brodjonegoro, 2001). IFLS data provide information on the revenue and spending of the community on di¤erent accounts for both years. We use the spending allocation of the local community government to construct the measure of local development. Total spending has the following components: (i) development spending refers to spending on new social (e.g., local schools and health centres) and physical (e.g., roads and transport) infrastructure as well as maintenance of existing infrastructure; (ii) non-development spending includes spending on sta¤ salary, o¢ ce maintenance, o¢ cial trips and contingencies; (iii) some under-developed communities also receive grants for various developmental programs, e.g., IDT for under-developed communities. We measure local development by the share of total development spending (Shdev) in total spending for each community. Given that community spending can be manipulated by the local government, we also check the robustness of our estimates by employing a more objective measure that pertains to an underdevelopment status of the community measured by whether the community is selected for the government poverty alleviation program Inpres Desa Terttingal or IDT for short. The selection of communities for the IDT program is made by the National Development Planning Agency (BAPPENAS) and the Ministry of Home A¤airs (MoHA) and it is based on the PODES socioeconomic survey by the Central Bureau of Statistics. The 1994 selection considered three basic indicators: village facilities and potential, housing and the environment and population characteristics. The selection criteria however had some ‡aws for identifying consumption levels; for instance, too much weight was given to infrastructure de…ciencies that do not always re‡ect lowest consumption levels-and thus the selection for the 1995 program gave more prominence to expenditure levels. Also number of households in the community determined how often these communities would be selected for IDT funds. Under IDT, the government, by virtue of presidential instructions, provided selected poor communities (or villages) with lump-sum grants designated for small business loans. These 10
Given the small number in this category, we were unable to split them into new democracy with old leader and new oligarchy with old leader.
10
selected villages were then instructed to choose relatively poor households that would be eligible for IDT loans based on village-level meetings. These were facilitated by the village head and a local government agency called Lembaga Ketahanan Masyarakat Desa (LKMD, Village Community Resilience Board). Accordingly, we construct a binary variable called ’U ndev’that takes a value 1 if the community is selected for IDT and 0 otherwise in our sample. The mean value of the variable is about 0.40, thus indicating that about 40% of sample communities have been selected for the anti-poverty programme IDT in our sample. Table 3 shows the simple mean comparisons of key spending and revenue variables in 1997 and 2007, using t-tests. It is evident that the decentralization in 2001 has been accompanied by lower grant from the central government, which was compensated, to a large extent, by the higher grant from the district government. The latter justi…es our use of year 2007 as a proxy for FD in our analysis, after controlling for all other possible unobservable district/year level factors and also district-level time trends that we capture by including district*year …xed e¤ects. Interestingly the average share of local funds is slightly higher in 2007 with a lower standard deviation in our full sample, thus suggesting the growing importance of local funds in the post-decentralization years. Finally, Table 4 compares the mean share of central grant, district grant and development spending among sample communities classi…ed by the nature of political transition in 2007. Interestingly, the share of development spending is the highest (0.51) in the communities with new polity and new leader relative to those with new polity old leader or old polity new leader. Considering the communities with new polity and new leader, we then examine the relative performance of new democracy versus new oligarchy: the share of development spending was about 7 percentage points higher (0.55 as opposed to 0.48) in new democracies with new leader relative to new oligarchy with new leader. This is compatible with the alternative development measure ‘undev’: 445% of new democracies with new leader was considered underdeveloped in 2007 as opposed to 64% of new oligarchies with new leader. Since the treatment related to political transition caused by …scal decentralization is not unique, our analysis considers the possibility of multiple political treatments of FD to identify the e¤ect of FD on share of community development spending (see section 3) which is a non-parametric method. Parametrically, we also estimate a di¤erence-in-di¤erence model to identify the di¤erential e¤ect of each political treatment (relative to the control group) on local development in our sample, thus highlighting the heterogeneous e¤ects of FD on di¤erentially politically treated communities. We describe and analyze our empirical model in the next section with a view to test whether these simple bivariate comparisons hold, after controlling for all other factors that may in‡uence local development.
3
Linking theory to empirics
Our objective is to assess the impact of FD on political transition and local development. Since our theoretical position, as highlighted in Section 1, suggests that political transitions of communities 11
are unlikely to be random, we adopt a two-step method. Assuming 2001 introduction of FD to be an unexpected/random shock to the system in the time dimension, we …rst examine if local politics changed in a systematic (and statistically signi…cant) manner between the years 1997 and 2007 in uni…ed and divided communities. Second, we identify the causal e¤ect, if any, of change in grassroots politics on local development spending (relative to communities that neither changed in local polity nor in leader). This section explains the way we build the empirical model in order to bring our theory to the data, carefully ensuring that the estimators are statistically robust to various dependencies (both in time and in attributes) and speci…cation errors.
3.1
Attributes of communities as key variables
We model FD by the binary variable yr2007 that takes a value 1 for year 2007 and 0 for year 1997. We have argued that political transition following the introduction of FD would largely depend on whether the community in question found a common voice for collective bargaining in the postFD year. Thus our identifying variable for determining political transition pertains to measures of community homogeneity. There could be di¤erent dimensions to a community’s homogeneity ranging from its ethnicity based on language spoken by the largest population group, its dominant religion or predominance of traditional laws (or local customs) that bring them together. We thus generate a series of homogeneity measures using the available information: Ethnicity: the dominant ethnic group in the data set is identi…ed by the main language spoken in each community, rural or urban. In this measure therefore, a community is ethnically more homogenous if it has a very large population group (pop1) speaking a common mother tongue. We create an index ethhety_pop1, which is (1-sum of square of share of the largest population group). Clearly the higher is pop1, the smaller is the value of ethhety_pop1 and greater is the ethnic homogeneity of the community. Religion: IFLS data provides information of whether Islam is the dominant religion of the community, if more than half of the population are Muslims. It can be argued that the observed dominant religion may help a community to form a common voice on religious grounds; this can in‡uenced by the presence of a religious leader who resolves con‡icts in these communities. Adat law : We classify communities exhibiting high and low levels of participation in collective activities using their strict adherence to traditional adat laws. We accept that the adat rules specify the nature and extent of the contribution that community members ‘expected’ to make towards a collective cause. These rules refer to customs, traditions, rules or practices that guide social life and decision-making across a range of matters in community and family a¤airs, including marriage, inheritance, land-holding, and dispute resolution in Indonesia (ibid; Kato, 1988). The IFLS data place each community into one of four possible categories: (i) traditional laws are almost never broken; (ii) traditional laws are sometimes broken; (iii) 12
traditional laws are frequently broken and (iv) only a few people understand traditional laws. We classify a community as an adat community if adat laws are almost never broken and a non-adat community otherwise. Following on from (see Pal and Wahhaj, 2015), we argue that a adat community de…ned in this is way is likely to be more cooperative/collective than non-adat communities. Clearly, community ethnicity, dominant religion or strict adherence to adat law have been features for generations and as such can be treated as exogenously given; we also speci…cally test for the exogeneity of these measures.11 We also include the local residency of the leader to account for leader turnover following FD. The underlying argument is that being selected as a community leader despite not being a resident in the community is more likely to highlight their possible links with the external forces (e.g., district or central authorities). Hence we include the interactions of these homogeneity measures as well as local residency of the leader with the binary variable yr2007 that captures FD in our empirical exercise. Similarly, we include the level of education of the local leadership and classify the ‘chosen’local leaders as either being highly educated or not (headshs and headshs*yr2007). To address the fact that these homogeneity measures are likely to be non-random, that is correlated with other factors that can in‡uence …scal decentralization and subsequent local development, the political transition equation not only controls for homogeneity measures, but also other possible correlates of homogeneity such as community size, population, distance from the district head quarter (to test if the district authority had any in‡uence on political transition in the post-FD years), if rural, access to sea, district dummies and more importantly their interactions with the post-FD dummy. As such, this speci…cation forces these estimates to be identi…ed only from a systematic change in community homogeneity between politically transiting (i.e. which are homogeneous) and politically stable (i.e., less homogenous) communities from the year FD was implemented. Easy availability of local funds in some communities may play an important role by a¤ecting political stability/instability, especially after FD. The data set allows us to classify local communities in terms of availability of opportunities to raise local funds and as such we include the share of local funds and its interaction with Yr2007, which we believe is likely to a¤ect local development as well.12 Such di¤erences may have some political, economic or social causes that a¤ect the economic environment of the community. It is then reasonable to assume that the likelihood of FD 11
Some may argue that income inequality may also play a role here; as such, we tried to account for the e¤ect of income homogeneity, if any. In this respect, we constructed a measure of coe¢ cient of variation (cv) of community’s average monthly per capita expenditure de…ned as the ratio of standard deviation to mean of the of community expenditure. Clearly the lower the value of cv, the higher is the degree of homogeneity in the distribution of expenditure. While a community’s income may ‡uctuate from one year to next, it is expected that expenditure is unlikely to do so and as such could be regarded as a more long-term measure of community prosperity. However the variable was never signi…cant either on its own or when interacted with the FD dummy Yr2007, suggesting its insigni…cant role on the political transition likelihood in our sample. 12 As the share of local funds raised in a community is a potentially endogenous variable either in determining political transition or local development, we also need to …nd an instrumental variable for the share of local funds. See further discussion in the methodology and results section.
13
promoting leaders who are more e¢ cient in raising (and using) local funds can depend upon these opportunities as well. Once we establish the link between probability of political transition with community characteristics, our next goal is to compare local development, measured by the share of development spending in total community spending (or alternative likelihood of a community being underdeveloped), among politically treated (relative to control group with stable polity) communities before (1997) and after (2007) the introduction of FD with a view to identify the causal e¤ect of FD on local development. Our theory predicts that FD is more likely to enhance local development in politically transiting homogenous communities (which were successful in signalling to the politicians that they have a common voice through change in local politics). We therefore hypothesize that more homogeneous communities (as de…ned above) are more likely to experience a political change which in turn may ensure better development outcomes compared to the communities that failed to trigger a political change after FD. We also predict that developmental spending in 2007 is higher in communities that are new democracies rather than new oligarchies (and further envisage that local development would be the lowest in stable oligarchies). We shall now turn to data analysis to explore the validity of these di¤erent hypotheses pertaining to the causal e¤ects of FD on political transition and local development in our sample, after controlling for all other factors that may also in‡uence local development.
3.2
Empirical model
We use the 1997 (pre-FD) and 2007 (post-FD) IFLS data to study the impact of the 2001 introduction of FD on local development in Indonesia, after controlling for all other factors that may also in‡uence political transition and local development of the sample communities. In this respect, we exploit the variations in local development across di¤erentially politically treated sample communities after the introduction of FD to identify the causal impact of FD on local development (measured by community’s share of development spending) in our sample. We measure political transition in terms of the categorical variable STATUS (or STATUS1) that classi…es sample communities in terms of multiple political treatments in the post-FD year. While communities with stable local polity (oligarchy/democracy) with no change in leader (when STATUS=0) is considered as the reference category, treated communities are those when STATUS=1,2,3,4. The conventional method of propensity score matching would not be able to account for these multiple political changes simultaneously. Hence we use the multivalued treatment e¤ects (a la Cattaneo, 2010; 2013) and compare the value of the outcome variable, namely, the community’s share of development spending (SHDEV) for treated communities (when the variable STATUS taking values 1,2,3,4) relative to the control group (when STATUS=0), after controlling for possible observable factors that may also in‡uence the outcome; we do the same with the alternative measure STATUS1. Multivalued treatment e¤ects may, however, su¤er from the omitted variable bias arising from various unobserved (community, district and year-level) factors. So our preferred estimates come from an extended
14
di¤erence in di¤erence model of local development as a function of di¤erent political treatments and their interactions with the FD dummy, that controls for possible observable factors as well district-level unobservable factors (accounted for by including various district dummies); we also include Year dummies and (District*Year) dummies to account for district-level unobserved time trends. Note however that the likelihood of political transition after FD is unlikely to be random at the community level and hence we need to generate instrumental variables (IVs) for political transition likelihoods (see further discussion of DID estimates below). To summarize, both methods (multivalued treatment e¤ects and DID) estimate the development outcome of various treated communities relative to the control group. The di¤erence in the two methods arises from the fact that the multivalued treatment e¤ects method is based only on observable characteristics of these communities while the di¤erence-in-di¤erence (DID) model takes account of both the observed and unobserved community characteristics. As such a DID model with IV for political transition measures appears to be superior as the potential omitted variable bias is likely to be minimum here. 3.2.1
Multivalued treatment e¤ects
We look at a collection of population parameters capturing the impact of our treatment variable on an outcome variable when the treatment takes multiple values, which is a natural extension of binary treatment e¤ects (Cattaneo (2010, 2013)). Our unit of observation is the community i observed over 1997-2007. We have 5 possible treatments taking values 0; 1; 2; 3; 4 as captured by the categorical variable STATUS: (i) 0: stable democracy or stable oligarchy with no change in leader; (ii) 1: New democracy or new oligarchy with no change in leader; (iii) 2: stable democracy/oligarchy with a new leader; (v) 3: new oligarchy with a new leader; (vi) 4: new democracy with a new leader. We observe the outcome y (as measured by the share of development spending) as well as the treatments 0; 1; 2; 3; 4; for each i - th community in the post-FD year 2007. For example, consider the treatment value when STATUS=0 to be the reference group. If we observe y0 ; : : : ; y4 for each treatment 0; 1; 2; 3; 4 respectively, we can then obtain the treatment e¤ects as y1 y0 , y2 y0 etc. for each community i relative to the reference treatment group when STATUS=0. Finally, aggregating these individual treatment e¤ects, we obtain the average treatment e¤ect on the treated (ATET) for the sample communities. We repeat the same procedure for the alternative STATUS1 variable. We use both the regression adjustment (RA) and also the inverse probability weighted regression adjustment (IPWRA) methods to derive the average treatment e¤ects on the treated (ATET), after controlling for various observed community characteristics. In each case we also obtain the potential outcome mean (POM) for each of the political transition variables STATUS and STATUS1 where POM is the average of each potential outcome for a given treatment. RA uses the contrasts of the average of the treatment-speci…c predicted outcomes to estimate treatment e¤ects and uses a two-step approach: (i) They …t separate regression models of the outcome on a set of covariates for each treatment level. (ii) They compute the averages of the predicted outcomes for each subject and
15
treatment level. These averages re‡ect the POMs. The contrasts of these averages provide estimates of the ATEs. By restricting the computations of the means to the subset of treated subjects, we obtain the ATETs. RA estimators are consistent as long as the treatment is independent of the potential outcomes after conditioning on the covariates. IPWRA estimators use weighted regression coe¢ cients to compute averages of treatment-level predicted outcomes, where the weights are the estimated inverse probabilities of treatment. There are three steps involved: (i) They estimate the parameters of the treatment model and compute inverse-probability weights. (ii) Using the estimated inverse-probability weights, they …t weighted regression models of the outcome for each treatment level and obtain the treatment-speci…c predicted outcomes for each subject. (iii) They compute the means of the treatment-speci…c predicted outcomes. By restricting the computations of the means to the subset of treated subjects, we can obtain the ATETs. We compare these two sets of estimates to check the robustness of our ATET estimates. One may however be concerned about the estimation bias arising from the unobserved factors as identi…cation in these models is based on observables only (Cattaneo, 2010); we try to redress this concern by estimating an alternative di¤erence-in-di¤erence model with district …xed e¤ects within a panel framework. 3.2.2
Di¤erence-in-di¤erence (DID) estimates
Given our concerns that the treatment estimates may su¤er from unobserved district level (timeinvariant and time-varying) factors, we also estimate the following di¤erence-in-di¤erence model of development:
0 + T ST AT U S + D F Dt + T D ST AT U S
Yit =
F Dt +
L LFit + X Xit + d + d
F Dt +uit ; (1)
where Yit is the community level development index ST AT U S = 1; 2; 3; 4; depending on the particular treatment experienced by the community while it takes a value 0 when there is no change in polity F D = 1 if year = 2007 (post- FD), and 0 (pre-FD) otherwise; this is an index of …scal decentralization LFit is the community’s share of local funds in total revenue d d
are the district-dummies to capture district-level omitted factors F Dt are the district-level omitted time trends and uit are the remaining errors.
16
After controlling for all possible observables (X) and time invariant and time-varying unobservables ( D ; F Dt ; D F Dt ), the coe¢ cient of particular interest to us is T D , which captures the di¤erential e¤ect of FD on community development index Y among the politically transient communities (relative to politically stable ones). The DID model, by de…nition, looks at subsets of the data set to …rst identify and cluster all communities that are statistically comparable in all respects in 1997. This allows the model to attribute the di¤erential outcomes in 2007 to FD after controlling for all observable and unobservable community characteristics, time trends and also district-level time trends (see further discussion below) within a panel framework. As the likelihood of political transition after introduction of decentralization in 2001 in our sample is unlikely to be random and more importantly be partly caused by the FD, we need to obtain a valid instrument for this variable. In doing so, we …rst use the pooled data for 1997-2007 to estimate a multinomial Logit model of the categorical variable STATUS (or Status1) indicating di¤erent types of political transitions experienced by the sample communities using measures of community homogeneity and also leader’s education and local residency among other controls. Accordingly, we …rst determine the multinomial logit likelihood of political transition STATUS over 1997-2007 using the following model:
ST AT U Si =
0
+
H Homogeneityi
H Homogeneityi
+
D DISDHQi
+
I IN Vi
F Dt + F Dt +
+
I IN Vi Z Zit
+
S SHSi
+
F Dt + Z Zit
D DISDHQi S SHSi
F Dt +
d
+
+
F Dt +
F Dt d
F Dt + wit
(2)
where ST AT U Si takes values 1, 2,3 for di¤erent types of political transition and 0 otherwise (stable democracy/oligarchy with old leader); – takes a value 1 if a community has new democracy/oligarchy with old leader in 2007 – takes a value 2 if a community has stable democracy/oligarchy with a new leader in 2007 – takes a value 3 if a community has new polity with a new leader in 2007 Homogeneity indicates measures based on size of the largest population group in the community, if Islam is the dominant religion and also the ethic of mutual cooperation as captured by strict adherence to Adat laws IN V indicates if the leader is a local resident in the community; SHS indicates if the community leader has at least senior high schooling DISDHQ indicates the distance of the community (in km) from the district head quarters;
17
Z is the set of observed control variables (time-invariant and time-varying) including characteristics of the leader (the share of local funds raised) and those of the community (population, geographic size, if Islam is the dominant religion, if it is a rural/urban community; if it has access to sea). is the unobserved time-invariant district level characteristics as captured by district dummies; d
F D: Dummy for year 2007, which accounts for the e¤ects of FD d
F D: District-level time trends
wit are the remaining iid errors for the i-th community in year t; t =1997, 2007. Following on from our discussion in Section 3.1, we not only control for measures of community homogeneity, leader’s high schooling and local residency, but also include distance of the community from the district headquarter DISDHQ as an important determinant of a community’s relationship with the district authority as the centre of power moves from the government in Jakarta to the district authority located in the district head quarters after decentralization. This is further justi…ed by our observation that the size of district grant is higher, on average, for the oligarchic communities in 2007 (see Table 4). Second, the FD law 25/99 introduced a ‘…scal needs formula’for the transfer of funds from the government based on geographic and demographic characteristics (Brodjonegoro 2001). Accordingly, we control for community attributes like geographic size, population, if Islam is the dominant religion and also if it is a rural community to determine the likelihood of STATUS. We use the 2007 year dummy as the proxy for FD that accounts for the changing allocation of grants from the centre, state and district governments. We also interact all arguments discussed above with the FD dummy to identify the di¤erential e¤ect of any of these characteristics on the likelihood of political transition (measured by STATUS) in the post-FD year 2007. Finally we include a set of district dummies to account for the unobserved district level characteristics that may also in‡uence the likelihood of political transition; as such our estimates entail the withindistrict variation in political transition. In order to minimize omitted variable bias, we also include district level time trends as captured by (District*FD). Identi…cation of the e¤ect of community homogeneity relies on the fact that these homogeneity measures are independent of the introduction of …scal decentralization. We test this by regressing these homogeneity measures on the dummy for FD, district dummies and also (district*FD) dummy. Table A5 summarizes these estimates; insigni…cance of the FD dummy in each column supports the assumption of exogeneity of these homogeneity measures in our sample. Using the …rst stage estimates of STATUS, we then generate the predicted probabilities of political transition ST AT U S as p0; p1; p2; p3 corresponding to the values 0; 1; 2; 3 assumed by the variable STATUS, which are then inserted (leaving aside predicted probability p0 when STATUS=0
18
that acts as the reference category) in equation (1) to estimate their impact on share of local development spending. We also interact p1; p2; p3 with the Year 2007 dummy to identify their di¤erential e¤ects on share of development spending in the post-FD year 2007. We re-estimate the equation with the alternative ST AT U S1 variable to estimate the predicted probabilities of ST AT U S1 as p0; p1; p2; p3; p4 corresponding to the values 0; 1; 2; 3; 4 assumed by the variable ST AT U S1 As mentioned earlier, the share of local funds (LF) raised in a community as an argument in equation (1) and (2) is likely to be potentially endogenous to the prevalent economic environment of the community. So we also need to instrument this and it is here that we borrow from Section 3.1 our idea of how presence of local markets can a¤ect local politics and development. We …nd that a local government in Indonesia has the ability of raising funds from local businesses, land and forestry, which can arguably be facilitated by the presence of a market in the community. As such, we use the presence of a market in a community as a proxy for its degree of opportunity to raise local funds because a market is a place where trade takes place between buyers and sellers for all sorts of goods and services. It may be a daily or occasional market (weekly, monthly and/or special occasions) depending on the needs and propensity to buy/sell the particular good among the local population. We observe a signi…cant variation in the access to the market in our sample communities. While about the 43% of sample communities had a local market (in our sense of the term) the corresponding proportion falls to about 30% for rural communities and rises to 51% for urban communities. The correlation between the share of local funds and presence of a market is 0.07 and it is statistically signi…cant at 5% level (with a robust t-statistic of 2:21 and the corresponding R -square value of 0:20 ) for the full sample 1997-2007, while the corresponding bivariate correlation between share of local funds and presence of the market is 0.12 and it is statistically signi…cant at 1% in 2007 (post-FD year). Taken together, we infer that communities with prior access to markets tend to have signi…cantly higher share of local funds in our sample and the relationship is stronger in 2007 than that for the full sample 1997-2007. As expected, we …nd that access to local market is uncorrelated with the distance to the district head quarter (the correlation coe¢ cient is 0.004 and it is statistically insigni…cant too). Accordingly, we instrument the community’s share of local funds by a binary variable indicating if there is any ‘market’in the community after controlling for other observable community characteristics including community size, population, access to various public infrastructure (road, transport, sea, PTO, PO etc.) , if rural and also if Islam is the dominant religion. Estimates for the share of local funds are shown in Appendix Table A5, which we use to generate an instrument for share of local funds. Our …nal objective is to obtain the estimates of local development and in this respect too we need to ensure some identi…cation restrictions to obtain the consistent estimates. First, note that we do not include the …rst stage identifying variables, namely, size of the largest population group, adherence to adat, leader’s local residency and also the distance of the community from the district head quarter13 (and also their interactions with the FD dummy Yr2007) for the second 13
Clearly, the distance to district head quarter is geographically determined and is beyond the control of the com-
19
stage determination of equation (1) representing local development. The identifying variables at the second stage are the predicted probabilities of political transition p1 ; p2 ; p3n for ST AT U S and p1 ; p2 ; p3 ; p4 for ST AT U S1 respectively (relative to the reference group p0 ). The variables which are included in both stages (determination of political transition equation (2) and the share of development spending equation (1)) relate to leader’s education, and the share of local funds raised (its predicted value) and also some community characteristics including population size, geographic area, composite public infrastructure, rural dummy and also its access to the sea. An additional issue arises from the fact that local community leaders became directly accountable to the district government after FD. While we do not observe the characteristics of the districts, the latter may have also in‡uenced the nature of local politics and/or its political transition. Hence, we control for district dummies in equation (1). In addition to including the dummy F D for the year 2007, we also include district F D to account for the unobserved district-level time trends, if any, which may also in‡uence the local development outcome. Given that district government regulates their activities, especially after 2001, there could be correlations of errors over time for a community in a district; hence we cluster all standard errors at the district level to minimize any bias arising from these correlations over time for a community in a district. Accordingly, we revise equation (1) to obtain the estimates of local development spending within a DID framework with district …xed e¤ects as follows:
Yit =
0
+
2D
1 p1
+
2 p2
p2 F Dt +
X Xit
+
d
+
+
3 p3
+
4 p4
+
p3 F Dt +
3D
D F Dt
+
1D
p1 F Dt +
p4 F Dt +
4D
F Dt + uit ;
d
(3)
where The expected value of Y for the treated communities in the pre-FD year is Yit =
0
+
1
+
2
+
3
+
4;
The expected value of Y for the treated communities in the post-FD year is Yit =
0
+
1
+
2
+
3
+
4
+
D
+
1D
+
2D
+
3D
+
4D ;
munity authority and as such one can usually treat it as purely exogenous. Note that the IFLS provides information on district head quarter in both 1997 and 2007 that we use. Thus although some districts may split between 1999 and 2007, this is unlikely to a¤ect our results as we would have the time-varying distance in our sample for these a¤ected districts. Further, we control for a number of community characteristics including its size, population and rural location(see further discussion below) with a view to limit that the distance variable does not proxy for anothing other than the community’s closeness to the district authority.
20
The expected value of Y for the control communities in the pre-FD year is Yit = The expected value of Y for the control communities in the post-FD year is Yit =
0, 0
and +
D;
Thus the di¤erential e¤ect of various political treatments in the treated communities in the post–FD year relative to that in the control group equals 1D + 2D + 3D + 4D In other words, the estimated coe¢ cients of T D , T = 1; 2; 3; 4 in the DID model (3) above would capture the di¤erential e¤ects of di¤erent types of political transitions on the share of local development spending in the treated communities (relative to the control group STATUS1=0), after controlling for all possible observable and unobservable factors that may also in‡uence local development in our sample. Since the ATET estimate obtained from the multiple treatment e¤ects may su¤er from the omitted variable bias, the DID estimates for T D ; T = 1; 2; 3; 4 remain our preferred estimate of a causal e¤ect of FD on local development via di¤erent types of politically transition,14 which would account for the di¤erential e¤ects of political transitions (arising from FD) on local development spending after FD. Note that consistent estimates of the interaction terms T D ; T = 1; 2; 3; 4 requires that political transition and FD dummy are uncorrelated with the error term. An essential precondition for ensuring this is that there are parallel trends in …rm outcomes of treatment & control groups and hence we test for the parallel trend assumptions (see section 4). We also check the relevance and validity of these instruments p1 ; p2 ; p3 ; p4 ; p3n for the determination of local development using a likelihood ratio Chi-square statistics (see discussion in section 4).
4
Findings
The report in this section is divided into two parts. First we report the results pertaining to political transition. We then consider the e¤ect of various types of political transitions on the share of local development spending, using both multi-valued treatment e¤ects and di¤erence-in-di¤erence estimates.
4.1
Change in grassroots politics: the importance of sociocultural homogeneity
Table 5 shows the multinomial logit estimates of political transition using the variable Status while Table 6 shows those using the extended political transition measure Status1. Columns (1)-(3) in Table 5 show the estimates for new polity with old leader, old polity with new leader and new polity with new leader relative to the control group of stable polity with old leader. Similarly columns (1)-(4) of Table 6 shows the estimates of new polity with old leader, old polity with new leader, new oligarchy with new leader and new democracy with new leader respectively. We …nd that there is no evidence that population homogeneity in terms of Ethnicity has any signi…cant impact on 14
Appendix Table A1 shows the means and standard deviations of all regression variables.
21
political transition of any type in the pre/post FD years. However, there is signi…cant support that homogeneity de…ned in terms of community’s adherence to Traditional Adat Laws in‡uences the likelihood of political transition. Note that we include two variables Adat and Adat*Year2007. While Adat*Year2007 re‡ects the di¤erential e¤ect of adhering to Adat in 2007 (post-FD), Adat coe¢ cient by itself re‡ects that for 1997, before FD was introduced. The coe¢ cient of Adat is positive and statistically signi…cant not only for new polity with new leader (Table 5), but also for new democracy/oligarchy with new leader (see columns 3 and 4 in Table 6). Thus we conclude that communities with strict adherence to Adat laws in 1997 experienced signi…cantly higher likelihood of new polity (including both new oligarchy and new democracy) with a new leader in 2007 as summarised in Tables 5 and 6 respectively. However, and interestingly, the interaction term Adat*Yr2007 remains insigni…cant in column (3) of Table 5, but turns out to be negative for new oligarchy with new leader in columns (3) and (4) of Table 6. So the total e¤ect of strictly ahdering to Adat is positive for new democracy with new leader as only the individual e¤ects in 1997 (adat coe¢ cient by itself) is statistically signi…cant. However, for oligarchic communities these individual e¤ects for 1997 and 2007 are both signi…cant and are of opposite signs. This makes us conclude that while community homogeneity in 1997 gave birth to new democracies and new oligarchies at the time of FD, change towards democracy continued thereafter particularly in homogenous communities while change towards oligarchy was reversed. This brings out the direct e¤ect of FD on immediate changes in grassroots politics and at the same time lends support to John Stuart Mills hypothesis that community homogeneity and democracy would eventually go hand in hand. While community’s adherence to Islam is not statistically signi…cant for new polity with new leader, it is signi…cant for new polity with old leader. In particular, strict adherence to Islam in 2007 (as re‡ected in the coe¢ cient of Islam*Yr2007), holding other factors constant, signi…cantly enhances the likelihood of a new local political environment (with old leader) in our sample; the coe¢ cient of Islam by itself is positive though remains insigni…cant so that the total e¤ect of being Islamic communities is likely to be positive for new polity with old leader (this holds both in Table 5 and Table 6). Taken together, this evidence arguably provides support to our hypothesis that only socio-cultural dimension of community homogeneity matters for political transition in our sample. More importantly, communities that strictly adhered to adat laws in 1997, are more likely to experience any type of political transition- new democracy or new oligarchy with new leader. What explains the slightly di¤erential role of strict adherence to Adat and Islam on political transition likelihood? We argue that adat communities were more co-operative (Pal and Wahhaj, 2015) and also that a signi…cantly higher proportion of these communities were underdeveloped in 1997 in that they were selected for the government’s poverty alleviation programme IDT (see further discussion section 4.2.2); the latter in turn may provide greater impetus for political change, manifesting in new politics and a new leader in strictly adat (as opposed to Islamic) communities in our sample. It also follows that new polity with new leader in 2007 crucially depends on whether the leader 22
is a local resident in 2007 and not in 1997 (given that new leaders in 2007 are most likely to be elected after FD). Evidently, the leader turnover after FD is more likely when the leader is not a local resident in the community (e.g., see the coe¢ cient estimate of local resident*Yr2007 in columns (2) and (3) as opposed that in column (1)). However, local residency fo the leader by itself that signi…es local residency of the leader in 1997 remains insigni…cant to determine the likelihood of political transition with new leader; this is because all new leaders have been elected in the post-FD years. In a similar vein, only leader’s education in 2007 matters for political transition after FD; in particular, higher schooling of the leader boosts the likelihood of any political transition with new leader (see columns 2 and 3). There is however no evidence that distance from the district head quarter played any signi…cant role in initiating a change in local polity (democracy/oligarchy), a change in local leader or both. In other words, insigni…cance of the distance variable can be taken as evidence of freedom from the district authority in the post-FD years. Essentially we get rather similar results from both Table 5 and Table 6. Our interpretation of the interaction e¤ects (with respect various explanatory variables) in the determination of political transitions relies on two assumptions. First, our measure of community homogeneity must not be an outcome of FD. We support this by showing that FD had no signi…cant e¤ect on the time-varying community homogeneity measures as shown in columns (1)-(3) of Table A5. Second, we assume that, conditional on our baseline controls, the interactions of the FD dummy and community homogeneity measures are not jointly determined with local development. In other words, we assume that homogeneity measures are not correlated with other factors (beyond the control factors) that can in‡uence the e¤ect of FD on development. This is highly likely since the baseline controls for the interaction of homogeneity and FD dummy. Taken together, we …nd support to our conjecture that communities that strictly adhered to Adat laws and Islam were able to bring about a change in the local politics (democracy to oligarchy or oligarchy to democracy). In our sample, leader turnover was largely for existing absentee-leaders and also if the leaders were less educated - there was a preference for more educated leaders in the post-FD years.
4.2
Change in local development : the importance of political transition
In this section we consider the e¤ect of political transitions on local development, using both multiple political treatment e¤ects and di¤erence-in-di¤erence methods, after controlling for other possible factors that may also in‡uence local development. 4.2.1
Multivalued political treatment e¤ects
First we compare the share of development spending (outcome) for each treatment using both measures of political transition ST AT U S and ST AT U S1 > 0 relative to the reference category ‘old polity (democracy/oligarchy) with old leader’. In doing so, we not only control for measures of community homogeneity, but also leaders’ education and local residency determining leader 23
turnover. We also include control for other community characteristics including geographic size, total population, share of local funds (iv), if the community is rural , if it has access to sea, a composite index of access to various public infrastructure and also a dummy for the year 2007 to account for post-FD years. We start with the multivalued treatment e¤ects estimates using two alternative methods, namely, regression adjustment and also inverse probability weighted regression adjustment (IPWRA). These estimates are shown in Tables 7 and 8 respectively: in each case, column 1 shows the estimates using the political transition measure Status while column (2) shows the corresponding estimates using the alternative political transition measure Status1. For each case we show the average treatment e¤ects on the treated (ATET) and also the corresponding potential outcome of means (POM). Evidently, the ATET is statistically signi…cant only for new polity with old leader and this holds for both Status and Status 1. First ATET estimates are quite comparable for a given political transition measure, irrespective of the choice of method (RA and IPWRA): it is 0.256 and 0.213 for Status and 0.333 and 0.254 for Status1. Second, we compare the estimates using Status and Status1 for a given method (RA or IPWRA): 0.256 and 0.333 for regression adjustment method and 0.213 and 0.254 for the IPWRA method; in other words, ATETs are somewhat lower when we use IPWRA method, which can be attributed to the di¤erence in the methods. While the e¤ects are positive for new polity with new leader in all cases (even when we split it into new democracy with new leader and new oligarchy with new leader), it is never statistically sign…cant. Further, the potential outcome of means are positive and statistically signi…cant in each case; these are also comparable across the two political transition measures Status and Status1 for a given method of estimation. However identi…cation in these multiple treatment e¤ects estimates are based on observables only and as such these estimates are likely to be biased, especially if the unobservables are correlated with the errors. In order to redress this shortcoming, we next move to a panel di¤erence in di¤erence model with district …xed e¤ects where we are able to account for district level time-invariant and time-varying unobserved factors. 4.2.2
Di¤erence-in-Di¤erence IV estimates
Table 9 summarizes the di¤erence-in-di¤erence (DID) estimates of the share of development spending in our sample using predicted values of political transition and their interactions with the FD dummy. All estimates satisfy the instrument relevance, validity and over-identi…cation restrictions. We start by comparing mean shdev of various treatment groups with the control communities (C - where we had stable polity with old leader, i.e., when STATUS1=0) in 1997 using t-statistic. Table A2 summarizes the t-tests for mean comparisons for di¤erent types of political transitions in 1997. The mean shdev for the C group is 0:57 and for the T group is 0:50 and the associated t-statistic is 1:3224. So there is no signi…cant di¤erence in the outcome variable shdev between these two groups in 1997 (i.e., the pre-FD year). While the mean share of development spending is 0.51 for the control group, it varies between 0.47 to 0.63 for communities facing di¤erent types of political
24
transitions as captured by ST AT U S1=1,2,3,4. Clearly mean share of development spending is the highest (0.63) in communities with stable polity but a new leader while it is the lowest (0.47) in communities with new oligarchy with a new leader. Mean share of development spending is 0.49 in newly democratic communities with a new leader. More importantly, however, none of the t-statistic are statistically signi…cant in Table A2, thus suggesting that the sample communities were rather comparable in terms of share of development spending in 1997 irrespective of the type of political transitions they experienced. An essential identi…cation condition for the DID model to function well is that the treatment and control groups only di¤er in terms of the treatment: impact of time is constant across groups and also the impact of group is constant across time, commonly labelled as the parallel trend assumption. Thus estimates of DID model will be biased if control and treatment groups do not satisfy the parallel trend assumption. In order to test this we regress our share of development spending on dummies representing di¤erent political transitions (ST AT U S1_i, i=1,2,3,4), Year 2007 dummy measuring post-FD years and also the respective interaction terms (STATUS1_i*Year2007), i=1,2,3,4. These estimates are summarized in Table A3 in the Appendix. Although somewhatcrude crude, it follows that the estimated coe¢ cient of the interaction terms (STATUS_i*Yr2007) for i=1,2,3,4 remain insigni…cant, thus rejecting the case for signi…cant di¤erential trends in the treatment and control groups, which in turn, validates the use of the di¤erence-in-di¤erence model. Table 9 summarizes these estimates - column 1 shows the estimates using ST AT U S while column (2) shows those using ST AT U S1. We look at the DID estimates of share of development spending as a function of the predicted value of political transition (using estimates summarized in Tables 5 and 6 respectively for ST AT U S and ST AT U S1). Our objective here is to focus on the variation in local development measure among heterogeneously treated (relative to control) communities within a district before and after the introduction of FD. In this respect, we are particularly interested in the estimated coe¢ cients of the interactions terms T D ; T = 1; 2; 3 or T = 1; 2; 3; 4;depending on whether we use Status or Status1 as the relevant measure of political transition that highlights the di¤erential e¤ect of FD on share of development spending for a given type of political transition. So the coe¢ cients of interest for us are the following:
Political transition measured by Status: : Coef f ( p1 F D): Di¤erential e¤ect of FD on communities with new polity with old leader (predicted) 1D
2D : Coef f (p2 F D): Di¤erential e¤ect of FD on communities with old polity with new leader (predicted)
: Coef f (p3 F D): Di¤erential e¤ect of FD on communities with new polity with new leader (predicted) 3nD
25
Political transition measured by Status1: 1D : Coef f ( p1 F D): Di¤erential e¤ect of FD on communities with new polity with old leader (predicted)
: Coef f (p2 F D): Di¤erential e¤ect of FD on communities with stable polity with new leader (predicted) 2D
3D : Coef f (p3 F D): Di¤erential e¤ect of FD on communities with new oligarchy with new leader (predicted)
: Coef f (p4 F D): Di¤erential e¤ect of FD on communities with new democracy with new leader (predicted) 4D
Considering the estimates shown in column (2) of Table 9, we …nd positive and statistically signi…cant interaction coe¢ cients for (p1 F D) and (p4 F D) while other interaction coe¢ cients (p2 F D) and (p3 F D) remain statistically insigni…cant. In particular, relative to the control group, communities with a new polity with old leader (as re‡ected in the estimated coe¢ cient of (p1 F D)) experienced an average increase of 0.136 in share of development spending; the corresponding increase in share of development spending is 0.237 for new democracies with new leader (as re‡ected in the estimated coe¢ cient of (p4 F D)) . Interestingly and reassuringly, sign of the estimated coe¢ cient of (p1 F D) in Table 9 is broadly compatible with the corresponding multiple treatment e¤ects estimates shown in Tables 7 and 8. While new democracies with new leader turns out to be positive and signi…cant in determining the share of development spending within DID framework, the corresponding e¤ect was not signi…cant in Table 7 or 8. We also …nd that the marginal e¤ect of new democracies with new leader is greater than that for new polity (democracy/oligarchy) with an old leader: 0.237 > 0:136. Note however, that no other interaction terms including that for new oligarchy with new leader turns out to be statistically signi…cant; as such there is no evidence to suggest that new oligarchy with new leader performed better than any old polity with new leader or new polity with old leader. Further communities with larger geographical size tend to have signi…cantly lower share of development spending while rural and Islamic communities tend to have higher share of development spending. We then compare these estimates with those in column (1) of Table 9 that shows the corresponding estimates using the alternative political transition variable ST AT U S. Note that we still get a signi…cant positive development e¤ect for communities characterized by new polity with old leader (as re‡ected in the estimated coe¢ cient of (p1 F D)) . However, the development e¤ect of new polity with new leader is positive though remains insigni…cant in column 1 (as re‡ected in the estimated coe¢ cient of (p3n F D)), as we merge new oligarchy with new leader and new democracy with new leader into a single category called ‘new polity with new leader’. In this sense, the column (2) estimates obtained using the political transition variable Status1 appears clearer; these also highlight the fact that the opposite e¤ects of new democracy with new leader and new oligarchy with new leader may cancel each other, at least to some extent, when we merge them 26
together into new polity with new leader(see column (1) of Table 9).15 Note however that this does not happen for new polity with old leader (see columns 1 and 2 of Table 9). It also indicates that the contrasting e¤ect of new democracy and new oligarchy is most pronounced when the leader is new. One possibility is that a newly democratic (oligarchic) community with old leader may continue to function as old oligarchy (democracy) and as such the distinction between new democracy and new oligarchy gets blurred here. Noting that 86% of communities with new polity had a new leader in 2007, our most robust estimates pertain to communities with new polity with new leader as hypothesized. Given the potential concern (e.g., measurement error) about the community spending variable as a measure of development, we further examine the e¤ect of …scal decentralization on the community’s development status as measured by a binary variable U ndev; indicating if the community in question has been selected for the government’s anti-poverty programme IDT (see discussion in section 2). As before, we assess the impact of FD on the likelihood of community being designated as underdeveloped by exploiting the variation among politically treated (relative to control) communities. Given the binary nature of the dependent variable U ndev, we use a probit maximum likelihood model in this respect. These estimates are summarized in Table 10 using the two political transition variables ST AT U S and ST AT U S1of our choice. As with the share of development spending, we …nd that the likelihood of being underdeveloped is signi…cantly lower in communities with new democracy and new leader after FD, thus supporting our central …nding from Table 9. In summary, there is evidence that relative to the control communities with stable polity with old leader, treated communities with new polity experience signi…cantly higher share of development spending in 2007: in particular, there is suggestion that communities with a new polity (democracy/oligarchy) will experience higher development spending in the post-FD years, even when the leader remains the same. However the contrasting e¤ect of new democracy and new oligarchy is most pronounced when there is a change of leader: we …nd that new democracies with new leader will perform signi…cantly better than new oligarchies with new leader, thus highlighting the bene…cial roles of both arrival of democracy and new leadership in the post decentralization years. The latter holds even when we use the alternative measure of development U ndev. Our empirical …ndings suggest that there are three aspects of political transition play a positive role for local development: (i) that there is a change in the way a community selects its leaders, (ii) ability of the new selection mechanism to ‘employ’ a new leadership and (iii) that this new selection mechanism is indeed democratic. Given our theoretical narrative in Section 1 and Section 3, these …ndings are consistent with the position that change in local polity is not without a social purpose and in many sense encouraged by local communities who envisage, correctly on average, that political change is an important engine for development when …scal responsibilities go local. It works as a signal to potential leaders that the community has found a common voice for better governance (to ensure more development) and such signals are typically generated from communities 15
We were unable to obtain estimates of …ner political transition classi…cations because of the limited number of observations in certain categories, e.g., new polity with old leader.
27
that are socioculturally homogenous (rather than ethnically or economically) so that producing a common voice is relatively easy. Finally, if the signal of the existence of a common voice brings new leadership, it further indicates that the community is on average successful in generating an e¢ cient candidature. Of course this e¢ ciency is found to be more prominent in our data set when the community selects new leadership through new democratic means since on average (a) a switch to a new oligarchy can at best ‘warn’ an unchallenged leader of future consequences while (b) a switch to new democracy not only signals the chosen leader that his/her performance will be closely judged but also in addition allows better current period selection through electoral competition.
5
Conclusion
The paper explores whether and how the new freedom brought about by a nation-wide policy of …scal decentralization is used by di¤erent types of communities in order to improve their development prospects. Considering the introduction of …scal decentralization via Law 22/99 and 25/99 in 2001 as a natural experiment, we use the recent 1997 and 2007 rounds of the Indonesian Family Life Survey (IFLS) community-level data to address this question. We …nd that FD was followed by multiple changes in local politics with and without changes in the old leader. We argue that change is a necessary way for communities to signal their intent to political candidates and that such signals are therefore more likely to be exercised in socioculturally homogenous communities where political mobilization is easier to obtain. Our estimation strategy is then to compare local development in various types of politically transiting treated communities with that for the stable control group (which did not experience any change in local polity and had the same old leader even after FD). We compare the non-parametric multiple treatment e¤ects estimates with the parametric di¤erence-in-di¤erence estimates which is an improvement over the multiple treatment e¤ects. As such our best estimates come from the di¤erence in di¤erence model with district, year and district*year …xed e¤ects that accounts for year and district-speci…c unobserved time trends. These results suggest that more homogeneous communities which were able to strike a common voice were more likely to experience a change in local politics. Further, communities which transited out from oligarchies to embrace democracy with/without change in community leader experienced signi…cantly higher share of development spending and the development impact was the highest in new democracies with new leader. In contrast, local development spending shares were signi…cantly lower in communities that exited out of democracy. These results highlight that one possible way FD impacts local development is through change in local politics and it may generate heterogenous local development depending on the nature of political transition. This is an important feature of the economics of …scal decentralization that has largely been overlooked in the existing literature. Lastly and most importantly, we …nd that while community homogeneity is a precondition for the success of grassroots democracy in the long run, it is not a necessary precondition for the initiation of democracy itself. On the contrary, community homogeneity seems to be a precondition for a political change, including a 28
change towards non-democratic practices, at least in the short run. Of course, we do not expect this to be a steady state impact and we …nd evidence for that as well. Once communities successfully signal their resolve through political changes, Mill’s theory applies and help such homogeneous communities to eventually converge to democratic governance.
References [1] Acemoglu, D. and J. A. Robinson (2000), “Why did the West extend the Franchise? Democracy, Inequality and Growth in Historical Perspective,” The Quarterly Journal of Economics, 115, pp. 1167 - 1199. [2] Acemoglu, D. and J. A. Robinson (2001), "A Theory of Political Transition", American Economic Review 91(4): 938-63. [3] Acemoglu, D., S. Johnson, J. A. Robinson and P. Yared (2008), “Income and Democracy,” American Economic Review, June, 98(3), pp. 808 - 842. [4] Alesina, A. and E. L. Ferrara, (2000), ‘Participation in Heterogenous Communities,’Quarterly Journal of Economics, pp. 847-904. [5] Banerjee, Abhijit and Rohini Somanathan (2007). “The political economy of public goods: Some evidence from India,” Journal of Development Economics, 82 (2), pp. 287–314. [6] Banerjee, Abhijit, Lakshmi Iyer, and Rohini Somanathan (2005). “History, Social Divisions, and Public Goods in Rural India,” Journal of the European Economic Association, 3 (2-3), pp. 639–647. [7] Bardhan, P. (2002). ‘Decentralization of governance and development,’ Journal of Economic Perspectives, pp. 185-205. [8] Bardhan, P. and D. Mookherjee (2000). ‘Capture and governance at local and national levels,’ American Economic Review, 90, pp. 135 - 139. [9] Bardhan, P. and D. Mookherjee (2006). ‘Decentralization, corruption and government accountability’in Susan Rose-Ackerman ed. International Handbook on the Economics of Corruption, Edward Elgar. [10] Besley, T., R. Pande, L. Rahman and V. Rao. (2005). ‘Participatory democracy in action: Survey Evidence from India’, Journal of the European Economic Association, Papers and Proceedings, 3, pp. 648-57. [11] Brodjonegoro, B. (2001). ‘Indonesian Intergovernmental Transfer in Decentralization Era: The Case of General Allocation Fund’, Paper Presented at the International Symposium 29
on Intergovernmental Transfers in Asian Countries, Asian Tax and Public Policy Program, Hitotsubashi University, Tokyo, February 2001. (Accessed at http://www.econ.hitu.ac.jp/kokyo/symposium/PDF/Indonesia.PDF) [12] Cai, H. and D. Triesman (2004). ‘State corroding federalism,’ Journal of Public Economics, 88, pp. 819 - 843. [13] Cattaneo, M., D. Drukker and A. Holland. (2013). ’Estimation of Multivalued Treatment E¤ects under Conditional Independence’, Stata Journal 13(3): 407-450. [14] Cattaneo, M. (2010). E¢ cient Semiparametric Estimation of Multi-valued Treatment E¤ects under Ignorability’Journal of Econometrics 155(2): 138-154. [15] Dahl, Robert (1971).‘Polyarchy, Participation and Opposition,’ New Haven/London: Yale University Press. [16] Davoodi, H. and H. Zou (1998). ‘Fiscal decentralization and economic growth: a cross country study,’Journal of Urban Economics, 43, pp. 244 - 257. [17] De Luca, Miguel, Mark P. Jones and Maria Ines Tula. (2002). ‘Black rooms or ballot boxes: candidate nomination in Argentina,’Comparative Political Studies, 35, pp. 413 - 436. [18] De Mello, L. and M. Barenstein (2001). ‘Fiscal decentralization and governance: a crosscountry analysis,’IMF Working Paper No. WP/01/71. [19] Enikopolov and Zhuravskaya (2007). ‘Decentralization and political institutions,’ Journal of Public Economics, 91, pp. 2261 - 2290. [20] Fishman, R. and R. Gatti (2002). ‘Decentralization and corruption: evidence across countries,’ Journal of Public Economics, 83, pp. 325 - 346. [21] Frankenberg, E. and D. Thomas. (2000). ‘Study Design and Results from Waves 1 and 2,’ Santa Monica, CA, Rand, DRU - 2238, Volumes 1-7, NIA/NICH. [22] Hayek, F. A. (1948). ‘Individualism and economic order,’University of Chicago Press, Chicago. [23] Inman, R. and D. Rubinfeld (1997a). ‘Rethinking federalism,’ Journal of Economic Perspectives, 11 (4), pp. 43 - 64. [24] Inman, R. and D. Rubinfeld (1997b). ‘The political economy of federalism,’in: Mueller, D. C. (ed.), Perspectives on Public Choice: A Handbook. Cambridge University Press, Cambridge. [25] Kato, T. (1989). "Di¤erent Fields, Similar Locusts: Adat Communities, and the Village Law of 1979 in Indonesia", Indonesia, Vol. 47, pp. 89-114.
30
[26] Lerner, Daniel (1958). ‘The Passing of Traditional Societies’, The Free Press. [27] Lipset, S. M. (1959). ‘Some Social Requisites of Democracy: Economic Development and Political Legitimacy,’The American Political Science Review, 53(1), pp. 69 - 105. [28] Marinez-Bravo, M. (2014) ‘The Role of Local O¢ cials in New Democracies: Evidence from Indonesia,’American Economic Review, 104(4) (2014), 1-45. [29] Maskin, E., Y. Qian and C. Xu. (2000). ‘Incentives, information and organizational form.’ Review of Economic Studies, 67 (2), pp. 359 - 378. [30] Mill, John Stuart (1861) ‘Considerations on Representative Government’, London: Parker. [31] Monitola, G., Y. Qian and B. R. Weingast. (1995). ‘Federalism, Chinese style: the political basis for economic success in China,’World Politics, 48 (1), pp. 50 - 81. [32] Mookherjee, D. (2014). Political Decentralization, mimeo Boston University. [33] Munshi, Kaivan and Mark Rosenzweig. (2008). “The E¢ cacy of Parochial Politics: Caste, Commitment, and Competence in Indian Local Governments,”NBER Working Papers 14335. [34] Oates, W. (1972). ‘Fiscal Federalism,’Harcourt, New York. [35] Olken, B. (2004). ‘Monitoring corruption: evidence from a …eld experiment in Indonesia’NBER Working Paper, Cambridge, MA. [36] Olken, B. (2007). ‘Political Institutions and Local Public Goods,’mimeo, Harvard University, February. [37] Padro-i-Miguel, G., N. Qian and Y. Yao. (2014). ‘Social Fragmentation, Public Goods and Elections: Evidence from China’, CEPR Discussion Paper No. DP8975. [38] Pal, Sarmistha and Zaki Wahhaj. (2015) "Fiscal Decentralization, Local Institutions and Public Goods Provision", SSRN working paper. [39] Parsons, Talcott (1971) ‘The System of Modern Societies’, Englewood Cli¤s: Prentice Hall. [40] Prud’homme, R. (1995). ‘The dangers of decentralization’, World Bank Research Observer, 10 (2), pp. 21 - 26. [41] Qian, Y. and G. Roland (1998). ‘Federalism and the soft budget constraint,’ American Economic Review, 88 (5), pp. 1143 - 1162. [42] Qian, Y. and B. Weingast (1996). ‘China’s transition to markets: market-preserving federalism, Chinese style,’Journal of Policy Reform, 1, pp. 149 - 185.
31
[43] Rodríguez, C. & P. Meirelles, 2010. ‘Devolution and Accountability E¤ects in the Public Provision of Water Services in Indonesia,’CEDE Series 107395, Universidad de Los Andes. [44] Sonin, K. (2003). ‘Provincial protectionism,’ William Davidson Institute Working Paper No. 557, University of Michigan Business School. [45] Strauss, J., F. Witoelar, B. Sikoki and A. M. Wattie (2009). ‘The 4th Wave of the Indonesian Family Life Survey (IFLS4): Overview and Field Report,’WR - 675/1 - NIA/NICHD. [46] Tiebout, C. (1956). ‘A pure theory of local expenditures,’ Journal of Political Economy, 64 (5), pp. 416 - 424. [47] Tanzi, V. (1996). ‘Fiscal federalism and decentralization: a review of some e¢ ciency and macroeconomic aspects,’ Annual World Bank Conference on Development Economics, 1995. World Bank, pp. 295 - 316. [48] World Bank. (2004) World Development Report. Making Services Work for the Poor. World Bank and Oxford University Press.
32
Tables Table 1. Different methods of selection of community leaders
1997
2007
rural
urban
Rural
urban
Consensus
38
53
15
18
Voting
64
50
100
78
Oligarchy
18
89
5
96
Total
120
192
120
192
Note: ‘Consensus’= 1 if the community leader is selected by consensus building through meetings; ‘Voting=1 if the community leader is elected by voting; ‘Oligarchy’=1 if the community leader is selected by few elites. Table 2. Political transition matrix 1997-2007 Community Classification by type of political transition (1997-2007)
(1) Number of communities
(2) Observed probability
Stable polity, old leader: 0
78
12.5
New polity, old leader: 1
28
4.49
Old polity, new leader: 2
364
58.33
New oligarchy, new leader: 3
76
12.18
New democracy new leader: 4
78
12.5
Total
624
100
Note: In this table we define a polity to be democratic if the leader is selected by consensus building or voting and oligarchic if selected by few elites. Thus a stable polity refers to the cases where democratic (oligarchic) communities remained democratic (oligarchic) in 2007. A second dimension of political transition is captured by whether there was a change in leader in communities with stable/new polity (democratic/oligarchic). Combining these two dimensions, we consider 5 possible treatments of fiscal decentralisation as described by community classification in this table and captured by the categorical variable STATUS1 that takes a value 0, 1, 2, 3, 4. We also generate a slightly aggregative political transition indicator Status which merged category 3 and 4 into one as ‘new polity new leader’ though our preferred measure is STATUS1, which allows us to distinguish between new democracy from new oligarchy.
33
Table 3. Distribution of grants and spending in 1997 and 2007 Variables
(1) 1997 Mean (sd)
(2) 2007 Mean (sd)
(3) T-statistics
Central grant share
0.33 (0.31)
0.07 (0.17)
12.1759***
Provincial grant share
0.14 (0.29)
0.13 (0.27)
0.3708
District grant share
0.09 (0.18)
0.41 (0.35)
-12.1928***
All grants share
0.57 (0.36)
0.61 (0.33)
-1.2010
Share of local funds
0.38 (0.37)
0.39 (0.33)
-0.4489
Share of development spending
0.55 (0.41)
0.48 (0.29)
2.2706**
Source: IFLS data. This table shows the comparisons of means of selected revenue and spending amounts in 1997 and 2007.
34
Table 4. Distribution of grants and spending by nature of political transition in 2007 Political transition
Community’s share of
status
Central grant
District grant
Local funds
Development spending
Old polity, old leader: 0
0.04
0.38
0.44
0.48
Old democracy, old leader
0.05
0.41
0.47
0.50
Old oligarchy, old leader
0
0.33
0.22
0.39
New polity, old leader: 1
0.09
0.33
0.46
0.45
New democracy, old leader
0.12
0.23
0.61
0.46
New oligarchy, old leader
0
0.76
0
0.42
Old polity, new leader: 2
0.07
0.39
0.41
0.47
Old democracy, new leader
0.096
0.40
0.43
0.49
Old oligarchy, new leader
0.014
0.37
0.37
0.38
New polity new leader
0.06
0.47
0.31
0.51
New oligarchy, new leader: 3
0.03
0.48
0.33
0.48
New democracy new leader: 4
0.09
0.45
0.30
0.55
Note: Here we show the mean share of central grant, district grant and development spending for the sample communities classified by the nature of political transition.
35
Table 5. Multinomial logit estimates of multiple political transition measure Status
VARIABLES Post FD - Year 2007 Ethhety: Σi{1 square(Pop1)i} Ethhety*2007 Adat1 Adat1*2007 Islam Islam*2007 Head local resident Head local resident*2007 Distance from district HQ (DHQ) DHQ*2007 Head had at least high schooling (Headshs) Headshs*2007 Community size Size*2007 Share of people with junior high school (Jhs) Jhs*2007 Share of local funds IV Local funds_IV*2007 Rural
Dependent variable STATUS (1)New (2) Old (3) New polity old polity new polity new leader leader leader -2.709*** 0.0995 -1.236 (0.408) (2.414) (2.919) -0.0395 0.905 0.379 (1.645) (0.917) (1.047) 1.847 -1.284 -0.808 (2.368) (1.251) (1.434) 0.406 0.710 1.216** (0.939) (0.441) (0.522) 2.209 -0.126 1.624* (1.345) (0.706) (0.835) 0.604 0.143 0.340 (0.843) (0.508) (0.640) 1.353*** 0.129 -0.466 (0.124) (0.945) (1.101) -1.634 -1.024 -1.176 (1.035) (0.733) (0.756) 1.414*** -1.700 -2.101* (0.163) (1.226) (1.262) -0.00231 0.0127 0.00751 (0.0604) (0.0341) (0.0357) -0.0386 0.00963 -0.0119 (0.108) (0.0449) (0.0517) -0.272 -0.0181 0.135 (0.822) -1.107 (1.318) 0.181 (0.173) -0.130 (0.342) 5.964**
(0.418) 2.387*** (0.653) -0.115 (0.136) 0.138 (0.196) 1.388
(0.526) 3.740*** (1.213) 0.186 (0.144) 0.0268 (0.234) 2.030
(2.908) -2.157 (4.184) -0.897 (2.557) -0.167 (4.589) 0.274 (1.113)
(1.398) -1.452 (1.927) -1.322 (1.379) -1.181 (2.167) 0.614 (0.504)
(1.673) -0.889 (2.387) -4.291** (1.737) -1.083 (2.628) -0.869 (0.681)
36
Rural *2007
-1.139 0.00169 1.493 (1.403) (0.731) (0.973) Sea -1.670** -0.191 -0.394 (0.746) (0.426) (0.512) Sea *2007 -0.556 0.132 0.695 (1.222) (0.635) (0.764) Constant -2.292 2.479 1.183 (2.544) (1.668) (1.738) Observations 607 607 607 Robust standard errors in parentheses; *** p
