Document de treball de l IEB 2012/31

Transcription

1 Document de treball de l IEB 2012/31 PARTISAN TARGETING OF INTER-GOVERNMENTAL TRANSFERS & STATE INTERFERENCE IN LOCAL ELECTIONS: EVIDENCE FROM SPAIN Marta Curto-Grau, Albert Solé-Ollé, Plar Sorrbas-Navarro Fscal Federalsm

2 Documents de Treball de l IEB 2012/31 PARTISAN TARGETING OF INTER-GOVERNMENTAL TRANSFERS & STATE INTERFERENCE IN LOCAL ELECTIONS: EVIDENCE FROM SPAIN Marta Curto-Grau, Albert Solé-Ollé, Plar Sorrbas-Navarro The IEB research program n Fscal Federalsm ams at promotng research n the publc fnance ssues that arse n decentralzed countres. Specal emphass s put on appled research and on work that tres to shed lght on polcy-desgn ssues. Research that s partcularly polcy-relevant from a Spansh perspectve s gven specal consderaton. Dssemnatng research fndngs to a broader audence s also an am of the program. The program enjoys the support from the IEB-Foundaton and the IEB- UB Char n Fscal Federalsm funded by Fundacón ICO, Insttuto de Estudos Fscales and Insttut d Estuds Autonòmcs. The Barcelona Insttute of Economcs (IEB) s a research centre at the Unversty of Barcelona whch specalzes n the feld of appled economcs. Through the IEB- Foundaton, several prvate nsttutons (Applus, Aberts, Ajuntament de Barcelona, Dputacó de Barcelona, Gas Natural and La Caxa) support several research programs. Postal Address: Insttut d Economa de Barcelona Facultat d Economa Empresa Unverstat de Barcelona C/ Tnent Coronel Valenzuela, 1-11 (08034) Barcelona, Span Tel.: Fax: eb@ub.edu The IEB workng papers represent ongong research that s crculated to encourage dscusson and has not undergone a peer revew process. Any opnons expressed here are those of the author(s) and not those of IEB.

3 Documents de Treball de l IEB 2012/31 PARTISAN TARGETING OF INTER-GOVERNMENTAL TRANSFERS & STATE INTERFERENCE IN LOCAL ELECTIONS: EVIDENCE FROM SPAIN * Marta Curto-Grau, Albert Solé-Ollé, Plar Sorrbas-Navarro ABSTRACT: We examne whether state-level ncumbents dscrmnate n the allocaton of transfers n favour of local governments controlled by co-partsans, and whether the electoral prospects of local ncumbents mprove when they are algned wth the state ncumbent. Usng a new database coverng around 3,000 Spansh muncpaltes durng the perod and a Regresson Dscontnuty desgn, we document a very strong and robust effect: n close races, muncpaltes algned wth the regonal government obtan on average 83% more per capta transfers and ther ncumbents gan 10% more votes at the local electons. We also show that the effect of algnment s stronger: () when regonal and local electons are held on the same day, () n regons wth less compettve regonal electons, and () n regons wth more budget resources. JEL Codes: C2, D72 Keywords: Poltcal partes, nter-governmental transfers, pork barrel poltcs Marta Curto-Grau Unverstat de Barcelona & IEB Avda. Dagonal Barcelona (Span) E-mal: mcurto@ub.edu Albert Solé-Ollé Unverstat de Barcelona & IEB Avda. Dagonal Barcelona (Span) E-mal: asole@ub.edu Plar Sorrbas-Navarro Unverstat de Barcelona & IEB Avda. Dagonal Barcelona (Span) E-mal: psorrbas@ub.edu * Ths research has receved fundng from projects ECO /ECON (Mnstero de Educacón y Cenca) and 2009SGR102 (Generaltat de Catalunya).

4 1. Introducton In recent decades, a number of countres have decentralzed ther provson of publc servces (see, e.g., Shah and Thompson, 2004). Such measures are recommended by scholars and nternatonal organzatons alke as part of reform packages that can mprove the effcency and effectveness of publc servce delvery (e.g, Broso and Ahmad, 2009). A better matchng of preferences (e.g., Oates, 1972) and ncreased accountablty (e.g., Seabrght, 1996) are the arguments often used to support ths polcy. Decentralzaton s typcally recommended f these benefts can compensate for any neffcency generated by spllovers and/or the lmtatons of economes of scale. However, whether decentralzaton can actually delver these benefts s more controversal, wth falure often beng attrbuted to measures that are only partal n nature, a term coned to refer to stuatons where the devoluton of fscal power s lmted (e.g., Brueckner, 2009, and Devajaran et al., 2009). For nstance, some authors clam that debt-related moral hazard problems can arse as a result of an excessve relance on transfers (e.g., Rodden, 2002, and Wengast, 2009). Smlarly, corrupton s also sad to be more prevalent wth transfer-dependent sub-natonal governments, because of the dmnshed nterest of voters n holdng poltcans accountable (e.g., Wengast, 2009, and Brollo et al., 2012). Moreover, accordng to Kheman (2010a and 2010b), partal decentralzaton mght reduce ctzen s awareness of sub-natonal responsbltes thus fosterng clentelsm and rent-seekng. Varous authors also pont to the problems created by hgher layer partsan ncumbents that dscrmnate between algned and unalgned local governments when allocatng transfers, to the pont that they are even able to nfluence the results of sub-natonal electons (Daz-Cayeros et al., 2006, and Schener, 2005). Ths nterference n the workngs of local electons can ultmately undermne one of the very benefts of decentralzaton, namely the mprovement n poltcans accountablty 1. It s ths specfc ssue that we focus our attenton on n ths paper. We examne whether the control of a hgher layer of government by one party s benefcal for ts co-partsans holdng power at a lower layer. Specfcally, our man goal s to determne whether Spansh regonal governments (the so-called Autonomous Communtes) allocate more transfers to algned local governments.e., to muncpaltes n whch the mayor s afflated to the same party as that of the regonal presdent. We focus on 1 Some authors go further and suggest that the overall level of poltcal competton n the country can be reduced f holdng mayoraltes helps the hgher layer ncumbent to become entrenched (see Schener, 2005). Other authors clam that the mere structure of local government mght be endogenous to these practces, snce ncumbents mght be reluctant to push for full decentralzaton f ths fosters competton (see Kheman 2010b). 2

5 earmarked captal transfers, whch are deemed to be the most dscretonary of the transfers made n Span. Addtonally, we analyze whether partsan algnment has an effect on the votes obtaned by a mayor at the local electons, and whether ths effect s related to the larger amount of transfers allocated to algned mayors. Fnally, to shed some lght on the mechansms explanng these results, we study how these effects dffers across muncpaltes dependng on whether regonal and local electons are held on the same day or not, the compettveness of the regonal electon, or the amount of regonal budgetary resources. Our analyss s motvated by plenty of anecdotal evdence that suggests that the partes controllng hgher layers of government allocate more resources to local governments run by co-partsans, and that nter-governmental transfers are an mportant means of achevng ths goal. At least n Span, our case of study, voters and poltcans alke seem to beleve ths to be the case. A recent post n a Spansh blog s llustratve of ths: The other problem [wth transfers] s the old-boy network and the partsanshp of grantors. Nobody dares to meddle wth ths ssue, for fear of beng added to the black lst, and so rsk recevng less than s usually receved, but the realty s that havng a frend n the rght place and beng a member of the party wegh much more heavly than they should n the awardng of transfers. ( Other nformal evdence suggests that beng algned wth a party controllng the hgher layer mght help a canddate to wn more votes at the local electons, and that ths mght also be due to the hgher amount of resources channelled to that muncpalty. Here s an example of how partes campagned for votes at the last local electons held n Span n 2011: People should understand (when decdng ther vote) that t s the PP (Partdo Popular, the man rght-wng party) who wll be n control of the resources of the government of the Autonomous Communty. ( But can these examples be generalzed or are they just a Spansh anomaly, anecdotes that emerge n the mddle of a keenly contested electoral campagn? We argue that they are not merely anecdotal, and to demonstrate ths we undertake a more systematc analyss, drawng on a new database of regonal transfers to local governments and of votng patterns at local electons for around 3,000 Spansh muncpaltes for the perod 2000 to Lkewse, we do not beleve ths ssue to be lmted to Span, and so our results should be nformatve for other countres. For nstance, Schener (2005) descrbes cases of both developng (e.g. Inda, Brazl and Mexco) and developed countres (e.g. Japan, Austra and Italy) n whch transfers 3

6 to local governments are poltcally manpulated n favour of co-partsans. However, only a few papers provde quanttatve, emprcal evdence of ths effect. Usng US data, Grossman (1994) fnds that states algned wth the federal government do, n fact, receve more funds. Arulampalam et al. (2009) fnd that the effect of algnment n Inda s to ncrease transfers from central to state governments by up to 16%. Daz-Cayeros et al. (2006), focusng on the Mexcan case, fnd that under the PRI, the state governments controlled by ths party receved up to 40% more transfers than those controlled by the opposton. Solé-Ollé and Sorrbas-Navarro (2008), and Brollo and Nanncn (2012), the only papers to examne grants to local governments, fnd an algnment effect of between 30 and 40% for the cases of Brazl and Span, respectvely. A number of papers also examne the mpact of algnment between layers of government on electoral outcomes 2. There s evdence, for example, of the effects of the US presdental vote on state legslatve electons (see, e.g. Campbell, 1986). Smlar nteractons are found for Argentna by Gélneau and Remmer (2006). In a comparatve study of Argentna, Canada, Germany and the US, Rodden and Wbbels (2011) show that the nteracton between federal and state or provncal electons becomes more apparent the more centralzed the partes are. Bottom-up effects, from gubernatoral to natonal electons, are found by Samuels (2000) for Brazl. Our paper contrbutes to these two lnes of lterature n several ways. Frst, our focus on regonal-local nteractons provdes greater plausblty to the man lne of reasonng used n explanng the algnment effect,.e. the dffcultes n assgnng poltcal credt to the dfferent government layers. Note, for nstance, that the spendng responsbltes of these two layers of government tend to overlap to a greater extent than those of federal and state governments. Indeed, qute often the provson of basc nfrastructure (the specfc target of the transfers we study) s a jont task, shared by state and local governments. Second, by focusng on local electons we are able to present evdence not only of exstng dscrmnaton n transfer allocaton but also of the nfluence of hgher layer ncumbents on the results of electons at lower layers. It s worth notng that no prevous attempts have been made n the lterature to analyze ncumbency spllover effects between regonal and local electons. Thrd, the use of data from several regons allows us to explot nsttutonal and poltcal dfferences across these areas, whch mght shed some lght on the partcular 2 Most of the papers dealng wth ncumbency spllover effects examne nteractons between dfferent electons at the same level of government. There s evdence of US Presdental and Gubernatoral effects on the electons for the federal and state legslatures, respectvely (e.g. Campbell and Summers, 1990; Folke and Snyder, 2012). Smlar effects are found n Europe by Hanmueller and Kern (2008) and Ade and Freer (2011). 4

7 mechansm at work. In ths sense, we are able to examne whether the effect of algnment on transfers and votes depends on the avalablty of budget resources n the regon, the tmng of regonal and local electons, and the compettveness of regonal electons. Fourth, we are aware that algnment status mght well be correlated wth party popularty and so we use a Regresson Dscontnuty Desgn (RDD) n our analyses, thus focusng on canddates that barely won or lost a majorty of seats at the local electons. Several recent papers n the ncumbency advantage lterature use RDD as ther man dentfcaton strategy (e.g., Lee, 2008; Lee et al., 2004; Hanmueller and Kern, 2008; Brockman, 2009; Folke, 2010; Folke and Snyder, 2012; Trounstne, 2011, and Ade and Freer, 2011). More closely n lne wth our concerns, Brollo and Nanncn (2012) use ths procedure to study the effect of algnment on transfers n Brazl. However, the use of the tradtonal close electons RDD, where the threshold s located at 50% of the vote, s problematc n our case, for two reasons. Frstly, local councls n Span are elected usng a proportonal electoral rule, the d Hondt rule, whch generates many possble thresholds at whch an addtonal vote can result n a party ganng one more seat, and none of these thresholds s necessarly located at 50% of the vote. To deal wth ths problem, we use as our forcng varable the share of votes that the regonal ncumbent s bloc has to lose (wn) n the local electons n order to lose (gan) the majorty of seats on the local councl. Secondly, n a large proporton of muncpaltes, no party has more than 50% of the seats, whch means that n many cases the mayor s elected on the formaton of a coalton of partes. In ths paper, we document that usually these coaltons are formed along deologcal lnes. Ths means that the dscontnuty n the treatment probablty s lower than one, and, as such, requres the use of a fuzzy RDD (Van der Klauw, 2002; Lee and Lemeux, 2010). Ths method conssts bascally n nstrumentng the algnment status wth a dummy equal to one f, at the local electons, the deologcal bloc of the ncumbent grantor obtans more seats than those won by the opposton bloc. Ths also consttutes a contrbuton of ths paper to the RDD lterature. Earler papers have developed an RDD for proportonal electons (see Folke, 2010, and Ade and Freer, 2011), and we use these as a benchmark for our study. Usng the aforementoned fuzzy RDD, we fnd a hghly marked effect of partsan algnment between regonal and local governments on the allocaton of regonal transfers to local governments. Local governments controlled by the same party as the regonal government receve 83% more funds for earmarked captal transfers than s the case of smlar unalgned muncpaltes. Ths effect s more than twce that estmated by OLS or dfference-n-dfferences. Moreover, mayors belongng to the same party as the regonal 5

8 presdent receve around 10% more votes at the local electons. These effects are more marked when regonal and local electons are held on the same day. We also fnd that these effects are stronger n regons wth less compettve regonal electons, and wth more budget resources. Ths last fndng suggests that the effect of algnment on votes works, at least partly, through the allocaton of transfers. The paper s organzed as follows. The next secton revews the theoretcal arguments that explan why algnment between ncumbents at dfferent layers of government mght have an effect on the allocaton of transfers. Secton three provdes the background nformaton on Span (.e. local governments, transfers, and local poltcs) needed to set the stage for the subsequent analyss. Secton four descrbes the econometrcs and the data. Secton fve presents the results. The last secton concludes. 2. Theoretcal dscusson In ths secton, we revew the man theores that predct an algnment effect (.e., that muncpaltes controlled by the same party as that to whch the regonal presdent belongs wll receve larger transfers from ths layer of government). We brefly summarze the man theores of targeted publc spendng, then dscuss how predctons may vary n the case of nter-governmental transfers (as opposed to the ncumbent s drect spendng), and consder whether the outcomes n local vs. regonal electons matter to the hgher layer ncumbent, and whether the tmng of the two electons s also mportant. Swng voters, core voters, and pvotal dstrcts. Extant models of dstrbutve poltcs offer several explanatons as to the ways n whch publc spendng polces mght target dfferent groups of voters. Frst, hgher layer ncumbents mght seek to enhance ther probabltes of beng re-elected by allocatng more resources to consttuences wth many swng voters (Lndbeck and Webull, 1987; Dxt and Londregan, 1996), on the understandng that ther low party allegance mght make t easer to buy ther votes. Secondly, poltcans may choose to allocate transfers to places n whch ther partes core voters concentrate. There are several ratonales that mght account for ths behavor. Rskaverse ncumbents, for example, mght prefer the lower degree of vote varablty among core voters to the only potentally hgher average vote return n swng dstrcts (Cox and McCubbns, 1986). Addtonally, the vote returns of a core-voter strategy mght be hgher f ncumbents have a better understandng of the specfc needs of ther core supporters (Cox, 2009) or f transfers to these places are effectve n boostng turnout (Ansolabehere and 6

9 Snyder, 2006) 3. Thrdly, when there are many electoral dstrcts and the purpose s to secure a majorty of seats, the strategy mght be to allocate more resources to pvotal dstrcts,.e., those n whch the ncumbent won/lost by a narrow margn (Snyder, 1989; Case, 2001) 4. Transfers and algnment. However, none of the above approaches s able to capture one of the fundamental trats of ntergovernmental transfers. Contrary to other targeted spendng programs, whch are mplemented drectly by the ncumbent, ntergovernmental transfers are decded by the hgher layer grantor government but executed by the sub-natonal recpent government. Ths s especally true n the case of earmarked captal transfers, whch are the focus of ths paper. In ths case, the grantor selects the projects based on ts own prortes and partly funds them, but t s the local government that must propose specfc projects for fundng and who has to contrbute local funds to them and take responsblty for ther executon. Ths overlappng of responsbltes means that the grantor cannot expect to reap all the poltcal benefts from the tactcal allocaton of these transfers, snce some share n the benefts must seep back to the local government. Ths should not represent an mpedment for the hgher layer grantor f the local government s controlled by the same party as the upper layer grantor (.e., both layers are algned). However, f the local government s controlled by the opposton, such transfers mght not be that effectve n mprovng the electoral prospects of the hgher layer ncumbent. At least two dfferent explanatons mght be nvoked to explan ths seepage of electoral benefts across layers of government. Frstly, voters mght splt the poltcal credt derved from the provson of the nfrastructure between layers of government (Arulampalam et al., 2009). When credt s attrbuted to the grantor government, the party n control at ths layer can reap all the electoral benefts. If credt s dvded equally between all layers, no one party can obtan an advantage from the addtonal transfers allocated to a muncpalty. When the strategy of the ncumbent s to target swng voters, the dvson of poltcal credt between layers means that a larger proporton of transfers wll be allocated to algned governments wth larger numbers of swng voters (Arulampalam et al., 2009; Solé-Ollé and Sorrbas- Navarro, 2008). The bold and dotted lnes n Panels (a) and (b) n Fgure 1 plot a hypothetcal (and over-smplfed) lnear relatonshp between the electoral vote margn of the hgher layer 3 To date the emprcal evdence s not conclusve as to whch of these two hypotheses s most pertnent, some papers supportng the swng voter hypothess (Case, 2001; Johansson, 2003; Dhalberg and Johansson, 2002) and others the core-voter one (Ansolabehere and Snyder, 2006). 4 The emprcal counterparts of the pvotal dstrct and of the swng voter hypotheses are smlar, snce the proporton of swng voters s often proxed by the ncumbent s vote margn (Johansson, 2003; Case, 2001). 7

10 ncumbent and the transfers allocated to the muncpaltes under ths hypothess. The graph mplctly assumes that there are more swng voters n muncpaltes wth a narrow margn of vctory 5. If the muncpalty s algned wth the hgher layer ncumbent (a stuaton that occurs when the vote margn of the regonal ncumbent s postve) and voters splt credt between the two layers, the amount of transfers receved wll be hgher, as ndcated by the jump or dscontnuty n the relatonshp between transfers and vote margn. Ths jump vanshes when voters are able to assgn all the credt to the hgher layer of government. Secondly, t s concevable that partsan algnment between layers of government mght also confer some beneft on the hgher layer ncumbent enablng hm to reach hs core supporters. The mayor mght be partcularly adept at dentfyng who the party s core supporters are at the local level and what ther specfc needs are. Thus, controllng the mayoralty would ensure that the ntal goals of the projects funded by the hgher layer of government do not become dstorted. Such a scenaro suggests that the algnment effect mght also nteract wth the core voter strategy. The bold and dashed lnes n Panel (a) of Fgure 1 show the shape of a hypothetcal relatonshp between the ncumbent s vote margn and transfers under ths hypothess. In ths case, we assume that transfers grow wth votes at both sdes of the zero-margn threshold. As n the swng-voter case, algnment makes the amount of transfers jump at the threshold. Of course, the algnment effect vanshes f the grantor s able to montor the use of transfers fully wthout the help of the mayor. [Fgure 1] Regonal vs local electons. These two justfcatons of the nteracton between the algnment status and the ncumbent s vote margn at the hgher layer rely mplctly on the assumpton that ncumbents am to maxmze ther probablty of beng re-elected at the next hgher-level electons. Arulampalam et al. (2009) explctly acknowledge ths fact. The only paper that suggests that the ncumbent s strategy mght, n fact, be focused on wnnng local electons s Brollo and Nanncn s (2012). Ths paper argues that n Brazl the best strategy for the federal presdent pror to the local electons s to am to wn as many mayoraltes as he can, snce mayors are nfluental opnon leaders n ther communtes and by engagng n campagnng and rent-seekng actvtes on the presdent s behalf can help wn more votes for the presdent at the hgher layer electons. 5 As dscussed by Johansson (2003), ths wll be true f the dstrbuton of deologcal preferences n support of the ncumbent (and, hence, aganst the opposton) s symmetrc and sngle-peaked. Dahlberg and Johansson (2002) present results that suggest that the departure from these assumptons s not dramatc n practce. 8

11 In lne wth ths hypothess, to use the resources at hs dsposal effcently, the hgher layer ncumbent should focus hs attenton on algned pvotal muncpaltes,.e., those n whch the mayoraltes were won by the narrowest margns. Ths strategy would target more funds for these muncpaltes (compared to algned muncpaltes won by a larger margn) and punsh unalgned pvotal muncpaltes, whch would receve less money than algned ones wth a smlar vote margn as well as less money than unalgned muncpaltes that the hgher layer ncumbent lost by a greater margn. The dashed and bold lnes n Panel (b) of Fgure 1 llustrate ths dea. Brollo and Nanncn (2012) fnd mxed evdence n favor of ths tactc, whch they refer to as tyng your enemy s hands n close races. Concurrent vs. alternatng electons. Brollo and Nanncn (2012) focus on the case of Brazl, where local electons are held n the mddle of the federal term-of-offce. As we explan below, n Span regonal and local electons are concurrent n some regons and alternatng n others. Ths dstncton allows us to compare the strength of the algnment effect n both cases. The smultaneous occurrence of the electons may ether reduce or ncrease the algnment effect. On the one hand, t mght shft the attenton of voters towards the ssues that are most relevant at the regonal level, thus lmtng the tactcal use of transfers to localtes. Lkewse, f the algnment effect only occurs when the strategy focuses on capturng mayoraltes, then the urgency of wnnng the next regonal electon (typcal of concurrent electons) mght attenuate the algnment effect. Before regonal electons, the regonal ncumbent mght choose to focus on hs core voters and f he s able to montor the use of resources wthout the help of mayors, ths wll generate a core-voter type profle but wthout any dscrmnaton n favor of the algned mayors. On the other hand, the smultaneous holdng of regonal and local electons may ncrease the salence of local ssues durng the campagn for the regonal electons. For example, n concurrent electons, regonal canddates may well be oblged to speak about local nfrastructure durng campagn vsts to muncpaltes. Smlarly, even f transfers do not matter drectly for regonal electons, they mght matter ndrectly though ther effect on the local electons, and the smultaneous occurrence of both electons could thus generate a bandwagon effect, wth the mpact on the vote of the local ncumbent beng transferred to some extent to the vote of the algned regonal ncumbent. Fnally, note that even n concurrent electons, the strategy of capturng mayoraltes mght make sense f the regonal electons are not compettve. Intutvely, f the regonal ncumbent feels safe, there would be less need to try to ncrease the total number of votes. Instead, t mght be worthwhle pursung a longer-term strategy,.e., wnnng 9

12 addtonal mayoraltes. Ths would allow more perks to be dstrbuted to party supporters and mght prove helpful at future regonal electons. 3. Background nformaton on Span 3.1 Spansh muncpaltes Spansh government comprses three layers: central, regonal, and local ters. There are seventeen regonal governments, the so-called Autonomous Communtes (ACs), whch have farly wde-rangng spendng responsbltes ncludng, for example, the provson of health care, educaton and welfare. Span s local layer conssts of over eght thousand muncpaltes, most of whch are relatvely small. These muncpaltes are multpurpose governments, wth major expendture categores correspondng to the tradtonal responsbltes assgned to the local publc sector (envronmental servces, urban plannng, publc transport, welfare, etc.), wth the excepton of educaton, whch s the responsblty of the regonal government. Current spendng s fnanced out of the muncpaltes own revenues (approxmately two thrds) and uncondtonal grants (approxmately a thrd). The latter are allocated accordng to a formula, whch hnders ther use for pork-barrel poltcs. However, the fundng of captal spendng s heavly dependent on grants: n 2008, captal grants, on average, represented 38% of captal spendng. Most Spansh muncpaltes do not have the capacty to fund necessary nvestments from other sources: ther tax bases are qute lmted, extraordnary resources from asset sales are not always avalable, and some muncpaltes may even have problems to access credt. Captal grants are transferred prmarly from the regonal layer (64%) and take the form of project grants 6 : there s an open call at regular ntervals (usually yearly) and a muncpalty can apply by submttng ts nfrastructure projects (e.g., street and road pavng, sewage systems and water ppes, parks and recreatons, educatonal and sports facltes, etc.). These are evaluated accordng to prevously establshed crtera (typcally publshed n the call), whch are subject to the nterpretaton of the grantor. Provsons are usually made for fundng emergency stuatons or projects consdered a prorty concern by the regonal government. The call often does not specfy clearly the weght attached to each of the crtera or t fals to specfy the lnk between the score assgned to each crteron and an objectve varable, leavng ths very much at the dscreton of the grantor. 3.2 Local poltcs n Span 6 A 19% comes from upper-local governments and the rest from the central government or the European Unon. 10

13 Local electons are held every four years on the same day throughout all the Spansh muncpaltes. Voters choose between several closed party lsts. The electoral system s a proportonal one, votes beng allocated to seats usng the d Hondt rule wth a threshold. The mayor s subsequently elected by a majorty of the councl (see Colomer, 1995). The councl operates as a small representatve democracy, and has to reach a majorty vote to pass the ntatves and regulatons proposed by the mayor, who acts as the agenda-setter. The dscplne enforced by Span s poltcal partes means that the chances of amendng the mayor s proposals are qute low when the mayor s party or coalton controls a majorty of the seats. The proporton of coalton governments s hgh (around 30% durng the terms we analyze), and most are formed along deologcal lnes. There are, however, exceptons to ths rule due, for example, to the fact that the platforms of many local partes are based solely on local ssues and so they are under less compulson to reach an agreement on deologcal grounds or because of pressure form hgher party ranks. Nevertheless, the nfluence of the party on the behavour of local poltcans s substantal, the local poltcal system beng seen as a frst step to subsequent promoton at the regonal and natonal levels. Electons to the regonal parlament are also held every four years and on the same day than the local electons n thrteen out of the seventeen regons. We refer to these polls as Concurrent electons. In the remanng four regons (.e., Galca, Catalona, Basque Country, and Andalusa), regonal electons are held md-term n relaton the local governments term of offce. We refer to these polls as Alternatng electons. Voters also choose between several party lsts, and the electoral system s also based on the d Hondt rule wth a threshold. Representatves elect the regonal presdent who, n turn, decdes the composton of the Cabnet. Here, also, around a thrd of the admnstratons are coalton or mnorty governments. 4. Emprcal desgn 4.1. The fuzzy RDD Papers usng observatonal approaches to estmate the effect of party deology on votes and polcy outcomes may suffer from an omtted varables problem: party control can be correlated wth the ncumbent s popularty and ths, n turn, mght have an mpact on the outcome varable. To deal wth ths problem some papers have recently adopted the closerace Regresson Dscontnuty Desgn (RDD) framework (see Lee, 2008; Lee et al., 2004; Pettersson-Ldbom, 2008; Ferrera and Gyourko, 2009; Albouy, 2010, Folke, 2010, Trounstne, 2011, and Gerber and Hopkns, 2011). The reasonng behnd ths method s that 11

14 electons won by a narrow margn are n practce very smlar events to electons lost by a smlar narrow margn. Thus, by focusng on close races, the RDD generates quasexpermental estmates of the effects of nterest (see Hahn et al, 2001). In a recent survey, Green et al. (2009) show that RDDs are comparable n accuracy to expermental studes. As mentoned, Brollo and Nanncn (2012) use ths approach to estmate the effect of partsan algnment on the allocaton of federal transfers to local governments n Brazl. In ths case, the treatment varable s defned as a dummy ndcatng whether the party of the federal Presdent (or the coalton that supports hm) won the local electon. The authors restrct the analyss to two- and three-canddate races so as to avod problems generated by the fact that Brazl s a hghly fragmented, mult-party system wthout any stable party coaltons. In any case, the pluralty rule used n Brazlan electons allows the authors to apply the tradtonal close-electons RDD. Ths s not an opton n our case, snce local councls are elected n Span usng a proportonal electoral rule. Ths rule generates many thresholds at whch an addtonal vote brngs one more seat to a party, and these are not necessarly located at the 50% vote threshold. To deal wth ths problem, we proceed n two steps. Frst, we compute our forcng varable as the share of votes that the deologcal bloc (.e., left or rght) of the regonal ncumbent has to lose (wn) to lose (gan) the majorty of seats n the local councl (and, thus, change ts algnment status), henceforth referred to as the vote margn. The calculaton of ths vote margn s not trval and has requred the development of a specfc procedure based on the d Hondt rule. We provde more detals on ths method n secton 4.4 and n Annex A. Second, we show that f the deologcal bloc of the regonal ncumbent has a majorty of seats n the local councl t s more probable (although not certan) that ths bloc also holds the mayoralty, whch means that the two layers of government are algned. Ths reflects the fact, dscussed above, that, more often than not, coaltons are formed along deologcal lnes. Ths means a fuzzy RDD has to be used (Van der Klauw, 2002; Lee and Lemeux, 2010), snce ths allows the treatment (.e., algnment) to be determned only partly by whether the assgnment varable (.e. the vote margn) crosses a cut-off pont (from negatve to postve). Whle n the sharp RDD the probablty of treatment jumps from 0 to 1 when the assgnment varable crosses a threshold, the fuzzy RDD nvolves a smaller jump n ths probablty. Snce the probablty of treatment jumps by less than one at the threshold, the dscontnuty n the outcome varable (that s, votes or transfers) at ths pont can no longer be nterpreted as an average treatment effect. However, the treatment effect can be recovered ether by dvdng the jump n the outcome varable by the jump n the probablty of 12

15 treatment or by estmatng the effect of algnment on the outcome by 2SLS, usng the threshold dummy as an nstrument for algnment Equaton specfcaton In our case, we use the followng three-equaton model: t = α a + g( m ) + ε (1) v = β a + f ( m ) + u (2) a = γ d + h( m ) + υ (3) where t =per capta transfers receved by the local government before the local electon; a =1 f there s algnment between the regonal and the local government and zero otherwse; m = regonal ncumbent s vote margn at the prevous local electons; v = vote share of the local ncumbent at the local electons; d =1 f the regonal ncumbent s vote margn s postve (.e. d =1 f m >0); the terms f(m ), g(m ) and h(m ), nclude polynomal terms of orders one or hgher, ftted separately at ether sde of the threshold (see Lee et al., 2004; Lee, 2008, and Lee and Lemeux, 2010). The frst equaton s used to estmate the effect of algnment on transfers. The second estmates the effect of partsan algnment on the local ncumbent s vote. The thrd descrbes the dscontnuty n algnment that we then use to dentfy the effects of nterest. Substtutng (3) nto (1) and (2) we obtan the reduced form equatons: t v = ϕ ) + 1 d + k( m ω (4) = φ ) + 1 d + j( m ν (5) where ϕ=αγ and φ=βγ are the ntent-to-treat estmates, whch are equal to the product of the effects of algnment on votes and on the dscontnuty. The estmaton of equatons (3), (4) and (5) allows us to recover the effect of algnment on votes and transfers as ˆ α = ˆ ϕ / ˆ γ and ˆ β = ˆ φ / ˆ γ. We could also estmate (1) and (2) by 2SLS, usng d as an nstrument for a. Both procedures should delver the same estmate as long as the order of the polynomals h m ) and j m ) or k m ) s the same. The estmates obtaned can be ( ( ( nterpreted as a weghted Local Average Treatment Effect (LATE), where the weghts reflect the ex-ante lkelhood of beng near the threshold (see Lee and Lemeux, 2010). The specfcaton n (2) and (3) can easly be modfed to analyze possble heterogeneous effects. Beng z a dummy varable defnng two non-overlappng groups of muncpaltes, we have: t = η ) + 1 d + η2d z + η3z + k( m ) + p( m z ς (6) 13

16 v = λ ) + 1 d + λ2d z + λ3 z + j( m ) + l( m z ϖ (7) To deal wth the possble correlaton of ths dummy wth other trats that dffer across subsamples we ntroduce the nteractons between algnment and several of the varables that can affect dffer across muncpaltes and the dscontnuty dummy and the polynomal at the same tme. Thus, the nterpretaton of the dfferental effect of algnment across subsamples reles on an dentfcaton strategy based on controllng for observables. Furthermore, n order to shed further lght on the possble mechansms behnd the algnment effect we can examne the shape of the polynomal at ether sde of the threshold, comparng these results wth the predctons derved from the dfferent theores surveyed n secton two. We are, however, also well aware that the shape of the polynomal has no causal nterpretaton n an RDD. To attenuate ths problem, we dscuss the shape of the polynomal only after the ncluson of a set of controls. The graphs used for ths purpose plot the resdual of the dependent varable (ether transfers or vote share) aganst the forcng varable. Ths means that n ths case too our dentfcaton strategy reles on our controllng for observables and that the conclusons reached are not as relable as those derved from the man RDD estmates Econometrcs In mplementng the RDD we have taken varous methodologcal decsons. Frst, as shown above, our man estmaton method uses all the observatons whle controllng for a flexble polynomal. Followng Lee and Lemeux (2010), we explctly test for the optmal order of the polynomal wth the Akake nformaton crtera. Ths procedure allows us to retan the entre sample when estmatng the heterogeneous effects. A possble drawback of ths method s that our results mght be senstve to outcome values for observatons far away from the threshold (see Imbens and Lemeux, 2008). To cope wth ths problem we also provde addtonal results obtaned by restrctng the bandwdths to 25% and 12.5%. The reason for ths choce s that the optmal bandwdth sze (Imbens and Kalyanaraman, 2009), s very close to 25% both for transfers (26.3%) and for votes (23.8%). Thus, n lne wth Lee and Lemeux (2010), we present our results wth optmal and half optmal bandwdths. Second, n order to show the need for usng a fuzzy RDD, we verfy the dscontnuty n the treatment probablty. To verfy that there s a substantal dscontnuty s tantamount to havng a strong frst-stage relatonshp n an IV desgn. Thrd, we also check the contnuty of the forcng varable around the threshold by nspectng the hstogram and usng a more formal test (see McCrary, 2008). The contnuty test provdes a means for dscardng the 14

17 manpulaton of the forcng varable, an ssue rased n varous papers (see, e.g., Caughey and Sekon, 2011). For ths same purpose, we also test for the contnuty of some pre-determned covarates. Fnally, we also provde some results usng a set of control varables (see next secton), n order to provde an addtonal valdaton check for our estmates (coeffcents should not change greatly) and to mprove the precson of our estmates. Furthermore, the use of covarates helps n the nterpretaton of the shape of the polynomals, snce, as already mentoned, they have no causal nterpretaton n an RDD analyss. 4.4 Sample and data. Sample. We estmate the effects of partsan algnment between local and regonal governments on transfers from the regonal to the local level and on the votes cast for the local ncumbent usng data on Spansh muncpaltes. We use two cross-sectons of data, for the terms and , wth around 3,000 muncpaltes n each. The sample s determned by data on transfers taken from a survey on budget outlays conducted yearly by the Spansh Mnstry of Economcs. Ths database ncludes all the muncpaltes wth more than 5,000 resdents and a representatve sample of the smaller ones 7. Transfers. The man results we report are for the estmaton of the algnment effects on captal transfers allocated to local governments n the two years precedng the next local electon. As explaned n secton two, gven the characterstcs of these transfers, we expect them to matter more n the perod runnng up to local electons. Ths dstncton, however, s rrelevant for twelve out of the ffteen regons (.e., those wth Concurrent electons), as regonal and local electons are held on the same day. It s true, however, than even f local electons matter most, the effect of algnment mght dffer n those regons wth Alternatng regonal and local electons (see secton two for a dscusson) and ths s why we also present our results for each of the samples. Although not ncluded here for motves of space, we wll also dscuss the results obtaned when analyzng the effects of algnment on transfers two year before the regonal electons (n the case of Alternatng electons) and durng the frst half of the term (n the case of Concurrent electons). In any of these cases, the two-year aggregaton helps n reducng the volatlty of the varable and the use of yearly nformaton wll not provde any statstcal advantage, snce the algnment status does not change between years wthn these two year perods. As we 7 Due to problems n accessng the data, the analyss s restrcted to ffteen regons, excludng the Basque Country and Navarre. These are qute small regons and ther excluson should not represent a bg problem. 15

18 explaned above, we focus on captal grants orgnatng from the regonal government because of a presumably hgher dscretonarty n ther allocaton. However, to confrm ths ntuton we also present results for the effect of algnment on current grants and on grants orgnatng from other layers of government (Central and Upper-Local). Votes. The second outcome varable we analyze s the mayor s vote share n the 2003 and 2007 local electons. Our results usng the coalton s vote share and the probablty of mayoral re-electon are smlar and not reported here for the sake of brevty. Votes by party at the local electons of 1999 and 2003 are used to construct the forcng varable and the dscontnuty nstrument. See Table A.1 n Annex A for the source of the vote results. Algnment. As explaned above, algnment s measured as a dummy equal to one when the mayor and the regonal presdent belong to the same party, regardless of whether the government at both layers s a sngle party or a coalton 8. See Table A.1 for the sources of these varables. As robustness checks, we have also checked whether the results are affected by the use of more comprehensve algnment defntons: stuatons where one party, even f t s not the man one, s present at both layers, and stuatons where the mayor and the regonal presdent smply belong to the same deologcal bloc and not only to the same party. Forcng varable. As explaned above, our man forcng varable s the Regonal ncumbent s bloc vote margn, computed as the votes needed for the deologcal bloc of the regonal ncumbent to gan/lose the majorty of seats n the local councl, expressed as a percentage of total votes cast at the local electons. To defne deologcal blocs we classfy all the partes standng at the local electons n three groups: left, rght and local partes (see Table A.1 for more detals). When the regonal party s a left/rght poltcal party, all the categores except left/rght are ncluded n the regonal opposton s bloc. As a robustness check, we also provde results after excludng those muncpaltes wth representaton of local partes from the analyss. The results obtaned do not depend on the specfc treatment of these partes. To compute the votes needed to brng about a change n the majorty of seats from one bloc to another, we use a very smlar method to that developed by Folke (2010). He provdes an algebrac formulaton for ths dstance under the Sant-League system, the one n operaton n Sweden (hs country of study). Wth ths formulaton he s able to compute the number of votes that each party needs to wn (or lose) an addtonal seat. We develop a smlar algebrac formulaton for the d Hondt system used n Spansh local electons. What 8 The concrete defnton of algnment used determnes the sze of the sample, snce we exclude the observatons not ncluded on the treatment or the control group. 16

19 we compute s the number of votes that the deologcal bloc to whch the regonal presdent belongs must lose (gan) to lose (wn) a majorty of seats at the local electons. In order to do ths, we make a number of assumptons regardng vote mgraton. We consder that the margnal votes lost (won): ) go (come) to (from) the abstenton, or ) partly to (from) the abstenton and the other deologcal bloc. We also assume that these votes are dstrbuted among the partes of the bloc n lne wth ther ntal vote share n the bloc. The man results of the paper use the vote margn computed under assumpton ) 9. Intutvely, n ths case, our formulaton works as f we were subtractng small numbers of votes from the mayor s bloc, dstrbutng them among the partes accordng to ther vote share wthn the bloc, whle keepng the number of votes for the partes of the other bloc constant. As we subtract votes, seats shft from one bloc to the other. The procedure stops when we observe a shft n the seat majorty from one bloc to the other. The number of votes needed to reach ths pont dvded by the total number of votes ntally cast at the electon s our measure of vote margn. See Box A.1 n Annex A for the algebrac formulaton used to compute the vote margn 10. Control varables. In order to provde a further check on the relablty of the RDD results and to mprove the effcency of our estmates, we also present results when controllng for several covarates. In the case of the transfer equaton, we control for log(populaton), land area per capta, property tax rate, assessed value of the property, debt burden, and Regonal dummes term effects (see also Solé-Ollé and Sorrbas-Navarro, 2008). In the vote share equaton, we control for party of the mayor term effects, Regonal dummes term effects, ncumbent s hstorcal vote share, hstorcal turnout at local electons, local coalton dummy, local frst-term dummy, and populaton sze dummes. 5. Results 5.1. Explorng the dscontnuty Panel (a) n Fgure 2 plots the seat margn of the regonal ncumbent s bloc at the local electons aganst ts algnment status whch s gven a value of one f the mayor and the regonal presdent belong to the same party. The graph shows a consderable jump when the deologcal bloc of the regonal ncumbent moves from -1 seat to +1 seat (.e., when t requres one addtonal seat to gan/lose a majorty of seats). 9 As a robustness check we have also examned whether the computaton of the vote margn under assumpton () does change the results. 10 In Annex B (not for publcaton) we also provde a numercal example whch llustrates how ths method works n practce. 17

20 Although t mght seem approprate to perform the analyss by comparng the average value of transfers or votes for the muncpaltes located at the -1 and +1 values of the seats margn, ths would not be correct, snce ths s qute a large group wth consderable nternal varablty n the popularty of the regonal ncumbent. For ths reason, we use the vote margn as the forcng varable, computed as the percentage of votes needed for the regonal ncumbent s bloc to wn/lose a majorty of seats n the cty councl. Panel (b) n Fgure 2 shows the plot between ths forcng varable and the algnment status. The dots represent averages of the algnment dummy over 5% bns. The sze of the bn has been selected usng the bn test proposed by Lee and Lemeux (2010). The black lne s the flexble polynomal ftted separately on both sdes of the threshold. From the fgure t s evdent that there s a szeable jump n the probablty of algnment when movng from -1 to +1 seats. [Fgure 2] Table 1 shows the results obtaned when estmatng the dscontnuty wth dfferent bandwdths: 100% wth polynomals of orders 1 to 3, and 25% and 12.5% wth a local lnear regresson. In the full sample case, the Akake nformaton crteron suggests that t s optmal to ft a 2 nd order polynomal. In ths case, the estmated value of the dscontnuty s 85%. The results do not change much when other polynomal orders are used or when the bandwdth s restrcted. [Table 1] A possble concern wth the RDD s the possblty that the forcng varable mght be manpulated. Ths could occur, for example, f the electoral results have been manpulated or, n the case of mult-party governments, f the vote of the last representatve needed to form a wnnng coalton has been bought. We deal wth ths last problem by usng local votes for the deologcal bloc of the regonal ncumbent as opposed to votes obtaned by the actual coalton that supports the mayor. A way of verfyng that the forcng varable has not been manpulated s to examne ts hstogram or, more formally, to test for the contnuty of ths varable at the cut-off by runnng local lnear regressons of the log of the densty separately on both sdes of zero (see McCrary, 2008). We have performed both checks, and we have not found any evdence of manpulaton. Another valdty check conssts on testng for the presence of a dscontnuty n the pre-determned covarates. The results of ths exercse also suggest that none of these varables s dscontnuous around the threshold All these results are reported n Annex B (not for publcaton), Table B.1 and Fgures B.1 and B

21 5.2. Partsan algnment and transfers Fgure 3 shows the plots between the forcng varable and both the amount of captal transfers and of resdual transfers (.e., the resdual of a regresson between transfers and control varables). The graphs suggest that there s a clear dscontnuty: muncpaltes margnally on the rght of the cut-off (those whch are very lkely to be algned) do receve much greater sums n transfers than those margnally on the left (those whch are very lkely to be unalgned). The result s a lttle bt clearer n the when usng resdual transfers. Ths shape suggests that the strategy used by regonal governments revolves around tryng to nfluence close local races n places where the mayor s a co-partsan. [Table 2 & Fgure 3] Table 2 presents the RDD estmates. Panel (a) shows the Reduced form estmates whle Panel (b) reports the 2SLS results. Columns () to () show the results wth the full sample and wth polynomals of orders 1 to 3. The polynomal of order 2 s the optmal one (accordng to the AIC crteron). Column (v) repeats the results usng the optmal polynomal but ntroducng the control varables n the equaton. Columns (v) to (v) present the results wth the 25% and 12.5% bandwdths, usng a local lnear regresson and wthout (v and v) and wth control varables (v and v). The estmates are qute robust to the choce of bandwdth and polynomal order and to the ntroducton of covarates. The reduced form coeffcents are around 80 euro and those of the 2SLS are around 92 euro. Ths amount has to be compared wth the transfers receved by unalgned muncpaltes just at the left of the cut-off, whch are around 107 euro. Wth these numbers, an algned muncpalty would receve 83% more per capta transfers than a smlar unalgned one Partsan algnment and votes Fgure 4 shows the plot between the forcng varable and the mayor s share of the vote. The graph suggests that there s a dscontnuty n the vote share: local ncumbents margnally to the rght of the cut-off do receve more votes than those margnally to the left. The shape of the plot s as expected: to the rght of the cut-off the local ncumbent s vote share s postvely correlated wth that of the regonal ncumbent s deologcal vote share; to the left of the cut-off, both varables are negatvely correlated. [Table 3 & Fgure 4] Table 3 presents the RDD estmates. Here, also, the results are qute stable across specfcatons. The reduced form coeffcents are statstcally sgnfcant at the 99% level n all cases and dentfy a dscontnuty between 3.8% and 4.4%. The 2SLS results suggest that 19

22 the average treatment effect s hgher, between 4.3% and 5.8%. These are szeable effects, especally f we take nto account that a mayor s vote share at the left of the cut-off s just 42.7%, meanng that an algned mayor wll receve 10.07% (=4.3% over 42.7%) more votes than a smlar unalgned mayor. Addtonal results (not shown here) suggest that the effects on the votes for the whole coalton are a lttle lower, mplyng that the mayor s party s the one that benefts most from algnment wth the regonal government OLS and dfference-n-dfferences The estmated effect of algnment on captal transfers (83%) s twce as great as the effect estmated by Solé-Ollé and Sorrbas-Navarro (2008) usng dfference-n-dfferences. Ths dfferental s strkng, gven that both studes draw on very smlar data. Comparson of the respectve results, however, s dffcult, snce the samples and perods are dfferent. To determne the causes of ths dscrepancy, we have also estmated the algnment effect on transfers by OLS and dfference-n-dfferences (.e., ncludng muncpalty fxed effects) n our sample, controllng n both cases for the full set of control varables. The results, shown n Table 4, mply that algned muncpaltes receve 52% more grants than unalgned muncpaltes. Ths s hgher than the 40% reported by Solé-Ollé and Sorrbas-Navarro (2008), but stll much lower than our RDD estmates Other transfers. We have also estmated the effect of beng algned wth other layers of government (Upper-local government, and Central government) on the amount of captal transfers allocated by these layers to muncpaltes. The reason we do not focus on these transfers from the outset s the smaller quanttes nvolved. The results are shown n Table 5 and suggest that muncpaltes algned wth Upper-local governments receve around 60% more transfers than those unalgned. The effect on captal transfers allocated by the central government s much lower, around a 27% ncrease, and s not statstcally sgnfcant. A possble explanaton for ths result mght be the fact that t s qute dffcult for central government to dscrmnate n ts allocaton of resources gven the hgh number of Spansh muncpaltes (around 8,000) and the consequent lack of specfc knowledge about the local poltcal stuaton of each. Thus, t mght be the task of ntermedate governments (regonal and upper-local) to help channel the mones of central government to the most poltcally senstve places (see also Castells and Solé-Ollé, 2005, and Solé-Ollé, 2012). 20

23 We have also estmated the algnment effects on the current transfers allocated by each of the three upper layers of government. In each case the algnment effect s not statstcally sgnfcant. Ths s as expected, snce most current transfers to Spansh muncpaltes are formula-based and, as such, are much more dffcult to manpulate than earmarked transfers for captal projects. Overall, our results dentfy the nstruments and governments that are most prone to beng affected by poltcal tactcs n Span: captal transfers and ntermedate governments, manly regons and, to a lesser extent, also Upper-local governments Robustness checks. The results are robust to many changes n some key aspects of the methodology. We brefly dscuss the man conclusons of ths analyss 12. Frst, the results are very smlar when usng two other (more comprehensve) measures of algnment: () ncludng all the cases n whch the man party at one layer (the one holdng the mayoralty or the regonal presdency) s a mere partner n the coalton at the other layer, and () ncludng ncludes cases n whch the two layers are consdered to be algned f the mayor s party belongs to the same deologcal bloc than the party of the regonal presdent, but t s not necessarly the same party. Second, the results are also robust to the excluson of the muncpaltes n whch local partes are represented n local councls, and to usng only the muncpaltes n whch the two man partes obtan more than 80% of the vote. Fnally, the results are more or less the same when usng an alternatve measure of vote margn, computed on the assumpton that votes are transferred not solely from abstenton but also from the opposton bloc Heterogeneous effects Concurrent vs. Alternatng electons. Table 6 shows the RDD results (reduced form) obtaned when ncludng nteractons of the dscontnuty dummy and the polynomal terms wth the electon tmng dummes. The results suggest that the effect s much hgher (nearly twce as hgh) n the case of Concurrent electons than n the case of Alternatng electons. [Tables 6 & 7] To shed some lght on the mechansm that can derve these results, we nteract dscontnuty dummy and the polynomal terms, not only wth the electon tmng dummes, but also wth the other potentally dsturbng varables. We consder, for example, that the algnment effect mght also be affected by whether: () regonal electons are compettve or 12 The tables showng the complete results are ncluded n Annex B (no for publcaton), Table B

24 not, () the regon has a large amount of budget resources, meanng t can allocate more generous captal transfers and that the dfferences between algned and unalgned muncpaltes mght be more marked, () the muncpalty has greater needs or s n a poorer fnancal stuaton. The compettveness of regonal electons has been proxed by a dummy (Compettve) whch s equal to one f the regonal vote share of the regonal ncumbent n the prevous regonal electon s lower than the sample medan. The avalablty of budget resources has been measured by a dummy whch ndcates whether the regon has more resources than the medan (Hgh resources) 13. Muncpal needs and the muncpal fnancal stuaton are proxed by three dummes: Small, ndcatng whether the muncpalty has less than 5,000 resdents, Debt, ndcatng whether the debt burden per capta les above or below the medan, and Hgh fscal capacty, ndcatng whether the per capta assessed value of the property les above or below the medan. We fnd that Concurrent s qute strongly correlated wth Compettve (correlaton coeffcent equal to -0.52) and wth Hgh resources (correlaton coeffcent equal to 0.18) but not wth the other varables (correlaton coeffcents around , n absolute value). In Table 6, column (v) shows the results when ntroducng the nteracton wth the three varables at the same tme. The prevous results stll hold; the effect n Concurrent electons beng more marked than that n Alternatng electons, despte the relevance of the other nteractons. However, the dfference between Concurrent and Alternatng electons s now much smaller, probably as a result of the aforementoned correlaton between electon type and the degree of compettveness of the regonal electons. In results not shown n Table 7 (but avalable upon request), we fnd that all the fnancal needs and fnancal stuaton varables have a postve mpact on the algnment effect, but these nteractons are not statstcally sgnfcant and ther ncluson does not modfy our concluson regardng the dfference between Concurrent and Alternatng electons. In Table 7 we repeat the analyss but now for the local vote share. Once agan, the algnment effect n Concurrent electons s stronger than that n Alternatng electons even when we control for the other nteractons. [Fgure 5] The top panel n Fgure 5 shows the plot between resdual transfers and the vote margn for Concurrent and Alternatng electons. The dscontnuty s clearly larger n the frst case. The shape of the two plots s smlar, but n the case of Alternatng electons the 13 Ths varable s equal to one (zero) f per capta standardzed resources (transfers + standardzed tax revenues) s hgher (lower) than the sample medan. Regonal-level data to compute ths varable comes from BADESPE (Insttute for Fscal Studes, Mnstry of Economcs). 22

25 slope at the rght of the threshold s more clearly negatve. We wll return to ths when nterpretng the results n the next secton. Fnally, the avalablty of data for the Alternatng electons sample allows us to look at the effect of transfers two years before regonal electons (as opposed to two years before the muncpal ones). Our results (not reported here for motves of space) show that n ths case partsan algnment has no effect on the amount of captal transfers allocated. It seems therefore that these transfers matter mostly for local electons 14. Compettveness and Budget resources. Columns () and () n Table 6 present the results for the nteractons wth the Compettveness and Budget resources dummes, and column (v) shows the effect of these nteractons when they are ntroduced at the same tme and smultaneously wth the electon type nteractons. The results of ths last column show that the algnment effect s also stronger n Non-compettve electons and n regons wth Hgh budget resources. Snce there are just three regons wth Alternatng electons, and gven the correlaton between electon type and Compettveness and Budgetary resources, we repeated the analyss consderng only the subsample of muncpaltes n regons wth Concurrent electons. The results are shown n columns (v) to (v) n Table 6 and suggest that the dfferences persst: the effect of algnment on captal transfers s hgher n muncpaltes belongng to regons wth Non-compettve electons and n regons wth Hgh budget resources. The dfferences are statstcally sgnfcant and meanngful, especally for the Compettveness nteracton. The bottom panel n Fgure 5 shows the plot between Resdual transfers and the vote margn for Non-compettve and Compettve Concurrent electons. The dscontnuty s larger when regonal electons are non-compettve and the slope of the polynomal s clearly negatve only n ths case. We wll return to ths result below Interpretaton of the results. Our results can be nterpreted as follows. Frst, the greater algnment effect reported here for Concurrent than for Alternatng electons mght be due to a modfcaton n voter behavor (and, hence, n poltcans ncentves) occurrng n ths latter case due to the 14 Addtonally, we have used the sample of Concurrent electons to look at the effect of algnment durng the frst two years of the (regonal and local) term-of-offce. In ths case (results also avalable upon request) we fnd an effect of algnment on transfers whch s approxmately half the magntude of the effect found for the second half of the term. 23

26 smultaneous occurrence of local and regonal electons. In Concurrent electons, voters cast ther votes for local and regonal canddates at the same tme. In ths case, f captal transfers confer some sort of advantage to the local ncumbent, ths advantage mght automatcally be transferred to the canddate from the same party at the regonal level. Ths bandwagon effect between canddates from the same party standng at smultaneously held electons has been documented n the lterature (see, e.g., Ade and Freer, 2011). Smlarly, n Concurrent electons, the local and regonal campagns mght be more closely connected, wth regonal canddates havng to speak about local ssues durng vsts to muncpaltes due to the greater salence of such questons n the local campagn. Ths means that even f the nfrastructure funded by captal transfers from regonal governments plays a small role n party platforms at the regonal level, t mght have an ndrect effect on voters decsons at that level. The absence of an algnment effect on local votes n the Alternatng case can be smlarly explaned. Second, the fact that, n Alternatng electons, algnment only seems to matter before muncpal electons, but not before the regonal ones suggests that regonal ncumbents care most about these local contests. In ths case (and also n the case of Non-compettve Concurrent electons), the shape of the polynomal also ponts n the same drecton, suggestng that regonal ncumbents am at capturng as many mayoraltes as they can. Fgure 3 clearly shows that transfers decrease before the threshold and ncrease after, whch s the pattern dentfed n secton two for ths type of electoral strategy (recall Fgure 1). Although the effect estmated through RDD cannot be extrapolated to observatons far from the threshold, the shape of the polynomal can be nformatve about the strateges used by the regonal ncumbents. Among the algned muncpaltes, the regonal government would rather target pvotal muncpaltes than loyal ones, whle pvotal unalgned muncpaltes mght be specally punshed. Fgure 5 shows that ths strategy s most apparent when electons are Alternatng. However, Fgure 5 also shows that n the Concurrent electons, regons wth Non-compettve regonal electons also adhere to ths pattern. Moreover, the polynomal n regons wth Compettve electons s qute flat, and the slope s even postve to the rght-hand sde of the zero-margn threshold. Smlarly, the sze of the dscontnuty s much lower n ths case. Ths suggests that an electoral strategy centered on pvotal muncpaltes mght underle the results of the Concurrent electons sample, at least for regons wth Non-compettve electons. For the remanng regons n ths sample, ths strategy mght be attenuated by a strategy that focuses on locatons of core voters, wth the 24

27 am of mprovng the chances of wnnng a hghly compettve regonal electon by tryng to moblze the electorate. 7. Concluson In ths paper we have used a fuzzy RDD to estmate the effect of partsan algnment between regonal and local governments both amount of transfers receved and on the vote for the local ncumbent at the local electons. We have provded very strong evdence that voters gve more support to local ncumbents belongng to the party that controls the regonal government. Our results suggest that algned muncpaltes obtan 83% more per capta transfers than unalgned muncpaltes. Algned ncumbents also wn 10% more votes than unalgned ncumbents. These estmates are much hgher than prevous estmates for Span usng dfference-n-dfferences technques and much hgher than results reported for other countres, ncludng those usng an RDD. We have also documented that the effect of partsan algnment s stronger: () when regonal and local electons are held on the same day, () when regonal electons are less compettve, and () when the regonal government has more budget resources to fund these dscretonary transfers. Ths nteracton wth the amount of budget resources suggests that the effect of algnment on transfers ultmately has consequences n terms of votes. Some secondary evdence suggests that the algnment effect mght arse as a result of a regonal electoral strategy centered on the transfer of resources to pvotal and algned muncpaltes wth the am of wnnng as many mayoraltes as possble. Ths strategy seems more evdent n Alternatng electons and n Non-compettve Concurrent electons. It seems, therefore, at least n some cases, that the regonal ncumbent pursues a delberate strategy of nterferng n the outcome of local electons. As dscussed n the ntroducton, such practces mght erode accountablty at the local level and, thus, undermne the very benefts of decentralzaton. 25

28 References Ade, F. and Freer, R. (2011): Dvded government versus ncumbency externalty effect: quas-expermental evdence on multple votng decsons, DIW Workng Papers Albouy, D. (2010): Partsan representaton n Congress and the geographc dstrbuton of Federal funds, Revew of Economcs and Statstcs, forthcomng. Ansolabehere S. and Snyder, J. (2006): Party control of State government and the dstrbuton of publc expendtures, Scandnavan Journal of Economcs 108(4): Arulampalam W., S. Dasgupta, A. Dhllon, and B. Dutta (2009): Electoral goals and centerstate transfers: a theoretcal model and emprcal evdence from Inda, Journal of Development Economcs 88, Brollo, F., Perott, R., Nanncn, T. and Tabelln, G (2012): The Poltcal Resource Curse, Amercan Economc Revew, forthcomng. Brollo, F. and Nanncn, T. (2012): Tyng your enemy s hands n close races: the poltcs of federal transfers n Brazl, Amercan Poltcal Scence Revew, forthcomng. Broockman, D.E. (2009): Do congressonal canddates have reverse coattals? Evdence from a Regresson Dscontnuty Desgn, Poltcal Analyss 17, Broso, G. and Ahmad, E. (2009): Does decentralzaton enhance servce delvery and poverty reducton?, Edward Elgar. Brueckner, J.K. (2009): Partal fscal decentralzaton, Regonal Scence and Urban Economcs 39, Campbell, J.E. (1986): Presdental coattals and mdterm losses n State legslatve electons, Amercan Poltcal Scence Revew 80(1), Campbell, J.E. and Sumners, J.A. (1990): Presdental coattals n Senate electons. Amercan Poltcal Scence Revew 84, Case, A. (2001): Electon goals and ncome dstrbuton: recent evdence from Albana, European Economc Revew 45, Castells, A. and Solé-Ollé, A. (2005): The regonal allocaton of nfrastructure nvestment: the role of equty, effcency and poltcal factors, European Economc Revew 49, Caughey, D. and Sekhon, J.S. (2011): Electons and the regresson dscontnuty desgn: lessons from close US House races, Poltcal Analyss 19, Colomer, J.M. (1995): España y Portugal, n J.M. Colomer (ed.): La polítca en Europa: ntroduccón a las nsttucones de 15 países, Arel, Barcelona. 26

29 Cox, G.W. and McCubbns, M. (1986): Electoral poltcs as a redstrbutve game, Journal of Poltcs 48(2), Cox, W.G. (2009): Swng voters, core voters, and dstrbutve poltcs, n I. Shapro, S. Stokes, E.J. Woods, A.S. Krshner (eds.), Poltcal Representaton, pp , Cambrdge Unversty Press, Cambrdge, UK. Dahlberg, M. and Johansson, E. (2002): On the vote purchasng behavor of ncumbent governments, Amercan Poltcal Scence Revew 96, Devajaran, S., S. Kheman and S. Shah (2009): The Poltcs of Partal Decentralzaton. In E. Ahmad and G. Broso (eds.), Does decentralzaton enhance servce delvery and poverty reducton, Edward Elgar Publshers. Daz-Cayeros, A., Magalon, B. and Wengast, B. (2006): Tragc brllance: equlbrum party hegemony n Mexco, Workng Paper, Hoover Insttuton. Dxt, A, and Londregan, J. (1996): The determnants of success of specal nterests n redstrbutve poltcs, Journal of Poltcs 58, Ferrera, F. and Gyourko, J. (2009): Do poltcal partes matter? Evdence from US ctes, Quarterly Journal of Economcs 124(1), Folke, O. (2010): Shades of brown and green: party effects n proportonal electon systems, Columba Unversty, _of_ Brown_and_Green_Olle_Folke.pdf. Folke, O. and Snyder, J.M. (2012): Gubernatoral mdterm slumps, Accepted Amercan Journal of Poltcal Scence, forthcomng. Gélneau, F. and Remmer, K.L. (2006): Poltcal decentralzaton and electoral accountablty: The Argentne Experence, , Brtsh Journal of Poltcal Scence 36, Gerber, E.R. and Hopkns, D.J. (2011): When mayors matter: estmatng the mpact of mayoral partsanshp on cty polcy, Amercan Journal of Poltcal Scence 55(2), Green, D.P., Terrence, Y., Leong, H.L. and Gerber, A.S. (2009): Testng the accuracy of regresson dscontnuty analyss usng expermental benchmarks, Poltcal Analyss 17(4), Grossman, J. P. (1994): A poltcal theory of ntergovernmental grants, Publc Choce 69, Hahn, J., P. Todd, and W. Van der Klaauw (2001): Identfcaton and estmaton of treatment effects wth regresson dscontnuty desgn, Econometrca 69,

30 Hanmueller, J. and Kern, H.K. (2008): Incumbency as a source of spllover effects n mxed electoral systems: Evdence from a regresson-dscontnuty desgn, Electoral Studes 27, Imbens, G. and K. Kalyanaraman (2009): Optmal bandwdth choce for the Regresson Dscontnuty estmator, NBER Workng Paper No Imbens, G. and T. Lemeux (2008): Regresson dscontnuty desgns: a gude to practce, Journal of Econometrcs 142, Johansson, E. (2003): Intergovernmental grants as a tactcal nstrument: some emprcal evdence from Swedsh muncpaltes, Journal of Publc Economcs 87, Kheman, S. (2010a) Gerrymanderng decentralzaton: poltcal selecton of grants-fnanced local jursdctons. In Bosch, N., Espasa, M. and Solé-Ollé, A. (eds.): The Poltcal Economy of Inter-regonal fscal flows: measurement, determnants, and effects on country stablty, Edward Elgar. Kheman, S. (2010b): Poltcal capture of decentralzaton. Vote buyng through grantsfnanced local jursdctons, Polcy Research Workng Paper 5350, The World Bank. Lee, D. and Lemeux, T. (2010): Regresson Dscontnuty Desgns n economcs, Journal of Economc Lterature 48(2) Lee, D., Morett, E. and Buttler, M. (2004): Do voters affect or elect polces? Evdence from the US House, Quarterly Journal of Economcs 119(3), Lee, D.S. (2008): Randomzed experments for non-random selecton n US House electons, Journal of Econometrcs 142(2), Lndbeck, A. and J. W. Webull (1987): Balanced-budget redstrbuton as the outcome of poltcal competton, Publc Choce 52, McCrary, J. (2008): Manpulaton of the runnng varable n the Regresson Dscontnuty Desgn: a densty test, Journal of Econometrcs 142, Oates, W.E. (1972): Fscal Federalsm, Harcourt Brace Jovanovch, New York. Pettersson-Ldbom, P. (2008): Do partes matter for economc outcomes? A Regresson- Dscontnuty approach, Journal of the European Economc Assocaton 6, Rodden, J. (2002): The dlemma of Fscal Federalsm: grants and fscal performance around the world, Amercan Journal of Poltcal Scence 46/3, Rodden, J. and Wbbels, E. (2011): Dual accountablty and the natonalzaton of party competton: evdence from four federatons, Party Poltcs 17, Samuels, D.J. (2000): The gubernatoral coattals effect: federalsm and congressonal electons n Brazl, Journal of Poltcs 62(1),

31 Schener, E. (2005): Ppelnes of pork. Japanese poltcs and a model of local opposton party falure, Comparatve Poltcal Studes 38(7), Seabrght, P. (1996): Accountablty and decentralzaton n government: An ncomplete contracts model," European Economc Revew 40(1), Shah, A. and Thompson, T. (2004): Implementng decentralzed local governance, Polcy Research Workng Paper Seres 3353, The World Bank. Snyder, J.M. (1989): Electon goals and the allocaton of campagn resources, Econometrca 57, Solé-Ollé, A. (2012): Inter-regonal redstrbuton through nfrastructure nvestment: tactcal or programmatc?, Publc Choce, forthcomng. Solé-Ollé, A. and Sorrbas-Navarro, P. (2008): The effects of partsan algnment on the allocaton of ntergovernmental transfers. Dfferences-n-dfferences estmates for Span, Journal of Publc Economcs 92, Trounstne, J. (2011): Evdence of a local ncumbency advantage, Legslatve Studes Quarterly 36, Van der Klauw, W. (2002): Estmatng the effect of fnancal ad offers on college enrollment: A Regresson-Dscontnuty approach, Internatonal Economc Revew 43(4), Wengast, B. (2009): Second generaton fscal federalsm: The mplcatons of fscal ncentves, Journal of Urban Economcs 65(3),

32 Tables and Fgures Table 1: Dscontnuty n the probablty of algnment. () () () (v) (v) d (89.13) *** (55.02) *** (39.90) *** (102.78) *** (59.64) *** R AIC Bandwdth 100% 100% 100% 25% 12.5% Polynomal order Obs Notes: (1) and terms. (2) Dependent varable s Algnment, a = 1 f mayor and the regonal presdent belong to the same party. (3) Explanatory varables: dscontnuty dummy d and polynomal on the Regonal ncumbent s bloc vote margn; polynomal ftted separately on ether sde of the zero threshold; d s one f vote margn s postve and zero f vote margn s negatve. (4) Bandwdth = 100% ndcates that all the observatons have been used n the estmaton; 25% of vote ndcates a bandwdth of -25% to 25%, 25% beng (approxmately) the optmal bandwdth of both the transfers and ncumbent s vote share used n Tables 2 and 3 (see below). (5) t-statstc n parentheses, robust standard errors used; ***, ** & * = statstcally sgnfcant at the 99%, 95% and 90% levels. (6) AIC = Akake nformaton crteron. Table 2: Effect of algnment on captal transfers. RD results. () () () (v) (v) (v) (v) (v) a) Reduced form d (6.12) *** (5.07) *** (4.05) *** (5.95) *** (4.39) *** (4.86) *** (5.68) *** (6.02) *** R AIC b) 2SLS a (5.31) *** (5.08) *** (4.54) *** (4.97) *** (4.88) *** (4.97) *** (5.13) *** (5.76) *** Bandwdth 100% 100% 100% 100% 25% 25% 12.5% 12.5% Pol. order Controls NO NO NO YES NO YES NO YES Obs Notes: (1) See Table 1. (2) Reduced form = OLS regresson of captal transfers aganst d, whch s one f vote margn s postve and zero f vote margn s negatve, controllng for a two-sded polynomal of the vote margn; 2SLS = 2SLS estmaton of captal transfers aganst the algnment dummy, a, usng d as the nstrument, and controllng for the same polynomals. (3) Control varables ncluded: log(populaton), land area per capta, property tax rate, assessed value of the property, debt level and Regonal term effects. See Table A.1 n Annex A for defntons and data sources. (4) Optmal polynomal order used n column (v). (5) Local lnear regresson wth optmal bandwdth used n columns (v) and (v); ½ of optmal bandwdth used n columns (v) and (v). 30

33 Table 3: Effect of algnment on local vote share. RD results. () () () (v) (v) (v) (v) (v) a) Reduced form d (9.16) *** (5.00) *** (4.01) *** (3.64) *** (5.59) *** (2.56) *** (2.38) *** (2.67) * R AIC b) 2SLS a (6.44) *** (5.06) *** (4.78) *** (4.86) *** (5.71) *** (2.56) *** (3.10) *** (2.81) *** Bandwdth 100% 100% 100% 100% 25% 25% 12.5% 12.5% Pol. order Controls NO NO NO YES NO YES NO YES Obs Notes: (1) See Table 2. (2) Dependent varable: % vote share for the mayor. (3) Control varables: party of the mayor term effects, Regonal dummes term effects, ncumbent s hstorcal vote share, hstorcal turnout at the local and regonal electons, local coalton dummy, local frst-term dummy, and populaton sze dummes (see Table A.1). Table 4: Effect of algnment on transfers & local vote share. OLS & Dfference-n Dfferences () () () (v) (v) (v) a) Captal transfers b) Vote share a (10.34) *** (5.07) *** (5.11) *** (21.23) *** (7.17) *** (5.08) *** R Controls NO YES YES NO YES YES Muncpalty fxed effects NO NO YES NO NO YES Obs Notes: (1) See Tables 2 and 3. (2) Standard errors clustered at the muncpalty level n eq. () & (v). Table 5: Effect of algnment on other types of transfers. RD results. () () () (v) (v) Captal transfers: Current transfers: Provncal Central Regonal Provncal Central a) Reduced form d (3.34) *** 9.33 (1.23) 8.92 (0.78) 3.44 (0.45) 4.56 (0.27) b) 2SLS a (3.45) *** 8.54 (1.10) (0.66) 5.09 (0.37) 8.98 (0.12) % Increase Obs Notes: (1) See Table 2. (2) % Increase = 2SLS coeffcent over captal transfers evaluated at left lmt of the threshold. 31

34 Table 6: Effect of algnment on captal transfers. Electoral margn and fscal capacty. () () () (v) (v) (v) (v) All electons Concurrent d Concurrent (4.90) *** (3.77) *** d Alternatng (2.96) *** (2.20) ** d Compettve (3.29) *** (7.07) *** (3.78) *** d Non-compettve (6.73) *** (4.22) *** (8.00) *** (8.23) *** d Hgh resources (4.19) *** (1.58) (5.29) *** (2.23) ** d Low resources (5.15) *** (9.64) *** Dfference [F-test p-value] [0.000] [0.000] [0.121] [0.002] [0.012] [0.048] [0.000] Notes: (1) See Table 3. (2) Reduced form RD results. (3) Compettve/Non-compettve = vote share for the regonal ncumbent >(<) lower than the medan. (4) Hgh/Low resources = per capta resources (transfers + standarzed tax revenues) >(<) than the medan. (5) All equatons have been estmated usng the full sample, a two-sded second order polynomal for each of the nteracted varables, and the full set of control varables. (6) Dfference = dfference between the coeffcents of the two mutually exclusve categores (e.g., n column (v) concurrent vs. alternatng, and n column (v) Hgh margn vs. Low margn). (7) Standard errors clustered at the regonal level. Table 7: Effect of algnment on local vote share. Electoral margn and fscal capacty. () () () (v) (v) (v) (v) All electons Concurrent d Concurrent (4.01) *** (2.82) *** d Alternatng (1.29) (1.39) d Compettve (2.65) ** (4.53) ** (0.42) d Non-compettve (6.96) *** (1.78) * (4.02) *** (3.28) ** d Hgh resources (5.86) *** (2.15) ** (3.13) *** (2.38) ** d Low resources (1.78) * (4.40) *** Dfference [F-test p-value] [0.049] [0.565] [0.015] [0.047] [0.035] [0.040] [0.042] Notes: (1) See Tables 2 and 4. (2) Reduced form RD results. 32

35 Fgure 1: Transfers vs. vote margn n Swng voter, Core voter & Pvotal muncpaltes. a) Swng voter vs. Core voter b) Swng voter vs. Pvotal 0 Vote margn 0 Vote margn Swng or Core Swng Core Swng or Pvotal Swng Pvotal Fgure 2: Algnment vs margn a) Seat margn b) Vote margn Algnment Regonal-Local Algnment Regonal-Local Regonal Incumbent s Bloc Seat Margn Regonal Incumbent's Bloc Margn Notes: (1) and terms. (2) Algnment Regonal-Local = 1 f the mayor and the regonal presdent belong to the same party. (3) Regonal ncumbent s bloc seat margn = dstance n seats to a change n deologcal bloc s seat majorty; seats as obtaned at the 1999 and 2003 local electons. (4) Regonal ncumbent s bloc vote margn = dstance n percentage of votes to a change n deologcal bloc s seat majorty; vote shares as obtaned at the 1999 and 2003 local electons (see Box A.1 n Annex A). (5) Dots = Bn averages; Bn sze = 0.05 (40 bns); optmal bn sze selected usng a standard F-test for nested models (Lee and Lemeux, 2010). (6) Black lne = 2nd order polynomal, ftted separately on ether sde of the zero threshold, usng the full bandwdth. (7) Dashed lnes = 95% confdence nterval. (8) See Table A.1 n Annex A for varable defntons and data sources. 33

36 Fgure 3: Captal transfers vs vote margn. a) Captal transfers b) Resdual captal transfers Regonal Transfers Regonal Transfers (Resduals) Regonal Incumbent's Bloc Margn Regonal Incumbent's Bloc Margn Notes: (1) Regonal transfers = Captal transfers from the Regonal to the Local government durng the last two years of the and muncpal terms. (2) Resdual transfers = resduals from a regresson between captal transfers and controls. (3) Black lne = 2nd order polynomal, ftted separately on ether sde of the zero threshold, usng the full bandwdth. (4) See Fgure 2. Fgure 4: Local vote share vs vote margn. b) Local vote share Incumbent's Vote Share Regonal Incumbent's Bloc Margn Notes: (1) Local vote share = % vote share of the local ncumbent party at the 2003 and 2007 electons. 3) Black lne = 2nd order polynomal, ftted separately on ether sde of the zero threshold, usng the full bandwdth. 34

37 Fgure 5: Resdual captal transfers vs. vote margn. Electon tmng & compettveness. a.1) Concurrent electons a) All electons a.2) Alternatng electons Regonal Transfers (Resduals) Regonal Incumbent's Bloc Margn Regonal Transfers (Resduals) Regonal Incumbent's Bloc Margn b) Concurrent electons b.1) Non-compettve electons b.2) Compettve electons Regonal transfers (Resduals) Regonal ncumbent's bloc margn Regonal transfers (resduals) Regonal Incumbent's bloc margn Notes: (1) See Fgure 2. (2) Resdual transfers = resduals from a regresson between captal transfers and controls. 35

38 Annex A: Data and varables. Table A.1: Defntons of varables and data sources Defnton Source Captal transfers: - from the Regonal gov. - from the Central gov. - from the Upper-Local gov. Current transfers: - from the Regonal gov. - from the Central gov. - from the Upper-Local gov. Vote share: - Mayor - Coalton Algnment (a): - Regonal-Local - Central-Local - Upper-Local-Local Incumbent s bloc seat majorty (d): - Regonal-Local - Central-Local - Upper-Local-Local Incumbent s bloc vote margn (m): - Regonal - Central - Upper-Local Captal transfers from the Regonal, Central, or Upper-Local governments per capta (tems 7.5, 7.2 & of the revenue budget) Survey of local fnances undertaken yearly by the Spansh Current transfers from the Regonal, Central, or Upper-Local governments per capta (tems 4.5, 4.2 & of the revenue budget) Mnstry of Economcs (years ) Votes for the party of the mayor and for the coalton supportng hm at the local electons, n % of votes cast Local electon statstcs (votes and seats for all the partes) and Dummy equal to one f the party of the mayor s the same as that of the presdent of the Autonomous Communty, the Central government or the Upper-Local government Dummy equal to one f the deologcal bloc of the party of the presdent of the Autonomous Communty, the Central government or the Upper-Local government has more seats n the local councl than the other deologcal bloc % of votes cast at the local electons that have to be added to (subtracted from) the deologcal bloc of the Regonal, Central or Upper-Local ncumbent to wn (lose) a majorty of seats n the local councl. partsan dentty of the mayor, provded by the Spansh Mnstry of Interor & Mnstry of Publc Admnstraton. (2003 and 2007 local electons) Vote margn computed wth the same data usng an algorthm developed for ths purposes that replcates the workngs of the d Hondt rule (see Table A.2 n Annex A) Income per capta Resdents ncome level, as estmated from objectve ndcators (e.g., cars, bank deposts, etc.) Anuaro Económco de España, La Caxa (years ) Debt burden Debt servce (captal, tem 9 of the spendng budget, + nterests, tem 3) as a share of current revenues Land area per capta Urban land area per capta, ncludng both bult on area and un-bult land plots Property tax rate Nomnal property tax rate (IBI), % on assessed property value Property value Assessed property value per capta Populaton Resdent populaton % Old % resdent populaton older than 65 years % Young % resdent populaton younger than 18 years % Immgrant % resdent populaton non-eu mmgrant % Unemployed % resdent populaton unemployed Left mayor Coalton Local party Hstorcal turnout Hstorcal vote share Mayor belongs to a left-wng bloc party Mayor governs n coalton wth other partes Party of the mayor cannot be classfed as left or rght wng % of votng age resdents votng at the local electons held snce 1979 % vote share for the deologcal bloc of the mayor at the local electons held snce 1979 Mnstry of Economcs (years ) Centro de Gestón Catastral y Cooperacón Trburara, Spansh Mnstry of Economcs (years ) Padrón de Habtantes, Natonal Insttute of Statstcs (years ) Local electon statstcs (votes and seats for all the partes) and partsan dentty of the mayor, provded by the Spansh Mnstry of Interor & Mnstry of Publc Admnstraton. (all local electons snce 1979) 36

39 Explanaton: Table A.2: Computng the vote margn. The forcng varable for our RDD s the Regonal ncumbent s bloc vote margn, computed as the rato between the mnmum number of votes needed for the deologcal bloc of the regonal ncumbent to gan/lose the majorty of seats n the local councl and the total votes cast at the local electons.the computaton of ths measure s not straghtforward and requres a consderaton of the specfc allocaton system used to assgn votes to seats, n ths case the d Hondt rule. Under ths rule the votes for each party are dvded by 1, 2, 3, 4,, N, where N s the number of seats to be assgned. The resultng quotas or comparson numbers are ranked and N seats are allocated usng ths rankng. We have developed an algebrac procedure to compute the vote margn for each of the muncpaltes n the sample 1. Our procedure works by subtractng votes from the regonal presdent s deologcal bloc f t holds a majorty at the local level, or addng votes f t does not. We make some ntal assumptons regardng the mgraton of these votes. Frst, we assume that these votes ether ) go to (come from) the abstenton or ) go to (come from) both the abstenton and the partes n the opposton bloc. The formulaton we present here s for the frst approach ) and the formula used n the second approach and the Stata code are avalable upon request. Second, we assume that the votes lost by (added to) the regonal ncumbent s bloc are allocated between the partes belongng to ths bloc proportonal to ther ntal vote share n the bloc. Below we present the formulaton used for the close electon cases 2.e., cases where the seat margn s 1 or +1. Notaton and defntons: k v I & v O k α I & α O k s I & O c I si ) = vi si I ( si + 1) = vi ( si + 1) mn c I ( s I ) = mn( I ( si )) max c I ( s I + 1) = max I ( I ( si + 1)) k ( O k k k mn c O s ), O ( so +1) O ( s O : votes for partes and k., from the regonal ncumbent s (I) and opposton s (O) blocs, respectvely. : votes for partes and k as a proporton of the votes for the bloc they belong to. s : seats for partes and k. ( : comparson number for the last seat won by party. c : comparson number for the next seat to be ganed by party. c : smallest comparson number for the last seat ganed by a party n I. c : largest comparson number for the next seat to be ganed by a party n I. max c, c ) and c ( + 1) : comparson numbers for the opposton s bloc. Formulaton: O s O If the regonal ncumbents s bloc holds a majorty n the local councl and, so, a party from the opposton bloc max has to gan a seat, ts comparson number for the next seat to be ganed, c O ( s O + 1), must be larger than the comparson number for the last seat dstrbuted to a party n the regonal ncumbent s bloc, once υ votes are subtracted from that bloc. The condton for party z n the opposton ganng a seat s: mn* max c I ( s I ) < c O ( s O + 1) [A.1] mn* where c I ( s I ) s the smallest comparson number for the last seat orgnally ganed by a party, say party x, among the partes from the regonal ncumbent s bloc once υ votes have been subtracted. z s the party that has the hghest comparson number for the next seat to be ganed among all the partes of the opposton bloc. x x x z z x Expresson [A.1] can be rewrtten as ( vi υ ) / si < vo /( so + 1), where υ are the votes subtracted from party x. 3 Under the assumpton that all the partes from the regonal ncumbent s bloc lose votes accordng to the votes orgnally cast, expresson [A.1] determnes that the total amount of votes that the regonal ncumbent s bloc has to lose to lose one seat s equal to: x = ( υ x x mn max x υ / α I ) + 1 where υ = ( c I ( si ) - co ( so + 1)) si [A.2] If the regonal ncumbent s deologcal bloc s n a mnorty n the local councl, the votes to be added to the opposton bloc for a party, say part y, n ths bloc to gan a seat are such that: mn max* c O ( s O ) < c I ( s I + 1) [A.3] max* where c I ( s I + 1) s the largest comparson number for the next seat to be ganed by party y from the regonal ncumbent s bloc, once δ votes are added to the opposton bloc. Party y s the one that orgnally has the hghest comparson number for the next seat to be ganed. Expresson [A.3] can be re-wrtten as: = ( y y y mn max y δ δ / α ) + 1 where δ = ( c ( s ) c ( s + 1))( s + 1) [A.4] I Notes: (1) A numercal example llustratng the workngs of ths algebrac procedure has been ncluded n Annex B (not for publcaton). (2) Whenever the seat margn s larger that one, the procedure we now explan s smply terated untl there s a swtch n the bloc holdng the majorty. Then, the fnal measure of the vote margn s an aggregaton of votes needed to lose (wn) all these seats. (2) Party x s such that equaton [A.1] and mn M (v M - v )/s M hold. Party x wll typcally be the party that ganed the last seat. If there s another party that ganed a seat (but not the last one) and whch accrues a greater share of votes, ths party could be the one that has to be consdered n order to guarantee that the opposton bloc gans just one seat. O O I I I 37

40 Documents de Treball de l IEB /1, De Borger, B., Pauwels, W.: "A Nash barganng soluton to models of tax and nvestment competton: tolls and nvestment n seral transport corrdors" 2010/2, Chrnko, R.; Wlson, D.: "Can Lower Tax Rates Be Bought? Busness Rent-Seekng And Tax Competton Among U.S. States" 2010/3, Esteller-Moré, A.; Rzzo, L.: "Poltcs or moblty? Evdence from us excse taxaton" 2010/4, Roehrs, S.; Stadelmann, D.: "Moblty and local ncome redstrbuton" 2010/5, Fernández Llera, R.; García Valñas, M.A.: "Effcency and eluson: both sdes of publc enterprses n Span" 2010/6, González Alegre, J.: "Fscal decentralzaton and ntergovernmental grants: the European regonal polcy and Spansh autonomous regons" 2010/7, Jamett, M.; Joans, M.: "Determnants of fscal decentralzaton: poltcal economy aspects" 2010/8, Esteller-Moré, A.; Galmarn, U.; Rzzo, L.: "Should tax bases overlap n a federaton wth lobbyng?" 2010/9, Cubel, M.: "Fscal equalzaton and poltcal conflct" 2010/10, D Paolo, A.; Raymond, J.L.; Calero, J.: "Explorng educatonal moblty n Europe" 2010/11, Adt, T.S.; Dutta, J.: "Fscal federalsm and electoral accountablty" 2010/12, Arqué Castells, P.: "Venture captal and nnovaton at the frm level" 2010/13, García-Quevedo, J.; Mas-Verdú, F.; Polo-Otero, J.: "Whch frms want PhDS? The effect of the unversty-ndustry relatonshp on the PhD labour market" 2010/14, Calabrese, S.; Epple, D.: "On the poltcal economy of tax lmts" 2010/15, Jofre-Monseny, J.: "Is agglomeraton taxable?" 2010/16, Dragu, T.; Rodden, J.: "Representaton and regonal redstrbuton n federatons" 2010/17, Borck, R; Wmbersky, M.: "Poltcal economcs of hgher educaton fnance" 2010/18, Dohse, D; Walter, S.G.: "The role of entrepreneurshp educaton and regonal context n formng entrepreneural ntentons" 2010/19, Åslund, O.; Edn, P-A.; Fredrksson, P.; Grönqvst, H.: "Peers, neghborhoods and mmgrant student achevement - Evdence from a placement polcy" 2010/20, Pelegrín, A.; Bolance, C.: "Internatonal ndustry mgraton and frm characterstcs: some evdence from the analyss of frm data" 2010/21, Koh, H.; Redel, N.: "Do governments tax agglomeraton rents?" 2010/22, Curto-Grau, M.; Herranz-Loncán, A.; Solé-Ollé, A.: "The poltcal economy of nfraestructure constructon: The Spansh Parlamentary Roads ( )" 2010/23, Bosch, N.; Espasa, M.; Mora, T.: "Ctzens control and the effcency of local publc servces" 2010/24, Ahamdanech-Zarco, I.; García-Pérez, C.; Smón, H.: "Wage nequalty n Span: A regonal perspectve" 2010/25, Folke, O.: Shades of brown and green: Party effects n proportonal electon systems 2010/26, Falck, O.; Heblch, H.; Lamel, A.; Südekum, J.: Dalects, cultural dentty and economc exchange 2010/27, Baum-Snow, N.; Pavan, R.: Understandng the cty sze wage gap 2010/28, Molloy, R.; Shan, H.: The effect of gasolne prces on household locaton 2010/29, Koethenbuerger, M.: How do local governments decde on publc polcy n fscal federalsm? Tax vs. expendture optmzaton 2010/30, Abel, J.; Dey, I.; Gabe, T.: Productvty and the densty of human captal 2010/31, Gerrtse, M.: Polcy competton and agglomeraton: a local government vew 2010/32, Hlber, C.; Lyytkänen, T.; Vermeulen, W.: Captalzaton of central government grants nto local house prces: panel data evdence from England 2010/33, Hlber, C.; Robert-Ncoud, F.: On the orgns of land use regulatons: theory and evdence from us metro areas 2010/34, Pcard, P.; Tabuch, T.: Cty wth forward and backward lnkages 2010/35, Bodenhorn, H.; Cuberes, D.: Fnancal development and cty growth: evdence from Northeastern Amercan ctes, /36, Vulovc, V.: The effect of sub-natonal borrowng control on fscal sustanablty: how to regulate? 2010/37, Flamand, S.: Interregonal transfers, group loyalty and the decentralzaton of redstrbuton 2010/38, Ahlfeldt, G.; Feddersen, A.: From perphery to core: economc adjustments to hgh speed ral 2010/39, González-Val, R.; Pueyo, F.: Frst nature vs. second nature causes: ndustry locaton and growth n the presence of an open-access renewable resource 2010/40, Bllngs, S.; Johnson, E.: A nonparametrc test for ndustral specalzaton 2010/41, Lee, S.; L, Q.: Uneven landscapes and the cty sze dstrbuton 2010/42, Ploeckl. F.: Borders, market access and urban growth; the case of Saxon towns and the Zollveren 2010/43, Hortas-Rco, M.: Urban sprawl and muncpal budgets n Span: a dynamc panel data analyss 2010/44, Koethenbuerger, M.: Electoral rules and ncentve effects of fscal transfers: evdence from Germany

41 Documents de Treball de l IEB 2010/45, Solé-Ollé, A.; Vladecans-Marsal, E.: Lobbyng, poltcal competton, and local land supply: recent evdence from Span 2010/46, Larcnese, V.; Rzzo; L.; Testa, C.: Why do small states receve more federal money? Us senate representaton and the allocaton of federal budget 2010/47, Patacchn, E.; Zenou, Y.: Neghborhood effects and parental nvolvement n the ntergeneratonal transmsson of educaton 2010/48, Nedelkoska, L.: Occupatons at rsk: explct task content and job securty 2010/49, Jofre-Monseny, J.; Marín-López, R.; Vladecans-Marsal, E.: The mechansms of agglomeraton: Evdence from the effect of nter-ndustry relatons on the locaton of new frms 2010/50, Revell, F.: Tax mx corners and other knks 2010/51, Duch-Brown, N.; Parellada-Sabata M.; Polo-Otero, J.: Economes of scale and scope of unversty research and technology transfer: a flexble mult-product approach 2010/52, Duch-Brown, N.; Vlalta M.: Can better governance ncrease unversty effcency? 2010/53, Cremer, H.; Goulão, C.: Mgraton and socal nsurance 2010/54, Mttermaer, F; Rncke, J.: Do countres compensate frms for nternatonal wage dfferentals? 2010/55, Boglacno, F; Vvarell, M.: The job creaton effect or R&D expendtures 2010/56, Pacenza, M; Turat, G.: Does fscal dscplne towards sub-natonal governments affect ctzens wellbeng? Evdence on health /1, Oppedsano, V; Turat, G.: "What are the causes of educatonal nequaltes and of ther evoluton over tme n Europe? Evdence from PISA" 2011/2, Dahlberg, M; Edmark, K; Lundqvst, H.: "Ethnc dversty and preferences for redstrbuton " 2011/3, Canova, L.; Vaglo, A.: "Why do educated mothers matter? A model of parental help 2011/4, Delgado, F.J.; Lago-Peñas, S.; Mayor, M.: On the determnants of local tax rates: new evdence from Span 2011/5, Polatto, A.; Schuett, F.: A model of musc pracy wth popularty-dependent copyng costs 2011/6, Duch, N.; García-Estévez, J.; Parellada, M.: Unverstes and regonal economc growth n Spansh regons 2011/7, Duch, N.; García-Estévez, J.: Do unverstes affect frms locaton decsons? Evdence from Span 2011/8, Dahlberg, M.; Mörk, E.: Is there an electon cycle n publc employment? Separatng tme effects from electon year effects 2011/9, Costas-Pérez, E.; Solé-Ollé, A.; Sorrbas-Navarro, P.: Corrupton scandals, press reportng, and accountablty. Evdence from Spansh mayors 2011/10, Cho, A.; Calero, J.; Escardíbul, J.O.: Hell to touch the sky? prvate tutorng and academc achevement n Korea 2011/11, Mra Godnho, M.; Cartaxo, R.: Unversty patentng, lcensng and technology transfer: how organzatonal context and avalable resources determne performance 2011/12, Duch-Brown, N.; García-Quevedo, J.; Montolo, D.: The lnk between publc support and prvate R&D effort: What s the optmal subsdy? 2011/13, Breullé, M.L.; Duran-Vgneron, P.; Samson, A.L.: To assemble to resemble? A study of tax dspartes among French muncpaltes 2011/14, McCann, P.; Ortega-Arglés, R.: Smart specalsaton, regonal growth and applcatons to EU coheson polcy 2011/15, Montolo, D.; Trllas, F.: Regulatory federalsm and ndustral polcy n broadband telecommuncatons 2011/16, Pelegrín, A.; Bolancé, C.: Offshorng and company characterstcs: some evdence from the analyss of Spansh frm data 2011/17, Ln, C.: Gve me your wred and your hghly sklled: measurng the mpact of mmgraton polcy on employers and shareholders 2011/18, Banchn, L.; Revell, F.: Green poltes: urban envronmental performance and government popularty 2011/19, López Real, J.: Famly reunfcaton or pont-based mmgraton system? The case of the U.S. and Mexco 2011/20, Boglacno, F.; Pva, M.; Vvarell, M.: The mpact of R&D on employment n Europe: a frm-level analyss 2011/21, Tonello, M.: Mechansms of peer nteractons between natve and non-natve students: rejecton or ntegraton? 2011/22, García-Quevedo, J.; Mas-Verdú, F.; Montolo, D.: What type of nnovatve frms acqure knowledge ntensve servces and from whch supplers?

42 Documents de Treball de l IEB 2011/23, Banal-Estañol, A.; Macho-Stadler, I.; Pérez-Castrllo, D.: Research output from unversty-ndustry collaboratve projects 2011/24, Lgthart, J.E.; Van Oudheusden, P.: In government we trust: the role of fscal decentralzaton 2011/25, Mongran, S.; Wlson, J.D.: Tax competton wth heterogeneous captal moblty 2011/26, Caruso, R.; Costa, J.; Rccut, R.: The probablty of mltary rule n Afrca, /27, Solé-Ollé, A.; Vladecans-Marsal, E.: Local spendng and the housng boom 2011/28, Smón, H.; Ramos, R.; Sanromá, E.: Occupatonal moblty of mmgrants n a low sklled economy. The Spansh case 2011/29, Polatto, A.; Trotn, G.: Optmal tax enforcement under prospect theory 2011/30, Montolo, D; Polatto, A.: Fnancng publc educaton when altrustc agents have retrement concerns 2011/31, García-Quevedo, J.; Pellegrno, G.; Vvarell, M.: The determnants of YICs R&D actvty 2011/32, Goodspeed, T.J.: Corrupton, accountablty, and decentralzaton: theory and evdence from Mexco 2011/33, Pedraja, F.; Cordero, J.M.: Analyss of alternatve proposals to reform the Spansh ntergovernmental transfer system for muncpaltes 2011/34, Jofre-Monseny, J.; Sorrbas-Navarro, P.; Vázquez-Grenno, J.: Welfare spendng and ethnc heterogenety: evdence from a massve mmgraton wave 2011/35, Lyytkänen, T.: Tax competton among local governments: evdence from a property tax reform n Fnland 2011/36, Brülhart, M.; Schmdheny, K.: Estmatng the Rvalness of State-Level Inward FDI 2011/37, García-Pérez, J.I.; Hdalgo-Hdalgo, M.; Robles-Zurta, J.A.: Does grade retenton affect achevement? Some evdence from Psa 2011/38, Boffa, f.; Panzar. J.: Bottleneck co-ownershp as a regulatory alternatve 2011/39, González-Val, R.; Olmo, J.: Growth n a cross-secton of ctes: locaton, ncreasng returns or random growth? 2011/40, Anes, V.; De Donder, P.: Votng under the threat of secesson: accommodaton vs. represson 2011/41, D Petro, G.; Mora, T.: The effect of the l Aqula earthquake on labour market outcomes 2011/42, Brueckner, J.K.; Neumark, D.: Beaches, sunshne, and publc-sector pay: theory and evdence on amentes and rent extracton by government workers 2011/43, Cortés, D.: Decentralzaton of government and contractng wth the prvate sector 2011/44, Turat, G.; Montolo, D.; Pacenza, M.: Fscal decentralsaton, prvate school fundng, and students achevements. A tale from two Roman catholc countres /1, Montolo, D.; Trujllo, E.: "What drves nvestment n telecommuncatons? The role of regulaton, frms nternatonalzaton and market knowledge" 2012/2, Gesen, K.; Suedekum, J.: "The sze dstrbuton across all ctes : a unfyng approach" 2012/3, Foremny, D.; Redel, N.: "Busness taxes and the electoral cycle" 2012/4, García-Estévez, J.; Duch-Brown, N.: "Student graduaton: to what extent does unversty expendture matter?" 2012/5, Durán-Cabré, J.M.; Esteller-Moré, A.; Salvador, L.: "Emprcal evdence on horzontal competton n tax enforcement" 2012/6, Pckerng, A.C.; Rockey, J.: "Ideology and the growth of US state government" 2012/7, Vergoln, L.; Zann, N.: "How does ad matter? The effect of fnancal ad on unversty enrolment decsons" 2012/8, Backus, P.: "Gbrat s law and legacy for non-proft organsatons: a non-parametrc analyss" 2012/9, Jofre-Monseny, J.; Marín-López, R.; Vladecans-Marsal, E.: "What underles localzaton and urbanzaton economes? Evdence from the locaton of new frms" 2012/10, Mantovan, A.; Vandekerckhove, J.: "The strategc nterplay between bundlng and mergng n complementary markets" 2012/11, Garca-López, M.A.: "Urban spatal structure, suburbanzaton and transportaton n Barcelona" 2012/12, Revell, F.: "Busness taxaton and economc performance n herarchcal government structures" 2012/13, Arqué-Castells, P.; Mohnen, P.: "Sunk costs, extensve R&D subsdes and permanent nducement effects" 2012/14, Boffa, F.; Polatto, A.; Ponzetto, G.: "Centralzaton and accountablty: theory and evdence from the Clean Ar Act" 2012/15, Cheshre, P.C.; Hlber, C.A.L.; Kaplans, I.: "Land use regulaton and productvty land matters: evdence from a UK supermarket chan"

43 Documents de Treball de l IEB 2012/16, Cho, A.; Calero, J.: "The contrbuton of the dsabled to the attanment of the Europe 2020 strategy headlne targets" 2012/17, Slva, J.I.; Vázquez-Grenno, J.: "The ns and outs of unemployment n a two-ter labor market" 2012/18, González-Val, R.; Lanaspa, L.; Sanz, F.: "New evdence on Gbrat s law for ctes" 2012/19, Vázquez-Grenno, J.: "Job search methods n tmes of crss: natve and mmgrant strateges n Span" 2012/20, Lessmann, C.: "Regonal nequalty and decentralzaton an emprcal analyss" 2012/21, Nuevo-Chquero, A.: "Trends n shotgun marrages: the pll, the wll or the cost?" 2012/22, Pl Damm, A.: "Neghborhood qualty and labor market outcomes: evdence from quas-random neghborhood assgnment of mmgrants" 2012/23, Ploeckl, F.: "Space, settlements, towns: the nfluence of geography and market access on settlement dstrbuton and urbanzaton" 2012/24, Algan, Y.; Hémet, C.; Latn, D.: "Dversty and local publc goods: a natural experment wth exogenous resdental allocaton" 2012/25, Martnez, D.; Sjögren, T.: "Vertcal externaltes wth lump-sum taxes: how much dfference does unemployment make?" 2012/26, Cubel, M.; Sanchez-Pages, S.: "The effect of wthn-group nequalty n a conflct aganst a untary threat" 2012/27, Andn, M.; De Blaso, G.; Duranton, G.; Strange, W.C.: "Marshallan labor market poolng: evdence from Italy" 2012/28, Solé-Ollé, A.; Vladecans-Marsal, E.: "Do poltcal partes matter for local land use polces?" 2012/29, Buonanno, P.; Durante, R.; Prarolo, G.; Vann, P.: "Poor nsttutons, rch mnes: resource curse and the orgns of the Sclan mafa" 2012/30, Anghel, B.; Cabrales, A.; Carro, J.M.: "Evaluatng a blngual educaton program n Span: the mpact beyond foregn language learnng"

44 Fscal Federalsm