Clientelism and polarized voting: Empirical evidence

Transcription

1 Clentelsm and polarzed votng: Emprcal evdence by Klarta Gërxhan and Arthur Schram Abstract One must take country-specfc nsttutonal features nto account when analyzng former communst countres transformaton process to new poltcal nsttutons. We do so for postcommunst Albana, where the regonal and cultural polarzaton that has exsted for centures has evolved to clentelsm n the new democracy. We show how clentelstc partes gve rse to very partcular votng patterns. These reveal major dfferences across regons not only n party choce but also n voters responses to government polces. Ths s n sharp contrast wth results obtaned when applyng the same model to a large number of more advanced democraces wth smlar electoral nsttutons. Keywords: polarzaton, clentelsm, electoral democracy, vote functons Aprl 2009

2 . Introducton Over the past two decades, the poltcal and economc development of former-communst countres has been sensatonal. For scholars nterested n democratc processes, t s a challenge to try to understand the nuances of what s happenng. Both poltcal scentsts and scholars n the dscplne of publc choce have contrbuted substantally to our understandng of the transformaton process from communsm. From a poltcal scence perspectve, t s mportant to realze that the road to democracy mples not only free and legtmate electons but also deeper poltcal partcpaton and government accountablty. One concluson often put forward s that many transtonal countres are currently functonng n a gray zone between authortaransm and democracy. Publc choce scholars tend to focus on specfc aspects n the poltcal economc development of these countres (such as changes n the popular vote, certan government polces, or trends n the sze of government). Ths lterature provdes several applcatons of standard research methods to a cross secton of these countres for general nsghts or to specfc countres as case studes. 2 The conclusons n ths lterature are qute dverse, however, and t s not yet possble to dstll general mplcatons from t. In ths paper, we use technques nspred by the publc choce tradton to study a queston that s central n the poltcal scence lterature n ths area. We focus on the development of one country n partcular, to wt, Albana. More specfcally, we wll study how specfc features of Albana s poltcal nsttutons nteract wth voter behavor. Our focus on one country s motvated by the belef that t s mportant for the understandng of many phenomena n post-communst countres that country-specfc nsttutonal features are taken nto account (North 990; Carothers 2002; McFaul 2002). In Albana, one of the features that may be mportant s an alleged clentelsm between partes and voters (Klos 997) that runs parallel to a strong polarzaton n the Albanan socety. Here we focus on the consequences of ths clentelsm and provde an emprcal analyss showng that t affects votng behavor n a way that s very unlke the patterns observed n very many developed democraces. For our emprcal analyss, we estmate so-called vote functons 3 and are able to show that () clentelsm plays an mportant role n determnng how Albanans vote; () neglectng the exstence of the polarzaton caused by clentelsm can bas not only the conclusons drawn from the applcaton of standard technques, but also the polcy mplcatons based on deal nsttutonal frameworks.

3 2. Clentelsm n Albanan poltcs Albana has hardly ever had a democratc poltcal system. In about four decades of communst dctatorshp t was completely solated. The poltcal regme was dctatoral and the economc system was completely socalzed. In the early 990s, Albana was the last of the Central and Eastern European countres to allow poltcal pluralsm and ntroduce democratc nsttutons and market mechansms. Snce the break up of communsm, there have been two major partes n Albana: the Democratc Party of Albana (DPA, n offce and snce 2005) and the Socalst Party of Albana (SPA, n offce ). In the three electons consdered here, the two partes receved 70-80% of the votes cast. The remanng votes were spread over more than 25 partes, wth no party recevng a share of more than 5%. The fall of the communst regme n 992 brought the DPA to power. Untl 997, the poltcal regme was characterzed by compettve authortaransm (Levtsky and Way 2002), whle the economy appeared to be growng reasonably well. However, the economc numbers may be based and moreover the prvatzaton process was used to ncrease DPA s popularty through clentelsm (Shala 997; MPEP 997; UNDP 2000; Gërxhan 2006). In any case, all that was acheved was demolshed n 997, when a collapse of the Pyramd schemes led to complete poltcal and economc chaos (Sadraj 999; Bezemer 200). The SPA won the 997 electons and remaned n government untl the DPA reganed power n After a slow start, some democratc progress was acheved reflected n a shft from a compettve authortaran regme to an electoral democracy. 4 At the same tme, the economy has been growng at a constant rate but unemployment remans at hgh levels. In ths paper, we focus on the DPA-run years and the SPA-run years The post-communst poltcal system n Albana s a parlamentary representaton, wth a mxed frst-past-the-post and proportonal votng system. 5 The parlament elects the presdent. The presdent apponts and dscharges the prme mnster and the cabnet. Because the partes are structured very herarchcally (Sal Bersha of the DPA has been leadng the party snce the start of transton and Fatos Nano led the SPA untl 2005 and stll remans very nfluental today), poltcal power s heavly concentrated n the hands of a few top poltcans of the wnnng party. When a party wns an electon, the leader ether becomes presdent or prme mnster hmself or apponts someone fathful to hm. Moreover, mportant publc offcals are replaced by the wnner s protégés. In short, snce the start of the transformaton process the wnner of Albanan electons (ether the SPA or the DPA) has vrtually obtaned unchallenged control over publc polcy. 2

4 Most scholars agree that there s an mportant cultural and lngustc polarzaton n Albana (e.g., Shala 997; Haefner 999, La Cava and Nanett 2000). Ths polarzaton s geographcally based, dvdng the country nto a northern and a southern regon. Hstorcally, the polarzaton of socety runs parallel to the exstence of two man clans n Albana, to wt the Ghegs (northern-based) and the Tosks (southern-based). Doll (2003) argues that they go back far n hstory. A regonal dvson was frst notceable n 395 AD, and a clan-based polarzaton may have started as early as 08, wth the ntroducton of feudalsm by the Normans. Ths polarzaton contnued durng the Ottoman occupaton n the 6 th century, the turbulent years of Albana s ndependence n 92, the monarchy n the late 920s, and was stll detectable durng communsm. It has an mportant nfluence on varous aspects of the Albanan socety. Poltcs are one such aspect, the one we focus on. The two man poltcal partes geographc bases match closely the country's regonal polarzaton: the democrats (DPA) are culturally and poltcally closely related to the north and the socalsts (SPA) to the south (Klos 997; Vckers and Pettfer 997; Gërxhan and Schram 2000; La Cava and Nanett 2000). The lnks between the partes and cultural/regonal groups are much stronger than the usual representaton of nterests observed n most democraces (Vcker and Pettfer 997; Gërxhan and Schram 2000). The relatonshps between partes and regonal groups have many of the characterstcs that would make Lauth (2000) classfy the DPA and SPA as clentelstc partes. Followng Schmdt (992), Lauth descrbes clentelsm n general as.forms of protectve relatonshp of mutual beneft between a person or persons occupyng a hgher place n the socal herarchy (patron) and a followng concerned wth protecton and the acquston of certan advantages (clents) (p. 27). Lauth classfes dfferent types of clentelsm along two dmensons: () whether the relatonshp s based on the socal or poltcal system; and () whether poltcal partcpaton takes place drectly or ndrectly. One of the forms of clentelsm wth drect partcpaton n the poltcal sphere s dubbed clentelstc partes (p. 29). Ths s characterzed by () openness n partcpaton (as n free electons); (2) competng patron-clent relatons (as DPA-northern voters vs. SPA-southern voters); (3) most patrons (.e., Albanan poltcans) comng from the poltcal establshment (as n the appontment of key poltcal postons n Albana, as descrbed above). 6 The patron-clent relatonshp ths categorzaton yelds runs two ways. Frst, partes wll bestow gfts upon ther clents (.e., the regon they are connected to). An example s the way n whch the DPA used the prvatzaton process and numerous appontments to key governmental postons to gve poltcal and economc favors to the north. For example, n 3

5 994, the unemployment rates were approxmately equal n the north and south. Yet, 35.3% of the populaton n northern regons receved socal assstance and only.2% n southern dstrcts (Gajo 999). When the SPA came to power n 997, the reverse occurred: t apponted most ndvduals from the south and amed polces towards ths regon. Gërxhan and Schram (2000) and Case (200) provde more evdence on the behavor of the patron sde of the clentelsm. Here, we focus on the second channel, namely the one runnng from clentvoters to patrons. The man expectaton s that when ther own party s n government, voters wll support t no matter what. Because of the numerous ways the party helps them and s expected to keep on helpng them, ther votes are not affected by specfc (e.g., economc) outcomes of these polces. In partcular, we formulate the followng man hypothess: Hypothess When the DPA (SPA) s n offce, voters n the south (north) react negatvely to economc swngs, whch they attrbute to the government. Voters n the north (south) have a strong allance wth the governng DPA (SPA) rrespectve of ts economc polces. In the central regon, whch s much less affected by clentelsm, voters always evaluate the government by ts polces. 7 Note that we focus on economc polces. There are two man reasons to do so. Frst, economc ndcators are generally of a quanttatve nature and therefore sutable for formal testng of hypotheses. Second, there s a rch lterature (to whch we refer n the followng secton) relatng economc outcomes to voter behavor. Focusng on economc polces enables a comparson of our results to ths lterature whch, n turn, allows us to dstll the effects of clentelsm. To the best of our knowledge we are the frst to emprcally test the consequences of clentelsm n ths way. We do so n the followng secton. 3. Polarzed votng behavor For our emprcal analyss we consder the results for the Albanan general electons n 996, 997 and 200 and categorze the 35 electoral dstrcts nto the regons: northern (), central (3) and southern (). 8 Table summarzes the electon results and reports the unemployment rate per regon. Detaled nformaton about the data used s avalable from the authors. [TABLE ABOUT HERE] 4

6 Note the tremendous decrease n the popularty of the DPA between 996 and 997. The dfference across regons s also remarkable. Independent sample t-tests (wth unequal varance) show that par-wse dfferences are statstcally sgnfcant at the % level except (northern-central) n 996, (central-southern) n 997 and (central-southern) n 200. Support for the DPA s therefore sgnfcantly lower n the south than n the north. Ths renforces the noton of poltcal polarzaton between the north and south wth the northern voters supportng the DPA and the southern voters supportng the SPA. In order to show that ths polarzaton s founded on clentelsm we nvestgate the hypothess presented n the prevous secton, that the response to economc polces s dependent on the governng party n both regons. To test the hypothess, we need an ndcator of economc polces. In the publc choce lterature, there are numerous examples where the relatonshp s studed between economc polces and government popularty. 9 Varous ndcators have been used n these studes. Nannestad and Paldam (994: 26) conclude that the two that consstently affect votes are unemployment and nflaton. Unfortunately, data on nflaton per dstrct n Albana are not avalable. Moreover, the Phllps curve relates the unemployment rate to nflaton. For ths reason, some authors choose to exclude nflaton from the analyss even f they have the data avalable. Therefore, we focus on the unemployment rate as a key ndcator of government polcy. 0 Next, we need to descrbe how a government s polces (.c., unemployment) affect ts support n electons. To do so, we model the way n whch voters determne ther choces at an electon. The model dstngushes between three types of voters, to wt, clents of the DPA clents of the SPA and non-clents. For the reasons dscussed above, we assume that there are no SPA-clents n the north and no DPA clents n the south. In prncple, non-clents may lve n any part of the country, and there may be SPA as well as DPA clents n the central regons of Albana. To enable the development of an emprcally tractable model, we now assume that the number of non-clents n the north and south s neglgble compared to the number of clents and that the reverse holds n the center of the country. Then, we consder votng behavor n Albana s 35 votng dstrcts (denoted by ). Each dstrct unquely les n one of the three regons r {north, central, south}. Now defne by l U g the utlty that a voter l n votng dstrct attrbutes to government party g {DPA, SPA}. Smlarly, l U o denotes the utlty that ths voter attrbutes to opposton party o {DPA, SPA}, o g. 2 The utlty attrbuted to the governng and opposton partes s gven n eqs (): 5

7 ( ) U = α + β D Un -Un* + ε () l l g rg r rg g U = α + ε l l o ro o where Un denotes the unemployment level n dstrct and Un* some (unknown) target unemployment that voters compare actual unemployment to. Note that each voter l unquely lves n some dstrct. Followng the dscusson above, the government party s evaluated wth respect to the unemployment rate n the dstrct n whch the voter lves, whereas the opposton s not. In (), α rg, αro and β r are unknown parameters, wth α rg, α ro > 0 descrbng predspostons that are common to voters of the same type and regon and β r (<0) a parameter common to all voters wthn r that descrbes how voters respond to unemployment. The error terms l ε g and l ε o are assumed to be..d. The term D rg s a dummy varable determnng the extent to whch a government s economc polcy (wth respect to unemployment) enters the voters utlty functons. Ths s determned by our assumptons on the behavor of clents. More specfcally: D D D = D = 0; ng, = DPA sg, = SPA = D = ; ng, = SPA sg, = DPA = D =. cg, = SPA cg, = DPA (2) where n= north ; c= central ; s= south. 3 These equatons reflect the assumptons that clents do not consder unemployment caused by a government of ther patron relevant for ther vote (frst row), but do hold the government of the other party accountable (second row). The non-clents n the central regon hold both partes accountable for unemployment, when n government (thrd row). Next, consder the predspostons α rg and α ro. The assumptons that voters n the north (south) are clents of the DPA (SPA) (and therefore have a postve allegance to the DPA (SPA) rrespectve of whether or not ths party s n government) and that voters n the center have no predspostons mply: α = α > 0; α = α > 0; α n, g= DPA n, o= DPA s, g= SPA s, o= SPA = α = 0 cg, co, (3) A voter wll vote for the government party f the utlty attrbuted to g exceeds that of opposton party o. Usng () ths gves for the probablty that voter l wll vote for g, l p g : 6

8 { } { ( ) ε ε } l l { εo εg rg ro r rg ( )} ( rg ro r rg ( )) p = Pr U > U = Pr α + β D Un -Un* + > α + = l l l l l g g o rg r rg g ro o = Pr < α α + β D Un -Un* = F α α + β D Un -Un*, (4) where F denotes the cumulatve dstrbuton functon of l o ε ε l g. If l ε g and ε l o are (ndependently and) exponentally dstrbuted, the dfference (F) s a double exponental dstrbuton. Straghtforward calculatons (see the appendx) then gve for electons wth the DPA n offce: EDPA ln = da n n, g DPA (-dn- dc) as, o SPA dcβc( Un- Un* ) (-dn- dc) βs( Un- Un* ) = = + + (5) E SPA where E ( E DPA SPA ) denotes the expected number of votes for the DPA (SPA) and d n (d c ) s a dummy ndcatng that n (c). Smlarly, when the SPA s n offce: ESPA ln = da s s, g DPA (-dc- ds) an, o DPA dcβc( Un- Un* ) (-ds- dc) βn( Un- Un* ) = = + + (6) E DPA where d s s a dummy ndcatng that s. Equatons (5) and (6) descrbe the relatve number of votes for the governng partes as a functon of, respectvely, () the postve predsposton of ther own clents; () the negatve predsposton of the opposton party s clents; () the response of non-clents to the rate of unemployment; and (v) the response of the opposton party s clents to the rate of unemployment. For our emprcal applcaton, we ntroduce tme (t) and reduce (5) and (6) to the followng vote functon that can be estmated usng electoral data: 4 V ln = + Un + D +, =,..35; t = 96,97,0; j = n, c, s (7) V t g t tj t t t t β0 β t βg g ε o where V ( V ) denotes the vote share of the governng (opposton) party n dstrct n t g t o t=996, 997, 200; j denotes the regon (n=north, c=central, s=south) that les n, Un t s the unemployment rate n dstrct n t, t D g s a dummy varable ndcatng the regon to whch the governng party s lnked, and ε t s a whte nose error term. 7

9 The appendx shows how eq. (7) follows from (5) and (6). Most mportantly, the parameters tn tc β, β ts, and β n (7) reflect the way voters n the varous regons respond to unemployment and therefore provde estmates of the parameters β n, β c, and β s n (5) and (6). Note that equaton (7) ncludes unemployment n the regon of the governng party as a regressor. Ths allows us to drectly test our assumpton that voters n these regons do not respond to tn ts unemployment,.e., β =0 ( β =0) after a DPA (SPA) government. These tests provde support for the assumpton. t The coeffcents 0 tj β t β, and β 2 n (7) wll be estmated wth OLS. Formally, the hypothess presented n the prevous subsecton yelds a test of: tn tc ts H 0 : 0 > β = β = β, t=996, 997, 200 tn tc ts vs. H : 0 = β > β > β, t=996, 997; ts tc tn 0 = β > β > β, t=200 Under the null hypothess, voters n the three regons respond (negatvely) to unemployment n a way smlar to that observed n very many countres across the world (Mueller 2003). The response does not dffer across regons. The alternatve states that n 996 and 997 voters n the south (.e., those that oppose the governng DPA) react negatvely and most strongly to unemployment durng the DPA government. Voters n central regons (wthout a clentelstc relatonshp to a party) react negatvely but less strongly. Fnally, voters n the north (DPA-clents) do not let unemployment affect support for the DPA at all. In other words, because the polces of the clentelstc party DPA favor the north, northern voters are the most forgvng to ths government where economc consequences are concerned. The opposte s predcted n 200. The SPA clentelsm favors the south, whose voters are now expected to be the most forgvng. The strongest effect of unemployment s now predcted for the north. 5 Note that H predcts extreme, opposte effects under the two governments. A tradtonal applcaton of vote functons does not dstngush between regons (predctng H 0 ). Therefore, we frst estmate the model mposng the null that all voters respond to unemployment n the same way. The results are gven n the no polarzaton columns of table 2. The coeffcent for unemployment s negatve n all three electons (ndcatng that the governng party has less support n dstrcts where the unemployment s hgher), and statstcally sgnfcant n 8

10 two of the three cases. The explanatory power of ths model (R 2 ) s between 0.24 and 0.5. Taken by themselves, these results would support the tradtonal publc choce fndng that a government s electon result s negatvely affected by unemployment. [TABLE 2 ABOUT HERE] The explanatory power ncreases substantally when we allow the coeffcents to vary across regons ( polarzaton columns of table 2). These results show mportant dfferences across regons and electons. In the north, when the DPA s n government voters do not react statstcally sgnfcantly to unemployment. In the center and south, negatve responses to the unemployment level are observed. The strongest (negatve) response s found n the south. A formal test of H 0 versus H shows that the equaltes n H 0 are rejected n favor of the nequaltes of H (p<0.0) n both years. It s remarkable how much the results change n 200, when the SPA s n government. Though the aggregate ( no polarzaton ) result resembles the 996 and 997 results, the dsaggregated results dffer substantally, n the predcted drecton. The north s no longer forgvng. There s a negatve and statstcally sgnfcant effect of unemployment on support for the government. In contrast, voters n the center and south, who were not forgvng for the DPA government, show no statstcally sgnfcant effects. 6 Once agan, a formal test of H 0 versus H shows that the equaltes n H 0 are rejected n favor of the nequaltes of H (p<0.0). The rejecton of H 0 means that our results support our man hypothess, that clentelsm causes voters to respond asymmetrcally to government polces. The fact that ths result s observed n the (opposte) predcted drecton for the two dstnct governments makes t especally strong. In partcular, the noton of a poltcally polarzed country fnds support. Note that f we had not taken account of ths polarzaton, we would have concluded that the effect of unemployment on the vote s smlar to that found n numerous studes. Fnally, the results for the central regon are nterestng. In 996, ths s closer to that for the north than that for the south. Ths changes n the drecton of the results for the south n 997 and 200. It may be the case that (changng patterns of) nternal mgraton to Trana (n the center) s startng to affect the extreme north-south polarzaton (as suggested by ICG 200, for example). It s stll too early to judge, however. To get a better dea about the quanttatve effects of polarzaton and clentelsm, we use the estmated coeffcents to smulate the electon outcomes n varous regons. Table 3 summarzes the outcome of these smulatons. 9

11 [TABLE 3 ABOUT HERE] The results show large dfferences across regons n the support for the DPA (columns a). Note however, that these dfferences could smply be due to regonal dfferences n party support (.e., polarzaton caused by predspostons wthout clentelsm). The responses to unemployment (depcted n columns b) provde evdence of clentelstc votng, however. The effect of an ncrease n unemployment from 0% to 20% n a dstrct strongly depends on the regon n whch the dstrct falls. In 996, for example, ths would lead to a two percentage ponts decrease n support (3% of the orgnal support) f the dstrct s n the DPA-supportng north but a 7 percentage ponts decrease (35% of the orgnal) f t s n the south. Fnally, columns c allow us to vsualze the effects of pure polarzaton,.e., the predsposton of voters towards the two partes. Note from eq. (4) that we can estmate only the (net) dfference n the government s regon. The results show that ths effect can be up to 0%- ponts. Hence, even wthout clentelsm, the polarzaton of voter behavor n Albana s substantal. 4. Conclusons The specfc hstory of any naton -but especally of a naton gong through a transformaton process from communsm- s very mportant to understand ts poltcal development. A key element throughout Albana s hstory s a cultural and geographcal polarzaton. Ths paper argues that after the fall of communsm, ths dvde has contnued to exst. It has led to a poltcal envronment where clentelstc partes renforce socety s polarzaton. We have shown that ths clentelsm systematcally affects votng behavor. Moreover, we have shown that a proper evaluaton of democratzaton n Albana requres lookng beyond electons per se and takng ths clentelsm nto account. The fact that the polarzaton of Albanan poltcs ncludes clentelsm follows from the dfferental electoral response to unemployment that we observe. Regonal polarzaton n tself would mply dstnct votng across regons, for example wth northern voters supportng the DPA and southern voters votng for SPA. We observe ths n our data. The fact that northern voters are forgvng towards the DPA but not towards the SPA, whle the reverse holds for southern voters, cannot be explaned by regonal polarzaton per se, however. The observaton that the DPA favors the north n a patron-clent relatonshp and the SPA s a clentelstc party supportng the south does provde a ratonale for ths observaton. 0

12 An nterestng queston, of course, s why clentelsm plays such an mportant role n Albana. There are two possble explanatons. One s related to path dependence, namely the exstng regonal dvson, possbly hstorcally founded on the exstence of two mportant clans (Doll 2003), whch facltates the polarzaton that has exsted n Albana for centures. 7 The other explanaton s based on the fact that formal governmental nsttutons are stll underdeveloped. It has been argued that ratonal expectatons mply that poltcal canddates must make promses that they can actually fulfll f elected (Myerson 993). Ths does not necessarly hold for electoral democraces where government nsttutons have remaned underdeveloped, however. 8 Ths may gve rse to serous agency problems, makng promses of poltcal canddates more dffcult to enforce. In turn, ths may lead voters to rely more heavly on other socal structures. In other words, as long as formal governmental nsttutons do not functon properly, nformal nsttutons lke clentelsm wll actvely fll the vacuum (Gërxhan 2004). A better functonng of government nsttutons may be the key to overcomng polarzaton and clentelsm. As long as clentelstc partes are able to use the government to serve ther clents, t may be expected that voters wll behave the way they have been dong. In turn, ths gves the partes no reason to change. Only f formal nsttutons develop that may better gude and control government polces may t be possble to break the centures old nformal nsttuton of clentelsm that n the democratc era has taken the form of clentelstc votng.

13 Appendx In ths appendx, we start wth the probablty derved n eq. (4) n the man text, that voter of type k wll vote for the governng party: ( ( )) p = F α α + β D Un -Un*, g rg ro r rg l l l where F denotes the cumulatve dstrbuton functon of εo εg. If ε g and ε l o are (ndependently and) exponentally dstrbuted, the dfference, F, s a double exponental dstrbuton, whch gves: αrg aro + βr D rp (Un -Un*) e g = αrg aro + βr D rp (Un -Un*) p + e po = p g =, αrg aro + βr D rp (Un -Un*) + e (A) where po denotes the probablty that voter l, n dstrct wll vote for the opposton party. Now, let n denote the number of voters n dstrct. The expected number of voters n votng for g s gven by E g = n pg. (Recall that we assumed n the man text that there are no DPA- (SPA-)clents n the south (north), and that there are only non-clents n central and that the number of non-clents n the north and south s neglgble. From ths we can derve for electon results wth the DPA n government: an,g= DPA an,o= SPA e n an,g DPA a E = n,o= SPA DPA e an,g= DPA north: = + = e ESPA n an,g DPA a e = n,o= SPA + β c (Un -Un*) e n E β c (Un -Un*) DPA e β c (Un -Un*) central: = + = e ESPA n e β c (Un -Un*) + as,g= DPA as,o= SPA+ β s(un -Un*) e n a a β (Un -Un*) E = = + DPA south: = + e = e ESPA n as,g DPA as,o SPA β s (Un -Un*) e = = + + s,g DPA s,o SPA s as,o= SPA+ β s(un -Un*) (A2) Puttng these together, wth a DPA government, we have: EDPA ln = da n n, g DPA (-dn- dc) as, o SPA dcβc( Un- Un* ) (-dn- dc) βs( Un- Un* ), = = + + (A3) E SPA whch s eq. (5) n the man text. Eq. (6) can be derved n a smlar way. The regresson equaton for whch we estmate the parameters s gven by eq. (7). The relatonshp wth (5) (.e., A4) and (6) s as follows. The error term n (7) reflects the dfference between the 2

14 t t expected ( E g ) and observed votes ( V GOV ) for the government ( ε t = Vg Eg ln ln t t V o E ). The constant o term n the regresson equaton captures the term n Un* and allows for a non-zero mean of the error t dstrbuton (μ) (e.g., after the DPA governments n 996 and 997: β0 = ( dcβc + dsβs) Un* + μ ). The terms n β 2 nclude clents predspostons (e.g., after a DPA government: a ng, = DPA). tj Fnally, the terms n β are dscussed n the man text. 3

15 For examples from poltcal scence, see Carothers (2002); Damond (2002); Levtsky and Way (2002); and Schedler (2002). 2 Fne examples of such research may be found n Shlefer (997) and Fdrmuc (2000). For smlar research from a poltcal scence perspectve see Kostadnova (2003). 3 As mentoned above, ths method s commonly used n the feld of Publc Choce. For overvews, see Paldam (99) or Mueller (2003). 4 Damond (2002); n partcular, see Table 2, where the Freedom House score for Albana n 200 s For more detals about the Albanan poltcal system, see ICG (200). 6 Lauth (2000: 33) also argues that systems wth clentelstc partes are often observed n envronments where there are also knshp structures such as clans. Ths renforces the dea that Albana s currently characterzed by ths knd of clentelsm. 7 One mght be tempted to conclude from the hypothess that partes should support the other sde because that s where votes are to be ganed. Ths would mean a collapse of clentelsm, however, n the end leadng to a loss of votes n the own regon. 8 Unfortunately, we have not been able to obtan the economc data needed to do the same analyss for the 2005 electons. 9 For an overvew, see Nannestad and Paldam (994) or Mueller (2003). 0 Fdrmuc (2000) concludes that unemployment s the most acute consequence of economc transton from communsm. In addton, Bell (997) concludes for Poland that unemployment consttutes the most mportant economc varable for explanng votng patterns. Therefore, we beleve that our restrcton to unemployment rates as an economc ndcator used to evaluate government polces wll not serously bas our results. Wth data on the fracton of clents n varous parts of the country, we would be able to estmate a model where dfferent voter types exst n each part. Obvously, these data do not exst. But the dea that there are mostly clents n the north and south and mostly non-clents n central regons seems a good approxmaton. The valdty of ths assumpton s supported by our emprcal results. 2 For ease of presentaton, we start by consderng one electon only and droppng the tme ndex t. 3 As a slght abuse of notaton, we use g=dpa and o=spa to ndcate whch party s n government and opposton, respectvely. 4 Ths dervaton s smlar to that n Krchgässner (985). For more detals about the vote functon, see Schneder and Frey (988) or Mueller (2003). 5 When analyzng the 200 data, we had to deal wth one outler, the southern mountan dstrct, Skrapar. Ths s the brthplace and electoral zone of the socalst prme mnster n ths perod, Meta. In lne wth the clentelsm descrbed, hs government enacted varous polces favorng ths dstrct. Asde from passng out mportant jobs to supporters from hs natve dstrct, Skrapar (ICG 200), he gave prorty to the reconstructon of the man road to ths dstrct, decreasng the travel tme from Trana from about 5.5 to 3.5 hours. As a consequence, support for the SPA was extreme, wth a 85-5 splt between SPA and DPA. Nevertheless, unemployment n ths remote dstrct was by far the hghest n the south, at 25%. Due to these extreme numbers, we drop ths outler from the analyss below. A note, accompanyng table 2, presents the results ncludng Skrapar. 6 Note from the standard devatons reported n table that there does not appear to be any relatonshp between the sgnfcance of coeffcents n the regresson and the varance of unemployment rates n a regon. Hence nsgnfcance does not seem to be caused by very small varances. 7 In a recent study of 27 poltcal systems n postcommunst countres, Armngeon and Careja (2008) conclude that path dependency s an mportant aspect of the lmted nsttutonal change the poltcal systems of these countres have experenced snce the early 990s. 8 Schedler (2002: 37) argues that a characterstc of electoral democraces s that though they manage to run far electons they fal to nsttutonalze other vtal dmensons of democratc consttutonalsm, such as the rule of law, poltcal accountablty, bureaucratc ntegrty, and publc delberaton. 4

16 References Armngeon, K. and Careja, R., Insttutonal change and stablty n postcommunst countres. European Journal of Poltcal Research 47: Bell, J., 997. Unemployment matters: votng patterns durng the economc transton n Poland, Europe-Asa Studes 49(7): Bezemer, D.J., 200. Post-socalst fnancal fraglty: The case of Albana. Cambrdge Journal of Economcs 25/: -23. Case, A., 200. Electon goals and ncome redstrbuton: Recent evdence from Albana. European Economc Revew 45: Carothers, Th.,2002. The end of the transton paradgm. Journal of Democracy 3: 5 2; Damond, L., Thnkng about hybrd regmes. Journal of Democracy 3.2 (2002) Doll, B., The relatonshp between the clan system and other nsttutons n Northern Albana. Journal of Southeast European and Black Studes 3 (2): Fdrmuc, J., Poltcal support for reforms: Economcs of votng n transton countres. European Economc Revew 44 (8): Gajo, A., 999. An assessment of land management and polcy n Albana. Master thess. Insttute of Socal Studes, Den Haag. Gërxhan, K., Poltco - economc nsttutons and the nformal sector n Albana. In: Bezemer, D.J. (Ed.), On eagle's wngs. Ten years of market reform n Albana. New York: NovaScence Publshers: Gërxhan, K., Tax evason n transton: Outcome of an nsttutonal clash? Testng Fege s conjecture n Albana. European Economc Revew 48: Gërxhan, K., and Schram, A., Albanan poltcal-economcs: Consequences of a clan culture. Journal for Insttutonal Innovaton, Development and Transton 4: 5-4. Haefner, L., 999. The role of regons n Albana. Newsletter Albanen (Swss non-proft journal for the cooperaton wth Albana), 24 December 999. ICG (Internatonal Crss Group), 200. Albana s parlamentary electons 200. ICG Balkans Brefng Paper, 23 August 200. Krchgässner, G., 985. Ratonalty, causalty and the relaton between economc con dtons and the popularty of partes. European Economc Revew 28: Klos, A., 997. Albana: Koha Jone Journalsts analyze crss. From the Frankfurt/Man Frankfurter Rundschau, 24 Aprl 997. Kostadnova, T., Voter turnout dynamcs n Post-communst Europe. European Journal of Poltcal Research 42: La Cava, G. and Nanett, R.Y., Albana: fllng the vulnerablty gap. World Bank Techncal Paper no The World Bank. Lauth, H., Informal nsttutons and democracy. Democratzaton 7: Levtsky, S. and Way, L The rse of compettve authortaransm. Journal of Democracy 3 (2): McFaul, M., The fourth wave of democracy and dctatorshp: Noncooperatve transtons n the Post-communst World. World Poltcs 54: MPEP (Mnstry of Publc Economcs and Prvatzaton), 997. The prvatzaton strategy of the companes wth state captal n sectors wth a specal mportance for the economy. Mnstry of Publc Economcs and Prvatzaton, unpublshed manuscrpt. Mueller, D., Publc Choce III. Cambrdge Unversty Press, Cambrdge, U.K. Myerson, R.B., 993. Incentves to cultvate favored mnortes under alternatve electoral systems. Amercan Poltcal Scence Revew 87/4: Nannestad, P. and Paldam, M., 994. The VP-functon: A survey of lterature on vote and popularty functons after 25 years. Publc Choce 79:

17 North, D.C., 990. Insttutons, Insttutonal Change and Economc Performance. Cambrdge: Cambrdge Unversty Press. Paldam, M., 99. How robust s the vote functon? A study of seventeen natons over four decades. In: Norpoth, H., M. Lews-Beck and J.-D. Lafay (Eds.), Economc and Poltcs: The Calculus of Support. Ann Arbor: Unversty of Mchgan Press, 9-3. Sadraj, K., 999. Albana: transton to a market economy. Ph.D. thess. Unversty of Amsterdam, Thess publsher, Amsterdam. Schmdt, M De Westlchen Länder (Lexcon de Poltk, Vol. 3). München: Beck): Schneder, F. and Frey, B.S., 988. Poltco-economc models of macroeconomc polcy. In: Wllet, Th.D. (Ed.), Poltcal busness cycles. Durham: Duke Unversty Press, Shala, A., 997. The Albanan crss A vew from wthn. A Transnatonal Insttute publcaton, Den Haag, The Netherlands Schedler, A The menu of manpulaton. Journal of Democracy 3 (2): Shlefer, A., 997. Government n transton. European Economc Revew 4 (3-5): UNDP (Unted Natons Development Program), Albanan human development report Unted Natons Development Program, Trana. Vckers, M. and Pettfer, J., 997. Albana: From anarchy to a Balkan dentty. Hurst & Co., London, UK 6

18 Tables Table : Electon Results for DPA relatve to SPA Result Unemployment Result Unemployment Result Unemployment North 66.2 (8.).0% (4.2) 48.0 (2.6) 6.5% (8.) 59.5 (6.8) 23.9% (8.6) Central 62.3 (2.4) 3.3% (7.) 26.2 (6.) 8.4% (.5) 40.3 (.4) 4.8% (8.4) South 48.4 (9.6) 9.2% (4.3) 7.8 (7.6) 3.7% (8.0) 35. (9.0) 5.3% (6.0) Total 59. (2.5).3% (5.6) 30.4 (7.7) 6.3% (9.4) 44.7 (3.9) 7.8% (8.6) Notes: standard devatons are n parentheses. Result : votes for DPA as a percentage of votes for DPA and SPA. 7

19 Table 2: Estmated Vote Functons Varable Constant (β 0 ) north β central β south β north β 2 DPA Government SPA Government no polarzatozatozaton no polar- no polar- polarzaton polarzaton polarzaton (0.79) (3.86)* (4.0)* (2.703)** (.729) (0.798) (0.30) (0.66) (2.65)* (3.336)* (.65) (2.048)** (2.460)** (.74) (.455) (3.032)* (3.52)* (0.422) (5.950)* (0.789) (5.373)* (0.836) south β (2.35)** (0.609) R Notes: The coeffcents β are defned n eq. (7). β s the coeffcent related to unemployment, β2 s the coeffcent related to the dstrct the government party s lnked to. For the analyss n 200, we dropped the outler dstrct Skrapar. Ths s explaned n footnote 5. Incluson of Skrapar ncreases the coeffcent β south to (.854). Though large, the coeffcent s stll not sgnfcant at the 5%-level. The coeffcent β south 2 reduces to 0.07 (0.267) and the other coeffcents reman unchanged; *=statstcally sgnfcant at the %- level; **=statstcally sgnfcant at the 5%-level. 8

20 Table 3: Estmated Vote Share of DPA DPA Government SPA Government a b c a b c a b c North 67% 2% 3% 46% +3% 0% 52% +5% -- Central 63% 4% -- 27% 6% -- 42% 4% -- South 49% 7% -- 22% 8% -- 38% 2% +6% Notes: We used the coeffcents n table 2 to estmate the (two-party) vote share of the DPA assumng an unemployment rate of 0% (columns a). We then estmated how ths share would change (n percentage-ponts) f unemployment were to rse to 20% (columns b). Fnally, columns c show the effects of droppng the polarzaton (.e., settng the coeffcents β 2 equal to zero). All coeffcents (ncludng non-sgnfcant ones) were ncluded n the estmatons. 9