Insttut für Halle Insttute for Economc Research Wrtschaftsforschung Halle Sharng Competences: The Impact of Local Insttutonal Settngs on Voter Turnout Claus Mchelsen, Peter Bönsch, Martn T. W. Rosenfeld September 2010 No. 21 IWH-Dskussonspapere IWH Dscusson Papers
Sharng Competences: The Impact of Local Insttutonal Settngs on Voter Turnout Claus Mchelsen, Peter Bönsch, Martn T. W. Rosenfeld September 2010 No. 21
IWH Authors: Claus Mchelsen Department of Urban Economcs cmn@wh-halle.de Phone: +49 (0) 345 7753-837 Peter Bönsch Martn- Luther- Unversty Halle-Wttenberg peter.boensch@ww.un-halle.de Martn T. W. Rosenfeld Department of Urban Economcs mrd@wh-halle.de The responsblty for dscusson papers les solely wth the ndvdual authors. The vews expressed heren do not necessarly represent those of the IWH. The papers represent prelmnary work and are crculated to encourage dscusson wth the author. Ctaton of the dscusson papers should account for ther provsonal character; a revsed verson may be avalable drectly from the author. Comments and suggestons on the methods and results presented are welcome. IWH Dscusson Papers are ndexed n RePEc-Econpapers and n ECONIS. Edtor: HALLE INSTITUTE FOR ECONOMIC RESEARCH IWH Prof Dr Dr h. c. Ulrch Blum (Presdent), Dr Hubert Gabrsch (Research Drector) The IWH s a member of the Lebnz Assocaton. Address: Klene Maerkerstrasse 8, D-06108 Halle (Saale), Germany Postal Address: P.O. Box 11 03 61, D-06017 Halle (Saale), Germany Phone: +49 (0) 345 7753-60 Fax: +49 (0) 345 7753-820 Internet: http:www.wh-halle.de 2
IWH Sharng Competences: The Impact of Local Insttutonal Settngs on Voter Turnout Insttutons are common predctors of voter turnout. Most research n ths feld focuses on cross-country comparsons of votng systems, lke the mpact of compulsory votng or regstraton systems. Fewer efforts have been devoted to understand the role of local nsttutons and ther mpact on poltcal partcpaton. Especally the mpact of dvded competences n relaton to publc good provson and ts mpact on voter turnout has been wdely gnored. In the present paper, we analyze the effects of dfferent nsttutonal settngs for nter-muncpal cooperaton on voter turnout. We use data from local electons n Germany, held n 2003 and 2004. Overall, we analyze aggregate voter turnout of 1661 muncpaltes and fnd strong evdence for our hypothess that local nsttutonal settngs are nfluental n ths context. Further, our results ndcate that the better competences correspond to the spatal dmenson of local publc goods, the hgher should be the voter turnout. Keywords: voter turnout, local nsttutons, nter-muncpal cooperaton JEL classfcaton: D70, D72, H11, H40 3
IWH Der Enfluss lokaler Insttutonen auf de Betelgung an Gemenderatswahlen Set geraumer Zet gbt es de Tendenz, kommunale Insttutonen zu bündeln um damt Größenvortele n der Lestungserstellung zu errechen. In Deutschland haben sch auf der Gemendeebene unterschedlche Modelle durchgesetzt, de sch zwschen den Extremen der gemensamen Aufgabenwahrnehmung von Gemenden oder vollständgen Gemendefusonen bewegen. Das Dskussonspaper geht der bsher weng untersuchten Frage nach, we sch des auf de Betelgung an lokalen Wahlen auswrkt. Im Zusammenhang mt Insttutonen werden n der vorlegenden Lteratur überwegend Auswrkungen des Wahlrechts länderverglechend dskutert. Vor allem der Enfluss ener Wahlpflcht oder de Notwendgketen der Regstrerung zu ener Abstmmung waren Gegenstand der Untersuchungen. De her aufgeworfene Fragestellung geht auf lokale nsttutonelle Aspekte en. Im Spezellen werden dabe unterschedlche Formen der nterkommunalen Zusammenarbet und der Vertelung von Entschedungskompetenzen betrachtet. Aufgrund der velfältgen Formen der kommunalen Zusammenarbet n der Beretstellung öffentlcher Güter, snd dese Fragen vor allem für de Mtgledsstaaten der Europäschen Unon relevant. Betrachtet werden n dem vorlegenden Paper deutsche Gemenderatswahlen n den Jahren 2003 und 2004. Der Enfluss unterschedlcher Gemendetypen wrd modellert und ökonometrsch untersucht. Dabe fnden sch deutlche Anzechen für enen Enfluss der nsttutonellen Ausgestaltung nterkommunaler Kooperatonen auf de lokale Wahlbetelgung. Darüber hnaus zegt de emprsche Analyse, dass de Wahlbetelgung dort höher ausfällt, wo de Nutzneßer öffentlcher Lestungen den Wahlberechtgten entsprechen und dese de entschedenden Gremen drekt wählen können. De nsttutonelle Ausgestaltung kommunaler Zusammenarbet n Verbandsgemenden nach Rhenland-Pfälzschem Vorbld schent desen Erfordernssen am ehesten zu entsprechen. Schlagworte: Wahlbetelgung, lokale Insttutonen, Gemendeneuordnung JEL-Klassfkaton: D70, D72, H11, H40 4
IWH 1 Introducton Snce Downs (1957) ntroduced the economc theory of democracy, enormous scholarly efforts have been made to understand what affects voter turnout. Untl today, an extensve lterature emphasses the mportance of socoeconomc aspects, poltcal competton and nsttutonal features on aggregate turnout (for revews of the theoretcal sde see Dhllon and Peralta 2002, Blas 2006, Geys 2006a; for a comprehensve revew of the emprcal lterature see Geys 2006b). The work on nsttutonal settngs manly focuses on cross-country comparsons of the votng system, for example the mpact of compulsory votng or regstraton systems. Fewer efforts have been devoted to understand the role of local nsttutons: most studes examne the mpact of nonpartsan electon or the splt of local electons from the tradtonal electoral calendar (for a revew see Caren 2007). Partcularly for the US exsts another strand of lterature, whch focuses on the mpact of the councl-manager form of government and the outsourcng of local publc goods to prvate provders on turnout. The common fndng s that erodng poltcal power of elected muncpal councls negatvely affects voter turnout (see Wood 2002, Hajnal and Lews 2003). The present paper focuses on other local nsttutonal aspects that, as far as the authors are aware, have not been nvestgated by scholars so far. Varous forms of ntermuncpal cooperaton, manly represented n Western European Countres, have so far been neglected n the lterature. Drven by the dea to realze economes of scale and to nternalze spll-over effects, muncpaltes often merge or cooperate n the provson of local publc goods. Ths affects procedures of poltcal decson makng and ther poltcal legtmatng whch can, as we argue, reduce or ncrease complexty n the votng decson. How these arrangements nfluence voter turnout, depends, n our vew, on how decson makng and electoral processes correspond to the scope of the provded publc servces. We argue that Mancur Olson s (1965) basc dea of an optmal provson of publc goods and, n ths sense, an optmal desgn of local nsttutons s drectly lnked to turnout: the better the electorate corresponds to the collectve of fnancers, the less complex should be the votng decson and, the hgher should be voter turnout. Of course, ballots on the provson of every sngle publc good are not desrable snce ths would ncrease the complexty of poltcal decson makng and cause unnecessary frustraton among voters. However, we argue that when desgnng new local nsttutonal arrangements, ther doubled functon as complexty reducng enttes needs to be consdered. In ths paper, we am to analyze the above outlned dea emprcally. We test our hypothess by usng data from local electons n Germany, takng advantage of the heterogenety of the German muncpal structure and assessng the nfluence of three types of nsttutonal arrangements (.e. centralzed, sem-centralzed and decentralzed munc- 5
IWH paltes) on voter turnout. Therefore we use cross sectonal data for local electons n 1661 muncpaltes and cooperatve muncpaltes from the years 2003 and 2004. The paper s organzed as follows. The next secton provdes a short dscusson of the exstng lterature on the mpact of local nsttutons on turnout. Wthn ths secton, we theoretcally dscuss the relatonshp between concentraton of competences and poltcal legtmatng of nsttutons that motvate our hypothess. Secton three contans an overvew on Germany s muncpal consttuton. We descrbe several nsttutonal features to derve at the types of muncpaltes whch should be compared. Secton four states the tested hypothess followed by the emprcal part of the paper n secton 5. After a short descrpton of the respectve methodology, we ntroduce our model and provde tables of descrptve statstcs. Fnally, we present and dscuss our results. 2 What do we know about voter turnout so far? Theoretcally, most of the present research refers to the economc theory of democracy ntroduced by Downs (1957). Downs argued that ndvduals partcpate n polls when benefts from votng exceed costs. Benefts from votng are defned as the dfference between utlty of polcy alternatves multpled by the ndvdual probablty to affect aggregated votng outcome. Snce the probablty to have an mpact on the result of an electon s very low and declnes wth the sze of the electorate, large scale abstenton n polls s the predcted result of Downs consderatons. Addtonally, t s argued that snce the utlty of partcpaton s very small, any costs of votng larger than zero must keep ndvduals away from the votng box. In realty, however, n most electons voter turnout s remarkably hgh, whch led researchers to conclude that there exsts a paradox of votng (see Rker and Ordeshook 1968, p. 31). Ths fndng has been adopted by scholars as a startng pont to add further explanatons to the ratonal choce model (for a bref revew see Geys 2006a) and to fnd alternatve explanatons for voter turnout. There s a large strand of lterature dealng wth ndvduals decson to vote usng survey data (for a revew of emprcal work see Prewtt and Ne 2009). Others are amng to explan aggregate voter turnout between states, countes or on the local level. For example, researchers found that natonal polls attract more voters than regonal or local electons (see Marsh 1998), whch s explaned by the model of Second-Order Electons (see Norrs 1997). Further, researchers attempted to ncorporate nsttutonal features nto people s votng ratonalty (see e.g. Jackman 1987, Merrfeld 1993). As we draw on the mpact of local nsttutons on turnout, we concentrate our theoretcal dscusson on these aspects. For other commonly tested varables we take advantage of prevous studes and refer the reader to the revew of Geys (2006b). Scholars have extensvely dscussed the mpact of soco-economc aspects on voter turnout focusng on the mpact of populaton sze, populaton concentraton, an electorate s age structure, ncome or homogenety of populaton. Secondly, poltcal competton has been proven 6
IWH to be nfluental on turnout. Here, for example, the closeness of an electon or the competton between partes has beenthe focus of scholary attenton. Thrdly, nsttutonal aspects have been wdely dscussed. The followng secton outlnes theoretcal approaches n ths context and provdes an extenson of the dscusson on the role of unversus multcameralsm poltcal systems on a local level. 2.1 The role of nsttutons n local electons The relatonshp between nsttutons and local voter turnout has only been selectvely appled n the lterature. Most of the studes avalable focus on the structure of government and electoral procedures (for a revew see Caren 2007). Studes on electoral procedures or the votng system manly address aspects lke nonpartsan electons (see e.g. Karng and Walter 1983, Schaffner, Wrght and Streb 2001), the tmng of an electon (Hajnal and Lews 2003), regstraton procedures (see e.g. Brans and Grofman 1999), votng dstrct szes (see Bullock 1990) or term lmts (see Copeland 1997). In the structural context, an often dscussed hypothess reles on the work of Jackman (1987), who ntroduced the argument of poltcal power and ts dvson between several chambers as a determnant of voter turnout. The argument of poltcal power s mostly dscussed n the context of a formal reallocaton of responsbltes between poltcal nsttutons lke strengthenng a cty manager n dsfavor of the elected muncpal mayor or the number of poltcal nsttutons whch are nvolved n decson makng (see e.g. Merrfeld 1993). In ths case, Jackman argues that an nsttutonal settng that concentrates poltcal power on one legslature leads to hgher voter turnout than a settng that spreads responsbltes between several chambers (see Jackman 1987 p. 408, Blas 2006). Some scholars found supportng evdence, showng that weakenng the power of poltcal decson makers leads to declnng partcpaton n polls (see e.g. Can et al. 2001, Wood 2002) Others on the other hand, dd not succeed n verfyng Jackman s hypothess (Radclff and Davs 2000, Pérez-Lñán 2001). In ths context we argue, that the twofold emprcal results may be explaned by ntroducng a further aspect to the dscusson. In partcular, we beleve that the complexty of a votng decson s nfluenced by the nsttutonal settng, especally by the cameralstc system of a jursdcton. As already ntroduced by Mancur Olson (1965), an optmal allocaton of local publc goods occurs when the collectve of benefcares corresponds to the collectve of fnancers and decson makers. Consequently, to fulfll ths optmalty restrcton one had to hold polls for nearly every sngle publc good, whch would allow voters to clearly dstngush between alternatves and to state ther preferences. Although ths would clearly add transparency to the votng decson, t also fosters complexty n poltcal decson-makng. It can be argued that jursdcton sze should therefore be optmzed wth respect to the spatal dmenson of tasks and that the scope of nsttutonal competences should be optmzed respectvely. In realty ths restrcton s, however, commonly volated whch may lead to a loss of poltcal partcpaton (e.g. voter turnout) or a loss of effcency n poltcal decson makng. 7
IWH On the local level, many countres therefore allow for nter-muncpal cooperaton (see Kelly 2007), where muncpaltes can provde publc goods and servces jontly. In ths case we argue that dfferent nsttutonal settngs, n partcular the (de)concentraton of competences, can reduce or foster complexty n decson-makng and therefore drectly nfluence voter turnout.1 We brefly dscuss these arguments n the followng paragraphs. For smplcty we use the followng assumptons to theoretcally nvestgate the effects of a concentraton of competences on the muncpal or the jont level and.e. the effects of dfferent nsttutonal settng on voter turnout: a) Two muncpaltes (A and B) produce two bundles of goods: x s a bundle of locally dstnct goods whle the bundle y has to be provded jontly. b) Preferences n A and B dffer concernng both, the bundles of good x and y. c) The competences to decde about how and how much of the goods are provded are concentrated on the legslatve (the elected muncpal councl and mayor). d) There s no asymmetry of poltcal power between A and B (e.g. n the sze of the electorates). Case 1: Muncpaltes reman ndependent - Good y s provded n A, whle B s ncorporated n fnancng wthout any competence to decde about the provson of good y In ths very smple case, nsttutons n Muncpalty B are less powerful than n A. Followng Jackman (1987), ths should decrease turnout n B, whle output n A should ncrease. Ths fndng s consstent wth emprcal research of Hajnal and Lews (2003), who nvestgated the effect of outsourcng of muncpal tasks on turnout. They concludthat an eroson of poltcal power, measured by the number of servces provded by cty staff, negatvely nfluences turnout (see Hajnal and Lews 2003, p. 657). Case 2: Muncpaltes reman ndependent and decde n consensus about the provson of good y In ths case, electons are held separately n A and B. Poltcal actors optmze ther offers wth respect to the local preferences to maxmze electoral outcome. For elgble voters, the ratonalty to vote for a bundle of goods provded locally remans unaffected. But the prncple of consensus between the muncpaltes for the provson of goods y adds uncertanty to the votng decson. Whatever the locally chosen level of y s gong to be, the actual outcome dffers when preferences between the muncpaltes vary. The 1 An approach to ncorporate nformaton nto the Ratonal Choce framework s offered by Matsusaka (1995). He concludes, that nformaton tself does not necessarly foster turnout. Instead he argues that confdence about the votng decson s a good predctor for turnout. One can understand complexty as one determnant of confdence. 8
IWH probablty to acheve the desred level of y can be seen as a decreasng functon of dfferng preferences. Further when allowng more than two muncpaltes to partcpate n the cooperatve producton of the good, t also reflects a decreasng functon of the number of jursdctons nvolved. Relaxng the assumpton of equally dstrbuted power between muncpaltes can also negatvely affect voters expectatons of outcome n less powerful muncpaltes. Compared to the stuaton wthout any cooperaton, these arguments suggest that ndvduals face a sgnfcantly more complex votng decson. Case 3: Muncpaltes merge all nsttutons, ncludng competences and polls When nsttutons and competences are completely merged, outcome-uncertanty related to the amount of y (due to consensual decson makng) dsappears. Furthermore, f the number of poltcal stakeholders remans constant (e.g. when there s a fxed number of partes actve n A and B and nonpartsan votng s excluded), complexty n the votng decson s reduced compared to case two.2 Poltcal stakeholders optmze ther offers concernng the bundle y wth respect to the aggregated preferences of A and B. But ncreasng complexty for the decson on the bundle of goods x can be assumed. When fnancal resources of the muncpalty are lmted, a tradeoff between x A and x B occurs, whch reflects locally dstnct bundles n A and B. Poltcal offers have to consder these restrctons and have to make an offer on y, x A and x B, whch may lack n spatal accuracy. Further, voters n A have to decde about locally dstnct goods n B (x B ) and vce versa nstead of decdng about two bundles of goods, poltcal offers contan a thrd bundle whch negatvely nfluences transparency of votng alternatves. Ths clearly adds complexty and therefore negatvely affects votng turnout. Compared wth case 2, t s uncertan whch nsttutonal arrangement leads to larger voter turnout snce the strength of the opposng effects of ncreasng and reducng complexty are unknown. Case 4: Muncpaltes merge competences and polls for good y and preserve ndependency concernng the provson of good x If competences are dvded between one jont muncpal councl for A+B and two others n A and B (correspondng to the types of goods x and y), electons are smultaneously held and decsons are ndependently, the advantages of concentraton (for good y) and de-centralzaton (good x) concernng turnout can be preserved. Voters can decde whch amount of x should be provded on the local level wthout consderng the level of publc servces n other muncpaltes. Moreover, decsons on y are made jontly reducng uncertanty about the actual outcome. Addtonally, voters are able to splt ther vote between partes, when partcpatng n both electons. Ths can enable voters to maxmze utlty compared to the other nsttutonal settngs (Case 1-3). To summarze, one can argue, that voter turnout s a decreasng functon of complexty of the votng decson. Complexty s determned by the nsttutonal arrangement of a 2 The effects of Multpartysm are already addressed by Downs (1957), p. 155 and are brefly dscussed by Jackman (1987), p. 408. 9
IWH muncpalty and thus largely depends on the degree of concentraton of competences on dfferent muncpal levels. For locally dstnct goods (x) the relatonshp between the concentraton of competences on the local level and complexty can be assumed to be negatve: the more competences are concentrated on the local level, the less complex s the decson n regard to the bundle of goods x. For the bundle of goods that are locally ndstnct (y), ths relatonshp can be assumed to be postve as complexty ncreases the more competences are concentrated on the jont level. The jont complexty curve (the sum of the complexty curve x and the complexty curve y) can be approxmated as an nverted u-shaped. Turnout s maxmzed at the mnmum of the jont complexty curve. A stylzed graphcal soluton s presented n fgure 1 where t s assumed that the effects of concentratng competences on the local or the jont level are equal. Fgure 1: Effects of concentraton of competences on complexty of votng decsons and turnout Source: Complaton by the authors. When the assumptons hold, one can expect nsttutonal settngs to be nfluental on voter turnout n a way that has not been dscussed n the lterature so far. Further, one can argue that nsttutonal settngs that concentrate competences accordng to the types of local publc goods mnmze complexty of the votng decson and therefore maxmze voter turnout. 10
IWH Fgure 2: Results for a domnatng effect of concentraton on the local level Source: Complaton by the authors. In fgure 2 t s assumed that the advantages of concentratng competences on the local level domnate the advantages for the provson of y on the jont level. In ths case, turnout n muncpaltes descrbed n case 2 should be larger than turnout n muncpaltes descrbed n case 3. Theoretcally, t s uncertan whch effect domnates the other and needs to be scrutnzed emprcally. 3 The Insttutonal Settng n the Feld of Local Government n Germany To test the above stated relatons emprcally, the authors chose Germany s muncpal structure, as ts nsttutonal arrangements cover the descrbed features of concentratng competences on dfferent local levels. Further, other nsttutonal features whch are mentoned to be nfluental n the lterature, lke the votng system, nonpartsan votng, regstraton requrements, term lmts and others are n ther core dentcal between German muncpaltes. Thus, when these nsttutonal effects are neglgble for Germany, we can observe the pure effect of the concentraton of competences on the complexty of the votng between muncpal levels and therefore the effect of nsttutonal settngs on voter turnout. In the followng we ntroduce the types of muncpaltes that are observable n Germany. 11
IWH 3.1 Germany s federal organzaton Legslatve competences n Germany are federally organzed and dvded between natonal and federal-state level. From a legal perspectve, local authortes are understood as self-regulatory bodes wthn the federal states admnstraton and whch ndrectly fulfl federal state admnstratve tasks. Nevertheless local admnstratons can act autonomously, guaranteed by the second paragraph of Artcle 28 of Germany s Consttuton (Art. 28, Abs. 2 GG). Ths means (followng the prncple of of subsdarty) that the competences to provde publc goods and servces are concentrated on the local level unless legal regulatons delegate the provson to hgher admnstratve bodes.3 The local self-admnstraton s organzed n countes ( Krese ), free towns ( kresfree Städte ), muncpaltes ( Gemenden ) or cooperatve muncpaltes ( Verwaltungsgemenschaften ) (see Fgure 3). The nsttutonal settng and scope of the provson of publc goods and servces are determned n the federal states legal framework, whch dffers sgnfcantly between states. Fgure 3: Vertcal separaton of Powers n Germany Source: Complaton by the authors. 3.2 Muncpal Tasks and Competences On the local level, there are three types of publc goods and servces provded by muncpaltes:4 a) Oblgatory tasks of the transferred sphere of responsbltes nclude the publc goods and servces, whch are provded by muncpaltes on behalf of the federal state. Muncpaltes act as f they were part of the federal states admnstraton. Muncpaltes cannot decde f, how much or how a certan good s provded. Local au- 3 For a more detaled descrpton of Germany s federal system see for example Behl (1994). 4 A detaled descrpton of muncpal tasks s gven by Zmmermann (1999), pp. 112 ff. 12
IWH thortes can only decde about the admnstratve effort, as long as the legally defned standards are fulflled. b) Oblgatory tasks of the own sphere of responsbltes nclude the publc goods and servces, whch have to be provded by muncpaltes, where Federal states legal settngs defne mnmum standards of the provson. Muncpaltes can furthermore decde to provde adtonal servces and how the local admnstraton should provderespectve goods. For example, muncpaltes are responsble for provdng fre departments, sewerage systems and urban land-use plannng. c) Voluntary tasks of the own sphere of responsbltes nclude the publc goods and servces, whch can be provded by the local admnstraton. Local authortes decde f, how and how much of a certan good s provded. For example, muncpaltes can mantan sports or cultural facltes lke orchestras, theatres or stada. For provdng goods and servces, muncpaltes have a substantal range of competences. Muncpal councls have the rght to determne the level of local taxes and to ssue local statutes. Further, muncpaltes have an own admnstraton to provde the respectve goods and servces. 3.3 Insttutonal arrangements on the muncpal level In Germany the federal states legslatons decde on local government nsttutons. As a consequence, several muncpal types exst n the states. The LAU1 (local authorty level) classfcaton by eurostat contans seven terms for German muncpaltes, 5 whch dffer n ther organzaton, ther competences and slghtly n ther votng system. Nevertheless the man features of local government are qute smlar throughout Germany. In some states, muncpaltes are dvded nto sub-unts. Therefore, local authortes can be classfed by ther dstrbuton of competences whch can be concentrated at one level for all member unts of a muncpalty or whch can be locally shared out to the member unts. Thus, muncpal nsttutonal arrangements fall roughly nto three categores. 6 a) Muncpaltes wth competences and nsttutons concentrated on the jont level Type A Most commonly, muncpaltes fulfll ther admnstratve tasks n own responsblty. So called kresangehörge Gemenden, whch are not member of cooperatve muncpaltes, decde (wthn the federal state s legal framework) on ther own about f, how and how much of a certan good or servce they want to provde. Decsons are made n cty councls and are mplemented by the local admnstraton. Mayors have a dual poston: they act as both, hghest representatve of a muncpalty and head of the admn- 5 The data of the European Statstcal Offce contans the terms: Krchspelslandgemende, Amt, Samtgemende, Verbandsgemende, Verwaltungsverband, Verwaltungsgemenschaft and kresangehörge Gemende. See EUROSTAT. 6 For a detaled comparson of muncpal types see Rosenfeld et al. 2007, pp. 75 ff. 13
IWH straton. The cty councl and the mayor are elected drectly by the ctzens of the respectve muncpalty. Local electons ( Kommunalwahlen ) are held every fve or sx years. In the past, there was a tendency to ncrease the sze of muncpaltes to acheve economes of scale n the provson of local publc goods and servces. Generally, federal state law defnes a mnmum sze for local authortes.7 To acheve ths, formerly ndependent muncpaltes can merge to larger muncpalty (so called Enhetsgemende ) wth one cty councl, mayor and admnstraton. In many cases, a second body of poltcal representaton s ntroduced; so called Ortschaftsräte are elected smultaneously to the cty councl and delberate on local affars of the member unts of the centralzed muncpalty. Yet, these bodes have no competences to decde about the provson of goods they are purely mplemented as advsory boards. Centralzed muncpaltes can be found n every federal state of Germany, except for the cty states, whch are muncpaltes and states at the same tme. b) Muncpaltes wth competences and nsttutons concentrated on the local level Type B A second type of local governmental arrangements s the locally concentrated muncpalty, whch can be found n nne federal states. LAU1 classfcatons are Amtsgemende (Schleswg-Holsten, Mecklenburg-Western Pomerana, Brandenburg), Krchlandspelgemende (Schleswg-Holsten), Verwaltungsgemenschaft (Bavara, Saxony- Anhalt, Thurnga) or "Verwaltungsverband (Baden-Wuerttemberg, Saxony). In contrast to Type A, the member unts of cooperatves (or admnstratve collectvtes) reman legally ndependent although wde ranges of competences are assgned to a centralzed level. These competences are executed n borrowed responsblty. The man characterstc of a Type B muncpalty s the double-layered decson process. The frst stage of decson-makng s located on member unts (the local) level where the local councl decdes, whch of the oblgatory and voluntary tasks of the own sphere of responsbltes are transferred to the centralzed level. Furthermore the local councl defnes the amount (how much) of the local publc good should be provded. Regularly, the oblgatory tasks of the transferred sphere of responsbltes are completely transferred to the centralzed level. 7 The sze and other nsttutonal features are regulated n the Gemendeordnungen or Kommunalverfassungen of the federal states. For the regulaton of the mnmum sze see for example Kommunalverfassung des Landes Brandenburg (BbgKVerf) 133. 14
IWH Table 1: Insttutons and Competences on the muncpal level an Overvew Type A Muncpaltes Jont Level Local Level electon electon electon centralzed (how) how much & how how much & how Baden-Wuerttemberg Bavara Brandenburg Hesse Lower Saxony Mecklenburg-Western Pomerana North Rhne-Westphala Rhneland-Palatnate Saarland Saxony Saxony-Anhalt Schleswg-Holsten Thurnga Type C Muncpaltes Jont Level Local Level electon electon Insttutons and democratc legtmaton Cty Councl Local Representatve Mayor Admnstratve Councl Admnstraton electon electon centralzed (how) how much & how how much & how how much & how how much & how Rhneland-Palatnate Lower Saxony Type B Muncpaltes Jont Level Local Level electon electon Members delegated by cty councl sem-centralzed (how) how how much & how Competences to decde about... how how much & how Baden-Wuerttemberg Bavara Brandenburg Mecklenburg-Western Pomerana Saxony Saxony-Anhalt Schleswg-Holsten Thurnga Insttutons Oblgatory tasks of transferred sphere of responsbltes Oblgatory tasks of own sphere of responsbltes Voluntary tasks of own sphere of responsbltes Muncpal Tasks () ndcates no presence of the respectve feature Source: Complaton by the authors. 15
IWH On the second stage, the jont admnstratve councl decdes on how the goods and servces are produced and coordnates comprehensve muncpal tasks for poltcal decsons n the local councls. In practce, consensus about the level of publc servces (how much) s found on the jont level. Dependng on the scope of the transferred tasks, the admnstraton s concentrated and s normally located n the largest member unt of the admnstratve collectvely. Other member unts stll have the competence to mantan own admnstratve facltes for tasks that are not transferred to the central level. In addton, local councls have a veto rght, when decsons about how a good should be provded do not meet the local requrements or preferences. The members of the jont admnstratve councl are the mayors of the member unts and addtonally (dependng on the sze of a muncpalty) deputes, elected by cty councls. Cty councls and mayors are drectly elected every fve or sx years. On the jont level, no separate electons are held. c) Muncpaltes wth competences and nsttutons concentrated on both levels Type C The thrd form of local government can be seen as a hybrd between the two types descrbed above. Type C muncpaltes are mplemented n Rhneland-Palatnate ( Verbandsgemenden ) and n Lower-Saxony ( Samtgemende ). Lke n admnstratve collectvtes, there exsts a double-layered decson process and the member unts of sem-centralzed muncpaltes reman legally ndependent. However, n contrast to Type B muncpaltes, both levels are poltcally legtmated by polls whch allow for a hgher degree of concentraton of competences on the jont level. On the hgher level, a jont muncpal councl and a mayor are elected. Lke n Type B muncpaltes, the jont level s responsble for the provson of oblgatory tasks of the transferred sphere of responsbltes. Addtonally, federal states legal settngs defne the mnmum scope of transferred Oblgatory tasks of the own sphere of responsbltes from the local to the jont level of the muncpalty, whle (n contrast to Type B muncpalty) the central level can decde about how much and how a certan good or servce s provded. For example, n Rhneland-Palatnate the jont level s responsble for schoolng, fre protecton, sports facltes, socal facltes, water supply and sewerage by law (see GemO Rhneland-Palatnate 67). If not transferred to the jont level, the local authortes can decde about how and how much of a certan good s provded. As n Type B muncpaltes, there are cty councls and mayors on the local level. Compared to the Type B enttes, the scope of local responsblty s smaller though poltcal legtmaton for the aggregate muncpalty s larger (see Rosenfeld et al. 2007, p. 93). 16
IWH 4 Hypotheses Aganst the background of the local nsttutonal settng n Germany we derve the followng hypotheses from the theoretcal dscusson n secton 2. Thereby we dstngush a weak and a strong hypothess. The frst weak one states that the dvson of competences between dfferent admnstratve levels of local poltcal enttes to decde about f, how and how much of a certan good s provded affects voter turnout. In other words, despte the mere sze-effect on voter turnout, the local nsttutonal settng of admnstratve enttes does matter. (H1) The local nsttutonal settng does affect voter turnout. The weak hypothess s nested n an emprcally more powerful one whch explctly refers to the structural dfferences observed at the German muncpal. As dscussed n secton 2, we expect muncpaltes of type A to have the lowest turnout snce there s no possblty to optmze the allocaton of competences between dfferent admnstratve levels wthn ths organsatonal settng (see Table 1). On the contrary, muncpaltes of type C have two elected admnstratve levels and hence should have the hghest turnout.thus, (H2) The muncpaltes of type A have the lowest and muncpaltes of type C have the hghest voter turnout. Regardng the dfferences between type C and type B muncpaltes we therefore mplctly assume that the flexblty of shftng competences on a centralzed admnstratve level, whch s possble n muncpaltes of type B, s overcompensated by the lack of poltcal legtmaton n ths organsatonal settng. 5 Emprcal Implementaton We test the hypotheses formulated above by modellng the turnout n = 1,, German muncpaltes and cooperatve muncpaltes as follows. turnout = α + ϕ D + ϕ D + β X + υ A C A C (1) Where α, ϕa, ϕc, β denote parameters and υ denotes an..d. error component. Besdes the control varables X, whch are dscussed n the next secton, we nclude two A dummy varables D and D C, whch ndcate the nsttutonal settng observed to be of type A or type C respectvely. In Equaton (1) the weak hypothess (H1) states that there s no effect of the local nsttutonal settng on turnout, whch can be tested by a jont test of the parameters ϕ A and ϕ C to be commonly zero. The more ambtous strong hypothess (H2) s also captured A C by the dummy varables D and D n Equaton (1). The parameters ϕ A and ϕ C meas- 17
IWH ure the dfference n turnout relatve to the base category whch s a type B nsttutonal settng n our set up. Accordng to the strong hypothess (H2) we expect the estmated coeffcents ϕ A and ϕ C to be sgnfcantly dfferent from zero and to have a postve and a negatve sgn respectvely. 5.1 Data and Descrptve Statstcs For estmatng equaton (1) we use a data set provded by the statstcal offces of the federal states on voter turnout n local electons supplemented wth data on the socoeconomc and spatal structure of the respectve poltcal enttes. Altogether, n Germany exst about 4,550 muncpaltes and cooperatve muncpaltes wth dfferent nsttutonal arrangements as dscussed n secton 3. Table 2: Structure of the Sample, date and purpose of the electon Federal State Date of Electon Concurrent Electons Insttutonal Type A B C Schleswg-Holsten (SH) March, 2 nd 2003 101 118 - Rhneland-Palatnate (RP) June, 13 th 2004 EU 37-163 Baden-Württemberg (BW) June, 13 th 2004 EU 179 272 - Brandenburg (BB) October, 26 th 2003 143 54 - Mecklenburg-Western Pom. (MV) June, 13 th 2004 EU 52 97 - Saxony-Anhalt (SHT) June, 13 th 2004 EU 40 157 - Thurnga (TH) June, 13 th 2004 EU, state 157 91 - Source: German Federal Statstcal Offce, calculatons by the authors. For our analyss we dropped all observatons from larger free towns, so-called kresfree Städte, and from the cty states Berln, Bremen and Hamburg, snce these admnstratve enttes addtonally carry out tasks of hgher admnstratve levels (county and state) as depcted n Fgure 3. In order to ensure comparable poltcal condtons at the natonal level we further restrct our sample to local electons, whch took place wthn the natonal legslatve perod between 2002 and 2005.8 Thus the observatons from Bavara and Lower-Saxony are dropped.9 Fnally, due to data avalablty ssues, our 8 Ths ssue s strengthened by the fact that the Natonal electon n 2005 was actually scheduled for 2006. The ntenton to ask for a vote of confdence and therewth to reschedule the natonal electon was announced by chancellor Schröder n May 2005. Hence, the electon campagn of the natonal electon 2005 dd not affect the electons held n 2004 analyzed n our study. 9 Muncpal electons n Lower-Saxony took place n 2001 and 2006. In Bavara the electons on the local level took place n 2002 and 2008. 18
IWH data set conssts of 1,661 muncpaltes and cooperatve muncpaltes from seven German federal states. (see Table 2). As depcted n equaton (1) we use several soco-economc control varables to dentfy the mpact of local nsttutons on voter turnout. We therefore take advantage of the exstng lterature that provdes an elaborate pool of covarates, whch should affect the utlty or the cost of votng (for an extensve revew see Geys 2006). Varable defntons and descrptve statstcs are provded n Table 3. Frst of all and n lne wth prevous studes we nclude the total number of all elgble voters n order to capture the sze-effect of the electorate, whch we expect to have a negatve mpact on voter turnout (Setz 2008, Matsusaka and Palda 1999, Blas and Dobrzynska 1998). Another mportant factor on turnout s populaton densty, whch appromxmates dfferng socal pressure, nterpersonal bonds and consensus on norms between rural and urban areas (Rker and Ordeshook 1968, Overbye 1995). In rural areas these factors are assumed to be more effectve. Hence, we expect a negatve mpact of populaton densty on voter turnout. Accordng to several emprcal studes, another relevant determnant of voter turnout s the stablty of the populaton wthn a jursdcton. Stablty n most studes s related to mgraton and moblty. A more stable populaton s expected to affect voter turnout postvely because of a hgher degree of group soldarty (Hoffman-Martnot 1994; Ashworth et al. 2002). There are several dfferent measures for stablty ntroduced n the lterature, ncludng the mgraton balance of a jursdcton or the rato of homeowners, whch are assumed to be less moble (see Geys 2006b). We nclude populaton moblty whch s defned as the sum of n- and outmgraton dvded by the number of nhabtants. We expect to fnd a negatve mpact of ths factor on voter turnout. Furthermore, t has been argued that populaton s homogenety wthn an electorate could lead to hgher costs of non-votng nduced by socal pressure (see Cohen 1982). On the other hand, heterogenety could lead to hgher voter turnout because voters have a strong nterest to defend ther postons aganst other socal groups (see Zmmer 1976). We ncorporate the long term unemployment rate and the Herfndahl ndex of the age structure nto our model n order to account for homogenety effects(verba and Ne 1972, Geys 2006b). Fnally, most emprcal studes fnd a sgnfcant mpact of educaton on voter turnout (Ashenfelter and Stanley (1975), Brody and Page (1973), Guttmann). That s why we nclude the share of hghly qualfed and the share of low qualfed nhabtants n our emprcal model. Accordng to prevous studes we expect to fnd postve and negatve sgned parameter-estmates respectvely. 19
IWH Table 3: Varable defnton and descrptve statstcs Varable Defnton Summary Statstcs Varable Descrpton Mean Sd. Mn; Max Endogenous Varable turnout Proporton of elgble to actual voters. 0.5483 0.0827 0.28; 0.77 Control Varables Dummy_typeA Dummy_typeC Populaton Populaton Densty Populaton Moblty HHI_age Unemployment Educaton_hgh Educaton_low Dummy_EU Dummy_state Dummy_east Dummy_RP Dummy_BW Dummy_MV Number of Observatons 1661 Source: calculatons by the authors. Equals one f the nsttutonal settng ot the muncpalty s of type A. Equals one f the nsttutonal settng ot the muncpalty s of type C. Number of regstered voters at the date of the electon. Number of people lvng on a square klometre of muncpal space. Sum of n- and outmgrants dvded by the number of nhabtants. Normalzed Herfndahl-Index regardng the age-structure Share of people who have been jobless for more than 12 months. Share of the populaton wth unversty degree. Share of the populaton wthout vocatonal tranng and wthout A-level degree. Equals one f the EU-electon 2004 took place concurrently. Equals one f the state-electon took place concurrently (only Thurnga). Equals one f the muncpalty s located n the area of the former GDR. Equals one f the muncpalty s located n Rhneland-Palatnate. Equals one f the muncpalty s located n Baden-Württemberg. Equals one f the muncpalty s located n Mecklenburg-Western Pomerana. 0.4269 0.4948 0;1 0.0981 0.2976 0;1 9912 8573 282; 80047 247.7 310.9 10.0; 2526 0.1144 0.0344 0.02; 0.54 0.0698 0.0034 0.06; 0.11 0.0282 0.0541 0; 0.73 0.0143 0.0197 0; 0.53 0.0334 0.0257 0; 0.25 0.7495 0.4334 0;1 0.1493 0.3565 0;1 0.4762 0.4996 0;1 0.1204 0.3255 0;1 0.2715 0.4449 0;1 0.0897 0.2858 0;1 20
IWH Last but not least, n order to capture possble effects of concurrent EU or State-level electons, we nclude an EU-electon and a State-electon dummy. The EU-dummy addtonally controls for a level shft n voter turnout between the years 2003 and 2004 (see Table 2). Remanng unobserved level-effects are accounted for by an East- Germany-dummy, whch accounts for all muncpaltes located n the area of the former GDR as well as three added country dummes. Addtonal country-dummes cannot be ncorporated to avod lnear dependency (see Table 3). 5.2 Econometrc Issues In equaton (1) the dependent varable turnout s bounded between zero and one. Thus, we need to employ an estmaton method whch s able to deal wth fractonal response varables. The frequently used lnear, tobt or logt models do not solve the specfc problems of fractonal dependent varables satsfactorly. For example, usng Ordnart Least Square (OLS) methods mplctly gnores the bounded nature of the dependent varable and mplctly assumes a constant effect for all explanatory varables on turnout over ts entre range. Moreover, the predcted values from OLS regresson can never be guaranteed to le wthn the nterval (Papke and Wooldrdge 1996). Applyng a two-lmt Tobt to proportonal data ndeed ensures the predcted values to be wthn the unt nterval but ths s true by defnton not caused by censorng (Keschnck and McCullough 2003). Several prevous studes commonly use Logt models to overcome the lmtatons of the aforementoned models. However, after logt-transformng the dependent varable t s not straghtforward to recover the orgnal condtonal mean functon, whch s of man nterest n our study (Papke and Wooldrdge 1996). In our case, wthout any observaton at the boundary values zero and one (see Table 3) two man approaches for estmatng Equaton (1) have been proposed n the lterature. The frst approach s the quas-maxmum lkelhood procedure proposed by Papke and Wooldrdge (1996). The second one s to estmate the condtonal mean functon by maxmum lkelhood assumng a beta-dstrbuton. The latter approach s consstent, asymptotcally normally dstrbuted and fully effcent f the assumed condtonal densty s correctly specfed. Most researchers use a meandsperson parameterzaton of the beta-densty suggested by Paolno (2001) and Ferrar and Crbar-Neto (2004) n order to smplfy the nterpretaton of the parameter estmates. In ths specfcaton a functonal form s modeled separately for the mean and the dsperson. The beta-densty then can be parameterzed as: wth ( µ φ) Γ( φ) ( φ ) ( 1 ) ( µ φ) ( ) ( ) f turnout = turnout turnout Γ Γ µφ 1, 1 1 µ φ 1 (2) 21
IWH ( ) E turnout = µ (3) and ( ) Var turnout ( 1 ) µ µ = 1+ φ (4) where Γ ( ) s the gamma functon, 0< turnout < 1, 0< µ < 1 and φ> 1. The mean of the dependent varable turnout now can be modeled usng dfferent lnk-functons, whch ensure the expected value of turnout to be bounded by 0 and 1. We follow prevous studes and use the convenent logt-lnk: ( z ) E turnout z ' π e = µ = 1 + e (5) z ' π here z stands for a matrx of all explanatory varables n Equaton (1), ncludng the dummy varables. π subsumes the correspondng parameter vector. As mentoned above, the essental drawback of the fully parametrc approach s that t yelds nconsstent parameter estmates f the condtonal densty of the dependent varable s ncorrectly specfed. Ths s the reason why Papke and Wooldrdge (1996) suggest a quas-parametrc regresson model, whch just assumes that: E( turnout z ) = G( zπ ) (6) The known nonlnear functon G ( ) satsfes 0 G ( ) 1. Typcally, the functon G ( ) s chosen to be a cumulatve dstrbuton functon. For our analyss we follow the majorty of prevous research and chose G ( ) to be the logstc functon. We use the partcular quas-maxmum-lkelhood method based on the Bernoull log-lkelhood functon proposed by Papke and Wooldrdge (1996), whch s consstent und asymptotcally normal regardless of the true dstrbuton of turnout on z gven that E( turnout z ) s correctly specfed. Ths quas-parametrc approach does not rely on the specfcaton of the full dstrbuton of f ( turnout z, π ). In order to test the hypotheses stated above we employ both estmaton procedures but rest our nterpretaton prmarly on the QML-results. We do so because there s some evdence that even f the beta assumpton s vald the ML approach only outperforms the QML-estmator under certan crcumstances (Ramalho and Ramalho 2010). The betaregresson results are gven n the appendx. 22
IWH 6 Results We estmate the model specfed n Equaton (1) usng the QML-procedure dscussed above. The results are provded n Table 4. We get qute smlar results usng the betaregresson model whch are gven n the appendx. The r-squared depcted at the bottom of Table 4 s defned as the explaned sum of squares of turnout dvded by ts total sum of squares based on the unweghted resduals and ndcates the hgh explanatory power of our model. In order to test the crucal assumpton that the condtonal mean s correctly specfed we perform a lnktest. We cannot reject the Null hypothess of a correctly specfed condtonal mean functon wth a p-value of 0.680. To begn wth we refer to the weak and the strong hypotheses stated above. The sgnfcant effects of the dummy varables ndcate that the observed nsttutonal settng at the local level has a strong mpact on voter turnout. Furthermore, the estmated coeffcents show the negatve and the postve sgns as stated n the strong hypothess. We also calculate margnal effects at the sample mean n order to assess the magntude of the nsttutonal settng on voter turnout. As can be seen n the fourth column n Table 4, shftng the nsttutonal settng of a muncpalty from type B to type A whle keepng all other explanatory varables at ther sample mean would reduce voter turnout by about 5.5%. On the contrary, a shft to the nsttutonal settng of type C would ncrease the voter turnout by about 3.4%. Regardng the ncluded control varables, the results are n lne wth prevous studes. We fnd a very weak affect of sze and populaton densty whch proxes more urbanzed areas shows the expected negatve sgn. Furthermore, educaton seems to be an mportant determnant for voter turnout. Somewhat surprsngly, the longterm unemploymentrate turns out to have a substantal postve mpact on voter turnout, whch underpns the hypothess of organzed nterests and contradcts the sometmes stated lack of partcpaton among ths group. Ths mportant ssue should be the subject of further research. Fnally, except for the ncluded dummy varables all other control varables are not sgnfcantly dfferent from zero. The dummy varables at the bottom of Table 4 are as expected and reflect the postve effect of concurrent state-electons and the ongong skeptcsm about poltcal partes n the former socalst eastern part of Germany. Yet, one result deserves a closer nvestgaton. The negatve sgn of the dummy varable ndcatng a concurrent EU-electon actually contradcts the economc theory statng that concurrent electons ncrease the beneft of votng. However, n our dataset the generated EU-dummy also captures a tme effect snce n our sample all local electons n 2004 took place at the same tme as the EU-electon (see Table 2). Hence, we cannot separate ths tme effect from the effect of the EU-electon. 23
IWH Table 4: QML-estmaton results varable parameter estmates margnal effects coeffcent std. error P> z dydx std. error Dummy_typeA -0.2222 0.0134 0.000-0.0550 0.0033 Dummy_typeC 0.1402 0.0356 0.000 0.0347 0.0088 Populaton 0.0000 0.0000 0.000 0.0000 0.0000 Populaton Densty -0.0001 0.0000 0.006 0.0000 0.0000 Populaton Moblty -0.1291 0.1716 0.452-0.0320 0.0425 HHI_age 1.9653 2.0253 0.332 0.4866 0.5014 Unemployment 0.4186 0.0935 0.000 0.1036 0.0231 Educaton_hgh -0.0901 0.3443 0.794-0.0223 0.0852 Educaton_low -2.1212 0.4505 0.000-0.5252 0.1116 Dummy_EU -0.2775 0.0235 0.000-0.0687 0.0058 Dummy_state 0.4947 0.0224 0.000 0.1225 0.0055 Dummy_east -0.3616 0.0245 0.000-0.0895 0.0061 Dummy_RP 0.2790 0.0422 0.000 0.0691 0.0105 Dummy_BW 0.2411 0.0308 0.000 0.0597 0.0076 Dummy_MV 0.1592 0.0251 0.000 0.0394 0.0062 constant 0.5481 0.1460 0.000 - - R 2 0.6250 umber of Observatons 1661 Source: Calculatons by the authors. 24
IWH 7 Concluson Apart from the rules of votng, the mpact of local nsttutons on voter turnout has been wdely gnored n emprcal research. Our paper s to shed lght on that ssue. We show the dversty of local nsttutons n Germany and dstngush three man types of nsttutonal arrangements on the local level. We argue that the dfferences between these nsttutonal settngs regardng the dstrbuton of competences affects the complexty of the votng decson and hence voter turnout. We derve two hypotheses regardng the mpact of local nsttutons on voter turnout and test them usng a dataset on local electons n Germany. Furthermore, we employ a Quas-Maxmum-Lkelhood approach proposed by PapkeWooldrge n order to deal wth the specfc problems of fractonal response varables. We fnd strong emprcal evdence for the mpact of local nsttutons on voter turnout. As stated n our theoretcal dscusson, besdes the mere exstence of dfferent admnstratve levels the optmal dstrbuton of competences between them ncreases turnout sgnfcantly. Fnally, our nqury shows that regardng the reform of local government t s mportant to take the mpact on turnout nto consderaton. To gnore ths effect and to purely focus on effcency n producng of local publc goods and servces at the local level may lead to a lack of allocatve effcency n the long run. 25