Fscal Decentralzaton and Development: How Crucal s Local Poltcs? Sarmstha Pal, Brunel Unversty & IZA * Jadeep Roy, Unversty of Brmngham September 2010 Abstract Does fscal decentralzaton n a poltcally decentralzed less developed country help strengthen democratc nsttutons at the grass root level? And s the mpact of such decentralzaton on local poltcs mportant n determnng local development? Our study on Indonesa suggests that fscal decentralzaton enhanced free and far local electons, though the ncdence of elte capture, and the consequent breakdown of local democracy, was also present n sgnfcant proportons. Fscal decentralzaton promoted development mostly n communtes whch transted out from elte capture to embrace free and far electons. Ths was followed by communtes that experenced the emergence of elte capture. Communtes that contnued to reman under ether elte capture or free and far electons dd the worst. These fndngs suggest that whle the emergence of elte capture exsts, t may not necessarly be the most harmful. Instead, and surprsngly so, stablty of local polty hurts development the most. Keywords: Local poltcs, Less developed naton, decentralzaton. JEL classfcaton: D72, H77, O18 * Correspondng author: Department of Economc s and Fnance, Brunel Unversty, Kngston Lane, Uxbrdge UB8 3PH, Mddlesex, UK. E-mal: Sarmstha.pal@brunel.ac.uk. We are grateful to Facundo Albornoz-Crespo Sddhartha Bandyopadhyay, Anndya Banerjee, Abgal Barr, Erlene Berg, John Fender, Sugata Ghosh, Anrban Mtra and semnar partcpants at CSAE Oxford for comments on an earler draft. Any errors are ours. Department of Economcs, Unversty of Brmngham, Edgbaston, Brmngham B15 2TT, UK; E-mal: j.roy.1@bham.ac.uk.
1 1 Introducton Fscal Decentralzaton and Development: How Crucal s Local Poltcs? Decentralzaton s by and large a poltcal decson of natonal leadershp that has mportant economc consequences. It nvolves devoluton of poltcal or fscal powers to local governng bodes. Economsts address largely the queston of whether decentralzaton enhances economc performance. 1 Whle the dstncton between poltcal decentralzaton and fscal decentralzaton s not always made precse n the lterature, t s generally argued that although decentralzaton may ncrease the accountablty of local governments and strengthen the voce of the poor and the margnalsed, t may also enhance the nfluence of the local eltes (see Bardhan (2002)) that then typcally erode local democratc rules of preference aggregaton. There s also an mplct consensus n ths lterature that such elte capture s necessarly harmful for local development. Recently Besley and Coate (2003) and Enkolopov and Zhuravskaya (2007) look at poltcal aspects of fscal decentralzaton. Besley and Coate study the mpact of fscal decentralzaton on local electoral competton n developed natons where free and far local electons s granted. They argue that the sharng of the costs of local publc spendng n a centralzed system s lkely to create a drect conflct of nterest between ctzens n dfferent jursdctons. On the other hand, when spendng decsons are made by a legslature of locally elected leaders, such conflcts transfer to the legslature tself where the group of elected leaders from dfferent jursdctons have to bargan amongst themselves. Ths dfference then translates tself nto dfferent equlbrum outcomes of poltcal competton n local electons. Enkolopov and Zhuravskaya on the other hand argue that the success of fscal decentralzaton depends crucally upon how centralzed s the overall poltcal scenaro n a gven naton. Ther data from a large number of transton economes suggests that fscal decentralzaton s more successful n those economes whch are more poltcally 1 Delegaton of ths nature has several benefts stemmng out of nter-jursdctonal competton a la Tebout (1956), local nformatonal advantages a la Hayek (1948) or preference matchng a la Oates (1972). See also, Montnola et al. (1995), Qan and Wengast (1996), Qan and Roland (1998), Maskn et al. (2000) that argue that decentralzaton s good, and Prud'homme (1995), Tanz (1996), Ca and Tresman (2004) that argue for the opposte effect.
2 decentralzed. None of these studes however explctly nvestgate the possble mechansm(s) as to how fscal decentralsaton may affect local poltcs tself and how that mpacts local development. We am to brdge ths gap n the lterature. We look at an economy where even though the degree of poltcal decentralzaton s hgh, local level electons may not be free and far. Our focus s naturally on developng natons where local communtes have been allowed tradtonally to choose the poltcal process by whch communty level decsons are made. These processes may range from a set of fully democratc local nsttutons that mplement free and far electons of local leaders, to stuatons where a handful of eltes defy democratc processes to promote vested nterests. It s mportant to menton here that poltcal decentralzaton does not necessarly mply grass root democracy n the developng world. Whle not all developng natons are poltcally decentralzed n ths sense, many are. 2 We then ask whether fscal decentralzaton changes the nature of local poltcs (and the magntude of elte capture) and whether (and how) these possble poltcal changes are crucal n understandng the mpact of fscal decentralzaton on local development, whch remans rather unexplored. Indonesa s an mportant case n pont whch has a long hstory of decentralzed local governance. The governance of local communtes (vllages and townshps) has been regulated tradtonally by customary ``adat'' laws ever snce the Dutch colonal rule that recognzed vllage governments as lawful enttes and encouraged self-rule. Local government structures throughout the naton are made up of a headman asssted by an apponted vllage assembly (LMD). The headman s generally elected by communty resdents every 5 to 8 years, although there are also many nstances where the headman s a member of a group of few powerful ndvduals who enforce leadershp va non-democratc means, a case smlar to the noton of elte capture. In any case, development projects and assstance are managed by communty reslence boards (LKMD) whch work under the leadershp of the headman who allocate obtaned development grants across households and projects. All such local governng bodes are accountable to the dstrct government. 2 See for example the UNDP Local Governance webste at http://www.undp.org/governance/focus_local_governance.shtml and the Commonwealth Local Government Forum webste at http://www.undp.org/governance/focus_local_governance.shtml. Also, as suggested by Bardhan and Mookherjee (2000), the extent of elte capture at the local level s lkely to be context and system specfc.
3 Wth the demse of Soeharto, Indonesa experenced an extensve decentralzaton process that has ts roots n Law 22/99, enacted n January 2001. It gave local communtes more autonomy n rasng local revenues, enforced strct budgetary cuts on the central leadershp to supply development grants to these communtes and nstead ncreased the budget of the dstrct governments for ths purpose. It s therefore easy to dentfy the year 2001 as the pont n tme when Indonesa experenced fscal decentralzaton for the frst tme n ts modern hstory whle, as s clear from the dscusson above, the naton s tradtonally a poltcally decentralzed one. We study these local communtes between 1997 and 2007, two years separated by the ntroducton of Law 22/99 n 2001. We also accept that there has been no other major polcy change of related concern durng ths perod n Indonesa that could affect the local polty, barrng of course the clam that there has been a general trend n the world towards democracy (as we shall see, ths possblty can only renforce our results). Many developng countres have been expermentng wth decentralsaton of publc servce delvery to elected local governments rather than bureaucrats apponted by a central government. It s thus mportant to understand the effect of fscal decentralsaton on local poltcs and development, though our knowledge remans rather lmted prmarly due to lack of data avalablty. Our focus on local communty has partcular advantage over the exstng emprcal lterature that predomnantly analysed the effects of decentralzaton (n general) on publc polcy usng cross country data (see for example, Davood and Zou (1998), De Luca et al. (2002), De Mello and Barensten (2001) and Fshman and Gatt (2002)). Enkopolov and Zhuravskaya (2007) use cross-secton and panel data from about 75 developng and transton countres to suggest that decentralzaton works better wth strong natonal partes who then allow for local leaders to be chosen locally, rather than when the natonal partes are weak and to avod poltcal competton, these weaker partes appont local leaders as admnstratve subordnates. In partcular, they use share of sub-natonal revenues n total revenues as the man measure of fscal decentralzaton. Ths s however not deal as there are dfferent strata of sub-natonal regons, whch they themselves acknowledge. We focus on the lowest level of admnstratve unt whch stll has an ndependent polty, called a communty (that s, a vllage or a townshp) and our measure of fscal decentralzaton relates to shares of total
4 communty revenue from central and provncal governments, before and after the announcement of Law 22/99 that ntated ths drve towards fscal decentralzaton. There are only a handful of studes that use communty level sngle cross-secton data to assess the development effect of fscal decentralsaton (Besley et al., 2005). Our advantage s that we have access to two-perod communty-level panel data, whch enables us to dentfy the causal relatonshps better. In addton, unlke Besley et al. (2005), we consder the effect of fscal decentralsaton on local poltcs, an aspect that s not addressed n the past. The emprcal analyss suggests that fscal decentralzaton does mpact local polty and ths mpact s crucal n determnng whether fscal decentralzaton enhances or dmnshes local development. In partcular, we fnd that fscal decentralzaton dd typcally foster local democraces, though the ssue of elte capture where exstng local democraces are uprooted, was also wdespread. Moreover, t promoted development mostly n communtes whch transted out from elte capture to embrace free and far electons. Ths was followed by communtes that experenced the emergence of elte capture, whle communtes that contnued to reman under ether elte capture or free and far electons dd the worst. These fndngs suggest that whle the emergence of elte capture was wdespread, t may not necessarly be the most harmful. Instead, and surprsngly so for the case of local democracy, stablty of local polty hurts development the most. Our results have wder mplcatons for other emergng countres expermentng wth smlar decentralsaton programmes. 2. Data Our analyss s based on the communty level data obtaned from 1997 and 2007 Indonesan Famly Lfe Survey (IFLS) 3 from 314 rural and urban communtes, drawn from 13 provnces ncludng Jakarta, Bal, Java (central, east and south), Sumatra (north, west and south), Lampung, Wntenara and south Kalmantan. Ths s a partcularly rch data set that provdes communty level nformaton on a whole range of demographc characterstcs and access to publc goods, local governance and ts publc fnances, ctzens' partcpaton n plannng and mplementaton of local development projects, as well as a range of publc utltes, 3 Although IFLS data are avalable for the years 1993, 1998 and 2000 as well, nformaton on local polty could be found only n the 1997 and 2007 surveys.
5 nfrastructure and transport, health and educaton facltes (see Frankenberg and Thomas (2000) and Strauss et al. (2009) for study desgn and overvew of the data set). Although IFLS data are avalable for the years 1993, 1998 and 2000 as well, nformaton on local polty could be found only n the 1997 and 2007 rounds of the survey. 1997 and 2007 rounds of the IFLS asked communty leaders about how a leader was selected, whch we use to classfy these communtes. Answers to ths queston are coded as: () votng, () all resdents, 4 () local eltes, (v) local nsttutons and (v) others. Under votng (code ()), the standard noton of pluralty was employed where the voters were local communty resdents. Otherwse, local leadershp was determned by the choce of local eltes (code ()) or by exstng offcals of local government bureaucraces (code (v)). However t s not clear as to how all resdents (code ()) or others (code (v)) selected the local leader, and so we exclude these communtes from our analyss. Accordngly, we classfy local polty as follows: Democratc f a leader s selected by free and far electons wth voters beng all communty members, and Olgarchc f a leader s selected by communty eltes (codes () and (v)) who then reman uncontested. We assume that the ncdence of olgarches reflect the sze of elte capture. Access to both 1997 and 2007 rounds of IFLS data also allows us to classfy poltcal transton nto the followng types (V for votng n local democraces and O for olgarchy or elte capture): (a) VV [Stable Democracy], f the communty was democratc n 1997 and remaned so n 2007; (b) VO [New Olgarchy], f the communty was democratc n 1997 but turned olgarchc n 2007; (c) OV [New Democracy], f the communty was olgarchc n 1997 but turned democratc n 2007; and (d) OO [Stable Olgarchy], f the communty was olgarchc n 1997 and remaned so n 2007. For each communty, we dentfy three sources of government revenue, namely, revenue from central government, provncal government and dstrct government. The rest of the communty revenue s generated from local funds. 4 We observe a thrd category of local polty called consensus buldng where a communty leader s selected by all resdents va vllage meetngs. We exclude ths communty from our analyss due to lack of clarty of ths mechansm arsng out of varous ambguous rules and defntons.
6 Table 1 summarzes the mean shares of grants from dfferent sources as well as development spendng for the four dfferent communty types dentfed before and after the ntroducton of law 22/99. It hghlghts the aspect of dwndlng source of central grants for communty development n the post decentralzaton perod. The latter has been accompaned by a correspondng ncrease n grants from the dstrct government. In other words, fscal decentralsaton shfts the centre of power from the central to the dstrct government. In order to devse an approprate ndex of fscal decentralsaton, we carefully revewed the exstng lterature. In many works lke Davood and Zou (1998), De Mello and Barensten (2001), Zhuravskaya et al. (2007), the share of sub-natonal revenues n total government revenues s used as the man measure of fscal decentralzaton n ther cross-country study. These papers also argue that the results are robust to usng the share of sub-natonal expendtures n total government expendtures as an alternatve measure of fscal decentralzaton. They also acknowledged that an mportant shortcomng s that they could not dstngush between state and muncpal expendtures and revenues. Ln (2000) performed a provnce-level analyss of fscal decentralsaton and economc growth and used margnal retenton rate of the locally collected budgetary revenue. In contrast to these studes, we consder the lowest level of admnstratve unt who hardly have any tax-rasng power. So for these communtes, a relevant measure of fscal decentralsaton pertans to the share of central grant they receve. As mentoned earler, and most crucally for us, we recognze that wth fscal decentralzaton, the development funds avalable from the dstrcts rose sgnfcantly and across all communty types. Hence from our data t s certan that decentralzaton moved the fund source from the central government to the dstrct government so that a fallng share of central grant s equvalent to a rsng share of grant from the sub-natonal authortes. In addton, we keep the communty share of local funds whch assumes growng mportance after fscal decentralsaton as s evdent from Table 1. We also observe local government spendng of the communty on dfferent accounts. () Development spendng refers to spendng on new nfrastructure as well as mantenance of exstng nfrastructure; () non-development spendng ncludes spendng on staff salary, offce mantenance, offcal trps and contngences; () some under-developed communtes receve grants for varous developmental programmes, e.g., IDT for under-developed communtes, whch we control for n our analyss separately (see further dscusson n secton 4). Our analyss partcularly focuses on the share of development spendng n total
7 spendng for each communty. Table 1 suggests that changes n the average values of development spendng depend on the type of communtes. It follows that development spendng ncreased only for the new democraces and new olgarchs; the trend has been rather opposte for the stable democraces and olgarches alke, wth stable olgarches dong the worst. 3. How Fscal Decentralsaton affects Local Poltcs? Intal statstcal evdence reveals the followng. Fscal decentralzaton ndeed caused changes n communty level poltcs n Indonesa. Whle about a two-thrd of the communtes seems to have mantaned the same local poltcal envronment over the perod under analyss, from the remanng one-thrd, close to 70% transted out from beng olgarches to new democraces, and the remanng 30% of them (roughly 10% of the entre populaton) gave up democracy, pavng way for new olgarches (see Table 3). In ths secton we suggest some plausble lnks between decentralzaton, poltcal nfluence, local poltcs and communty development. In dong so, we classfy local leaders by two characterstcs: hghly educated and local resdent. The IFLS data allows us to clearly see ths classfcaton. It s then lkely, and partcularly so n a less developed country, that hghly educated leaders are typcally more lkely to be wealthy eltes and eltes n turn, are more lkely to have better connectons wth central and dstrct authortes. Also, selecton as a communty leader despte not beng a resdent n the communty s more lkely to hghlght ther possble lnk wth the central or the dstrct authortes. The followng assumpton then seems reasonable. Assumpton A: [Poltcal Contacts]: Pck a communty and consder ts leader n 1997. Ceters parbus, () Hgher educaton of the leader ncrease the chances of more poltcal nfluence wth both Dstrct and Central governments; () Local resdence of the leader and a larger geographc dstance of the communty from the Dstrct Headquarters reduces the chances that the leader has poltcal nfluence wth the Dstrct government; gven ths, () It s easer to establsh poltcal nfluence wth the Dstrct government than the Central government.
8 One may suspect that communtes whch had hgh central fundng n 1997 had leaders wth poltcal nfluence wth the central government as pror to decentralzaton t was the central government that decded all about fund allocaton across communtes. It could then be the case that such communtes were typcally governed by hghly educated leaders who used ther poltcal contacts wth the centre to sustan power and oppose local electons. We then have the followng hypothess regardng the year 1997. Hypothess I Ceters parbus, communtes recevng larger central grants n 1997 were more lkely to be olgarchc n that year. Moreover, n each such olgarchc communty, leaders were more lkely to be hghly educated. We now move to poltcal transton. Suppose Hypothess I s true. Pck such a communty that had very large central grants n 1997 and had a leader who was hghly educated. When decentralzaton took place n 2001, these educated leaders lost the mportance of ther poltcal connectons wth the central government but could mantan and possbly nvest more n enhancng ther contacts wth the Dstrct authortes (a possblty that s lkely gven part () of Assumpton A). Thus n such communtes there s an nerta to reman olgarchc. However, an opposte force cannot be ruled out. The possblty of enhancng dstrct level contacts by the exstng olgarchs could have been reduced f the leader was a local resdent of a communty that s far from the dstrct headquarters. 5 In the latter cases, current leaders were unable to mantan ther local poltcal power and ths could have caused a rse n local democraces. Hence we have our second hypothess. Hypothess II Consder a communty that was an olgarchy n 1997. Ceters parbus, decentralzaton has the followng effects on transton of local poltcs: (a) Hgher educaton of local leaders n 1997 decreases the ncdence of local democracy n 2007 and (b) Local resdence of leaders n 1997 ncreases the ncdence of local democraces n 2007. We then classfy local communtes n terms of avalablty of opportuntes to rase local funds, callng them accordngly, locally productve and locally unproductve. Such dfferences may have many economc or socal causes that affect the economc envronment 5 Notce that local resdence and dstance from the centre could nfluence the sze of central grants n 1997 as well. Ths s perhaps reflected n the dstrbuton of central grants n 1997. However, the mpact wth dstrct grants seems to us more senstve to resdence and dstance.
9 of the communty. The data set allows us to make ths dstncton qute easly by lookng at the sze of local funds n 1997 so that communtes wth larger local funds are presumably more productve locally. It s then reasonable to assert that decentralzaton promotes leaders who are more effcent n usng local funds n locally productve communtes. Ths s because economc development n productve communtes s certan to be one of the man electoral agendas, so that decentralzaton s more lkely to promote electoral competton and local democraces n these communtes. Based on these possbltes, we formulate our thrd hypothess. Hypothess III Ceters parbus: (a) Decentralzaton s lkely to ncrease the probablty of local democraces n communtes that are locally productve as reflected n large shares of local funds n 1997 and (b) n communtes whch do not have leaders wth poltcal connectons (meanng they are ether not hghly educated or are local resdents), decentralzaton cannot harm the process of local democracy. Although Hypothess III s plausble, one should be careful n two aspects. Frst, ncumbency of elected leadershp may be harmful. That s, even n locally productve communtes, f local democracy s tradtonal (n our case ths means t s already democratc n 1997), ncumbent leaders may enjoy certan poltcal advantages that then allow them to perform well n electons wthout havng to perform too well n the economc sphere. Hence, t could be that economc performance of new local democraces s better than n democraces whch are sustaned. As these negatve features of ncumbency are more lkely to be stronger when the ncumbent s an olgarch, one may assert that the negatve mpact of sustenance of olgarchy s the hghest. However, f there are communtes that were democratc n 1997 but become olgarchc n 2007, t s not clear f ths s because now there emerged local leaders wth good poltcal lnks wth the dstrct, or s the ant-democratc move drven by other unproductve factors. Hence, the relatve rankng of economc performance between new olgarches and stable democraces becomes a truly emprcal queston. We then formulate the followng hypothess to check f ths holds emprcally n our sample. Hypothess IV Developmental spendng n 2007 s hghest n communtes that are new democraces and lowest n those that are stable olgarches. The relatve performance of stable democracy versus new olgarchy n the sphere of developmental spendng remans ambguous.
10 In what follows we shall employ emprcal analyss to see f these theoretcal predctons are consstent wth the data. 4. Methodology We buld our emprcal model based on the theoretcal possbltes hghlghted n Secton 3. Snce we want to nvestgate whether the effect of decentralsaton on local development depends on the underlyng nature of poltcal transton (defnng the local polty), one needs to redress the potental selectvty bas (a la Heckman, 1979) when estmatng the development spendng equaton for each type of local polty that we dentfy, namely, VV, VO, OV, OO. When the selecton s over a large number of exclusve choces, the multnomal logt specfcaton s attractve due to ts smplcty. Let Z j denote that communty j (a subscrpt that we drop subsequently for notatonal smplcty) s of type. Then, Z 0 1 2 3 f OO f VV f VO f OV. We post that Z s determned as follows: Z ' W u, (1) where W s the set of lagged explanatory varables (from 1997) for the selecton and u s the random error term. Thus, the Multnomal Logt probablty of a gven type Z s: k e W Pr[ Z k W ] 3 ' t e W t 0 ', k 0,1,2,3. The estmated multnomal Logt equatons provde a set of probabltes for the four types of local polty n queston. Snce the dependent varable s dscrete n nature, we also calculate the margnal effects for each explanatory varables for the selecton equaton. Gven the selecton mechansm (1), the equaton of our prmary nterest s
11 Y ', (2) X where Y s the changng share of development spendng for the -th communty type over 1997-2007. As before, X s the set of lagged explanatory varables explanng Y (see further dscusson later n ths secton). The rule s that Y s observed only when Z 0. Thus, E( Y Z 0) ' X ( u ) v, (3) where u ' W u. The parameter s the Inverse Mll's Rato for the -th type communty, whch corrects for the potental selectvty bas. Accordngly, the least square estmates of (3) would produce consstent estmates of and. Note also that the estmated coeffcent accounts for the estmated margnal effect, ceters parbus, of the predcted probablty of Z. Our key varables n determnng (1) and (3) relate to the measures of fscal decentralzaton. Followng our dscusson n secton 2, we consder share of central grant and local funds as the key measures of fscal decentralsaton, whch marks a departure from the exstng lterature. Ths s clearly lnked to our analytcal argument that the sze of the central grant s a measure of the communty leader s lnk to the centre. Our second argument relates to the nfluence of the leader n the local communty (see secton 2). In the absence of any better alternatve (e.g., leader s ncome or wealth, whch are unobservable), we use leader s educaton (.e., whether the leader has senor hgh school educaton or hgher), whch s lkely to be correlated wth wealth and also communty resdency as a proxy for the leader s nfluence. An essental requrement for estmatng (1) and (3) s that there needs to be some dentfyng varables n X (as opposed tow ). In addton to the Inverse Mll's Rato, X ncludes ntal (1997) share of development spendng and an ndex of ethnc fractonalsaton n (3), whch we defne below. The ntal development spendng s ncluded to account for the possblty of Barro convergence (Barro, 1999) n spendng across communtes, f any. Also, ethnc fractonalsaton has been shown to be another mportant determnant of publc polcy (see for example, Barro (1999) and Alesna et al. (2000)). The ethnc fractonalsaton ndex
12 equals 1 p 2, where p s the proporton of the -th ethnc group n the populaton, usng the sze of three largest ethnc groups n each communty as provded by the IFLS. Smlarly, we also need an dentfyng varable that s present n the poltcal transton equaton (1), but not n the development equaton (2). To ths end, we nclude a varable measurng the dstance of the communty from the dstrct captal. Proxmty to the dstrct headquarters may nfluence the nature of the poltcal lnk and therefore the probablty of poltcal transton. The varables whch are ncluded n both (1) and (3) relate to communty populaton sze, relgon and rural/urban locaton. These demographc factors are known determnants of poltcal nsttutons and development (see for example, Barro (1999)). Communtes wth greater populaton sze may or may not be more lkely to be democratc, or that hgher degrees of urbansaton may or may not foster free and far electons. In the absence of any pror, we thus emprcally explore the role of these varables on local polty. Whle characterstcs of the leader (secondary and hgher levels of schoolng as well as resdency n the communty), namely, are known to nfluence publc polcy (see for example, Besley et al. (2005)), these characterstcs are also of relevance for the poltcal transton equaton. 4.1. Endogenety and robustness Some may argue that fscal decentralzaton was endogenous to expectatons over local poltcal transtons and local development. Statstcally speakng, ths s because varaton n these varables may not have been orthogonal to unobservable factors that jontly determne the outcomes studed here. The most wdely appled approach to dentfyng treatment effects employs nstrumental varable technques, whch rely on dentfyng an event or a varable that s orthogonal to the unobservable factors lkely to affect the outcomes of nterest. However, the randomness assumpton made wth respect to those nstruments s subject to much sceptcsm. Alternatvely, economsts have also attempted to make use of experments that objectvely randomze treatments to asses ther effects n the presence of unobserved heterogenety. These randomzed experments too are subject to crtcsms that they lack generalzablty and often do not adhere to the requrement of treatment randomness. Accordngly economsts and other socal scentsts have used natural experments random treatments that arse truly by chance. The recent poltcal hstory of Indonesa suggests that the fall of Soeharto and the mmedate poltcal decson of the new central leadershp of
13 fscal decentralzaton were nearly natural n our context of nvestgaton; thus choce of sample years facltates dentfcaton n our case. Ceters parbus, we determne the effect of share of central grant (our chosen measure of fscal decentralsaton) n 1997 (predecentralsaton perod) to predct the poltcal transton (rather than the poltcal outcome n a gven year) or the changng share of development spendng (rather than development spendng per se n a gven year) over 1997-2007. For potental smultanety to bas our estmates of poltcal transton or changng share of development spendng, communty leader needs to antcpate share of central grant ten years n advance, whch can be ruled out wthout much loss of generalty. A further ssue s that local communty leaders become drectly accountable to the dstrct government, especally after decentralsaton. Whle we do not observe the characterstcs of the dstrcts, the latter may have nfluenced the nature of local polty. Consequently, we control for communty fxed effects to mnmze any potental estmaton bas arsng out of unobserved communty-level heterogenety. There may also arse some problems concernng reverse causalty wth respect to some of the explanatory varables used n our analyss. For example, urban areas are lkely to be more prosperous and hence may have hgher share of development spendng. Also they may be more populated wth hgher dversty and hence democracy could be a more convenent polty. Note however that both X and W nclude one-perod (.e., a decade) lagged values of these varables; for potental smultanety to bas our estmates, the leader needs to antcpate the values of these varables ten years n advance, whch can be ruled out. Fnally, one must be aware that over the past two decades, there has been a global force towards upholdng and nurturng democratc nsttutons. Perhaps our estmates are then based n terms of such a worldwde trend. Nevertheless, as our results show the reverse trend n many local communtes, such global trends actually strengthen such estmates aganst democracy. 5. Results and Predctons Our dscusson n ths secton s dvded nto two parts. Frst we report our results on how fscal decentralzaton affected local poltcs. We then show the mpact of ths changng local
14 polty on the changng share of development spendng n our sample. All Tables and Fgures are provded at the end of the text. 5.1 Impact on Local Poltcs From Table 1 (and the observed proportons of each communty type from Table 3) we can see that the share of central grant for communtes whch were democratc n 1997 was about 24%, whle that for communtes whch were olgarchc was about 40%. Hence, t s clear that the frst part of Hypothess I holds, that s, communtes recevng larger central grants n 1997 were less lkely to be democratc n that year. So now we restrct attenton to communtes whch were olgarches n 1997. Let P headshs O; Q ] denote the probablty [ k that the leader of a communty n 1997 was hghly educated, gven that n that year the communty was an olgarchy and that t had a share of central grant greater than the k-th quartle Q k. Our data shows that P headshs O; Q ] 0.82 P[ headshs O; Q ] 0.87. Thus [ 1 3 as we move from the lowest to the hghest quartle, the probablty of head beng hghly educated ncreases monotoncally n olgarchc communtes. Ths lends support to the second part of Hypothess I. Therefore, the consequent theory that s bult n Secton 3 remans plausble. We then move to the next hypotheses. Table 2 summarzes the multnomal Logt estmates of the baselne regresson model (1) whle Table 3 compares observed probablty wth predcted probablty of poltcal transton derved from the estmates shown n Table 2. Appendx Table A1 shows the correspondng margnal effects of the key varables of our nterest. Table 2 suggests that the probablty of any type of poltcal transton depends on the share of local funds avalable n 1997. In partcular, hgher shares of local funds n 1997 enhance the probablty of democracy, both for new (vz. column 3) and exstng ones (vz. column 1), thus provdng some support to Hypothess III. Ceters parbus, hgher share of local funds s however nsgnfcant n determnng the lkelhood of transton from democracy to olgarchy (column 2). In contrast, the mpact of central funds rased n 1997 on ths poltcal transton s dfferent from that of the share of local funds. In partcular, the probablty of a communty transtng
15 from olgarchy to a new democracy s lower where the share of the central funds avalable n 1997 was hgher. Second, communtes wth less central funds n 1997 are more lkely to retan democracy. In contrast, the share of the central funds avalable n 1997 remans nsgnfcant when a communty transts to new olgarchy. Thrd, characterstcs of the communty leader also play an mportant role n the poltcal transton n our sample. Note that the communtes where the leader s more educated tend to have hgher probablty of transtng to a new olgarchy whle t remans nsgnfcant for new or exstng democraces. Also, a leader s resdency n the localty boosts the lkelhood of democraces (new and exstng ones) whle t remans nsgnfcant n determnng the lkelhood for new olgarches. Fnally, communtes further from the dstrct headquarters (DHQ) are more lkely to foster democraces (new and exstng ones). Hence, there s ample evdence n our data that support to Hypothess II. Some other control varables also nfluence the process of poltcal transton. In partcular, communtes where Islam s the man relgon tend to have a hgher probablty of sustanng democracy. In contrast, underdeveloped communtes are less lkely to transt out of democracy. We now move to Table 3. Column 1 shows the actual number of communtes n each class, whle column 2 shows ther relatve frequences. Fnally, column 3 summarzes the means and standard devatons of the predcted probabltes of each class derved from Table 2. About 71% of the sample communtes tend to be poltcally stable. Among them, 62% contnue to be democratc even after decentralzaton, whle the remanng 38% reman olgarchc. Among the transent communtes, 66% are new democraces whle the rest are new olgarchs. Interestngly, about 10% of the total sample communtes turn out to be new olgarchs. The predcted probabltes closely reflect these relatve frequences observed n the sample, confrmng the effcency of the multnomal logt estmates shown n Table 2. Note however that although the standard devatons of these predcted probabltes are compatble wth the correspondng means, there s some ntra group varaton that we turn to next. Whle Table 3 reflects the average predcted probablty of poltcal transton, Fgures 1 to 4 depct the dstrbuton of these probabltes n the sample. Fgure 1 shows that communtes sustanng democracy were truly and dstnctly b-modal; almost 50% of them were hghly lkely to reman democratc whle the rest were hghly lkely to become
16 olgarchc. In contrast, the probablty dstrbutons of new democraces (Fgure 2) and new olgarches (Fgure 3) tend to hghlght a skewed (to the left) pattern; n each case, more than 50% had less than 10% chance of swtchng ther polty. For sustanng olgarchs (Fgure 4), about 60% had only about 10% chance of mantanng the polty whle about 10% had more than 80% chance of dong so. 5.2. Impact on Local Development Spendng Table 4 summarzes the selectvty corrected estmates of changng share of development spendng over the decade (1997-2007) for each type of local poltcal transton, namely, VV, VO, OV and OO. 6 Snce the nverse Mll's Rato s a functon of the respectve predcted probablty of transton, sgn and sgnfcance of nverse Mll's Ratos allow us to explore the varaton n changng share of development spendng, ceters parbus, wth predcted probablty of poltcal transtons. In partcular, these estmated coeffcents are negatve and sgnfcant for VV, VO and OO, whle t s postve and sgnfcant for OV. In other words, whenever the probablty of a new democracy rses n a gven localty, decentralzaton leads to an ncrease n development spendng, whch lends support to Hypothess IV; otherwse, decentralzaton lowers local development spendng. Some other fndngs are nterestng too. In all cases, a hgher share of ntal development spendng s assocated wth lower development spendng over the decade, thus confrmng the condtonal convergence hypothess of Barro (1999). Characterstcs of the communty leader tend to nfluence ths change n development spendng. In partcular, the educaton of the communty leader matters (and s benefcal) only n stable democraces (.e. VV). Whether the communty leader s a resdent member seems mportant (postvely) only n stable olgarches, suggestng the case of preference matchng. Degree of ethnc fractonalzaton only matters for stable democraces n that local communtes wth a hgher fractonalzaton tend to experence a lower spendng n our 6 Note that we also tred to nclude the share of central grant and local funds n the determnaton of share of development spendng. Snce these varables were never sgnfcant n any specfcaton, we drop them n the parsmonous specfcaton shown here.
17 sample. Ths s compatble wth results found by Alesna et al. (2000) and Devarajan et al. (1996) for example. In sum, our analyss suggests that poltcal connectons of local leaders wth central and dstrct authortes, some decsve features of these leaders and geographcal locatons of the communtes n queston do matter n determnng the mpact of fscal decentralzaton on the nature of local poltcs and development. 6. Concluson The paper attempts to explore whether fscal decentralzaton can foster local democracy n an economy whch s already poltcally decentralzed, and to see f the economc consequences of such decentralzaton are sgnfcantly determned by ts poltcal mpact. Indonesa s an mportant case n pont whch has a long tradton of poltcal decentralzaton nferred by the Dutch colonal rule, whle the country has ntated ts fscal decentralzaton drve early n the Mllennum. Usng recent 1997 and 2007 rounds of the Indonesan Famly Lfe Survey (IFLS) communty-level data, the paper fnds that fscal decentralzaton dd not necessarly promote local democracy. The most nterestng fndng s that fscal decentralzaton fostered local development only f t changed the local polty, as a stable local polty, even f democratc, reduced development after the decentralzaton. Consequently, t follows that the emergence of elte capture was not necessarly the worst scenaro for local development spendng. Instead, a long lastng rule of the elte hurt development the most n Indonesa.
18 Tables and Fgures Communty classfcaton by type of poltcal transton VV VO OV OO Share of 1997 2007 1997 2007 1997 2007 1997 2007 Central grant 0.23 0.08 0.25 0.07 0.43 0.09 0.34 0.01 Provncal grant 0.04 0.07 0.13 0.09 0.07 0.15 0.41 0.29 Dstrct grant 0.04 0.34 0.35 0.40 0.05 0.41 0.15 0.34 Local funds 0.65 0.50 0.19 0.44 0.41 0.35 0.01 0.36 Development spendng 0.67 0.54 0.56 0.57 0.52 0.54 0.53 0.40 Table 1: Descrptve statstcs of central fundng and local spendng across dfferent types of communtes n 1997 and 2007. Note on communty classfcaton VV: Votng n 1997 and 2007 (stable democraces) VO: Votng n 1997, but olgarchy n 2007 (new olgarches) OV: Olgarchy n 1997, votng n 2007 (new democraces) OO: Olgarchy n 1997 and 2007 (stable olgarches)
19 (1) (2) (3) VARIABLES votng-votng votng-olgarchy olgarchy-votng Central grant 97-2.888** -9.148-2.566* (1.462) (5.160) (1.510) Local funds 97 6.272* 0.595 7.560* (3.689) (4.235) (3.992) Leader resdent 0.0488** -0.0304 0.0191 (0.0197) (0.0197) (0.0149) Leader hgh sch 97-1.239 21.68*** -1.361 (1.323) (2.525) (1.088) Populaton 97 0.0851 1.852 0.160 (0.750) (1.222) (0.684) Hhs wth sch. 97-3.752 17.99*** -3.435 (5.201) (6.794) (4.126) Ethnc hety 97-1.205-8.233 1.230 (3.491) (5.348) (3.002) Rural 97 1.426 3.936-0.0497 (1.642) (2.459) (1.332) Islam 97 3.150** 7.07 2.268** (1.605) (5.471) (1.146) undev97 1.068-45.05*** 1.177 (1.131) (5.580) (0.839) Dstance from DHQ 0.244** 0.712 0.360* (0.109) (0.634) (0.202) Intercept Yes Yes Yes Dstrct FE Yes Yes Yes Log-pseudeoL -78.6040 Pseudo R 2 0.6358 Observatons 180 180 180 Robust standard errors n parentheses; *** p<0.01, ** p<0.05, * p<0.1 Olgarchy-Olgarchy s the base category Table 2: Multnomal Logt Estmates of the Probablty of Poltcal Transton 1997-2007
20 Communty Classfcaton by type of poltcal transton (1997-2007) (1) Number of communtes (2) Observed probablty (3) Predcted probablty Mean (sd) Votng-Votng 89 0.44 0.45 (0.38) Olgarchy-Votng 38 0.19 0.19 (0.17) Votng-Olgarchy 20 0.10 0.10 (0.11) Olgarchy-Olgarchy 55 0.27 0.26 (0.34) Table 3: Observed and Predcted Probablty of Poltcal Transton. Note on communty classfcaton VV: Votng n 1997 and 2007 (stable democraces) VO: Votng n 1997, but olgarchy n 2007 (new olgarches) OV: Olgarchy n 1997, votng n 2007 (new democraces) OO: Olgarchy n 1997 and 2007 (stable olgarches)
21 (1) (2) (3) (4) VARIABLES VV OV VO OO Intal share of dev -1.106*** -1.008** -1.676-1.135*** spendng_97 (0.163) (0.340) (3.068) (0.302) Populaton sze 97-2.18e-06 1.16e-05* 1.68e-05-1.45e-05 (5.07e-06) (6.91e-06) (3.20e-05) (9.43e-06) Ethnc heterogenety 97-0.144* -0.0116 1.078 0.0696 (0.0754) (0.147) (2.538) (0.537) Rural 97 0.193** -0.0871 (dropped) -1.031 (0.0937) (0.153) (0.932) Islam 97 0.435*** 0.416* 0.304-0.435 Head s educaton>= senor hgh school 97 Head s a resdent n the communty 97 (0.108) (0.199) (0.475) (0.351) 0.158** -0.130 (dropped) 0.0168 (0.0648) (0.178) (0.196) -0.0732-0.213-0.388 0.169* (0.0792) (0.169) (0.290) (0.101) Mll s rato - 0.0616* 0.0263** - 0.234* - 0.124** (0.0302) (0.012) (0.123) (0.05) Intercept YES YES YES YES Dstrct FE YES YES YES YES R-squared 0.449 0.630 0.793 0.619 Observatons 89 38 20 55 Robust standard errors n parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 4: Selectvty-corrected estmates of changng share of communty-level development spendng. VV: Votng n 1997 and 2007 (stable democraces) VO: Votng n 1997, but olgarchy n 2007 (new olgarches) OV: Olgarchy n 1997, votng n 2007 (new democraces) OO: Olgarchy n 1997 and 2007 (stable olgarches)
22 Predcted probablty of V-V Percent 0 10 20 30 40 0.2.4.6.8 1 Pr(change07==1) Fgure 1: Dstrbuton of the Predcted Probablty of Stable Democraces Predcted probablty of O-V Percent 0 10 20 30 40 50 0.2.4.6.8 1 Pr(change07==3) Fgure 2: Dstrbuton of the Predcted Probablty of New Democraces.
23 Predcted probablty of V-O Percent 0 20 40 60 80 0.2.4.6.8 1 Pr(change07==2) Fgure 3: Dstrbuton of the Predcted Probablty of New Olgarches. Predcted probablty of O-O Percent 0 20 40 60 0.2.4.6.8 1 Pr(change07==4) Fgure 4: Dstrbuton of the Predcted Probablty of Stable Olgarches.
24 References [1] Alesna, A. R., R. Baqr and W. Easterly (2000). `Redstrbutve publc employment,' Journal of Urban Economcs, 48, pp. 219-241. [2] Bardhan, P. (2002). `Decentralsaton of governance and development,' Journal of Economc Perspectves, pp. 185-205. [3] Bardhan, P. and D. Mookherjee (2000). `Capture and governance at local and natonal levels,' Amercan Economc Revew, 90, pp. 135-139. [4] Barro, R. J. (1999). `Determnaton of democracy,' Journal of Poltcal Economy, 107(6), pp. S158 - S183. [5] Beard, V. (2007). 'Household contrbuton to communty development n Indonesa', World Development, 35(4), pp. 607-25. [6] Besley and Coate (2003). `Centralzed versus decentralzed provson of local publc goods: a poltcal economy approach,' Journal of Publc Economcs, 87, pp. 2611-2637. [7] Besley, T., R. Pande, L. Rahman and V. Rao. (2005). 'Partcpatory democracy n acton: Survey Evdence from Inda', Journal of the European Economc Assocaton, Papers and Proceedngs, 3, pp. 648-57. [8] Ca, H. and D. Tresman (2004). `State corrodng federalsm,' Journal of Publc Economcs, 88, pp. 819-843. [9] Davood, H. and H. Zou (1998). `Fscal decentralzaton and economc growth: a cross country study,' Journal of Urban Economcs, 43, pp. 244-257. [10] De Luca, Mguel, Mark P. Jones and Mara Ines Tula. (2002). `Black rooms or ballot boxes: canddate nomnaton n Argentna,' Comparatve Poltcal Studes, 35, pp. 413-436. [11] De Mello, L. and M. Barensten (2001). `Fscal decentralzaton and governance: a cross-country analyss,' IMF Workng Paper No. WP/01/71. [12] Devarajan, S. V. Swaroop and H. Zou. (1996) `The Composton of Publc Expendture and Economc Growth', Journal of Monetary Economcs 37: 313-344.
25 [13] Enkopolov and Zhuravskaya (2007). `Decentralzaton and poltcal nsttutons,' Journal of Publc Economcs, 91, pp. 2261-2290. [14] Fsman, R. and R. Gatt (2002). `Decentralzaton and corrupton: evdence across countres,' Journal of Publc Economcs, 83, pp. 325-346. [15] Frankenberg, E. and D. Thomas. (2000). ``Study Desgn and Results from Waves 1 and 2,'' Santa Monca, CA, Rand, DRU - 2238, Volumes 1-7, NIA/NICH. [16] Hayek, F. A. (1948). `Indvdualsm and economc order,' Unversty of Chcago Press, Chcago. [17] Heckman, J. (1979) `Sample Selecton Bas as a Specfcaton Error', Econometrca 47, 153-61. [18] Inman, R. and D. Rubnfeld (1997a). `Rethnkng federalsm,' Journal of Economc Perspectves, 11 (4), pp. 43-64. [19] Inman, R. and D. Rubnfeld (1997b). `The poltcal economy of federalsm,' n: Mueller, D. C. (ed.), Perspectves on Publc Choce: A Handbook. Cambrdge Unversty Press, Cambrdge. [20] Lockwood, B. (2002). `Dstrbutve poltcs and the benefts of decentralzaton,' Revew of Economc Studes, 69 (2), pp. 313-338. [21] Maskn, E., Y. Qan and C. Xu. (2000). `Incentves, nformaton and organzatonal form.' Revew of Economc Studes, 67 (2), pp. 359-378. [22] Mansur, G. and Rao. (2004). 'Communty Based and Drven Development: A Crtcal Revew', World Bank Research Observer 19(1) 1-39. [23] Montola, G., Y. Qan and B. R. Wengast. (1995). `Federalsm, Chnese style: the poltcal bass for economc success n Chna,' World Poltcs, 48 (1), pp. 50-81. [24] Oates, W. (1972). `Fscal Federalsm,' Harcourt, New York. [25] Olken, B. (2007). `Poltcal Insttutons and Local Publc Goods,' mmeo, Harvard Unversty, February.
26 [26] Prud'homme, R. (1995). `The dangers of decentralzaton,' World Bank Research Observer, 10 (2), pp. 21-26. [27] Qan, Y. and G. Roland (1998). `Federalsm and the soft budget constrant,' Amercan Economc Revew, 88 (5), pp. 1143-1162. [28] Qan, Y. and B. Wengast (1996). `Chna's transton to markets: market-preservng federalsm, Chnese style,' Journal of Polcy Reform, 1, pp. 149-185. [29] Strauss, J., F. Wtoelar, B. Skok and A. M. Watte (2009). ``The 4th Wave of the Indonesan Famly Lfe Survey (IFLS4): Overvew and Feld Report,'' WR - 675/1 - NIA/NICHD. [30] Tebout, C. (1956). `A pure theory of local expendtures,' Journal of Poltcal Economy, 64 (5), pp. 416-424. [31] Tanz, V. (1996). `Fscal federalsm and decentralzaton: a revew of some effcency and macroeconomc aspects,' Annual World Bank Conference on Development Economcs, 1995. World Bank, pp. 295-316.
27 Appendx Table A1. Multnomal logt margnal effects of poltcal transton, selected varables 1 2 3 Varables votng-votng votng-olgarchy olgarchy-votng Central grant 97-0.88064** -0.00001-0.896678** 0.36726 0.00006 0.36365 Local funds 97 0.0752* -0.00005 0.091777** 0.03774 0.00004 0.037087 Leader Resdent 97 0.22478* -0.00019* -0.09239 0.13306 0.00011 0.21687 Leader schoolng 97 0.015549 0.0006** -0.02706 0.26417 0.00003 0.26377