Essays on Party System Institutionalization in East-Central Europe

Transcription

1 Essays n Party System Institutinalizatin in East-Central Eurpe Dissertatin Presented in Partial Fulfillment f the Requirements fr the Degree Dctr f Philsphy in the Graduate Schl f The Ohi State University By Jasn William Mrgan, M.A. Graduate Prgram in Plitical Science The Ohi State University 2015 Dissertatin Cmmittee: Janet M. Bx-Steffensmeier, C-Advisr Irfan Nruddin, C-Advisr Luke J. Keele William Minzzi

2 c Cpyright by Jasn William Mrgan 2015

3 Abstract This dissertatin presents three essays explring the nature f party system institutinalizatin in pst-cmmunist East-Central Eurpe. While the develpment f rbust party systems is ften seen t be critical t healthy demcracies, party systems in this regin have been slw t develp. Instead, party plitics in these cuntries is characterized by high levels f fragmentatin, electral vlatility, and a lack f partisan attachment n the part f bth vters and elected elites. The essays cntained in this dissertatin tuch n several f these issues, cntributing t the theretical and empirical understanding f party plitics in East-Central Eurpe. In ding s, the verarching theme in these essays is that mre prgress is being made than is ften recgnized in the literature, but that the nature f this prgress is different than what the experience f Western Eurpe wuld have us expect. In the first essay, I lk at the phenmenn f party switching in the Plish Sejm. In ding s, I develp a new statistical methd fr analyzing dynamic scial netwrk data. The latent path mdel I prpse differs frm previus mdels fr lngitudinal netwrks by allwing the explicit mdeling f trends in the mvement f actrs in the latent scial space. I apply this mdel t the netwrk f party switching by Plish members f parliament and shw that, while there were nearly 1,100 instances f party switching by sitting members f parliament during the first five parliamentary terms, rather than being a symptm f cntinued party system fragmentatin, switching ii

4 has played a cnstructive rle in the Plish party system by allwing pliticians the flexibility t srt themselves int mre idelgically hmgeneus parties. The secnd essay, c-authred with Paul DeBell, addresses the nature f plitical idelgy and its relatinship t individual psychlgical mtives in the Czech Republic, Hungary, Pland, and Slvakia. Applying item respnse thery and a seemingly unrelated regressin mdel, we shw that while left-right self-placement in these cuntries des nt adhere t expectatins gleaned frm the Western plitical experience, the meaning f left and right are far frm randm. Instead, we shw that the meaning f the left-right idelgical labels is cnditined by the specific appeals made by plitical parties. In the third and final essay, I investigate the link between lcal electral institutins and party building in new demcracies. I argue that, as is the case with natinal electral systems, lcal institutins can play an imprtant rle in party system institutinalizatin. Specifically, by shaping the incentives fr lcal pliticans t jin natinal parties and fr natinal parties t get invlved in lcal plitics, different lcal electral institutins can inhibit r encurage natinal party penetratin int lcal plitics. I supprt this argument with an analysis f an unique electral system discntinuity in Pland. My analysis, which applies a regressin discntinuity design apprach, shws that a change frm plurality electin rules t prrtinal representatin had a significant effect n natinal party perfrmance in lcal electins in Pland. iii

5 This thesis is dedicated t my daughter, Evelyn, and t my parents, Pamela and Richard. iv

6 Acknwledgments This dissertatin is nly pssible because f the supprt and encuragement f numerus peple at The Ohi State University. I especially want t thank my c-chairs, Jan Bx-Steffensmeier and Irfan Nruddin fr their cnsiderable patience, and fr giving me the latitude t pursue an eclectic and decidedly nnlinear research prgram. I wuld als like t thank Luke Keele fr his excellent intrductin t the statistical analysis f plitics. I can t imagine a better intrductin. He taught me the skills necessary t explre the variety f statistical appraches utilized in this thesis. I want t extend a heartfelt thanks t William Minzzi fr spending an inrdinate amunt f time cunseling me thrugh the research prcess and prviding me supprt and encuragement when I mst needed it. I am als extremely grateful fr all f the supprt and inspiratin I have received frm my clleagues in the plitical science department at Ohi State. I am particularly indebted t Paul DeBell, my c-authr n Chapter 3, wh has been an invaluable cllabratr and astute critic thrughut this prcess. I thank Peter Tunkis fr his insight int the arcane wrkings f Plish plitics as well as his willingness t brave the Plish bureaucracy in a futile search fr lcal electins data. My thanks als g ut t Austin Knuppe and Drew Rsenberg fr their cnsiderable feedback n Chapter 2 and related wrk. Austin and Drew, alng with Jsé Frtu and Anna Meyerrse, shuld als be thanked fr their cnfunding ability t ask me the questins I wuld v

7 rather nt be asked. Answering them, r at least attempting t, has made me a better teacher and schlar. Lastly, n dissertatin can be cmpleted withut a healthy share f cmmiseratin. Fr this I we a debt f gratitude t my friends and fellw happy hur crew members, including Dustin Carnahan, Alex Castill, Jessy Defenderfer, Lauren Ellitt, Carlina and Kevin Fresman, Margaret Hansn, Carlyn Mrgan, and Rbert Mulcahy. I wuld have never achieved this milestne withut the uncnditinal supprt f my family and friends. I particularly want t thank my parents, Pamela and Richard Mrgan fr their unquestining supprt f my academic pursuits, even when there appeared t be n end in sight. Likewise, I need t thank Devn Mrgan, my brther, Tim Mrgan, and my sister-in-law, Jan Mrgan, fr supprting me thugh this prcess. A special thanks ges t Rni and Steve Gdstein, wh have prvided a slid fundatin f encuragement and cheer ver the years. Finally, I need t thank my daughter, Evelyn, fr spending the first fur years f her life patiently waiting fr me t be finished. T her, I say: Yes, my dear, it s Dr. Daddy nw. vi

8 Vita Brn Salem, Oregn USA B.A. in Ecnmics and German Studies, Lewis & Clark Cllege M.A. in Plitical Science, Bstn Cllege Fields f Study Majr Field: Plitical Science Studies in: Quantitative Methds Cmparative Plitics vii

9 Table f Cntents Page Abstract Dedicatin Acknwledgments Vita List f Tables ii iv v vii xi List f Figures xiii 1. Intrductin Explaining Parties and Party Systems Existing Theries Explanatins f Pst-Cmmunist Party System Develpment Weakness f Existing Theries Chapter Overview The Latent Path Mdel fr Dynamic Scial Netwrks with an Applicatin t Party Switching in Pland Intrductin Pland s Party System Party Switching as a Netwrk in Pland The Latent Space Mdel Netwrks: Terminlgy and Representatin The Latent Space Mdel fr Netwrk Data The Prblem with Dynamic Netwrks The Latent Path Mdel viii

10 2.4.1 Identificatin and Estimatin Cmparisn t ther Mdels An Analysis f Party Switching in the Plish Sejm The Data Mdel Specificatin Parameter Identificatin Mdel Estimatin Results Cnclusin Valuing Left and Right: Party Cmpetitin and Plitical Idelgy in East- Central Eurpe Intrductin Fundatins f Plitical Idelgy and the Cncepts f Left and Right Plitical Idelgy and Left-Right in Pst-Cmmunist Eurpe A Heuristic Explanatin fr Left-Right Orientatins in ECE The Psychlgical Crrelates f Left and Right Data, Measurement, and Mdel Specificatin Results Testing the Heuristic Hypthesis Data, Measurement, and Mdel Specificatin Results Cnclusin The Cheap Seats: Party Develpment and Lcal Electral Refrm in Pland Intrductin Lcal Plitics and Natinal Parties in Pst-Cmmunist Cuntries Lcal Elite Incentives Natinal Party Incentives Observable Implicatins The Test Case: Lcal Electral Refrm in Pland The 2002 Plish Lcal Electins PR and Lcal-Level Natinal Party Perfrmance Design Results Rbustness Checks Reginal Party Participatin in Lcal Electins Cnclusin ix

11 5. Cnclusin Appendices 158 A. Appendix: Latent Path Mdel A.1 The Plish Electral System & Electin Results A.1.1 Summary f Electral Rules A.1.2 Plish Electin Details and Results: A.2 Cnvergence Statistics A.3 Surce Cde fr Stan Mdel B. Left-Right B.1 Schwartz PVQ B.2 Descriptive Statistics B.3 IRT Mdel and Results B.4 Cding Values frm the CMP B.5 Stan Cde fr Basic SUR Mdel x

12 List f Tables Table Page 2.1 Overview f Plish Parliamentary Electins Results, Number f Active Parties, MPs, and the Number f Switches by Term in the Plish Sejm, Parliamentary Terms Distributin f Shared Parties by Time Perid Bx Cnstrains n the Latent Lcatins fr the Reference MPs Prirs fr Intercepts, Decay Parameter, Latent lcatins, and Trajectries fr all ther Parameters in the Latent Path Mdel Descriptive Statistics fr Plish Municipalities Distributin f Party Activity Effects f a Change in Lcal Electral Institutins n Natinal Party Share f Lcal Cuncil Seats Effects f a Change in Lcal Electral Institutins n Natinal Party Share f Lcal Cuncil Seats Ppulatin and Ppulatin Grwth f Municipalities Clse t the Cut- Pint, Activity f Reginal Parties and Perfrmance Relative t Natinal Parties by Municipal Ppulatin, 2002 Plish Lcal Electins A.1 Summary f Electral Rules fr Electins t the Plish Sejm, xi

13 A.2 Dates f Lcal, Parliamentary, and Presidential Electins and Natinal Referenda in Pland A.3 Electin Results fr the Plish Sejm, A.4 Electin Results fr the Plish Sejm, A.5 Electin Results fr the Plish Sejm, A.6 Electin Results fr the Plish Sejm, A.7 Electin Results fr the Plish Sejm, A.8 Electin Results fr the Plish Sejm, A.9 Electin Results fr the Plish Sejm, B.1 Schwartz Prtrait Values Questinnaire, Eurpean Scial Survey B.2 Summary Statistics f Left-Right Self-Placement, ESS B.3 Summary Statistics fr the Main Variables f Interest xii

14 List f Figures Figure Page 2.1 Number f Changes in Party Affiliatin by Day in the Plish Sejm, Parliamentary Terms Netwrks f party switching fr each parliamentary term: Pling a Dynamic Netwrk Analyzing a Dynamic Netwrk Separately Estimated Latent Psitins fr Reference MPs, Piechta, Kmrwski, and Pawlak Psterir Distributin f Trajectry Decay Parameter Estimated Latent Lcatins as Prduced by the Latent Path Mdel Cmparing Estimated Latent Lcatins f MPs t Measures f Idelgy as Prvided by the Chapel Hill Expert Survey (2002, 2006) The Relatinship Between Left-Right Self-Placement and Estimated Values Orientatins in Eurpe, ESS Explaining Respndent-Party Left-Right Agreement in Czech Republic Explaining Respndent-Party Left-Right Agreement in Hungary Explaining Respndent-Party Left-Right Agreement in Pland Explaining Respndent-Party Left-Right Agreement in Slvakia xiii

15 4.1 Plish Municipalities Affected by the Change in Electral System, Distributin f Natinal Party Cuncil Seat Share, 2002 Plish Lcal Electins Regressin Discntinuity Plts McCrary Test fr Ppulatin Srting at the Cut-Pint Lcal Cuncil Seat Share Plts fr Lcal, Reginal, and Natinal Parties137 A.1 Trace and Density Plts fr Lg-Prbability A.2 Trace and Density Plts fr Decay Parameter A.3 Distributin f the Ptential Scale Reductin Factr fr Estimated Latent Psitins and Trajectries B.1 Cntinuus Measure f Respndent-Party Agreement n Left-Right Scale171 B.2 Mean Factr Scres fr the Fur Values Dimensins, ESS xiv

16 Chapter 1: Intrductin Plitical parties play a fundamental rle in mdern demcracy, mderating and channeling public pinin, mbilizing vters, scializing citizens and pliticians t the demcratic system, acting as valuable prgrammatic labels fr vters, decreasing the cst f running fr electin, and facilitating gverning (Aldrich 1995; Lewis 2000, p. 157). Cnsidering their multifaceted rles, strng parties and party systems are cnsidered integral t the histric develpment and stability f mdern demcracy (Huntingtn 1968). Sme have gne s far as t say that mdern demcracy is unthinkable save in terms f parties (Schattschneider 1942, p. 1) This dissertatin presents three essays n party system develpment in pstcmmunist East-Central Eurpe. In the cmparative parties literature, the verall view f party systems in this regin is that they remain fluid and inchate, characterized by high levels f electral vlatility, gvernment turnver in parliament, a lack f partisan attachment n the part f plitical elites and vters, a lack f party develpment, and general uncertainty (e.g., see Bakke and Sitter 2005; Bielasiak 2002; Cnrad and S. N. Glder 2010; Kpecký 1995; Lewis 2000; Shabad and Słmczynski 2004). While these stylized facts are well-established in the literature, the brad theme in the essays that fllw is that things are nt as bad as they may at first appear. Instead, in studies lking at three phenmena typically indicative f weak 1

17 party systems in pst-cmmunist demcracies pervasive party switching by sitting members f parliament, the cnfused nature f plitical idelgy, and the lack f party building in the regin I shw that, in fact, mre prgress is being made than is ften recgnized in the literature. In ding s, these essays cntribute t the theretical and empirical understanding f party plitics in East-Central Eurpe, and plitical parties mre generally. They shw that while the nature f prgress may be smewhat different than what the experience f Western Eurpe wuld have us expect, it remains prgress nnetheless. These studies are described in mre detail in Sectin 1.2; hwever, befre prviding a mre detailed summary f each chapter, the next sectin prvides sme backgrund n the appraches t studying parties and party systems, bth in general and mre specifically in East-Central Eurpe, and then ges n t discuss sme f the limitatins f current research with respect t understanding party systems in East-Central Eurpe. 1.1 Explaining Parties and Party Systems This sectin prvides a brief verview f the research explaining parties and party systems. This is nt meant as a cmprehensive literature review; rather, it is meant t set the stage fr the research presented in Chapters Existing Theries The literature n the rigins and develpment f plitical parties and party systems is anchred by three theretical appraches. 1 These appraches, which I refer 1 This review glsses ver the imprtant distinctin between plitical parties as singular rganizatins and party systems as patterns f interactins between parties (Mair 1997, pp. 5 6). A mre cmplete review wuld make this distinctin explicit, but ding s is nt critical t identifying the key theretical appraches present in the literature. 2

18 t as the scilgical, institutinal, and individual-ratinal mdels f party develpment, 2 were riginally develped t explain the prcess f party system develpment in Western demcracies, thugh their applicatin t new demcracies, including pstcmmunist cuntries, has been imprtant in shaping the literature in these cntexts as well. The scilgical mdel, best articulated by Lipset and Rkkan (1967), sees plitical parties and party systems as reflectins f underlying scial cleavages. In the West these cleavages were the result f centuries f plitical and scial cnflict, ultimately resulting in the class-based party systems that came t define Western demcratic plitics. In ther scieties, plitically salient cleavages may be defined alng ethnic r religius lines. The scilgical mdel is a macr-level thery that des nt distinguish between the plitical preferences f individuals and the preferences f scial grups as a whle. These grups are assumed t be made up f hmgeneus individuals which, frm an analytical perspective, means each grup can be represented by a single set f plitical preferences. In this mdel, plitical parties are the means thrugh which these hmgeneus grup interests are represented in gvernment. Cnsequently, the scilgical mdel allws little rm fr electral strategy. Party prgrams are centered n plicies seen t benefit the grup they represent. This prgram is knwn and unambiguus, making vters chices als knwn. Fr parties, electral plitics 2 In his review f the schlarship n party develpment, Bix (2007) divides the literature int tw grups f theries: histrical-scilgical and ne-institutinal. The prblem with this categrizatin is that it reflects an assumptin that the institutinal and ratinal appraches are necessarily tied tgether. The review belw shws that this assumptin mischaracterizes the literature, and that it is nt unusual fr schlars t pair the scilgical and institutinal (Duverger 1954) r the scilgical and ratinal (Zielinski 2002) mdels in their attempts t explain party system develpment. 3

19 revlve arund mbilizatin, nt n articulating a prgram that will attract the mst vtes (Katz and Mair 1995, pp. 6 7). The institutinal mdel fcuses instead n the rle f plitical institutins, such as electral rules and cnstitutinal arrangements, in shaping party systems. Fllwing the seminal wrk by Duverger (1954), it is cmmn fr schlars lking at the imprtance f electral rules t assess the mechanical and strategic r psychlgical effects different rules have n the develpment f party systems (fr a detailed discussin, see Clark and M. Glder 2006). In cntrast t scilgical arguments, the institutinal apprach prvides scant insight int the rle f parties in integrating individuals int the demcratic system r even explaining hw individuals and pliticians chse between different parties (Amrim Net and Cx 1997, p. 152). On the ther hand, institutinal theries d prvide clear expectatins abut hw the patterns f party cmpetitin e.g., the number f parties cmpeting in electins depend n a cuntry s electral rules and plitical institutins (see Cx 1997). The individual-ratinal apprach t explaining party systems fcuses n individual vters and pliticians, and the underlying assumptin f this apprach is individual ratinality. Develped largely in the American plitics literature, this apprach has been applied t investigate such questins as why pliticians decide t start r jin parties (Aldrich 1995), and why they chse t maintain a tw-party system (Riker 1982, p. 764). The dminant explanatin f vte chice is prvided by Dwns (1957) 4

20 thery f spatial vting: the idea that vters chse candidates r parties clsest t their individual ideal pints. 3 In the individual-ratinal mdel, parties are explained as prviding imprtant benefits t pliticians and vters. Fr instance, Aldrich argues that parties prvide a mechanism fr vercming the prblem f cllective actin in the legislature: while jining a party may require pliticians t smetimes frg shrt-term gains, ver time the benefits f belnging t a stable legislative calitin ut-weigh the csts (Aldrich 1995, p. 28). Fr vters, parties act as valuable prgrammatic labels that decrease the infrmatin csts assciated with vting. The scilgical, institutinal, and individual-ratinal mdels f party cmpetitin are sme f the mst sphisticated and well-develped in plitical science, drawing n decades f careful theretical and empirical research investigating party develpment in bth demcratic and nn-demcratic settings. Hwever, these theries are nt necessarily stand-alne explanatins that exclude the imprtance f their intellectual cmpetitrs. Schlars rarely apprach the study f party develpment frm such an extreme mncausalist perspective (Amrim Net and Cx 1997, p. 151). Instead, features f tw r mre appraches are ften cmbined t mre clsely fit with plitical reality and t address the inherent limitatins f the different theries. Fr instance, Duverger s (1954) wrk itself reflected a cmprmise between the scilgical and institutinal appraches. In particular, he saw scilgical factrs specifically, ethnic hetergeneity as prviding a set f plitical pssibilities, which institutinal arrangements then translated int bserved plitical 3 While dminant in the ratinal chice literature, the spatial mdel f vte chice des nt g unchallenged. Fr instance, see Iversen (1994) fr a mdel that cmbines spatial and directinal appraches. 5

21 utcmes (see the discussin in Clark and M. Glder 2006, p. 685). Subsequently, this interactive relatinship between ethnic hetergeneity and plitical institutins has becme an imprtant line f research (see, e.g., Ordeshk and O. V. Shvetsva 1994; Amrim Net and Cx 1997; Clark and M. Glder 2006). Chhibber and Kllman (2004) als prvide an excellent example f research that melds the three general theries. In their investigatin f the natinalizatin f party systems in the United States, Canada, India, and the UK, they argue that the distributin f authrity acrss different levels f gvernment interacts with scial cleavages, electral rules, [and] plitical entrepreneurs t explain the change in party systems ver time (2004, pp ) Explanatins f Pst-Cmmunist Party System Develpment The scilgical, institutinal, and individual-ratinal theries explaining party system develpment have played an imprtant rle in the study f pst-cmmunist demcracies. Hwever, the direct applicatin f these theries riginally frmulated t explain Western demcracies t the pst-cmmunist cntext was nt withut cntrversy, as can be seen in the debate between Bunce (1995) and Schmitter and Karl (1994). Despite this early cntrversy, hwever, schlars have fund the insights these theries bring t the study f party systems in pst-cmmunist demcracies t be useful. In the aftermath f the cmmunist cllapse, schlars fllwing the scilgical mdel were particularly quick t begin searching fr the cleavages that wuld shape pst-cmmunist party system develpment. Tucker (2002, p. 292) summarizes this search as an attempt by schlars t refute a null hypthesis that the extreme 6

22 uncertainty f the plitical and ecnmic transitin wuld be s strng as t verwhelm the imprtance f any existing cleavages. Tw views were cmmn amng schlars. In the first grup are thse schlars that believed the lifting f the ppressive hand f cmmunism wuld allw the reemergence f latent cleavages suppressed during cmmunist rule and that this wuld lead t plitical cnflict. Latent religius and ethnic cleavages in particular were expected t shape pst-cmmunist plitics. Wittenberg (2006), fr instance, ties pst-cmmunist vte patterns in Hungary t religius cleavages present in pre-cmmunist Hungary. Jasiewicz (1993) similarly finds the secular-religius divide t be a key cleavage in Pland. The secnd grup f schlars expected that new scial divisins wuld emerge t structure party plitics. Fr example, Kitschelt (1992) prvides a discussin f the dimensins alng which cleavages in new demcratic scieties can take rt; namely, arund the issue f wh is admitted t participate, what the rules f plitical cmpetitin will lk like, and what assets players are endwed with. Kitschelt argues that cnflict ver the first tw f these differences will be mre detrimental t plitical stability than cnflict ver material assets (i.e., redistributin). In Western demcracies, it is the latter f these cleavages that mst plitics revlves arund. In Central and Eastern Eurpe, at least at the mment f transitin, plities face questins arund all three. Evidence fr the emergence f an assets-based cleavage structure is prvided by Słmczynski and Shabad, where they argue that [t]he mst bvius aspect f change is a new class cmpsitin resulting frm privatizatin f the ecnmy and its cnsequences (1996, p. 187). 7

23 The rle f institutins in shaping party system develpment in pst-cmmunist cuntries has likewise been an active area f research almst frm the mment cmmunist regimes cllapsed. This research has, fr instance, lked at the rle f electral institutins in determining the number f parties participating in natinal plitics (Ordeshk and O. V. Shvetsva 1994; Amrim Net and Cx 1997) and n vter vlatility (Tle 2000). Brader institutinal cntexts, such as the territrial rganizatin f a cuntry (Bschler 2010), have als been fund t cntribute t party system natinalizatin. The ecnmic vting literature prvides the clearest example f the applicatin f the individual-ratinal apprach t party system develpment in pst-cmmunist cuntries, and there is great deal f evidence that vters in the regin use party labels and the assciated ecnmic platfrm as a useful heuristic in deciding which party t vte fr (Pwers and Cz 1997; Tucker 2006; Jacksn, Klich, and Pznańska 2005). Other literature fcuses n the rle f elites in shaping pst-cmmunist party systems, investigating the degree t which pliticians are able t avid being accuntable t vters fr pr perfrmance (Zielinski, Słmczynski, and Shabad 2005), as well as the imprtance f elites in shaping plitical discurse t emphasize particular underlying cleavages and de-emphasize thers (Zielinski 2002) r t frmulate brad integrative narratives in rder t facilitate party cnslidatin (Hanley et al. 2008, p. 408). In additin t the three main theries discussed abve, schlars f pst-cmmunist party develpment ften emphasize the imprtance f histric legacies in shaping pst-cmmunist demcratizatin. Legacy arguments cme in tw flavrs: thse that 8

24 emphasize the detrimental effects f cmmunism the s-called Leninist legacy and thse that emphasize the circumstances f the demcratic transitin. Jwitt prvides an early and pessimistic accunt f what the Leninist legacy held in stre fr the regin when he argues that the cmmunist perid engendered a wrld view in which plitical life is suspect, distasteful, and pssibly dangerus; t be kept at bay by dissimulatin, made tlerable by private intimacy, and transcended by private virtues r charismatic ethics (1992, p. 215). The ultimate results f this distaste fr and fear f plitics, Jwitt argues, wuld be the rise f demaggues, priests, and clnels, which wuld lead the regin t adpt a liberal authritarianism rather than liberal demcratic frms f gvernance (1992, pp ). In cntrast t Jwitt, Grzymała-Busse (2002) presents a mre psitive legacy argument when she describes the ability f frmer cmmunist parties t reinvent themselves after the transitin t demcracy. Grzymała-Busse specifically argues that the ability f cmmunist parties t reinvent themselves was predicated n the actins and the structure f the cmmunist parties during the cmmunist perid. In thse cuntries where the cmmunist party was clser t sciety, such as in Pland and Hungary, parties were themselves mre pen t refrm after the transitin. They likewise had mre legitimacy in the eyes f the public and, thus, culd becme legitimate cmpetitrs in subsequent demcratic electins. In additin t Grzymała-Busse s wrk, ther schlars have fund cmmunist legacies t be imprtant in explaining gegraphic patterns f supprt fr successr parties in Pland (Lubecki 2004) and fr defining the main cleavages shaping party cmpetitin in pst-cmmunist cuntries mre generally (Rivera 1996). 9

25 The experience f transitin has als been imprtant in explanatins f party system develpment. The ecnmic vting literature, fr instance, has established a clear cnnectin between the hardships created by the ecnmic transfrmatin and individuals later vte chice. Cnsistent with expectatins, this research has shwn that mre vulnerable ppulatins were less likely t supprt parties pushing rapid ecnmic and scial refrms (see Tucker 2006; Owen and Tucker 2010). Schlars have als fund that perceptins f the transitin shape views twards the new demcratic regime (Pwers and Cz 1997). Influenced by the experiences f Suthern Eurpe, schlars have als lked at the relatinship between the mde f transitin e.g., whether the transitin was negtiated r the result f regime cllapse and party system develpment in pst-cmmunist cuntries (Kitschelt 1995). Tavits (2005) prvides an alternative view f party system institutinalizatin emphasizing what she called demcratic maturatin the idea that we shuld expect greater vlatility in the early years after transitin and that this vlatility shuld decline ver time. Similarly, Tavits and Annus (2006) find that strategic vting, as measured by the share f wasted vtes in natinal electins, decreases ver time in pst-cmmunist cuntries, suggesting time and party system institutinalizatin are clsely linked. And finally, Tavits (2007) finds that the number f new parties being created in these cuntries declines ver time, due t increasing entry csts fr new parties Weakness f Existing Theries Existing appraches t explaining party system develpment suffer frm sme significant weaknesses, which becme particularly apparent when put in the cntext 10

26 f pst-cmmunist demcracies. A general weakness afflicting the scilgical and individual-ratinal mdels is the tendency in each case t treat scial grups and plitical parties as unitary actrs. Fr scilgical theries, treating scial grups and parties as unitary actrs is a natural extensin f the thery: class cnsciusness guides individual actin fr elites and vters which is directly reflected in the actin f parties. Fr individual-ratinal theries, treating scial grups and parties as unitary actrs is a cnvenience adpted fr tractability. In Dwnsian spatial mdels f vting, fr instance, parties are assumed t have a single psitin n plicy, where any differences in psitin amng party members is assumed away by treating the median psitin f members as the verall party psitin. Withut this simplifying assumptin it is difficult t determine if vters vte accrding t their interests. In frmal mdels, tractability becmes even mre imprtant since analytic slutins are rarely available nce many hetergeneus actrs and strategic behavir are cnsidered. Unfrtunately, treating scial classes and parties as unitary actrs is a simplificatin that risks impsing a degree f structure n plitics that did nt exist in pst-cmmunist cuntries. As Przewrski nted, citizens f pst-cmmunist cuntries were atmized and cynical (Przewrski 1991, p. 2), which likely played an imprtant rle in the prcess f party system institutinalizatin. Atmized citizens are unlikely t have clse ties t either classes r parties; cynical citizens are unlikely t believe what elites tell them. Cmbine these traits and yu are likely t find an uninfrmed and unpredictable electrate. Such an electrate shuld nt be treated as if it can be easily divided int cherent subgrups fr cnvenient analysis. Mving n t mre specific criticisms f existing theries, there are three reasns t be wary f the direct applicatin f the Western scilgical mdel t the 11

27 pst-cmmunist cntext. First, latent cleavage explanatins seldm accunt fr the flattening f sciety that tk place under the cmmunist regimes (Mair 1997, ch. x). 4 When making such an argument schlars shuld prvide an explanatin f hw pre-cmmunist scial cleavages managed t remain intact during the cmmunist perid despite the eliminatin f the scial and ecnmic institutins that under-girded them. This is seldm dne (an exceptin being Wittenberg 2006). Secnd, it is likewise prblematic t assume that the scial cleavages s imprtant t party system develpment in Western demcracies will necessarily be imprtant t party system develpment and pst-cmmunist cuntries. Fr example, scilgical explanatins f party system develpment in the West rely heavily n particular patterns f industrializatin, demcratizatin, and develpment f the market ecnmy. These patterns cannt be matched in pst-cmmunist demcracies (Mair 1996; Birch 2003). Third, the scilgical apprach depends n the existence f sme mechanism thrugh which members f a grup r class identify their interests. In the West, class membership came with a set f clear scial and institutinal ties, which made it relatively easy t identify ne s interests; e.g., unins, religius rganizatins, and business relatins were key in ding s. In pst-cmmunist cunties, where citizens were atmized and made cynical by decades f cmmunism (Przewrski 1991, p. 2), scial cnnectins are weak and membership in institutins rare. As a result, it is difficult t see hw individuals can identify thse parties that best serve their interests. 4 This is nt t say that the flattening f sciety during the cmmunist perid resulted in perfectly hmgeneus scieties. As Słmczynski and Shabad pint ut, making this assumptin bscures the nature, degree, and cnsequences f scial differentiatin in these scieties, bth befre and after the nset f systemic change (Słmczynski and Shabad 1996, p. 188). Hwever, I think it is clear that relative t Western cases, where the scilgical mdel was first elabrated, pst-cmmunist scieties were largely flattened. 12

28 There are three reasns t dubt the ability f the institutinal apprach t prvide an adequate explanatin fr party system develpment in pst-cmmunist cntexts. First, institutinal arguments suffer frm their inability t prvide specific guidance n what shuld be expected frm a party system; in ther wrds, which parties are likely t prsper and which are likely t be eliminated frm the party system. Being able t identify the parties mst likely t survive is imprtant because it prvides an imprtant clue as t whether centripetal r centrifugal frces will be mst at wrk in the party system. The existence f anti-system parties an example f a highly centrifugal frce may suggest an inherent weakness f the party system (Sartri 1976). Secnd, while institutinal explanatins may be useful in prviding bunds n party system parameters such as the number f effective parties they cannt explain imprtant characteristics f party system institutinalizatin, such as party turnver and electral vlatility. This is because the institutinal apprach des nt prvide any insight int the cntent f plitics; i.e., the apprach des nt tell us why peple vte fr particular parties r why pliticians take particular plicy psitins. This is a critical missin because it means that the thery cannt explain why, given the same institutins, ne party system may exhibit centrifugal tendencies while the ther des nt. 5 Third, and perhaps mre imprtant t the study f party system develpment in develping demcracies, the assumptin that cnstitutinal and electral institutins can be treated as exgenus factrs shaping party systems is highly prblematic (Kitschelt 1992, p. 9). Treating institutins as exgenus ignres the fact that in 5 See Sartri (1976) fr a discussin f party systems demnstrating centrifugal and centripetal tendencies. 13

29 new demcracies institutins are mre malleable and are likely t be changed by pliticians seeking t imprve their plitical frtunes. The direct applicatin f the individual-ratinal apprach t pst-cmmunist party develpment is als nt withut prblems, the chief issue being the vast uncertainties that pervaded transitin plitics. As Bunce and Csanádi described the situatin: what Eastern Eurpeans gained frm 1989 was uncertainty in plitical, ecnmic, and scial utcmes the cntributins f liberalizatin but what they did nt gain were all f thse elements f certainty that make risk tlerable and functinal (Bunce and Csanádi 1993, p. 267). These uncertainties make it extremely difficult fr schlars t judge whether r nt plitical actrs have made utility maximizing decisins cnsistent with the available infrmatin. There are tw prblems with hw legacy explanatins are ften presented. First, schlars ften neglect t identify the mechanism cnnecting past and future. In ther wrds, they d nt ffer an explanatin f hw what happened in the past is transmitted int the future. Sme analyses effectively regress an utcme f interest here it is party system utcme n a particular legacy such as the flattening f sciety that ccurred during the cmmunist perid. But a lt can happen between the histric cause and future effect, and withut identifying the mechanism it is difficult t assess whether a particular legacy is driving a particular future utcme r whether smething ccurred in the meantime, making the apparent relatinship between the legacy and utcme spurius. Withut a clear mechanism, a lng lag time between histric cause and future effect is particularly prblematic. 6 6 When a cnvincing mechanism is prvided, such as in Grzymała-Busse s (2002) detailed tracing f the evlutin f cmmunist party rganizatins and their relatinships with sciety, legacy arguments can be quite persuasive. 14

30 The secnd prblem with legacy explanatins is that they rarely specify when past histric circumstances shuld n lnger matter. This prblem is clsely related t the prblem f failing t identify the mechanisms f transmissin: when an analyst presents a legacy argument with a clear mechanism, it is, at least in principle, pssible t identify when the legacy n lnger matters: when the transmitting mechanism can n lnger be traced, the legacy is n lnger a critical factr in shaping future plitics. Hwever, it may nt be the case that the terminatin f the mechanism is perfectly crrelated with the end f the imprtance f legacy. The mechanism may cntinue t carry n, but it is imprtance may be limited by ther factrs. When making an histric legacy argument schlars need t indicate the circumstances under which the legacy will n lnger matter. Otherwise they may run int the prblem f attributing everything t it. 1.2 Chapter Overview Chapter 2 f this dissertatin lks at the phenmenn f legislative party switching by sitting members f the Plish Sejm. A high rate f party switching by pliticians is typically expected t inhibit party system institutinalizatin by reducing demcratic representatin, accuntability, and the heuristic value f party labels (Despsat 2006; Heller and Mershn 2005, 2009; Mainwaring 1998). Hwever, in this chapter I argue that in new demcracies, where scial cnnectins and party labels are weak, party switching allws pliticians the flexibility t srt themselves int mre chesive grups, ultimately cntributing t an increased likelihd f lng-term party system stability. 15

31 In rder t test this hypthesis and investigate the phenmenn f party switching in the Plish Sejm mre clsely, this chapter develps a new latent variable mdel fr dynamic netwrk data. The prpsed latent path mdel is a natural extensin f the latent space mdel fr static netwrks develped by Hff, Raftery, and Handcck (2002) and is in the spirit f the dynamic netwrk mdel f Ward, Ahlquist, and Rzenas (2013), Sewell and Chen (2015), and Sarkar and Mre (2005). In applying this mdel t party switching in Pland, which has seen mre than 1,100 instance f party switching since the first demcratic electin in 1991, I shw that instead f being an unqualified indicatr f persistent party system weakness, switching during the first five parliamentary terms has resulted in greater idelgical cherence f parties, and that a cre grup f 199 lng-serving MPs has been at the leading edge f this cnvergence. This cunterintuitive result suggests party switching can smetimes play a mre cnstructive rle in party system institutinalizatin than is typically realized, while it als suggests that the Plish party system may be develping the fundatins f a strnger and mre stable party system. Chapter 3, authred with Paul DeBell, shifts the fcus frm the actins f elected elites t the psychlgical mtivatins f individuals by investigating the nature f plitical idelgy and its relatinship t individual psychlgical mtives in the Czech Republic, Hungary, Pland, and Slvakia. Self-placement n the left-right spectrum is ne f the mst cmmnly used variables in studies f cmparative plitical behavir. Hwever, in the first stage f ur analysis, we use graded item respnse mdels and the Schwartz Prtrait Value Questinnaire embedded in the Eurpean Scial Survey t shw that the crrespndence between underlying psychlgical mtives and left-right self-placements adhere t expectatins in Western demcracies but vary markedly 16

32 in East-Central Eurpe. These results cmprt with previus analyses (Aspelund, Lindeman, and Verkasal 2013; Piurk, Schwartz, and Davidv 2011) and raise serius questins regarding the nature and rigins f left-right plitical idelgy amng mass publics in East-Central Eurpe; specifically, whether these idelgical labels functin t link vters t elites in the cntexts f pst-cmmunist demcratizatin. Hwever, we cntend that this variatin is nt randm. Instead, we argue that vters cme t assciate the idelgical terms left and right with the values prmulgated by the majr plitical actrs in their cuntry. In a secnd analysis, we test this Heuristic hypthesis by emplying a seemingly unrelated regressin mdel (Zellner 1962) and cmbining data frm the Chapel Hill Experts Survey (Bakker et al. 2012) and Cmparative Manifests Prject (Vlkens et al. 2014) t shw that the meaning f left and right in the Czech Republic, Hungary, Pland, and Slvakia crrespnds t the specific appeals made by the majr parties f the left and right. In ther wrds, peple identify the values appeals f majr parties, then select their idelgical psitin t match the plitics f their cuntry. This result has imprtant implicatins fr the study f cmparative plitical behavir as it expses a need fr schlars t incrprate mre flexible measures f idelgy int crss-natinal research, instead f relying n a simple and hmgeneus uni-dimensinal left-right cncept. In Chapter 4, I present a theretical and empirical explratin int the link between lcal electral institutins and party building in new demcracies. While recent research has demnstrated the imprtance f lcal party rganizatin t the electral success f natinal parties (Tavits 2012), the imprtance f lcal electral institutins has been largely neglected in the literature. In this chapter, I argue that by prviding different incentives fr lcal pliticians t jin natinal parties which, in 17

33 turn, incentivizes natinal parties t becme invlved in lcal plitics, lcal electral institutins d matter t natinal party develpment in new demcracies. T test my hypthesis, Chapter 4 leverages an unique electral system discntinuity in Pland. In 1998, the Plish gvernment passed a brad package f refrms t the structure and gvernance f lcal gvernments. Fr a select set f municipalities thse between 20,000 and 40,000 residents these refrms included a change in the way they elected their municipal cuncils, mving frm single-member districts with plurality vting t pen-seat prprtinal representatin. Using a regressin discntinuity design apprach (Thistlethwaite and D. T. Campbell 1960), I shw that this change in electral institutin was assciated with an increase f apprximately 18% in natinal party seat share n lcal cuncils. This significant results which is rbust t changes in mdel specificatin, prvides strng supprt fr the idea that lcal electral institutins can affect the ability f natinal parties t penetrate int lcal plitics. 18

34 Chapter 2: The Latent Path Mdel fr Dynamic Scial Netwrks with an Applicatin t Party Switching in Pland Intrductin Schlars have lng viewed a stable party system t be an imprtant cmpnent f a healthy demcracy (Huntingtn 1968; Schattschneider 1942). A key t develping such a party system is the develpment f stable partisan attachments between pliticians and vters (Cnverse 1969; Mainwaring 1999; Mainwaring and Scully 1995). Cnsequently, high rates f party switching by pliticians is ften seen as a clear indicatr f a lack f verall party system institutinalizatin. Party switching can reduce pliticians accuntability t vters, suggest a lack f party rganizatin and discipline, and lessen the heuristic value f party labels fr vters (Despsat 2006; Heller and Mershn 2005, 2009; Mainwaring 1998). 7 I wuld like t thank Jan Bx-Steffensmeier, Paul DeBell, Luke Keele, Austin Knuppe, William Minzzi, Irfan Nruddin, Santiag Olivella, Andrew Rsenberg, and Peter Tunkis fr their numerus helpful cmments. This prject was supprted by the Ohi Supercmputer Center and the Institute fr Ppulatin Research at The Ohi State University. Earlier versins were presented at the Sciety fr Plitical Methdlgy s Annual Summer Methds Meeting ( ) and the Annual Meeting f the Plitical Netwrks Sectin f the American Plitical Science Assciatin ( ). 19

35 The empirical evidence suggests that party switching is damaging t the prcess f party system institutinalizatin in yung demcracies. When plitics are uncertain and vters have yet t learn the cnturs f the new demcratic regime, cnstant switching frustrates attempts by vters t hld their elected fficials accuntable fr pr perfrmance (Zielinski, Słmczynski, and Shabad 2005). Furthermre, high levels f switching can cntribute t disrganizatin in parliament, verall party system fragmentatin (Kreuzer and Pettai 2003), and has been shwn t encurage the self-serving gals f pliticians (Despsat 2006). As Despsat (2006, p. 77) argues, [s]witching effectively destrys the meaning f party labels, raises vters infrmatin csts, and eliminates party accuntability. There are, hwever, theretical reasns t believe that party switching per se may nt be inherently detrimental t party systems. As Heller and Mershn pint ut, the ntin that switching is damaging t demcratic representatin assumes that vters select candidates based slely n their party affiliatin. In ther wrds, party labels are meaningful and, as such, prvide a great deal f valuable infrmatin t vters abut candidates and parties plicy psitins (Heller and Mershn 2009). Yet, switching may play a cnstructive rle if it prvides pliticians the flexibility t take psitins n plicy that mre clsely match the views f their cnstituents. 8 Likewise, when labels carry little meaning, as is the case in new demcracies, switching may play a cnstructive rle by allwing pliticians the freedm t srt themselves int mre chesive grups that reflect the idelgical cnturs f sciety, thus ultimately cntributing t an increased likelihd f lng-term party system stability. 8 As an empirical example, cnsider the partisan realignment in the U.S. after the passage f the Civil Rights Act in 1964, which made the party labels mre clsely reflect the values f the vters in the Suth (Levendusky 2009). 20

36 In this chapter, I take a dynamic netwrk apprach t assessing whether patterns f party switching in Pland indicate a grwing cherence f parties in the cuntry r whether this switching indicates cntinued party system weakness. Pland makes an interesting and difficult case fr the prpsitin that party switching can play a psitive rle in party system develpment. Ever since the first demcratic parliamentary electin in 1991, the Sejm has been plagued by a chrnic tendency fr elected members f parliament t switch parties. During the first five parliamentary terms, there were nearly 1,100 instances f intra-term party switching in the Sejm, the lwer huse, with almst 30% f members f parliament (MPs) switching parties at least nce. As a cnsequence f this switching, mre than 70 different parties served in parliament during this perid. The fragile nature f elite partisan attachments has cntributed t persistent gvernment instability: during the first seven parliamentary terms, Pland had 17 different gvernments, 11 prime ministers, and nly ne gvernment survived t cmplete a full fur-year term (Cnrad and S. N. Glder 2010). 9 By ther cmmn measures f party system institutinalizatin, hwever, the Plish party system has begun t shw sme encuraging signs f stabilizatin. Fr instance, despite the upheaval caused by MP party switching, nly ne new party has been elected t the Sejm during each f the last three electral cycles. Furthermre, in 2011, Civic Platfrm (PO) wn its secnd cnsecutive electin, a first fr Pland, nearly duplicating its 2007 perfrmance. At the same time, Law and Justice (PiS) 9 These figures assume the present gverning calitin between Civic Platfrm (PO) and the Plish Peasant Party (PSL) survives the current term that ends in

37 retained its psitin as the main ppsitin. This greater electral certainty is reflected in declining electral vlatility and lw levels f fragmentatin, and seems t suggest that an enduring and meaningful divisin has emerged in Plish plitics. 10 Frm the perspective f extant thery, as nted abve, the Plish party system raises interesting questins abut the state f institutinalizatin in the cuntry and abut the prcess f party system institutinalizatin in new demcracies mre generally. Lng established theries f party system develpment hld that the advent f stable partisan cmmitments n the part f pliticians and vters is critical t the prcess f institutinalizatin (Huntingtn 1968). At the same time, these theries emphasize the imprtance idelgy plays in structuring party cmpetitin, with stable party systems exhibiting a clse identificatin between particular parties and idelgical psitins (Mainwaring and Scully 1995). The level f party switching seen in Pland clearly suggests a lack f party system institutinalizatin; hwever, mre recent stability in the partisan makeup f electral plitics, and the relatively stable idelgical psitins f thse parties (Markwski 2008; Szczerbiak 2013), suggests grwing cherence in the party system. 11 What explains the discrepancy between plitics at the elite level and that at the aggregate electral level? Why has nt the lack f partisan cmmitments by members f parliament and resultant intra-term party system instability translated int even greater party fragmentatin? Finally, what can past patterns f party switching tell us abut the pssibility f party system institutinalizatin in Pland? 10 Szczerbiak (2013) ntes that this divisin appears t be real, with PO and PiS having becme the main pints f reference fr each ther (Szczerbiak 2013, pp ). 11 Lking at an earlier perid, Shabad and Słmczynski (2004) als nte the presence f switching alngside indicatrs f party system institutinalizatin. 22

38 T date, party switching has been treated as an individual-level phenmenn, whereby switching as an utcme is determined by legislatrs perceptins f the csts and benefits assciated with ding s (Despsat 2009; Heller and Mershn 2005; Laver and Benit 2003; Zielinski, Słmczynski, and Shabad 2005). These benefits f switching include increasing the likelihd f being reelected, btaining rent frm hlding ffice, r achieving sme plicy utcme. While these appraches are fcused n individual-level decisin making, there is an implicit relatinal aspect t the theries in that the benefits t be gained frm switching is cnditinal n the rest f the party system remaining cnstant. Empirically, these appraches have assumed such independence. Hwever, there are reasns t believe that such independence des nt exist. Fr instance, a party switch by ne MP may increase the likelihd that allies in her ld party switch in the future; party disslutin may make it necessary fr many MPs t switch parties simultaneusly; and, in a new party system, a learning prcess may ccur which decreases the prbability f switching ver time, perhaps due t slidifying partisan lines r increasing party discipline. In ther wrds, party switching shuld be seen as a relatinal, dynamic prcess; thus, empirical analyses seeking t understand this prcess require the use f mdels apprpriate fr such data. In this chapter, I develp a new latent variable mdel suitable fr tracing the mvement f members f parliament thrugh a latent scial space ver time. The prpsed mdel, which I call the latent path mdel, builds n the latent space mdels f Hff, Raftery, and Handcck (2002) and is similar t the dynamic latent space mdel f Sewell and Chen (2015). 12 The mdel differs, hwever, frm prir research 12 Als see Sarkar and Mre (2005) fr an earlier develpment alng these lines. 23

39 in that it allws the explicit mdeling f nn-linear trends in the mvement f actrs in the latent scial space; can accmmdate directed, undirected, and weighted netwrks; and fits mre neatly int generalized linear mdels familiar t plitical scientists. I prvide a Bayesian implementatin f the mdel in Stan (Stan Develpment Team 2013). I then apply the prpsed latent path mdel t a cre subset f 199 Plish MPs. I find that switching during this perid, rather than being a symptm f cntinued party system fragmentatin, has resulted instead in greater idelgical cherence f parties. In ther wrds, switching seems t have played a cnstructive rle in the Plish party system by allwing pliticians the flexibility t srt themselves int mre idelgically hmgeneus parties. Overall, these cunterintuitive results suggest that party switching may nt necessarily be a detriment t party system institutinalizatin and demcratic cnslidatin mre generally. Furthermre, frm a brader theretical perspective, the patterns f change seen in Pland prvide insight int the prcess thrugh which idelgically hmgeneus and stable parties develp frm the dynamic interactin f pliticians in new demcracies. In the next sectin, I briefly describe the verall trends in party plitics in Pland since the demcratic transitin in 1991, while als prviding a mre detailed discussin f party switching in the Plish Sejm frm a netwrk perspective. Sectin 2.3 discusses the prblems psed by relatinal data, such as the party switching netwrk, fr standard statistical mdels, and intrduces the latent space mdel previusly develped fr static netwrks (Hff, Raftery, and Handcck 2002). In Sectin 2.4, I then prpse a latent space mdel fr dynamic netwrks. This mdel is applied t the Plish party switching data in Sectin 2.5. Sectin 2.6 cncludes. 24

40 2.2 Pland s Party System Cmpared t ther third wave demcracies, party systems in pst-cmmunist Eastern Eurpe have been slw t institutinalize. Overall, these party systems can be characterized by their cmparatively high levels f fragmentatin, electral vlatility, and general uncertainty (Bakke and Sitter 2005; Bielasiak 2002; Epperly 2011; Lewis 2000). At first glance, this vlatility is unsurprising given the unique bstacles these scieties faced in their effrts t shed the ecnmic and scial legacies f cmmunism (Mair 1997, ch. 8; Offe 1993). Hwever, by many measures, it remains unclear whether party systems in these cuntries are mving in the right directin and whether schlars can yet talk abut general trends in party system institutinalizatin in the regin. On the ne hand, there have been sme psitive signs that demcratic maturatin is ccurring (Tavits 2005; Tavits and Annus 2006), and expected patterns f ecnmic vting are emerging (Duch 2001; Tucker 2006). On the ther hand, verall indicatrs f institutinalizatin suggest that the cnsistent patterns that we wuld usually expect frm institutinalized systems have nt yet emerged (Casal Bérta and Mair 2012). The party system in Pland has been particularly resistant t stabilizatin. In many ways, this is surprising. In cntrast t ther cuntries that saw the thrugh flattening f sciety by the cmmunist regimes, Pland managed t maintain sme semblance f civil sciety, 13 as evinced by the imprtance f Slidarity in the demcratic transitin, while als preserving a largely autnmus Cathlic Church 13 Of curse, the cmmunist perid did nt result in perfectly hmgeneus scieties. As Słmczynski and Shabad pint ut, making this assumptin bscures the nature, degree, and cnsequences f scial differentiatin in these scieties, bth befre and after the nset f systemic change (Słmczynski and Shabad 1996, p. 188). 25

41 and resisting large-scale cllectivizatin f the agricultural sectr. Furthermre, tw cmmunist successr parties the Demcratic Left Alliance (SLD) and the Plish Peasants Party (PSL) survived the demcratic transitin relatively intact. 14 These parties managed t maintain their rganizatinal structure and were headed by lngstanding members (Grzymała-Busse 2002), which lent a degree f pre-existing structure t the party system. 15 By anchring the plitical spectrum idelgically, the presence f Slidarity and the Cathlic Church n the right and SLD and PSL n the left shuld have aided Pland in develping a rbust and stable party system. But this has nt been the case. Instead, the party system in Pland has been characterized as being cmpletely under-institutinalized (Casal Bérta 2012, p. 5), which is reflected in Pland having sme f the lwest levels f partisan attachment in pst-cmmunist Eurpe (van Biezen, Mair, and Pguntke 2012; Whiteley 2011), persistently high levels f electral vlatility (Epperly 2011), and sme f the highest turnver in gverning calitins in the regin (Casal Bérta and Mair 2012; Cnrad and S. N. Glder 2010; Grtz and Weber 2012). Pland als has the lwest level f turnut in pst-cmmunist Eurpe, averaging 47.7% in natinal parliamentary electins. Only tw electins (in 1993 and 2005) recrded turnut abve 50%, and the first fully demcratic electin in 1991 recrded a turnut f just 43.2%, an astnishingly lw figure given Pland s leading rle in the reginal transitin. By cmparisn, in the Czech Republic and Hungary, 14 SLD was the successr t the cmmunist-era ruling Plish United Wrkers Party (PZPR), while PSL was the successr t United Peple s Party (ZSL), a cmmunist era agrarian satellite party. 15 Strictly speaking, SLD did nt cnslidate int a single party until 1999, when Scial Demcracy f the Republic f Pland (SdRP) and the Plish Scial Demcratic Unin (PUS) merged. They mstly cmpeted as a single entity in electins prir t this, hwever. 26

42 turnut averaged ver 70% and 60% during the same perid (Birch 2003, pp ; Kstadinva 2003). 16 Nne f this is t say that Pland has nt made nticeable prgress twards develping a mre stable party system. Sme clear and psitive indicatrs are available. Table 2.1 presents sme general electral trends in Pland ver the last tw decades. Frm the data presented in this table, it is tempting t say that the Plish party system has settled int a relatively stable pattern f party cmpetitin. Fr example, as measured by the effective number f parties serving in parliament it appears that the party system has resisted extreme levels f fractinalizatin. 17 Furthermre, nly ne new party has been elected t the parliament in each f the last three electins, which indicates that existing parties are beginning t attract stable levels f supprt and themselves becming mre institutinalized. Finally, in the last tw electins, the same ruling party, center-right Civic Platfrm (PO), has wn and frmed the same calitin with PSL Party Switching as a Netwrk in Pland As presented in the previus sectin, there is sme questin as t whether r nt the Plish party system has been making prgress twards a mre stable pattern f cmpetitin. Traditinal measures f party system institutinalizatin are rather ambiguus: the effective number f parties in parliament has been relatively stable since 1993; electral vlatility has declined since the 1997 electin, and in the 2011 electin 16 In the partially-free electin f 1989 turnut was smewhat ver 60%, thugh this was still lwer than the first electins thrughut pst-cmmunist Eurpe (Kstadinva 2003). 17 In the first demcratic electin, there was n electral threshld, which resulted in a large number f parties (29) winning seats in the Sejm. A 5% threshld was instituted fr the secnd electin in 1993, which cntributed significantly t the decline in the number f parties in parliament. 18 Gwiazda (2009) cnsiders the Plish party system t be quasi-institutinalized. 27

43 Table 2.1: Overview f Plish Parliamentary Electins Results, Ttal elected New elected Ttal cmpeting Effective parties Electral vlatility Turnut (%) Vte share tp tw parties (%) Ntes: Electral vlatility cmes frm Pwell and Tucker (2014) and crrespnds t their Ttal Vlatility measure. Vlatility fr 2011 was calculated by the authr. Effective number f parties calculated as in Laask and Taagepera (1979). was half the level it was during the previus electin f 2007; and fr the first time in pst-cmmunist Pland, the same calitin f PO and PSL wn in tw cnsecutive electins. Hwever, ne f the clearest signs f a lack f institutinalizatin in Pland has been the extreme fluidity and lack f stable partisan attachments at the elite level (Shabad and Słmczynski 2004). Table 2.2 shws the verall trend f party switching during the first five parliamentary terms in the Sejm. 19 During the perid cvered in the table, there were almst 1100 instances f intraterm party switching. There have been mre than 100 switches during each term, with the lwest number f switches (105) ccurring during the 2005 term and the mst (481) during the previus 2001 term. On average, mre than 21% f MPs have changed their party affiliatin at least nce during each term, and f the 1603 MPs that have served in the Sejm ver the perid, 28.5% have switched parties at least nce during their time in ffice. Anther 19 The party switching data discussed here cme frm McMenamin and Gwiazda (2011). In their analysis, they prvide an event histry analysis f party switching in each term, with the bjective f identifying the individual mtivatins fr switching. These data are discussed further belw. 28

44 Table 2.2: Number f Active Parties, MPs, and the Number f Switches by Term in the Plish Sejm, Parliamentary Terms Parties elected Parties existing MPs Switches Number switched Pct. switched Max. MP switches Surce: McMenamin and Gwiazda (2011). These data include switches t unregistered status. characteristic f this switching is that when MPs have changed their party affiliatin, it was ften t new parties instead f existing parties. Cnsequently, 74 different parties have served in the Sejm during this perid. Figure 2.1 prvides anther perspective n MP party switching in the Plish Sejm, reprting the number f changes in party affiliatin by day ver the first five parliamentary terms. The dashed vertical lines mark the dates f parliamentary electins, while the slid lines delineate the frmatin f a new gvernment as reprted by Cnrad and S. N. Glder (2010, table 8, p. 143). Tw things are remarkable abut the patterns shwn in this figure. First, while there have been sme significant spikes in switching which cincided with recnfiguratins f majr parties switching is nt cnstrained t such perids f acute vlatility; instead, switching has been a cntinual feature f plitics in the Sejm, with the nly extended lull ccurring in Secnd, while the summary f switching included in Table 2.2 seems t indicate that, with 105 ttal switches and 8.2% f MPs switching during the term, there was a decline in 29

45 switching during the 2005 term, Figure 2.1 clearly shws that this decline in apparent switching is an artifact f the term being limited t tw years. If the rate f switching is extraplated ut t a full fur-year term, we wuld expect upwards f 200 switches during the term. Figure 2.1: Number f Changes in Party Affiliatin by Day in the Plish Sejm, Parliamentary Terms. Number f switches Here I investigate this questin with a descriptive analysis f party switching in the Plish Sejm. Unlike ther studies that have lked at party switching frm the perspective f the incentives facing individual pliticians (Despsat 2006; Heller and Mershn 2005; Laver and Benit 2003; Mershn and O. Shvetsva 2008), I take a scial netwrks apprach, which emphasizes the relatinal nature f party membership in parliaments. 30

46 The data fr the fllwing analyses cme frm McMenamin and Gwiazda (2011), and were cllected directly frm the recrds f the Plish parliament. 20 These data are unusually detailed, cvering all intra-term party switching by MPs, including the exact day they switched as well as their destinatin party, 21 during the first five parliamentary terms f the Plish Sejm (Oct Oct. 2007). Imprtantly, and what makes this dataset s interesting, is that it includes switches t and frm parties that never cmpeted frmally in electins. It was nt uncmmn during this perid fr grups f MPs t leave their party and frm a new ne, nly t merge with anther party at a later date. Such ephemeral parties wuld nt be included in analyses that nly lked at recrds f switching at the time f electins. Hwever, these shrtlived parties shuld be interesting t schlars interested in the dynamic evlutin f scial relatins in the Sejm. These shrt-lived parties carry imprtant infrmatin abut persnal allegiances and idelgical subfactins present in parliament. Indeed, it is cmmn fr party subfactins t express their displeasure with their party by splitting frm it. Lking just at the aggregate trends in party switching ver the first five parliamentary terms, as was dne in Sectin 2.2, it is easy t be pessimistic abut the state f party system institutinalizatin in Pland. Hwever, such summaries d little t reveal the structure f scial relatins between MPs in parliament. This is where a netwrk apprach begins t demnstrate its value. 20 crrespndence with Anna Gwiazda ( ). In their riginal study, McMenamin and Gwiazda (2011) analyzed the data by term; cnsequently, sme additinal prcessing was needed befre it culd be analyzed as a single dataset. This included nmalizing the names f all pliticians acrss time perids s that cmplete histries f switching culd be cnstructed fr each MP. Sme additinal cleanup f the data was als required. All data and detailed ntes n the changes made t it are available upn request. 21 MPs are allwed t be unregistered, and it was quite cmmn fr MPs t abandn their party and remain unaffiliated fr sme time, thugh it was nt pssible t get elected withut a party affiliatin. In this analysis, I ignre unregistered members. Thus, if tw MPs frm different parties leave their respective parties, they are nt cnsidered t share a party f unregistered MPs. 31

47 Figure 2.2 presents the netwrk f relatinships between all Plish MPs and plitical parties as it is derived frm the pattern f MP party switching in the Sejm. 22 During this perid, 1603 MPs served and 74 parties perated in parliament. Fr interpretive clarity, fur f the majr parties are highlighted: SLD (red), PSL (green), PiS (blue), and PO (range). All ther parties are pltted as the larger dark-gray ndes and individual MPs are presented as small, light-gray ndes. As these are pltted as 2-mde, affiliatin netwrks, ties represent MP membership in the parties during each f the terms. Thus, when an MP is tied t multiple parties, this indicates that the MP switched parties at least nce during that term. 23 These netwrk plts prvide an interesting perspective n party plitics in Pland. The first thing that is readily apparent frm the graphs is that SLD, the pstcmmunist successr party, has managed t remain relatively chesive during this perid. This is clear frm the party s relative islatin and lack f cnnectins t ther parties in the verall netwrk. In ther wrds, members f SLD have been less likely t switch parties, and the party has nt suffered frm the number f party splits 22 The netwrks depicted in Figure 2.2 were laid ut using the fllwing prcedure. First, lcatins fr all ndes in the full netwrk dataset were calculated using the algrithm f Fruchterman and Reingld (1991). This frce-directed algrithm identifies psitins t minimize edge verlap and distribute the ndes relatively evenly acrss the plt surface, while als maintaining the structure f the netwrk. Secnd, the nde lcatins were rtated s that SLD, the pst-cmmunist successr party, was t the left f PO and PiS. This cnfrms t what schlars knw abut these parties. Finally, t facilitate cmparisn acrss time perids, nde lcatins in the full netwrk graph where used t psitin the ndes in the specific time perids; in ther wrds, MP and party nde lcatins are static acrss time. It is imprtant t emphasize that the nde psitins and the magnitude f distances between ndes depicted in the graphs are nt meant t be used fr inference. 23 Fr the purpse f this analysis, all switches were treated equally; in ther wrds, I did nt distinguish between, fr example, switches by individual MPs t ther existing parties r induced by party splits. Other schlars have emphasized the imprtance f the different types f switches (Kreuzer and Pettai 2003; Shabad and Słmczynski 2004). Furthermre, multiple switches between parties are nt accunted fr. Fr example, if an MP switches frm party A t party B and then back t party A, the data as analyzed simply recrd this as the MP having a membership tie t each f thse parties. Future research, hwever, may be able t take advantage f the sequencing f changes in ties and the directin f the switching. 32

48 that many f the parties n the right have experienced. Indeed, the lack f partisan cmmitments by pliticians n the right is readily apparent in the graphs. Members f right-leaning parties in the Sejm have been much mre likely t switch parties and the parties themselves have been much mre susceptible t splits. This difference between the stability f the left and right is smething that has been bserved by ther schlars (McMenamin and Gwiazda 2011). 33

49 Figure 2.2: Netwrks f Party Switching fr each Parliamentary Term: Large gray ndes indicate parliamentary parties, small gray ndes are individual MPs. Majr parties are labeled and indicated by clr: SLD (red), PSL (green), PiS (blue), and PO (range) PSL 1997 PSL SLD PSL SLD SLD Cmbined PSL PSL PiS SLD PSL PiS SLD PO PiS SLD PO PO

50 Secnd, the netwrk plts als reveal the grwing assciatin f PSL with the right side f the plitical spectrum. Over the first three parliamentary terms, PSL remained relatively islated in terms f the party switching netwrk, with the majrity f switching that did ccur being between SLD and PSL. This was likely a symptm f members f PSL, a successr party, maintaining scial and idelgical ties t SLD in the years fllwing the cllapse f the cmmunist regime. Beginning in the 2001 term, hwever, the majrity f switching PSL has experienced has been with the right. This change likely reflects tw things. First, turnver in PSL s membership has meant the number f persnal ties between members f PSL and SLD have declined. Secnd, PSL has been a party f pprtunity, playing the rle f pivtal party in parliament. Indeed, PSL has been a member f the gverning calitin in 5 f 7 parliamentary terms 7 f 17 gverning calitins since 1991 (Cnrad and S. N. Glder 2010, p. 143, table 8). Despite the usefulness f these netwrk graphs in prviding a general idea abut hw the party system in Pland has evlved, they are f limited use fr determining whether any apparent patterns in party switching are indicative f grwing cherence in the party system. Fr ne thing, because f the high levels f party switching in Pland, the netwrks are simply t cluttered t allw fr anything but the bradest patterns t be readily discernable, and even then the plts d nt allw fr any srt f frmal inference t be perfrmed. Fr anther thing, while the verall rientatin f the ndes has been specified in a way that makes sense frm an idelgical standpint, the distance between pairs f ndes is nt meaningful. In fact, fr the purpse f presentatin, the algrithm that psitins the ndes intentinally limits the amunt 35

51 f nde verlap. 24 Substantively, this wuld incrrectly suggest that tw pliticians culd never hld the same psitin in the latent scial space, which is an assumptin that wuld be vilated if any tw MPs had the same pattern f party affiliatin (smething that is quite cmmn in the Plish data). In the fllwing sectins, I develp a mdel capable f rigrusly analyzing dynamic, relatinal data. 2.3 The Latent Space Mdel In this sectin, I present a discussin f netwrk data and the prblems such data pse fr standard statistical mdels. I fllw with a review f the latent space mdel fr scial netwrks, which prvides the fundatin fr the dynamic netwrk mdel I prpse in Sectin Netwrks: Terminlgy and Representatin A netwrk is a cllectin f actrs and pssible pairwise relatins between thse actrs. 25 The simplest netwrks cnsist f a single type f actr, where ties are binary and nn-directinal. Fr example, in a netwrk f internatinal trade agreements, the actrs wuld be states and the ties wuld represent the existence f a trade agreement between each state in the dyad. Directed ties are als pssible. In directed netwrks, asymmetric relatinships are pssible. In a netwrk f friendship ties, fr instance, ne member f a dyad may indicate a friendship with the ther persn, but this friendship may nt be reciprcated. Netwrks need nt be restricted t a single type f actr. Mre cmplex netwrks can include multiple types f actrs with valued 24 Details f hw the ndes were psitined are prvided in fn Actrs in the netwrk are als knwn as ndes r vertices, while relatins are knwn as ties r edges. The cannical intrductin t netwrk methds is Wasserman and Faust (1994), while Kadushin (2012) prvides a mre current review f a range f substantive applicatins. 36

52 ties between them. In the Plish party switching netwrk analyzed belw, there are tw types f actrs members f parliament and a party where ties represent MP membership in the parties. In this case, ties are undirected. Netwrks with this type f structure are knwn as affiliatin netwrks. Mathematically, a netwrk can be represented by a matrix, Y, knwn as an adjacency (r sci-) matrix. Each element f the matrix crrespnds t a relatinship between tw actrs in the netwrk. In a simple netwrk with a single type f N actrs and binary relatins, Y is an N N matrix, with each element f the matrix, y ij {0, 1}, indicating the existence f a tie between actrs i and j. 26 In an undirected netwrk, Y is symmetric; i.e., y ij = y ji i j. In a directed netwrk, symmetry need nt hld. In these netwrks, i represents the sender f a relatinship, while j is the receiver. Weighted netwrks are als pssible. In this case, y ij can take n any value. Statistical analysis f netwrk data fcuses n explaining patterns f ties between actrs, either at the dyadic level r frm a brader structural perspective. Such data are ften expected t have strng interdependencies. Fr instance, ties received by an actr are ften reciprcated (reciprcity), actrs with similar characteristics are mre likely t have ties with each ther (hmphily), and friends f a friend are als mre likely t be friends (transitivity). These interdependencies cmplicate the analyses f netwrk data with standard statistical appraches, as they vilate the cmmn assumptin in regressin mdeling that bserved utcmes are independent cnditinal n the mdel and the included cvariates. Cnsequently, by using lgistic regressin t mdel tie frmatin in binary netwrks, fr example, while ignring 26 Self-ties, r lps, are nt typically allwed and diagnal elements f the adjacency matrix are zer by definitin; i.e., y ii = 0 i. 37

53 strng degrees f dependence in the prcess that generates these ties, schlars risk significant bias in estimates f cefficient and standard errrs. Standard mdels are simply nt apprpriate fr data with high levels f dependence between ties The Latent Space Mdel fr Netwrk Data The methdlgical cmplicatins netwrk data pse have encuraged the develpment f numerus statistical appraches t analyzing such data. Methds range in apprach frm actr-based, decisin-theretic mdels, which explain the bserved netwrk structure as the result f the cumulative decisins f actrs in the netwrk (Snijders, Bunt, and Steglich 2010), t mre hlistic mdeling strategies like the expnential randm graph mdel (ERGM), which aims t estimate the likelihd f bserving a netwrk in its entirety given particular structural characteristics (see Cranmer and Desmarais 2011, p. 222). The latent space apprach t mdeling netwrk data, first prpsed by Hff, Raftery, and Handcck (2002), takes smething f a middle grund between the actr-based and hlistic ERGM appraches. This mdel takes a netwrk and psits that the presence (r strength) f a tie between each pair f actrs in the netwrk is a functin f their psitins in a latent scial space. The fundamental assumptin f this mdel is that actrs lcated mre clsely tgether in the latent scial space are mre likely t have ties with each ther. Fr example, in the applicatin t party switching in the Sejm, the latent space culd be interpreted as an idelgical space, where MPs near each ther in that space are mre likely t share parties. 27 Viewed this way, the 27 In their riginal study, Hff, Raftery, and Handcck (2002) analyze the Flrentine marriage data f Padgett and Ansell (1993), which recrds relatins between majr Flrentine families during the 15th century. In that dataset, a tie between families is recrded if there is a marriage between them. In using the latent space apprach t mdeling netwrk interdependencies, the authrs are saying that marriage ties between families are representative f their psitins in sme latent scial space. 38

54 latent space mdel is analgus t many ideal pint mdels mre cmmnly seen in plitical science (Ple and Rsenthal 1997; Clintn, Jackman, and Rivers 2004). In these mdels, a vte fr a bill is seen t be mre likely when it reflects a pint clse t a legislatrs ideal plicy psitin; thus, legislatrs that ften vte the same way are seen t have similar psitins n sme idelgical scale. 28 Sme mathematical ntatin shuld help further clarify things. As discussed abve, a netwrk can be defined as a set f pairwise ties between actrs in the netwrk. This set f ties defines a respnse vectr, y ij, where each element f the vectr indicates whether r nt there is a tie between actrs i and j. In the latent space mdel, this respnse vectr is mdeled as a functin f the pairwise distances between actrs: y ij = F{β T x ij d(z i, z j )}. (2.1) Here, z i and z j represent the k-dimensinal vectr f psitins fr actrs i and j in the latent scial space, while d(z i, z j ) is sme distance functin specified by the analyst that satisfies the triangle inequality. T ease interpretatin, the Euclidean distance is ften used; e.g., d(z i, z j ) = K (z ik z jk ) 2. (2.2) k=1 Hwever, ther distance mdels are pssible. Fr example, Hff, Raftery, and Handcck (2002) als include a prjectin mdel that maps actr lcatins t crdinates n a unit circle, and Schweinberger and Snijders (2003) extend the apprach t use 28 In the latent space mdel, the scial space may be smewhat less well defined than in ideal pint mdels, since the lcatin f bills in the idelgical space is ften als estimated. In the latent space mdel, we nly see the cnnectins between legislatrs, which can be affected by factrs ther than idelgy. 39

55 ultrametric distances and a hierarchical structure. Finally, the mdel can als include a set f (ptinal) cvariates and assciated cefficients, specified here as β T x ij. 29 In their riginal frmulatin, Hff, Raftery, and Handcck (2002) demnstrated the latent space mdel n netwrks with binary ties, bth directed and undirected. In that case, a lgistic regressin mdel was used. Hwever, the latent space apprach is much mre general and can readily accmmdate netwrks with mre cmplex tie structures. Krivitsky et al. (2009), fr example, demnstrate a cunt mdel that assumes a Pissn data generating prcess, using it t assess shared peridical readerships in Slvenia. Generally speaking, the latent space mdel can be easily incrprated int the generalized linear mdeling framewrk, thugh t date there has been relatively little research int hw well these mdels perfrm in mdeling real-wrld data. Furthermre, several extensins t the mdel have been develped. Fr instance, Handcck, Raftery, and Tantrum (2007) extend the latent space mdel t include actr-level clustering, making it pssible t identify grups f similar actrs based nly n their ties. In mre recent wrk, Krivitsky et al. (2009) specify mdels t include s-called sciality randm effects terms in undirected netwrks and sender and receiver randm effects in directed netwrks. 30 Such terms are meant t capture the tendency fr sme actrs t frm ties mre readily than thers (i.e., sme are mre sciable than thers). Finally, Hff (2005) prvides a bilinear mixed-effects mdel that includes the crss-prduct f latent sender and receiver psitins. The flexibility f the latent space apprach set it apart frm ther netwrk methds, such 29 In specifying cvariates, it has t be remembered that the dependent variable in these mdels is a tie between tw ndes. Cnsequently, cvariates are ften defined n the dyad. 30 Krivitsky et al. (2009) als include clustering in their randm effects latent space mdel. 40

56 as ERGMs, which have nly recently been extended t value-edged netwrks and are cmputatinally mre demanding (Krivitsky 2012; Desmarais and Cranmer 2012). Smething shuld be said abut the fundamental assumptin underlying this mdel. The latent space mdel carries with it a strng cnditinal independence assumptin; i.e., ties are assumed t be independent given the nde psitins in the latent space and any cvariates included in the mdel. In ther wrds, the latent psitins (alng with the cvariates) fully capture the cmplex dependencies that affect tie frmatin in the netwrk. Hwever, while stringent, this is n different than the assumptins made in traditinal regressin mdels, where assumptins f cnditinal independence are als made. That said, unlike standard mdels, little is knwn abut the sensitivity f the latent space mdel t deviatins frm the cnditinal independence assumptin r n their perfrmance in small netwrks The Prblem with Dynamic Netwrks The latent space mdel was riginally frmulated fr single realizatins f a netwrk, which limits its applicability t dynamic netwrks. As riginally cnceived, schlars had tw ptins fr mdeling dynamic netwrks with the latent space apprach: they culd pl all bservatins int a single netwrk r they culd analyze each netwrk separately. Neither f these ptins is particularly satisfactry. As with ther types f data, pling a dynamic netwrk int a single realizatin means masking ptentially interesting prcesses that drive structural change. Furthermre, when netwrks are pled, structures may appear different than they are. Depending n the research questin being explred, the inferences we make by pling a dynamic netwrk may be quite misleading. 41

57 t = 1 Figure 2.3: Pling a Dynamic Netwrk. t = 2 pled A A A B B B C C C D D D Cnsider the simple netwrk depicted in Figure 2.3. This netwrk cnsists f fur ndes and tw perids. In the first perid, there is a transitive relatinship between ndes B, C, and D. While in the secnd perid the ties between B and D and between C and D have brken dwn, while at the same time, new ties were frmed between A and B and between A and C. If we pl these netwrk realizatins int a single netwrk suitable fr analysis with the standard latent space mdel (as depicted in the third netwrk in the figure) the structural change in the netwrk between these tw perids is n lnger apparent. Instead, the netwrk appears t be nearly fully cnnected, nly lacking a tie between A and D. 31 Depending n the questin being asked, inferences drawn frm this pled netwrk may be wrng. Fr example, suppse the netwrk represented military alliances. In this case, the first and secnd netwrks represent very different wrlds. In t = 1, cuntry A was excluded frm alliances with the ther cuntries, suggesting it may play the rle utside the system f internatinal security r culd even be a cmmn enemy f the ther cuntries during this perid. As such, the latent psitin f A wuld be lcated far frm the ther ndes. In t = 2, D is 31 In this simple netwrk, B and C wuld als have identical psitins in the latent space. 42

58 the utlier, while A has nw been incrprated int the alliance netwrk f B and C. Obviusly, these are very different scenaris, which wuld be erased in an analysis that pled the netwrks tgether. The secnd ptin available t schlars wanting t apply the latent space mdel t dynamic netwrks wuld be t mdel each netwrk realizatin separately. Ding s wuld make change in netwrk structures mre apparent; hwever, estimating separate mdels raises its wn prblems. Cnsider the netwrk f three ndes and three perids in Figure 2.4. In this netwrk, ties are valued, which we may interpret as being the number f interactins between ndes during each perid. In the first perid, t = 1, B and C reprt three interactins with each ther, while bth B and C reprt ne interactin each with A. The latent space mdel wuld, in this case, place B and C clse t each ther in the latent scial space and far frm A. In t = 2, the number f interactins between A and B increases t tw and the number f interactins between B and C decreases t tw. Fr this reasn, the latent space mdel, knwing nthing abut the psitins f the actrs in the previus perid, wuld shift the latent psitins s that the distance between A and B and between B and C were equal. Finally, in the last perid, t = 3, the number f interactins between the ndes returns t the values bserved in t = 1; thus, the latent psitins estimated fr each nde by the latent space mdel will als revert t thse f the first mdel. Taken as individual netwrks, it makes sense that the psitin f B wuld change ver time; hwever, when lked at as a dynamic netwrk, the relatively large changes in the psitin f B seems less apprpriate. First, estimating three separate mdels ignres the knwledge abut previus perids, making the estimate inefficient. Fr instance, given T realizatins f a netwrk with n ndes, a k-dimensinal latent space mdel, 43

59 estimated fr each realizatin f the netwrk, wuld require T n k estimated parameters. Hwever, by putting sme structure n actrs mvement in the scial space ver time, it may be pssible t greatly limit the number f parameters that need t be estimated. In the example f Figure 2.4, = 18 parameters wuld need t be estimated in three latent space mdels, while a linear trajectry mdel (intrduced in Sectin 2.4) requires 3 4 = 12 parameters. 32 Secnd, estimating three separate mdels intrduces a risk f verfitting r inferring mre change in nde lcatin in the latent scial space what may be the case. Figure 2.4 shws this ptential quite clearly. As discussed abve, the change in psitin f B in the secnd perid lks t extreme given the bserved ties in the first and third perids. If ties are a stchastic prcess, the pattern f ties seen in the secnd perid wuld nt be unusual even if the pattern in the first and third perids was the expected ne. Finally, estimating separate mdels culd make it difficult t cmpare latent psitins acrss bserved netwrks. As discussed belw, mdel identificatin can be an bstacle in these mdels. Depending n hw it is achieved, the latent spaces culd be n different, incmparable scales acrss mdel estimates. This is wuld be particularly the case if there was nde turnver in the netwrks, which culd greatly affect nde psitining. 2.4 The Latent Path Mdel Here I intrduce a new mdel suitable fr mdeling dynamic netwrks, such as the party switching netwrk analyzed in the next sectin. The prpsed latent path mdel builds n the latent space mdel f Hff, Raftery, and Handcck (2002) utlines in the 32 The relative efficiency f the prpsed mdel increase as the number f times perids increase. 44

60 Figure 2.4: Analyzing a Dynamic Netwrk Separately. t = 1 t = 2 t = 3 A 1 A 2 B A 1 1 B B 3 3 C C C previus sectin and is clsely related t ther dynamic latent space mdels (Sarkar and Mre 2005; Sewell and Chen 2015; Ward, Ahlquist, and Rzenas 2013), thugh there are imprtant differences that will be discussed belw. 33 This mdel has a few key features. First, instead f assuming that each actr is lcated at a single pint in the latent scial space, the mdel treats the lcatin f each actr as lying n a path in that space. In ther wrds, actr psitins are allwed t shift ver bserved netwrk realizatins, and the directin and magnitude f these changes is inferred frm the dynamic evlutin f ties in the netwrk. Secnd, by explicitly linking the bserved realizatins f a netwrk thrugh the estimatin f trajectries fr each actr in any f the bserved netwrks, the mdel prvides a natural way t accmmdate changes in the ndal cmpsitin f the netwrk ver time. 34 Missing data, either in the frm f unaccunted fr ties r missing r changing cmpsitin f mdes, 33 In their cnclusin, Hff, Raftery, and Handcck (2002) mentin the pssibility f extending the mdel t dynamic netwrks, thugh they d nt prvide detailed guidance n ding s. 34 Greater levels f nde turnver will, at a minimum, increase the uncertainty in the estimates f the estimated latent trajectries fr all ndes in the netwrk. This is similar t a situatin in item respnse mdels, where missing respnses increase the standard errrs arund the estimates f latent abilities. 45

61 is a significant prblem in the statistical study f netwrks (Rbins, Pattisn, and Wlcck 2004; Kssinets 2006; Brgatti, Carley, and Krackhardt 2006; Huisman and Steglich 2008). Yet, in scial netwrks, it is cmmn fr there t be significant change in the cmpsitin f the netwrk. Fr example, in the applicatin t party switching in the Plish parliament, a great deal f MP turnver in parliament ccurs with each term. If change ver time is f interest, this makes existing mdels inapprpriate. The ability f the latent path mdel t accmmdate netwrks that experience sme level f turnver in ndes shuld make it useful fr analyzing a brader range f netwrks. The frmal definitin f the latent path mdel is directly analgus t the latent space mdel f Hff, Raftery, and Handcck (2002) described abve. The difference is that the respnse vectr in the latent path mdel includes repeated bservatins fr each dyad fr each time perid, while the distances between actrs in the netwrk are als allwed t change ver time as actrs psitins in the latent scial space change. Mathematically, this suggests the additin f time subscripts as well as a redefinitin f the psitin vectrs frm thse f Eq. (2.1): y tij = F{β T x tij d(z ti, z tj )}. (2.3) Here, the subscripts i and j cntinue t refer t the tw ndes in a dyad in the netwrk, while t = 1, 2,..., T indicates the bserved realizatins f the netwrk f interest. The nticeable difference between the mdel f Eq. (2.3) and the latent space mdel f Eq. (2.1) lies in the definitin f the latent psitins, z ti. In the latent path mdel, z ti n lnger represents a single pint in the k-dimensinal scial space; rather, it represents a path defining a trajectry f mvement fr actr i thrugh 46

62 the latent scial space ver a set f temprally-rdered netwrk realizatins. 35 Fr example, the simplest, nn-trivial functin wuld assume that actrs in the netwrk change psitins in the latent scial space fllwing a linear trajectry (i.e., ndes are mdeled as mving frm pint A t pint B in T equivalent steps). In this case, z ti can be defined as fllws: z ti = g(t, z 0 i, z s i ) = z 0 i + t z s i. (2.4) Here, z 0 i is a vectr representing the starting psitins fr actr i in the latent scial space and z s i is a step vectr indicating the directin and magnitude f mvement taken by the actrs in each perid. Of curse, mre elabrate functins are pssible. One can imagine, fr instance, a quadratic r therwise curved path being specified. Alternatively, as a smaller innvatin n the linear trajectry, a path functin culd be specified that allwed an increasing r decreasing degree f mvement during each perid. Such a mdel is used in the analysis f party switching presented in Sectin Identificatin and Estimatin Parameter identificatin is a significant bstacle ne faces when attempting t estimate latent variable mdels. The lack f identificatin in these mdels takes tw frms. 36 The first issue lies in the invariance f the likelihd t reflectin, rtatin, and translatin f the latent psitins. In ther wrds, while maximizing 35 It is nt necessary fr the netwrk t be bserved at regular pints in time; hwever, if the perid f time between bservatins des vary, changes t the definitin f the trajectry functin may need t be made t accunt fr this, depending n hw imprtant the gap between realizatins is t the substantive interpretatin f the prblem at hand. 36 The issues f identificatin described here are analgus t that faced by schlars wanting t estimate ideal-pint r ther latent variable mdels (Clintn, Jackman, and Rivers 2004; Bafumi et al. 2005). 47

63 the likelihd with respect t distances is pssible and relatively straightfrward (Hff, Raftery, and Handcck 2002, p. 1092), maximizing with respect t the lcatins is nt. Fr example, suppse Z is a matrix f n latent lcatins in a k-dimensinal space. Cnsider these the true psitins f these ndes. There then exists a k k transfrmatin matrix, T, such that Ẑ = ZT and L = F(d(Ẑ)) = F(d(Z)), where d() is the distance functin and L is the likelihd given the inputs. Cnsequently, the mdel is nt identified with respect t nde lcatins: there exist an infinite number f sets f lcatins in the latent space that map t the same set f distances between ndes. The secnd surce f nn-identificatin is due t additive aliasing. Suppse d = d(z i, z j ) i j, a vectr f distances between each pair f ndes in the netwrk, b = β1 is an intercept term (perhaps representing the verall, default cnnectivity f the netwrk), and the likelihd can be seen as sme functin f their difference: L = F(b d). In this case, ne can add sme value δ t bth b and d and recver the same likelihd: L = F(b d) = F ((b + δ) (d + δ)). (2.5) There are tw appraches t dealing with the prblem f identificatin with respect t the invariance f distances t latent lcatins. The first apprach is t first define sme particular set f latent psitins, Z, as a reference class, 37 and then during each draw frm the psterir (in a Bayesian cntext) use a Prcrustes transfrmatin t minimize the difference between the psterir draw and the latent psitins. 38 This is the methd used by Hff, Raftery, and Handcck (2002), and subsequent studies 37 Hw these psitins may be chsen will be discussed belw. 38 See Brg and Grenen (2005, ch. 20) fr a descriptin f the Prcrustes transfrm. 48

64 have used a similar apprach (e.g., Krivitsky et al. 2009; Shrtreed, Handcck, and Hff 2006). The secnd apprach t mdeling identificatin is t fix the lcatins f a small set f reference ndes in the latent scial space (as in ideal pint mdels, k +1 ndes need t be fixed t assure identificatin; see Clintn, Jackman, and Rivers 2004). Similarly, in a Bayesian apprach, strng prirs can be used t cnstrain the latent lcatins f sme f the ndes t assure cnvergence t a single mde f the psterir. Each f these appraches carries with it technical and substantive advantages and disadvantages. In the first apprach, the use f a Prcrustes transfrm shuld greatly decrease the cmputatinal burden f estimating the mdel, since every draw f the latent lcatins frm the psterir is frced, via the transfrm, t reflect a single mde f the psterir. In cntrast, using fixed reference ndes fr identificatin will likely be mre cmputatinally demanding, since the mdel will take sme time t cnverge t the mdel psterir (i.e., a lnger burn-in perid will be required cmpared t the Prcrustes transfrm apprach). Finally, by relying n strng prirs, the fully Bayesian apprach is likely t be even mre cmputatinally nerus. On the ther hand, since the Prcrustes methd f transfrming the latent lcatins used in these analyses relies n a prcedure that minimizes the difference between the reference class, Z, and the transfrmed psitins, ZT, there is sme risk that the draws frm the psterir using this methd will nt reflect the variability in the actual latent lcatins. This is because the Prcrustes transfrm frces each draw t be as clse as pssible t the reference cnfiguratin. Instead, fixing a set f reference ndes and estimating the ther ndes latent lcatins directly, withut 49

65 such a transfrm, shuld result in the psterir marginal variances fr the ndes that accurately reflects the infrmatin cntained in the netwrk. A secnd advantage f the fixed-nde methd is that it allws the analyst t bake-in an rientatin f the latent space that is cnsistent with thery. Schlars ften knw, fr instance, that sme parties r pliticians are t the plitical left r right f thers; thus, it makes sense t rient estimated psitins in the latent scial space t reflect this prir knwledge. Fixing particular, influential ndes prvides a straightfrward substantive interpretatin f mdel results withut the need fr pstprcessing. The prblem f additive aliasing can be addressed, in the Bayesian cntext, by using tight prirs n the distances and the intercept term. In a maximum likelihd apprach, aliasing is a bit mre difficult t deal with; hwever, it is pssible t pstprcess the psitins and intercept after maximizatin. This is necessary even when yu fix the lcatins f sme ndes, since the simultaneus estimatin f the intercept and the lcatins tends t increase the distances between ndes (als bserved by Shrtreed, Handcck, and Hff 2006, p. 27). Alternatively, in the case where ndes can be gruped by sme characteristic (fr instance, by gender in a study f adlescent friendship ties r by time perid in a tempral mdel) a hierarchical apprach t specifying the prirs culd be used t assure identificatin. If taking a maximum likelihd apprach, setting ne f the intercept terms culd be set t zer (Bafumi et al. 2005, pp ). Because f the inherent dependencies in netwrk data, the large number f parameters that need t be estimated, as well as the identificatin issues, establishing gd 50

66 starting values fr any f the ptimizatin rutines can greatly reduce the cmputatinal burden f these mdels. Hff, Raftery, and Handcck (2002) and subsequent researchers have used a multi-step apprach. In the first step, gedesic distances between each pair f ndes are calculated between ndes. Secnd, latent starting psitins in the k-dimensinal space are generated using classical mulitdimensinal scaling (Gwer 1966). Hff, Raftery, and Handcck (2002) then use these psitins as starting values t get pint estimates fr the latent lcatins via maximum likelihd. Their subsequent MCMC algrithm then uses the maximum likelihd estimates as the reference psitins fr their Prcrustes transfrm, which establishes the psterir distributin f the nde lcatins Cmparisn t ther Mdels Tw ther latent space mdels have been prpsed fr dynamic netwrks. Ward, Ahlquist, and Rzenas (2013) build n the bilinear mixed effects mdel f Hff (2005) t mdel dynamic internatinal trade netwrks. That mdel accunts fr the thirdrder dependencies f netwrk data thrugh the inclusin f the crss-prducts f sender and receiver psitins in the latent space. When a dyad s sender and receiver psitins are riented in the same directin, this is an indicatin that the ties between these actrs is strnger than ne wuld expect given the rest f the mdel. Ward, Ahlquist, and Rzenas (2013) extend the mdel t include lagged bilinear terms. These terms are nly included fr t = 2, 3,..., T, with the bilinear terms fr all perids after the first being interpreted as changes in latent psitins in the sender 39 The standard errrs derived frm maximizing the likelihd shuld nt be used fr inference given the nn-independence f the data. 51

67 and receiver spaces. 40 Fr mdeling directinal netwrk data, the bilinear mdel has sme advantages, particularly in cases where latent sender and receiver spaces have clear theretical interpretatins. This was the case in the authrs applicatin t internatinal trade netwrks. Hwever, fr mdeling affiliatin r ther undirected netwrks, the mdel is less useful: in these netwrks, there is n cncept f sender and receiver. Furthermre, their mdel als draws the latent sender and receiver lcatins frm different spaces; thus, if there is an interest in the lcatin f the ndes in the latent space, this makes cmparisn difficult. Finally, their lag term is n the crss-prduct, s unit mvements are nt explicitly mdeled. An earlier apprach t mdeling dynamic netwrks was prpsed by Sarkar and Mre (2005). One f the biggest differences between this mdel and the ne I prpse is that in their mdel, change in latent psitin is nt mdeled explicitly. Instead, ndes are simply allwed t drift in the latent scial space. Anther difference is hw they mdel ties. Each nde in the netwrk is assumed t have a level f sciality. This sciality defines a space within which ndes are t have a decreasing likelihd f a tie with anther nde within the space. The prbability f a tie with a nde utside this circle is defined t be small. Finally, the mdel was cnstructed fr binary ties and it is nt clear hw easily it can be extended t weighted edges. An advantage f their mdel, hwever, is that it can readily accmmdate large netwrks, as they shw in their Mnte Carl analysis, which includes netwrks with up t 11,000 ndes. Building n Sarkar and Mre (2005), Sewell and Chen (2015) have recently 40 Nte that it is nt pssible t drp a perid frm the mdel in rder t cnstruct the lagged bilinear effect; as a latent trait f the system, there is n way t cnstruct the lag withut first estimating the psitins. 52

68 intrduced a randm walk mdel incrprating an attractins parameter meant t mdel sciality f individual ndes. Randm walk mdels are attractive in their simplicity; hwever, fr the purpse f the substantive analysis f plitics, they are quite susceptible t verfitting. Fr instance, in the recent mdel by Sewell and Chen (2015), each nde nticeably shifts during each perid depending n whatever ties they have during that perid. If a large number f perids are included in the mdel, then these idisyncratic shifts in latent psitin shuld average ut. Hwever, when nly a few realizatins f the netwrk are available, these mdels culd be influenced t a great degree by nise in the data. 2.5 An Analysis f Party Switching in the Plish Sejm The latent path mdel presented in the previus sectins prvides a nvel apprach t assessing specific patterns f party switching and their unique effects upn party system institutinalizatin. Three features f the mdel make it particularly well suited t the task. First, the latent path mdel is dynamic. Thus, it is apprpriate fr evaluating the level f change in the party system ver time. Secnd, the latent path mdel is explicitly relatinal. An MP s decisin t switch parties is likely t affect ther members decisin. Perhaps it signals the weakness f party discipline r the nn-viability f the riginal party. Because MPs decisins t switch are nt independent, standard statistical mdels that assume independence f bservatins are nt apprpriate. Finally, the latent path mdel is spatial, which fits well with spatial theries f plitics (e.g., Dwns 1957). 53

69 In this sectin, I apply the latent path mdel t party switching in the Plish Sejm with the intent f assessing whether there is any cnnectin between switching and idelgy in Plish plitics. Idelgy is an imprtant factr in determining the nature f the party system, prviding structure t plitical cmpetitin and placing imprtant cnstraints n elected elites (Mainwaring and Scully 1995; Mainwaring 1999; Tavits 2005, p. 286). In yung, pst-cmmunist demcracies, the idelgical cnturs f plitics are ften unclear. Parties and pliticians ften fail t take clear psitins n issues, and fcus instead n ppulist appeals r persnal charisma t btain ffice (Innes 2002). This is certainly the case with Pland. But idelgy remains an imprtant determinant f party systems and, if institutinalizatin is ccurring, we shuld see greater idelgical cherence in parties ver time. Here I use the prpsed latent path mdel and the netwrk f party switching in Pland t shw that party switching has allwed Plish MPs t srt int mre cherent idelgical grups. In ther wrds, cntrary t cnventinal wisdm abut the destructive cnsequences f party switching, the evidence presented here suggests switching can play a psitive rle in the prcess f party system institutinalizatin in new demcracies The Data The netwrk f party switching intrduced in Sectin were used in this analysis. Because there was a great deal f attritin f in the netwrk f the 1603 MPs that served in the Sejm during the five terms analyzed here, 1096 (68%) served nly ne term, while 308 (19%) served tw terms nly thse 199 MPs that served three r mre terms were used in the analysis. 41 The netwrk was als cnverted frm 41 While the latent path mdel is able t accmmdate attritin frm and additin t the netwrks, estimatin f the mdel with s much vlatility in the netwrk wuld nt be feasible. 54

70 Table 2.3: Distributin f Shared Parties by Time Perid. # f shared parties Term a bipartite netwrk f party affiliatins t a unimdel netwrk recrding the number f shared parties between each MP in the netwrk. The distributin f the number f shared parties between the 199 MPs included in the netwrk is shwn in Table Mdel Specificatin Befre applying the latent path mdel t the Plish data, several decisins abut mdel specificatin need t be made. First, the number f dimensins fr the latent scial space has t be determined. In the absence f prir knwledge abut this space, the analyst wuld need t estimate multiple mdels with different dimensins and cmpare them t determine what prduces the best fit; fr example, by using sme ut f sample predictive diagnstic. In the case f the Plish Sejm, hwever, we are able t lean heavily n prir research and knwledge f Plish plitics. While there has been sme debate in the literature n Plish plitics abut the salience f particular dimensins at particular times, area specialists are largely in agreement that Plish plitics is structured arund tw primary dimensins f plitical cntestatin: a redistributive/pr-market ecnmic dimensin and a secularist/cnfessinal 55

71 dimensin (e.g., see Markwski 2008). Thus, I have chsen t use a tw dimensinal latent scial space. 42 Secnd, the frm f the trajectry has t be specified. Fr this analysis, I have chsen a linear trajectry (see Equatin 2.4) fr MP mvement in the latent scial space, with ne slight mdificatin. If party plitics in the Plish parliament are settling dwn party system maturatin is ccurring this wuld suggest that MPs wuld be shifting their psitins in the latent scial space mre slwly ver time. Fr this reasn, I have included in the mdel a decay parameter n the distance MPs mve during each term in parliament. If this decay parameter is estimated t be less than 1, this wuld suggest that, n average, MPs are mving mre slwly ver time. On the ther hand, if the parameter is greater than 1, this wuld imply that the Plish party system is increasingly chatic. The assumptin that MPs fllw a linear path thrugh the latent scial space ptentially a mre prblematic assumptin than assuming a tw-dimensinal latent scial space. It is quite pssible, fr instance, that plitical learning by different pliticians may be nn-linear. That said, the linearity assumptin is justified by the limitatins f the available data. The Plish party switching data cntains just five perids; cnsequently, it is likely that successfully estimating anything but a linear mdel wuld be difficult. 43 Furthermre, a mre cmplicated specificatin culd be susceptible t verfitting. 42 A ne-dimensinal mdel was als attempted; hwever, it never reached cnvergence (even when run fr twice the number f iteratins as the 2-dimensinal mdel), which is a strng indicatin that restricting the mdel t a single dimensin was nt reflective f Plish party plitics in the Sejm. 43 Recall that a separate trajectry is estimated fr each MP included in the data. In this case, 196 different trajectries, r = 392 trajectry parameters, are estimated fr the Plish data (the lcatins f the reference units are treated as fixed). With just five realizatins f the netwrk available, a mre cmplicated functinal frm wuld ask t much f the data. 56

72 As described abve, the unimdal netwrk analyzed here recrds the number f shared parties between each MP. As a discrete, cunt variable, a Pissn mdel was used. Given the trajectry discussed abve, the mdel is shwn here: y tij = Pis{β t d(z ti, z tj )} i j where z ti = z 0 i + t α z s i. (2.6) In Equatin (2.6), α is the decay parameter meant t capture the maturatin f the party system Parameter Identificatin T facilitate mdel estimatin and identificatin, restrictins n the latent psitins and trajectries f three MPs n the scale f the latent space was necessary. 44 Frm a technical standpint, restrictins culd be placed n any three MPs in the dataset; hwever, cnstraining particular MPs makes interpretatin f the results mre straightfrward. Fr this analysis, I have chsen t restrict the psitins f Jacek Piechta, Brnisław Kmrwski, and Waldemar Pawlak. Tw factrs make these gd candidates fr restrictin. First, each f these MPs is, r has been, a leader f imprtant factins in Plish plitics. Jacek Piechta f the Demcratic Left Alliance (SLD), successr t the cmmunist-era ruling party, has been a lng-time left-wing plitician, served as Minister f the Ecnmy in tw gvernments, and was a member f the ruling cmmunist party. Brnisław Kmrwski, currently serving as the President f Pland, was a prminent member f Civic Platfrm (PO), currently the majrity party in the gverning calitin, befre he left the party in 2010 when he was elected President. Kmrwski was als a member f Slidarity and 44 As discussed in Sectin 2.4.1, pairwise distances are invariant t reflectin, rtatin, and translatin. Withut these restrictins, it wuld nt be pssible t statistically identify unique psitins fr the MPs in the latent scial space. A similar restrictin is required fr identificatin in ideal pint mdels (see, e.g., Clintn, Jackman, and Rivers 2004; Bafumi et al. 2005). 57

73 is a gd representative f the plitical center-right. Finally, Waldemar Pawlak has been a lng-time member f the Plish Peasants Party (PSL) and was als a member f United Peples Party (ZSL), the agrarian satellite party during cmmunism. He has twice been Prime Minister and served as Deputy Prime Minister during first PO/PSL calitin ( ). The secnd, practical factr that makes Piechta, Kmrwski, and Pawlak gd candidates as reference units is that each f the three served in the Sejm fr all five terms analyzed here. Since the lcatins f all ther pliticians in the latent scial space are psitined relative t these three reference ndes, having them present in each f the time perids reduces the cmputatinal burden f estimating the mdel. Chsing pliticians fr whm their idelgical leanings are well knwn lends interpretive structure t the latent scial space. In this case, relative latent lcatins can be chsen fr each f these MPs in a way that agrees with prir knwledge abut their plitical leanings. Specifically, as a leader f the ex-cmmunist party, Piechta is knwn t be left f Kmrwski n ecnmic issues, while als being less scially cnservative. Between Kmrwski and Pawlak, we knw that Kmrwski was less scially cnservative than Pawlak and als smewhat t the right n ecnmic issues. Thus, by psitining these pliticians in a way that reflects this knwledge, the estimated latent psitins f the ther MPs becme interpretable n these idelgical dimensins. Hwever, instead f assigning Piechta, Kmrwski, and Pawlak arbitrary lcatins that satisfy their relative psitins in the scial space, which wuld have guaranteed mdel identificatin, their lcatins in the space were estimated simultaneusly with the lcatins and trajectries f all ther MPs, subject t sme strng prirs and bx cnstraints n their latent lcatins. These bx cnstraints 58

74 Table 2.4: Bx Cnstrains n the Latent Lcatins fr the Reference MPs. Ecnmic Scial Piechta [ 5, 0 ] [ 5, 0 ] Kmrwski [ 0, 5 ] [ 5, 0 ] Pawlak [ 5, 5 ] [ 0, 5 ] are shwn in Table 2.4. In additin, a nrmal prir with mean f zer and standard deviatin f 0.25 was placed n their lcatins. This is a cnservative specificatin, since it means any distance away frm the rigin wuld be a strng sign that the netwrk cntains infrmatin abut differences in their plitical leanings. Prirs fr all ther parameters estimated are reprted in Table 2.5. As shwn, all parameters were given relatively tight prirs t facilitate identificatin. Hwever, as with the lcatins f the reference MPs, this is a cnservative specificatin, since any estimated differences in latent lcatins wuld indicate clear differences in idelgical psitining. On the ther hand, if MPs estimated psitins are clustered arund the rigin, this wuld be an indicatin that the netwrk carries with it very little infrmatin Mdel Estimatin The mdel was estimated using Hamiltnian Mnte Carl in Stan (Stan Develpment Team 2013) using the N-U-Turn sampler (Hffman and Gelman 2013). Fur chains were simulated with 400,500 iteratins each, saving the last 500 draws frm the psterir fr each chain. Gelman-Rubin s (1992) ˆR was less than 1.05 fr these 59

75 Table 2.5: Prirs fr Intercepts, Decay Parameter, Latent lcatins, and Trajectries fr all ther Parameters in the Latent Path Mdel. Parameter Prir Intercept (β 0t ) Nrmal(µ = 0, σ = 1) Decay (α) Nrmal(µ = 1, σ = 1) Psitins (z 0 i ) Nrmal(µ = 0, σ = 1) Trajectry (z s i ) Nrmal(µ = 0, σ = 0.5) decay parameter was cnstrained t be psitive. as well. Summary f mdel cnvergence is available in Appendix A.2. Stan cde fr the mdel is lcated in Appendix A Results The lcatins f the reference MPs Piechta, Kmrwski, and Pawlak is key t the interpretatin f the mdel results. Figure 2.5 shws a randm draw f 250 pints frm the psterir distributin f estimated final lcatins fr these MPs. The large, slid pints represent the psterir means. As expected, the mdel psitins Piechta, Kmrwski, and Pawlak in the lwer-left, upper-left, and lwer-right quadrants, respectively. This is, f curse, a result f the relative cnstraints placed n their psitins in the mdel. What is ntewrthy, hwever, is the distances f the psterir means frm the rigin. Recall that the mdel des permit (and even encurages given the tight nrmal prir n their psitins) the psitins f these reference units t 45 Cnvergence was nt perfect. Of the 797 parameters estimated (includes the lg-prbability), 93 (11.6%) ˆR values were ver 1.05 (the usual threshld accepted as indicating cnvergence); hwever, nly 5 parameters had an ˆR abve It is nt expected that this lack f apparent cnvergence shuld change the results presented belw as it appears that three f the fur chains had cnverged, but that the furth was mre prblematic. A greater number f iteratins shuld slve this. 60

76 cllapse t the rigin, which wuld happen if the party switching netwrk cntained n infrmatin abut the idelgical lcatins f these actrs. Instead, structural patterns in the switching netwrk clearly indicate idelgical differentiatin. Figure 2.5: Estimated latent psitins fr reference MPs, Piechta (red), Kmrwski (range), and Pawlak (green). Open pints are 250 randm draws frm the psterir psitins fr each MP. Means fr the full marginal psterirs are shwn as dark, clsed pints. Pawlak liberal / cnservative Piechta Kmrwski redistributive / pr-market It was argued abve that if the Plish party system is settling int a mre stable equilibrium, the mvement f MPs in the latent scial space will slw ver time. 61

77 Figure 2.6 reprts the psterir distributin f the estimated decay parameter and prvides evidence fr this hypthesis. As shwn, the psterir mean fr the decay parameter is apprximately 0.9. A value less than 1.0 indicates that, n average, MPs mvements thrugh the latent space are, in fact, slwing. 46 Figure 2.6: Psterir Distributin f Trajectry Decay Parameter. Slid vertical line is the psterir mean, while the dashed lines define the 95% credible interval. 10 Percent f Ttal Decay parameter Finally, the full results frm the estimated latent path mdel are presented graphically in Figure 2.7. This figure plts the estimated psitins in the latent scial space fr each active MP in the Plish Sejm ver the five parliamentary terms included in 46 While the slwing is nt dramatic mvement between the 4th and 5th perids had decreased t just under 80% f what it was between the first and secnd perids it shuld be remembered that this is an average rate fr all MPs in the dataset. 62

78 the study, with each pint in the plt representing ne f the 199 MPs included in the estimatin. Since the estimated psitins f the MPs in the latent scial space can verlap when they have identical r similar ties, MP lcatins have been pltted with sme transparency, s that the darker the area the larger the number f MPs lcated at that lcatin. T aid interpretatin, labels are als included fr several grups (parties) f interest. Several things are apparent frm these results. First, despite the expected rganizatinal strength f Slidarity and the Cathlic Church at the beginning f the demcratic transitin, the plitical right appears t have been in disarray during much f this perid. This is apparent in the scattered latent psitins estimated frm the data, which is particularly nticeable in the first and furth terms. Als clear is the split between PiS and PO in The secnd thing t ntice is that SLD has remained fairly stable and chesive during this perid. Very few f its members have switched t r frm the party, which is apparent frm the lack f MPs being placed by the latent path mdel int psitins between the SLD MPs and ther grups. Even while in turmil during the 2001 term, members did nt seek ut parties utside the leftist, scial demcratic family f parties. Recall that the relative disrder n the right, the stability f SLD, and the grwing assciatin f PSL with the right was als reflected in the standard netwrk plts f Figure 2.2. Alng with qualitative knwledge f the Plish party system, this helps lend credibility t the latent path mdel. Hwever, the netwrk plts als bscured the degree t which the Plish parliament has cme t be dminated by fur relatively hmgeneus factins. While switching clearly cntinues, particularly n the right, 63

79 this switching has cme t take place almst exclusively between mre hmgeneus grups ver time. This is evident in the fact that very few MPs are psitined significantly far frm grups f ther MPs. In fact, by the 2005 parliamentary term, nly a cuple MPs are lcated in psitins that culd be cnsidered between grups. Overall, these results lend supprt t the hypthesis that party switching, even at the levels seen in Pland, are nt necessarily a lng-term liability t the prcess f party system institutinalizatin and demcratic cnslidatin. Instead, in Pland it appears that patterns f party switching reflects a prcess f plitical learning and the emergence f a mre idelgically cherent party system. 64

80 Figure 2.7: Estimated Latent Lcatins as Prduced by the Latent Path Mdel PSL 65 liberal / cnservative SLD Slidarity PSL PiS SLD PO redistributive / pr-market

81 While the estimated lcatins fr Plish MPs presented abve supprt the hypthesis that the Plish party system is n a path twards institutinalizatin, there may be sme cncern that the idelgical labels d nt cnfrm t thse actually salient in Plish plitics. T validate these lcatins as representing idelgy, Figure 2.8 presents a cmparisn f latent lcatins t thse prvided by the Chapel Hill expert survey fr SLD, PiS, and PO (Hghe et al. 2010). 47 Clearly, the idelgical lcatins prvided by the Chapel Hill survey cnfrm clsely t thse generated by the latent path mdel. In ther wrds, it is reasnable t cnclude that the structure f party switching in the Plish Sejm is related t the idelgical rientatin f MPs and that the latent path mdel is able t recver these rientatins independent f any ther exgenus infrmatin, such as party manifests r vting recrds. 2.6 Cnclusin This chapter has made tw cntributins, ne methdlgical and ne substantive. In terms f the methdlgical cntributin, I have presented a new netwrk mdel fr dynamic, relatinal data that will be mre attractive t thse plitical scientists hping t accunt fr the dependencies inherent t relatinal data, while remaining in the mre familiar the realm f GLMs. Based n the latent space apprach t mdeling netwrk interdependencies (Hff, Raftery, and Handcck 2002), this mdel, which I call the latent path mdel, supprts binary r valued ties as well as directed r undirected ties. Furthermre, unlike alternative methds recently develped in the 47 Scial idelgical lcatin refers t galtn measure; ecnmic idelgical lcatin refers t the lrecn measure. T put the numbers n the apprpriate scale frm [ 5, 5], 5 was subtracted frm the measures f idelgy and then shifting their centrid t equal that f the reference MPs. N ther transfrmatins r rescaling was perfrmed. 66

82 Figure 2.8: Cmparing Estimated Latent Lcatins f MPs t Measures f Idelgy as Prvided by the Chapel Hill Expert Survey (2002, 2006) liberal / cnservative PSL SLD PO PSL SLD PO redistributive / pr-market literature, the latent path mdel allws fr the specificatin f nn-linear mvement f actrs in the latent space ver time. The chapter als cntributes t the substantive literature n party system institutinalizatin in new demcracies, particularly with respect t the prblem f party switching. Mst extant literature argues that rampant party switching is detrimental t the develpment f a healthy demcracy. Hwever, an applicatin f the latent path analysis t the party switching netwrk in Pland shws that such switching is nt necessarily a prblem. Instead, cntrary t cnventinal wisdm, party switching in Pland has allwed pliticians t srt int mre hmgeneus grups. Fr schlars f Plish party plitics, this lends supprt t the idea that the party system in the cuntry is finally starting t shw signs f settling dwn (Markwski 2008). Fr 67

83 parties schlars mre generally, the results suggest we lk mre clsely at the rle f party switching in the develpment f stable party plitics. 68

84 Chapter 3: Valuing Left and Right: Party Cmpetitin and Plitical Idelgy in East-Central Eurpe Intrductin The terms left and right are ubiquitus in plitics. Since defenders f the anciens régime sat n the right side f the French Natinal Assembly f 1789, while supprters f the revlutin sat n the left, these terms have structured discussins f plitical ideals in bth ppular discurse and schlarly study. Subsequent histry and plitical develpment have prduced widely accepted definitins fr left and right in mst advanced industrial demcracies: the left advcates fr plitical change in supprt f a mre egalitarian sciety, while the right favrs the status qu and grants a rle fr certain inequalities in sciety. Hw well d the cnstructs f left and right travel t plities utside f the Western plitical experience? Gaining an understanding f hw peple in different scieties talk and think abut plitics is critical t develping brader theries f vte chice, 48 This chapter was cauthred with Paul DeBell, The Ohi State University. We wuld like t thank Jessy Defenderfer, William Minzzi, Bryn Mraski, Irfan Nruddin, Gldie Shabad, and Peter Tunkis fr numerus helpful cmments and suggestins. This prject was supprted by the Ohi Supercmputer Center. Versins f this research have been presented at the Annual Meetings f the Midwest Plitical Science Assciatin (Chicag, ) and the American Plitical Science Assciatin (Washingtn, D.C., 2014) under the wrking title: Wh s Left, What s Right: Party Cmpetitin and Plitical Idelgy in East-Central Eurpe. By chance, the title was t similar t that f Zechmeister (2006); thus, we have changed it. 69

85 public pinin, and plitical cmpetitin, bth in the well-studied demcracies f Western Eurpe as well as in newer demcracies. Furthermre, as a practical matter, left-right placement is a widely used indicatr f idelgical psitin in cmparative plitics; 49 thus, understanding hw peple use this cncept in their plitical decisin making has a direct bearing n hw schlars apprach many questins in cmparative plitics. This chapter examines the rigins f the meaning f left and right with an analysis f public pinin in the yung demcracies f pst-cmmunist East-Central Eurpe (ECE). T date, there is cnflicting evidence as t what anchrs left and right in this regin. Sme have fund that left and right are rted in similar underlying psychlgical mtivatins regarding the regulatin f uncertainty and tlerance f inequality in bth the advanced industrial demcracies f the West and the pst-cmmunist cuntries f Central and Eastern Eurpe (Jst, Glaser, et al. 2003). Cnversely, mre recent studies supprt this finding regarding Western demcracies, but shw n such cnsistency in pst-cmmunist Eurpe (Aspelund, Lindeman, and Verkasal 2013; Piurk, Schwartz, and Davidv 2011). Mrever, these later studies discern n cgent pattern f meaning in self-reprted left and right placement in ECE. What shuld we make f these cnflicting empirical results? Is it truly the case that left and right mean nthing t vters in the regin? We argue that in ECE left and right functin as heuristic labels identifying grups f plitical cmpetitrs. In an analysis incrprating measures f fur fundamental psychlgical mtivatins, we shw that as idelgical labels, left and right meaningfully reflect the plitical divisins f individual cuntries. In ther wrds, in cntrast 49 See, fr example, Huber (1989), Knutsen (1995, 1997), Kitschelt and Hellemans (1990), and Evans and Whitefield (1998). 70

86 t recent wrk, we shw that the meaning f left and right is neither cnsistent with the West, nr is it meaningless r randm. Left and right are flexible terms, acquiring meaning thrugh the interactin f individuals underlying psychlgical mtivatins and the specific actins and discurse f plitical elites in each cuntry. In this way, we build ff f wrk highlighting the partisan aspect f plitical idelgy (e.g., Inglehart and Klingemann 1976; Medina 2013; Zechmeister 2006), but draw n multiple data surces t dcument the prcess thrugh which partisan cmpetitin imbues the labels left and right with values meaning amng the electrate. 3.2 Fundatins f Plitical Idelgy and the Cncepts f Left and Right Plitical idelgy plays a key rle in demcratic plitics by helping t rganize the cmplex plitical wrld in a way that engenders cmmunicatin and accuntability between citizens and plitical elites. The cncept f idelgy is, hwever, s flexible as t be difficult t grasp. The term has been used t mean everything frm the cmprehensive wrldviews f Marx and Engels, t Burke s general duality between cntinuity and change, t specific individuals explanatins fr the scial wrld arund them. When it cmes t bth ppular discurse and scientific study, hwever, plitical idelgy is almst always cllapsed int a single dimensin: the left-right spectrum. This is an versimplificatin f plitical psitins, but it is useful t pliticians and ther elites in that by referring t psitins n a single left-right dimensin, they can make plitics mre understandable t vters. In this way, this left-right semantic space helps t rganize and simplify diverse plitical attitudes and evaluatins (Fuchs and Klingemann 1989, p. 229). Cnsequently, this spectrum is the predminant 71

87 space fr party cmpetitin (Dwns 1957) and structures the way peple perceive the plitical wrld (Ldge and Taber 2013). The scial and plitical psychlgy literature n idelgy has a lng traditin f prbing the meanings f left and right. Indeed, the ntins f the right being mre resistant t change (Huntingtn 1957), less cmfrtable with ambiguity r uncertainty (Tetlck 1983), and mre authritarian (Adrn 1950; Altemeyer 1998) have lng histries. The past decade has been a particularly fruitful perid f investigatin int the psychlgical rts f left-right plitical idelgy. Schlars fcusing n the advanced industrial demcracies have dcumented stark differences between thse n the left and thse n the right in terms f deeply held ntins f mrality (Graham, Haidt, and Nsek 2009), strategies fr dealing with new stimuli (Janff-Bulman, Sheikh, and Baldacci 2008; Shk and Fazi 2009), and basic elements f persnality (Caprara et al. 2006; Carney et al. 2008). There is even grwing evidence fr a genetic cmpnent underlying liberal-cnservative plitical idelgy in the United States (Alfrd, Funk, and Hibbing 2005; Hatemi, Eaves, and McDermtt 2012). Overall, research shws that the cnstructs f left and right (r liberal and cnservative in the US) are widely-used, deeply ingrained, and cmmnly understd in Western demcracies (Fuchs and Klingemann 1989; Huber 1989; Jst 2006). This ntin that idelgical rientatins are rted in basic human psychlgy has empirical supprt. In the mst cmprehensive cntemprary psychlgical mdel f left and right, Jst, Glaser, et al. (2003) argue that left and right plitical idelgy is cnsistently anchred in the same set f psychlgical mtives crss-natinally. They argue that plitical idelgy is rted in basic rientatins twards acceptance (r resistance) t change as well as acceptance (r rejectin) f inequality. Thus, 72

88 cnservatism is mtivated scial cgnitin driven by deference t and respect fr the status qu and a series f mral arguments that justify inequality and make it tlerable (Jst, Glaser, et al. 2003). Jst et al. argue that plitical idelgy prvides a valuable service t individuals by fulfilling basic psychlgical needs. These needs include individuals relatinal mtives t have meaningful cnnectins t thers and t lcate neself in sciety, epistemic mtives t deal with uncertainty and explain the scial wrld, and existential mtives t prvide security and predictability. They g further in psiting an electral affinity between citizens and the idelgy t which peple are psychlgically predispsed, arguing that peple can be said t chse ideas, but there is an imprtant and reciprcal sense in which ideas chse peple (Jst, Federic, and Napier 2009, p. 308). 3.3 Plitical Idelgy and Left-Right in Pst-Cmmunist Eurpe While the empirical research n lder demcracies shws that plitical cnservatism is rted in resistance t change and acceptance f inequality acrss cases, the evidence fr a direct link existing in the newer demcracies f East-Central Eurpe is mixed. In their crss-natinal meta analysis, Jst, Glaser, et al. (2003) argue that, given the inclusin f Pland in the analysis, the these cnclusins hld fr the pst-cmmunist wrld. A mre recent analysis by Thrisdttir et al. (2007) largely supprts these findings, shwing that resistance t change is predictive f right-wing rientatin in ECE. Interestingly, hwever, Thrisdttir et al. (2007) fund n relatinship between identifying with the left and attitudes twards inequality in ECE. Anther study by Ksswska and Van Hiel (1999) fcusing n Pland specifically finds resistance t change t be psitively assciated with cnservatism, as wuld be 73

89 expected by the Jst paradigm. This als cmprts with lng-established arguments in plitical science that cnnect right wing rientatin t favritism fr the status qu (see Huntingtn 1957). Ntwithstanding this evidence, there are several cmpelling reasns t believe that ntins f left-right plitical idelgy may be different in ECE. The cmmunist era was a crushing experience fr the scieties it ccupied, with hmgenizing pressures s strng cmmentatrs wndered whether real scietal divides existed t be represented in plitics (Linz and Stepan 2011). While this tabula rasa hypthesis was largely unfunded (Kitschelt, Mansfeldva, et al. 1999; Shabad and Slmczynski 1999), the fact remains that the cmmunist legacy left anemic civil scieties (Hward 2003) and flattened class divisins. Indeed, there are a number f challenges t develping distinct and cnsistent left-right idelgical psitins in ECE. The simultaneus plitical, ecnmic, and scial transitins created a situatin unseen (and thus unstudied) in the cases f the lng-established advanced industrial demcracies (Haggard and Kaufman 2008; Przewrski 1991). The system f state scialism als left cleavages n ecnmic dimensins muted, such that experts predicted little party cmpetitin r idelgical differentiatin n the ecnmic cleavages that s affected plitical cmpetitin in Western Eurpe (Zielinski 2002). In additin t the ptential weakness f certain traditinal idelgical divides in the regin, vters in ECE are expsed t crss-pressures when it cmes t idelgy in at least tw ways. First, whereas idelgy is inherently bund up with the nature f the status qu (Huntingtn 1957; Jst 2006), the status qu was very different in the wake f cmmunism than in the West. Instead f an aristcracy f the right and assciated inequality being an element f the status qu, state scialism and 74

90 relative equality was the nrm fllwing the transitin. Indeed, plitical change and refrm were t becme assciated with significant increases in inequality. While right wing rientatin is assciated with a distrust f change and a tlerance fr inequality, vters wh are skeptical f change may well tend twards the plitical left in ECE, while maintaining an intlerance fr inequality demnstrated by leftists in the West. Indeed, Greenberg and Jnas (2003) argue that it is the plitical right typically seen as traditinal and wary f change crss-natinally that advcates fr change when the status qu is f the left. Similarly, McFarland, Ageyev, and Abalakina- Paap (1992) shw that in the initial years f transitin in pst-cmmunist Russia, resistance t change was a crrelate f the left side f the plitical spectrum, nt the right. A secnd, related, surce f idelgical crss-pressure is the disjuncture between stated party idelgical psitins and their actins, bth in terms f rhetric and plicy utputs. Ecnmic cnditins in the yung demcracies f ECE invlved highly cmpeting pressures: n the ne hand enrmus unemplyment and the crumbling remnants f the blated cmmunist state ecnmies required intense austerity, while n the ther hand this very austerity created majrities f newly empwered vters feeling the brunt f these ecnmic changes (Ck 2007). Parties in ECE ften dealt with these cmpeting pressures in surprising ways given their stated idelgical psitins. Carey and Reynlds (2007) argue that parties have been extremely vague in terms f prgrammatic platfrms, and Lipsmeyer (2002) finds n relatin between stated ECE gvernment idelgy and verall spending, and a psitive relatinship between pensin spending and rightist gvernments. Additinally, left-right idelgy is nt an adequate predictr f hw calitin gvernments are 75

91 frmed in ECE (Savage 2012). Tavits and Letki (2009) shw that the actual plicy utputs f gvernments n the right and left are nt in line with their stated idelgical psitins. They present evidence that gvernments stensibly n the left have been mre likely t prmulgate the austere ecnmic plicies typically assciated with the right, while parties f the right twed a ppulist line and prduced plicies that are mre prtectinist. Even measuring party platfrms in terms f left and right is a fraught prcess in pst-cmmunist Eurpe. The widely-used RILE measure f left-right party idelgy frm the Cmparative Manifests Prject, which hlds up in a meaningful way when applied t Western Eurpean demcracies, runs int difficulty when applied t thse Eurpean demcracies that experienced cmmunism (Mölder 2013). Such crss-pressures are felt by vters, with strng implicatins fr the way they use (r fail t use) labels pertaining t plitical idelgy. Indeed, even in the United States, where party cmpetitin is highly institutinalized, peple are likely t resist placing themselves n the left-right (liberal-cnservative) plitical spectrum due t cmpeting idelgical pressures (Treier and Hillygus 2009). Recent research n the relatinship between psychlgical rientatins and leftright rientatins supprts the idea that ECE is different with respect t the relatinship between left and right and underlying values. Piurk, Schwartz, and Davidv (2011) search fr cmmn predictrs f left and right in the value structure f vters in twenty cuntries. While they find discernible trends in the advanced industrial demcracies f the West, they find that values d nt structure left-right plitical idelgy in pst-cmmunist Eurpe. Similarly, Aspelund, Lindeman, and Verkasal (2013) cnduct a cmparisn f left and right plitical idelgy in Western and Eastern Eurpe, 76

92 explicitly rting their analysis in Jst and clleagues thery. In examining whether plitical cnservatism is rted in rientatins twards change and inequality, they find a strng crss-natinal relatinship in Western Eurpe but a seemingly randm relatinship in pst-cmmunist Eurpe. They shw that the relatinships between resistance t change and acceptance f inequality and right wing rientatin varies widely acrss bth cases and ver time in ECE. Bth f these studies, therefre, find little rhyme r reasn t the meaning f left and right in ECE. 3.4 A Heuristic Explanatin fr Left-Right Orientatins in ECE If left and right are nt predicted by identical underlying psychlgical mtives acrss cases in ECE, des this indicate that they are simply nt anchred in the psychlgical dispsitins f vters? We argue that left and right serve as heuristics fr grups f plitical cmpetitrs in ECE. Thus, crss-natinal hetergeneity in the psychlgical attributes f individuals that identify with the left and the right causes the differences bserved in the crrelates f idelgical self-placement in the regin. In ther wrds, left and right are meaningful in the eyes f citizens and elites, but their specific meanings are determined by factrs specific t plitical messages and events within each cuntry. A heuristic is a mental shrtcut that enables a persn t make quick judgments and decisins withut expending time r mental resurces. Thugh imperfect and frequently the cause f significant misjudgments, these simplificatins are critical t human cgnitin and behavir acrss numerus realms, and plitics is n different. Indeed, Sniderman, Brdy, and Tetlck (1993) see plitical idelgy as ne f the mst imprtant plitical heuristics. This functin f idelgy is a basic tenet f spatial 77

93 mdels s prevalent in theries f plitical behavir dating back t the seminal wrk f Dwns (1957). If left and right are labels that cme t serve as heuristics fr understanding the cnturs f plitical cmpetitin in a specific cuntry, the actins f that cuntry s plitical elites are key t understanding hw these idelgical labels btain their meaning. This is backed up by a great amunt f schlarship, even within the traditin f examining left and right in the established Western demcracies. Indeed, Jst and clleagues argue that the main factr gverning mass acquisitin f idelgical cntent seems t be attentin t and cmprehensin f infrmatin flwing frm plitical elites (Jst, Federic, and Napier 2009, p. 317). Idelgy has lng been seen as deeply rted in partisanship, with Inglehart and Klingemann (1976) arguing that ne f the central meanings f left and right in lder established demcracies is a reflectin f partisan lyalties. Medina (2013) supprts this thery with recent data including cases frm bth Eastern and Western Eurpe, and adds that the degree f plarizatin in a plity helps accunt fr the strength f the partisan element t idelgical self-placement. 50 Relatedly, Zechmeister (2006) shws that elite packaging, valence issues, and plicy psitins affect the meanings f left and right in Mexic and Argentina. At the utset f multiparty cmpetitin, the actins f plitical parties and elites are the main surce f infrmatin fr vters abut the meaning f left and right. Prir 50 While bth extant scial cleavages and party cmpetitin will surely interact in a mutually reinfrcing manner, a tp-dwn prcess f elite actins affecting mass perceptins f the meaning f left and right seems mre likely. Parties in the regin did nt bubble up frm belw ut f civil sciety, but instead emerged instantly as cartel parties, prmulgating platfrms and appeals with an eye f attracting cnstituents and nt representing natural grups f cnstituents (Mair 1996). As Tavits (2008) demnstrates, the frmatins f the party systems in the regin seem much mre tied t elite actins than anything amng the masses. 78

94 t multiparty cmpetitin, vters have a limited idea abut hw t place candidates, parties, and even themselves n an idelgical spectrum that is cnsistent with their new plitical cntext. Vters will, hwever, have a strng incentive t define this new idelgical space. Judgments are mre easily made with regard t scial grups than n, fr instance, plicy prpsals. Furthermre, idelgies may mre easily attach t grups f peple such as cmpeting plitical factins than t specific plitical issues r bundles f abstract ideas. While specific plicies ften fail t prvide idelgical cnstraint even in a plitical system as established and stable as the US (Cnverse 2006 [1964]), scial grups are an integral aspect f plitical rientatin (A. Campbell et al. 1960). Indeed, citizens ften persnalize cncepts as abstract as the state in trying t understand the plitical wrld (McGraw and Dlan 2007). These factrs make the pssibility that left-right idelgical labels serve as heuristics fr lcating plitical cmpetitrs a likely scenari. Parties in ECE emerged as cartels f elites cmpeting fr vtes and placed themselves n the left-right spectrum with the gal f attracting vtes (Innes 2002; Mair 1996). As they wrked t maximize their vtes and represent certain cnstituents, their appeals, actins, and plicies affected the ntin f what left and right mean in the eyes f vters. This des nt preclude similar psychlgical mtives underpinning left-right plitical idelgy in the regin. On the cntrary, it seems likely that the same elective affinity utlined by Jst, Federic, and Napier (2009) wuld take place. The mtives may be largely the same, but their expressin int the plitical realm, and their labeling as left and right, may be a functin largely f dynamics unique t each state and its plitical actrs. 79

95 This Heuristic hypthesis that the psychlgical predictrs f left and right are hetergeneus acrss ECE, with their meanings being determined mainly by the actins and appeals f elites and parties that adpt the left-right labels helps bridge the gap between wrk n bth sides f the debate ver what left and right have cme t mean in pst-cmmunist Eurpe. It suggests that the answer is neither cmplete adherence t the patterns bserved in the advanced industrial demcracies f the West, nr the apparent randmness mre recent studies uncver. Instead, the left and right idelgical labels are meaningful representatins cnnecting psychlgical mtives t plitical divides in ECE; hwever, the specific manifestatins f these representatins are cntingent n the dynamics f plitical cmpetitin within each case. We turn t testing this hypthesis in Sectin 3.6; hwever, first we shw that the meaning f left and right are, in fact, different in ECE. 3.5 The Psychlgical Crrelates f Left and Right Given the cnflicting evidence in the literature, and t lay the fundatins fr ur argument that left and right serve as heuristics fr individuals understanding f plitical cmpetitin, we begin ur analysis with an investigatin int the relatinship between vters psychlgical rientatins and their left-right self-placement acrss Eastern and Western Eurpe Data, Measurement, and Mdel Specificatin T peratinalize psychlgical mtivatins, we use underlying human values as put frth by Schwartz 1992; We fllw the example f Piurk, Schwartz, and Davidv (2011) and Aspelund, Lindeman, and Verkasal (2013) by drawing upn the 80

96 Schwartz Prtrait Value Questinnaire (PVQ) 51 embedded in each rund f the Eurpean Scial Survey (ESS) t analyze the extent t which underlying values relate t left-right self-placement. In their analysis, Piurk, Schwartz, and Davidv (2011) used structural equatin mdeling t identify 10 different value dimensins in the first rund f the ESS. Aspelund, Lindeman, and Verkasal (2013) use estimates derived frm a separate study (Verkasal et al. 2009) t unifrmly prject PVQ respnses frm the third (2006) and furth (2008) runds f the ESS nt a 2-dimensinal values space. In cntrast, ur analysis emplys individual Item Respnse Thery (IRT) mdels and the PVQ t peratinalize fur principal values dimensins Cnservatin, Openness, Self-transcendence, and Self-enhancement fr respndents in 23 cuntries (15 Western Eurpean, 8 Eastern Eurpean) ver all six runds f the ESS. Fr ur purpses, item respnse mdels prvide key advantages ver traditinal data-reductin tls such as factr analysis. IRT mdels estimate factr scres directly instead f depending upn factr rtatins and ther assumptins (Treier and Jackman 2002, 2008), while als being able t accurately mdel the rdinal respnses in ur data. Mst imprtantly, thugh, is ability fr IRT mdels t accmmdate missing respnses, which are cmmn in the PVQ. 52 We use IRT t extract fur dimensins because ur analyses reveal this t be the best fit t the data. The 51 The 21 item PVQ battery in the ESS measures the 10 basic human values prpsed by Schwartz (1992). This scale and the underlying value mdel it measures have been validated in hundreds f samples in ver 60 cuntries (Schwartz 1992; Schwartz, Melech, et al. 2001). Mrever, elements f this values schema have als been widely used in studies f left-right idelgy in the West, with the values categries f cnservatin and self-enhancement being highly predictive f bth right-wing plitical preferences (Barnea and Schwartz 1998; Chrs et al. 2007) and ther varius psychlgical measures assciated with plitical idelgy, such as right-wing authritarianism and scial dminance rientatin (McKee and Feather 2008). 52 Between apprximately 10,000 (4%) and 12,000 (5%) f each PVQ questin went unanswered. Listwise deletin wuld have resulted is remving mre than 26,000 (10%) f respndents frm the sample. Mre details f the IRT mdels and the results are prvided in Appendix B.3. 81

97 2-dimensinal mdel, as used by Aspelund, Lindeman, and Verkasal (2013), is ptentially prblematic fr the ECE data because the questins that lad cnsistently with particular dimensins in Western Eurpe d nt cnsistently lad with thse dimensins in the pst-cmmunist wrld. 53 The ten-dimensinal analysis that Piurk, Schwartz, and Davidv (2011) use, n the ther hand, relies n nly tw questins fr determining individual s lcatin n each dimensin (three questins fr Universalism). Given nn-respnses and measurement errr, this asks a great deal f the data. We extract fur dimensins that crrespnd readily t Jst et al. s (2003) thery f the psychlgical determinants f cnservatism. 54 Openness and Cnservatin crrespnd t acceptance and avidance f scial change, while Self-enhancement and Self-transcendence relate t the acceptance r rejectin f a rle fr inequality in sciety. 55 In rder t understand the psychlgical crrelates f left and right acrss Eastern and Western Eurpe, we use the fur value measures estimated frm the IRT mdels t predict participants left-right self-placement. We estimate a linear mdel fr each cuntry, where left-right self-placement is regressed n the fur latent dimensins as well an indicatr fr the ESS rund. 56 The theretical and empirical 53 This finding is cnsistent with Piurk, Schwartz, and Davidv (2011). 54 Schwartz and Behnke (2004, p. 251) shw that using these fur higher-rder scales is ften a desirable and valid alternative t the full ten dimensin circumplex f values Schwartz prpses. This is particularly true in cases where statistical pwer is an issue, as it is here because the PVQ in the ESS cntains nly half f the full Schwartz value battery. 55 Hw each questin in the PVQ maps t these fur dimensins is prvided in Table B.1 f the Appendix. Questins in the PVQ were recrded n a scale frm 1 t 6, with 1 being the highest level f agreement with the statement. Fr the purpse f ur analyses, this rder was reversed. 56 Left-right self-placement was recrded n a scale frm 0 t 10. While a strictly rdinal, rather than interval, scale, there were enugh categries (11) that mving t an rdered lgit wuld nt prvide any advantage and wuld needlessly cmplicate the interpretatin f the mdels. That said, a set f rdered lgit mdels were estimated and the results were in line with the OLS estimated presented here. 82

98 research discussed previusly suggests that we shuld expect Cnservatin and Selfenhancement t be psitively assciated with right-leaning respnses, while the Openness and Self-transcendence dimensins shuld be negatively assciated with rightleaning respnses, standing in fr tlerance f change and intlerance f inequality, respectively Results Figure 3.1 presents the results f these mdels fr each f the 23 Eastern and Western Eurpean cuntries in the study. In the figure, the vertical axis represents the estimated cefficient value and the hrizntal axis indicates the cuntry. Pint estimates fr each value dimensin and 95% cnfidence intervals are prvided. As higher values f the measure fr left-right self-placement indicate psitins farther t the right, psitive cefficients indicate a value dimensin that predicts identifying n the right in a given case. Negative cefficients indicate that the value dimensin is crrelated with left leaning idelgical placement. Western Eurpean cuntries are psitined n the left side f each panel, while the shaded regin n the right f each panel highlights the estimates fr pst-cmmunist Eurpe. The results cmprt the findings frm the studies by Piurk, Schwartz, and Davidv (2011) and Aspelund, Lindeman, and Verkasal (2013). That is, the expected psychlgical mtivatins predict identificatin n the left-right idelgical scale in the lder demcracies f Western Eurpe, but nt in pst-cmmunist cuntries. With the exceptin f Cyprus (a relatively yung demcracy cmpared t ther Western Eurpean cuntries), the patterns that emerge are thse predicted by Jst s mdel f plitical idelgy. Cnservatin and Self-transcendence cnfrm t the mdel 83

99 mst clsely, with the frmer pwerfully predicting right-leaning rientatin and the latter crrelating with self-placement n the left side f the plitical spectrum. Selfenhancement als cnfrms t the expectatins, typically predicting self-placement n the right in the mdels fr Western Eurpe. In the case f Openness, n clearcut pattern emerges in the Western Eurpean data, but the cefficient values are small and the verall picture in Western Eurpe remains largely cnsistent with the expectatins f existing psychlgical theries f plitical idelgy The standard deviatin fr each f the latent values estimates was apprximately 0.9. Cnsequently, a cefficient value f 0.5 fr a value indicates an increase in self-placement f just under 0.5 fr each standard deviatin increase in the latent value. 84

100 Figure 3.1: The Relatinship Between Left-Right Self-Placement and Estimated Values Orientatins in Eurpe, ESS The figure presents OLS pint estimates and 95% cnfidence intervals fr a regressin f left-right self-placement n estimates f respndent psychlgical rientatins in Eurpe. Respndent psychlgical rientatins were estimated frm graded IRT mdels using the Schwartz PVQ embedded in the ESS. Cuntry Estimated cefficient Cnservatin AT BE CH CY DE DK ES FI FR GB GR IE NL NO SE BG CZ EE HU PL SI SK UA Openness AT BE CH CY DE DK ES FI FR GB GR IE NL NO SE BG CZ EE HU PL SI SK UA Self Enhancement Self Transcendence 85

101 In cntrast, there is n cnsistent relatinship between underlying value dimensins and left-right self-placement in the new demcracies f ECE. What is mre, where the relatin between values and an idelgical label are strngest (ntably in the Czech Republic and Slvakia), the relatinships run in the ppsite directin t what we wuld expect given the wrk n idelgy in the West. Fr instance, whereas Cnservatin shuld predict self-placement n the right, in the Czech and Slvak cases it is a strng predictr f self-placement n the left. The ppsite is true f Openness in these tw cases; greater adherence t this value shuld be an attribute f the left, but in the Czech and Slvak Republics it is in fact an attribute f thse that place themselves n the right. Our analysis des nt lend credence t Jst et al. s (2003) argument that the same psychlgical mtivatins anchr left and right in ECE as in the advanced industrial demcracies f Western Eurpe. This begs the questin, what, if anything, d left and right mean in the eyes f vters in pstcmmunist demcracies? This is a questin we seek t address in the next sectin with a test f ur heuristic hypthesis. 3.6 Testing the Heuristic Hypthesis T reiterate, ur heuristic hypthesis psits that left and right btain their meaning frm the actins and messages f plitical parties assciated with the left and right. Thus, left and right d express psychlgical mtivatins in ECE as in Western Eurpe (Jst, Glaser, et al. 2003), but their specific meanings depend upn idisyncratic patterns f plitical cmpetitin within each cuntry, helping t explain the lack f a pattern that Aspelund, Lindeman, and Verkasal (2013) and Piurk, Schwartz, and Davidv (2011) bserve. This hypthesis predicts that left and right are useful labels 86

102 fr cmmunicatin between vters and elites within each case, but des nt necessarily entail crss-natinal cnsistency in terms f hw left and right crrelate with psychlgical mtivatins. In ther wrds, the labels left and right indicate plitical parties facing ff against ne anther; hwever, the specific values that each label is assciated with are a functin f hw they appealed t underlying psychlgical needs. T test this hypthesis we mdel the level f agreement between parties and individuals n the left-right spectrum as a functin f their level f agreement n the values (Cnservatin, Openness, Self-enhancement, and Self-transcendence) discussed in previus sectins. When agreement is high between an individual and a particular party n their values, the heuristic hypthesis suggests that the self-placement f the individual n the left-right spectrum wuld likewise be in agreement with the psitining f the party. Cnversely, when agreement n values is lw, the individual s left-right placement is unlikely t be cnsistent with the psitining f parties. Fr example, we wuld typically expect a party perceived as right-leaning t prmte Cnservatin values. In this case, respndents that als scre highly n Cnservatin shuld be mre likely t place themselves n the right side f the left-right scale. In ther wrds, high agreement n a value is hypthesized t predict high agreement n left-right placement. Hwever, we stress that the heuristic hypthesis des nt imply a fixed relatinship between values and left-right. If a party perceived as leftleaning prmtes Cnservatin values, vters scring high n that value shuld place themselves n the plitical left. 87

103 3.6.1 Data, Measurement, and Mdel Specificatin T test the heuristic hypthesis we fcus n the Visegrad Fur: Hungary, Pland, and the Czech and Slvak Republics. These cases are ideal fr ur purpses because they share many attributes. They are all strng demcracies that started the transitin frm state scialism with rbust anti-cmmunist ppsitin mvements, and all fur cases jined the Eurpean Unin in Hwever, as ur results frm Sectin 3.5 shw, they als prvide ample variatin in terms f which side f the idelgical spectrum the values we measure predict. These similar and frequently cmpared cases thus feature the apparent randmness in the meaning f left-right self-placement emblematic f ECE. Our test requires that we supplement the Eurpean Scial Survey data used in the previus analysis f individual values with data apprpriate fr capturing party left-right placement as well as the values expressed by each party. In this analysis, we draw n the Chapel Hill Expert Survey fr the measurement f each party s left-right placement and the Cmparative Manifests Prject t capture party values. Dependent Variable: Prximity n the Left-Right Scale The heuristic hypthesis requires a measure f cngruence between an individual s psitin n the left-right scale and a similar measure fr the psitin f each party. A measure f individual s psitin n the scale is readily attainable frm the left-right self-placement respnses available in the ESS data analyzed in the previus sectin. A measure f parties left-right placement, n the ther hand, must cme frm an independent surce in rder t capture the idelgical psitins that each party claims fr itself. Fr this we rely upn expert left-right placement f plitical 88

104 parties frm the Chapel Hill Expert Survey (hencefrth, CHES; Bakker et al. 2012). Unfrtunately, the expert placement f parties n the left and right is nly an apprximatin f the latent value f hw these parties are perceived. Hwever, it is an accurate apprximatin as these plitical experts best understand the cnturs f party cmpetitin in each case and are themselves imprtant cntributrs t the discurse f left and right that we want t measure. Individual idelgical self-placement n the ESS and the expert party idelgical placement in the CHES were bth measured n an 11-pint scale ranging frm 0 t 10, with 0 indicating placement all the way t the left, 5 a psitin exactly in the middle f the idelgical spectrum, and 10 placement as far t the right as pssible. Frm these measures f left-right we calculated the abslute difference between the individual s self-placement and the CHES placement f each party. With this measure f cngruence which ranged frm apprximately 0 t 9.5, with an average f 2.6 an increase indicates a greater distance between the individual and the party n the left-right spectrum, while lwer values indicate that an individual places herself n the left-right spectrum clse t the psitin f a given party. 58 Explanatry Variable: Agreement n Values Our explanatry variables f interest are measures f agreement between individuals and parties n the fur Schwartz values. Fr individuals, we adpt the measures derived frm ur item respnse mdels in Sectin 3.5. Fr parties, we cnstruct measures frm the Cmparative Manifests Prject (hencefrth CMP; Vlkens et al. 2014) data n the cntent f the platfrms f each party. CMP cders break each 58 Details n the distributin f the measures used in the fllwing analysis can be fund in Appendix B.2. 89

105 party platfrm int individual statements and then categrize each as belnging t ne tpic ut f an extensive cding scheme. Cders then divide the ttal number f statements in the manifest by the number f statements dedicated t each tpic. The resulting data measure the prprtin f each party manifest dedicated t each tpic in a given electin. Frm these measures we created indices based upn the Schwartz thery f human values. While the CMP data have several critics and are nt apprpriate as measures f cncrete issue psitins, 59 these measures are useful as indicatrs f the salience f particular aspects f plitics (Budge 2001). This is in line with ur theretical argument, which hlds that as parties identified with ne r the ther side f the left-right spectrum make certain values salient, citizens identifying with this value will cme t self-identify with that side f the spectrum. Thus, they prvide an unbiased, thugh rugh, apprximatin f the party value appeals we seek t explre. Mrever, the bias intrduced by using the CMP data is against supprting the heuristic hypthesis. Because these measures pick up what is psted in fficial manifests, they are ften criticized fr missing ut n what parties actually d. While we hypthesize that bth actins and appeals influence the meaning f left and right in the eyes f vters, actins are arguably mre influential t mst vters. T generate measures frm the CMP data, we matched statement categries t the fur Schwartz values. We then cnstructed an additive index f these categries fr each f the parties and centered and scaled by their standard deviatins. This prcess required that we make numerus cding decisins. In ur cding scheme we erred n the side f inclusive indices in rder t prvide the mst cmprehensive analysis 59 See Gemenis (2013) fr an verview. 90

106 pssible (see Appendix B.4). This is largely because the frequency f any given individual CMP measure varied widely acrss bth cases and time. Stable measures required that we make the indices widely inclusive. Hwever, this had the drawback f sacrificing sme precisin fr stability. The inclusin f the ecnmic measures was a particularly difficult decisin given the expectatin that ecnmic and scial values may be at dds in the regin cmpared t Western expectatins (Zielinski 2002), but given the critical imprtance f these issues in the lng transitin away frm state scialism they merit inclusin. Analyses with different values indices yield identical results t thse presented belw. With the individual and party values indices in hand, we cnstructed measures f individual-party agreement n the fur values in a similar way t left-right agreement by calculating the abslute difference between individuals and parties values. As with the measure f left-right agreement, an increase in these measures indicates greater disagreement between the party and respndent. Mdel Specificatin and Estimatin Our analytic strategy is t estimate the relatinship between values-agreement and the agreement in left-right scale using a seemingly unrelated regressin (SUR) mdel (Zellner 1962). 60 Specifically, let i {1, 2,..., I} index individual respndents, p {1, 2,..., P } the parties active in a particular perid, while v indexes the fur values: Cnservatin, Openness, Self-enhancement, and Self-transcendence. Then the respnse variable is defined as y ip = y i y p, the abslute difference in left-right scre between respndent i and party p; i.e., we are mdeling the distance between an 60 Fr plitical science applicatins f the SUR mdel, see Tmz, Tucker, and Wittenberg (2002) and Jacksn (2002). 91

107 individual s left-right self-placement and the placement f the party. The explanatry variables f interest are defined analgusly as x vip = x vi x vp ; the distance between the individual and the party n value v (there are fur such distances given the fur values). Finally, let x vi = [x vi1 x vi2... x vip ] be a vectr f these distances fr individual i n value v, X i be a 4 P matrix stacking these vectrs f distances, while β represents a vectr f 4 cefficients t be estimated n these values distances. S defined, the linear SUR mdel is as fllws: [y i1 y i2... y ip ] T N P (X T i β, Σ P ), (3.1) where Σ P is a P P cvariance matrix fr the multivariate nrmal distributin. We rely upn the SUR mdel instead f a mre straightfrward OLS regressin mdel because it allws us t take int cnsideratin the dependence in the respnse variable. T understand the surce f this dependence, take fr example a case with tw parties and ne value. If the parties psitins n the values measure are knwn, say p 1 = 0.7 and p 2 = 0.1, while the distances between the parties and the individual are als knwn, say d 1 = 0.5 and d 2 = 0.1, then the value fr individual is uniquely determined; in this example, the individual s value n the measure wuld be 0.2. Practically speaking, what this means is that the individual equatins in Equatin (3.1) are nt independent; i.e., the ff-diagnals f Σ P are nt zer. Summary statistics n these distances are available in Appendix B.2. SUR mdels were estimated in Stan (Stan Develpment Team 2013), 61 ne fr each cuntry-rund f the ESS. Since the left-right distances were highly skewed, the square rt f the respnse variable the distance between individuals and parties n 61 An example f the mdel cde is available in Appendix B.5. Stan versin was used fr all mdels. 92

108 the left-right scale, y ip was used in the mdel. An intercept was included fr each party in the mdel. Relatively diffuse prirs were specified. 62 Twelve chains were run until the ˆR cnvergent diagnstic (Gelman and Rubin 1992) was belw 1.05 fr every parameter and the effective number f samples was abve 1000 (3600 raw samples were taken fr each parameter) Results Figures shw the results frm the SUR mdels estimating the relatinship between individual-party agreement n the fur value dimensins and n left-right. Each f the fur figures shws estimated psterir means fr the estimates slpe parameters pltted n the vertical axis, while the ESS rund is indicated n the hrizntal axis. 63 Pluses dente psitive and significant estimates, while minuses indicate significant, negative estimates. Red symbls indicate significance at the 95% level, grey at the 90% level. Light grey circles indicate nn-statistically significant results. 64 Bradly speaking, the statistical results are cnsistent with the prpsed Heuristic hypthesis. While the results are nt particularly strng, it is clear that number f 62 Nrmal prirs center at zer with a standard deviatin f 2.5 were used fr the cefficients (intercepts and slpes fr the distances n values). While nt being a typical flat prir, since the respnse variable had a range frm 0 t apprximately 3, the prirs n this scale remain quite diffuse. Mre diffuse nrmal prirs were tested n select cuntry-years but made n difference t the estimatin, while the taking cnsiderably mre time t estimate. Fr the standard deviatins, half- Cauchy prir with dispersin f 2.5 were specified, while an LKJ prir (Lewandwski, Kurwicka, and Je 2009) with parameter Recall that a separate SUR mdel was estimated fr each cuntry-year. 64 Significance was determined by cmparing the apprpriate quantile value fr the psterir distributin f the selected parameter t zer. Fr instance, if the 2.5% quantile was psitive, then the estimate was seen t be psitive and significant at the 95% level; i.e., it was pltted as a red plus sign. 93

109 Figure 3.2: Explaining Respndent-Party Left-Right Agreement in Czech Republic. Psterir means (vertical axis) pltted fr each party and rund. A red plus (minus) indicates a psterir mean that is psitive (negative) and significant at the 95% cnfidence level; a grey plus (minus) indicates significance at the 90% level. Light grey circles are nn-significant estimates. Cnservatin Openness β^ β^ Self enhancement Self transcendence β^ β^

110 Figure 3.3: Explaining Respndent-Party Left-Right Agreement in Hungary. Psterir means (vertical axis) pltted fr each party and rund. A red plus (minus) indicates a psterir mean that is psitive (negative) and significant at the 95% cnfidence level; a grey plus (minus) indicates significance at the 90% level. Light grey circles are nn-significant estimates. Cnservatin Openness β^ β^ Self enhancement Self transcendence β^ β^

111 Figure 3.4: Explaining Respndent-Party Left-Right Agreement in Pland. Psterir means (vertical axis) pltted fr each party and rund. A red plus (minus) indicates a psterir mean that is psitive (negative) and significant at the 95% cnfidence level; a grey plus (minus) indicates significance at the 90% level. Light grey circles are nn-significant estimates. Cnservatin Openness β^ β^ Self enhancement Self transcendence β^ β^

112 Figure 3.5: Explaining Respndent-Party Left-Right Agreement in Slvakia. Psterir means (vertical axis) pltted fr each party and rund. A red plus (minus) indicates a psterir mean that is psitive (negative) and significant at the 95% cnfidence level; a grey plus (minus) indicates significance at the 90% level. Light grey circles are nn-significant estimates. Cnservatin Openness β^ β^ Self enhancement Self transcendence β^ β^

113 psitive and significant results utnumber the negative cefficient estimates. 65 Substantively, this indicates that a greater distance between a party and an individual n a particular value is assciated with a greater distance in left-right placement. In ther wrds, if parties emphasize values that are distant frm the views f an individual, the individual is, n average, less likely t place themselves clse t that party n the left-right spectrum. Cnversely, clseness n a value is assciated with clseness n the left-right spectrum. While the estimated cefficients largely supprt the Heuristic hypthesis, in many cases the substantive significance f these variables is nt insignificantly. Recall that the respnse variable ranged frm 0 t apprximately 3.1 (0 t 9.47 in raw distances), while the cvariates ranged frm 0 t just abve With a cefficient estimate f 0.05, mving frm the minimum t the maximum distance n a cvariate is assciated with an increase n the left-right measure f agreement f 0.3, r arund 10% f the range in the respnse variable. Similarly, a cefficient f 0.10, is assciated with a change f 0.6, r 20% f the range f the respnse. Overall, frm the psterir means presented in the figures we can see that 36 f the psitive and significant results are abve 0.05 in magnitude. Turning t the specific cuntry results, we see that results fr the Czech Republic and Pland are the strngest. Frm the Czech results in Figure 3.2, it is clear that distance n Openness and Self-enhancement are clsely assciated with distance n left-right agreement. Fr Openness, this assciatin hlds fr nearly all ESS runds and parties, fr Self-enhancement the relatinship is strngest in perids 2 and Specifically, the number f psitive versus negative and significant (at 90% level) cefficients was: 17 psitive, 6 negative in the Czech Republic; 10 psitive, 7 negative in Hungary; 22 psitive, 13 negative in Pland; and 10 psitive and 8 negative in Slvakia. 66 See Appendix B.2. 98

114 There is als a clear psitive relatinship between distance n Cnservatin and leftright agreement, especially in the latter three ESS runds, while there is n evidence fr the imprtance f Self-transcendence. The results fr Pland, shwn in Figure 3.4, als prvide strng supprt fr the Heuristic hypthesis. In this case, distance n Cnservatin and Openness are mst assciated with distance n left-right, thugh sme evidence is als prvided by Self-enhancement (except in ESS rund 2). Selftranscendence in the Plish case is, hwever, smething f an utlier, demnstrating an verall pattern f being negatively assciated with agreement n left-right. It is hard t make clear sense f this, but it is likely related t the strng, crss-cutting nature f the scial liberal-cnservative cleavage in the cuntry (e.g., see Markwski 1997, p. 236). A significant pattern in the SUR results fr Czech Republic and Pland, specifically related t the previus IRT results, illuminates the Heuristic hypthesis in actin. While the received wisdm in study f the psychlgy f idelgy hlds that Cnservatin is an attribute f peple n the right side f the plitical spectrum, this is nt the case in the Czech Republic. As the IRT results and Figure 3.1 shws, Cnservatin strngly predict self-placement n the left, nt the right, in the Czech Republic. Hwever, it is als the case that Czech parties assciated with the left side f the plitical spectrum prmulgate plicies that fall within the realm f cnservatin values, while parties n the right are less likely t d s. Thus, the clear psitive relatinship between left-right agreement and Cnservatin shwn in Figure 3.2 cnfirms that thse wh value cnservatin values are likely t share their lcatin n the idelgical scale similar t that f parties that prmulgate these values; that is t say, bth these vters and parties are n the left. 99

115 Cngruence n the Cnservatin dimensin is als an aspect f idelgy in Pland, but in the ppsite substantive directin frm the Czech Republic. Here the values f Cnservatin are prmulgated by the parties assciated with the right. Likewise, Ples surveyed by the ESS wh identify with this value are mre likely t place themselves n the right. Why is this? The Plish Cathlic Church remained ne f the cuntry s mst imprtant institutins even thrugh the cmmunist perid. Thus, it anchred bth the anti-cmmunist element f the right as well as the right s mral elements such that t be n the right side f the idelgical spectrum in Pland cnfrms much mre clsely t Western precnceptins f right-leaning cnservative values. As ne f the wrld s mst secular cuntries, the Czech Republic lacks this religius ple. There the right is mre assciated with the anti-cmmunist, antiauthritarian mvement, leading t right mre grunded in the values f Openness. While Cnservatin as an element f leftist plitics situatin may appear backwards frm the utside, it is meaningful frm the perspective f Czech vters and parties. As these examples illustrate, it is pssible fr left and right t be meaningfully rted in the psychlgical mtives f vters within tw cuntries but fr the psychlgical crrelates f left and right t be idisyncratic t each case. The results fr Hungary and Slvakia are smewhat weaker than thse fr the Czech Republic and Pland thugh, bradly speaking, patterns in the results fr these cuntries als supprt the Heuristic hypthesis. Fr Hungary, the results are strngest fr Self-transcendence, thugh the rest f estimates shw little clear pattern. This is likely a reflectin f hw plitics have evlved ver the last decade and a half in the cuntry, particularly the cllapse f the plitical left and the strng shift t a greater fcus n natinalism. Fr Slvakia, Cnservatin prvides the strngest 100

116 supprt fr the hypthesis, thugh it is nt nearly as strng as in the Czech Republic and Pland. As with Hungary, the weaker assciatin f the psychlgical rientatins with left-right self-placement may be related t the peculiar plitics f the cuntry, specifically the stunting demcratic develpment during Mečiar perid. While the link between distance n values and that n left-right placement is nt as strng in Hungary and Slvakia as it appears t be fr the Czech Republic and Pland, the pattern seen in the SUR mdels is encuraging. Pst-cmmunist party plitics are messy. The fragmentatin f ECE party systems has meant that messages cming frm the multiple f parties n the left and right ften cnflict and cntradict ne anther. Mrever, the frequent emergence f ppulist and vanity parties which rarely fit neatly int the traditinal left-right space has further cmplicated the messages vters receive. In ther wrds, ECE is a hard cases fr the develpment f clear links between values and idelgical labels. Hwever, as the these results seem t indicate, such cnsistent links rted in psychlgical mtives d appear t be develping. Finally, a shrt cmment shuld be made regarding the variance we seen in the results ver time; fr example, why in the Czech Republic the distance n Cnservatin seems less assciated with left-right agreement in earlier surveys, but appear t becme mre clsely linked in the later perids. Simply put, this shuld nt be surprising. Alng with the emergence f new parties, the changing salience and emergence f different plitical issues (e.g., the EU), is likely t cmplicate party plitics. In this envirnment, we shuld expect t see sme vlatility in the results. 101

117 3.7 Cnclusin The link between left-right self-placement and idelgical rientatins in the yung demcracies f East-Central Eurpe is nt randm. On the cntrary, within individual pst-cmmunist cuntries, the ntins f left and right make sense. As we have argued, vters cme t assciate the idelgical terms left and right with the values prmulgated by the majr plitical actrs in their cuntry. Fr example, when parties psitining themselves n the left f the plitical spectrum prfess cnservatin values e.g., resistance t change citizens hlding thse same values will begin t assciate themselves with the left side f the spectrum. On the ther hand, if a party n the right emphasizes such values, citizens hlding thse values will cnsider themselves t be n the plitical right. In simple terms, this Heuristic hypthesis says that peple identify the values rientatins f parties and then select their psitin n the left-right spectrum t match the unique plitics f their cuntry. Our findings speak directly t the debate surrunding the meaning f left and right in the new demcracies f pst-cmmunist Eurpe. Our argument and evidence prvide a cmmn grund between cnflicting views f the meaning f idelgical labels in these cuntries. On the ne hand, there are thse wh claim that the same psychlgical rientatins mtivate the meaning f these labels in these new demcracies as they d in Western Eurpe (Jst, Glaser, et al. 2003). On the ther hand, there is mre recent wrk that shws this is nt the case; instead, these cmpeting studies shw great variatin acrss cases in pst-cmmunist Eurpe when it cmes t the meaning f left and right. Fr instance, in sme cases cnservatism is assciated with the right and in sme with the left. These studies have taken this as evidence that left and right d nt mean anything, but we have shwn the plausibility f a 102

118 middle grund (Aspelund, Lindeman, and Verkasal 2013; Piurk, Schwartz, and Davidv 2011). We have shwn that neither psitin is cmpletely crrect: while the meaning f left and right are nt the same in pst-cmmunist demcracies, ur analysis cmprts the basic elective affinity that Jst (2006) and Jst, Federic, and Napier (2009) argue fr; i.e., that citizens bring psychlgical mtives t their plitical understanding, and that the expressin f these mtives is cnditined n the actins f elites. Finally, this research has imprtant implicatins fr the study f cmparative plitical behavir mre bradly. The left-right idelgical spectrum is a useful cnstruct, allwing a cmplicated plitical wrld t be simplified int mre a manageable and understandable frm. Fr this reasn, it is n wnder that its use is ubiquitus in plitical discurse as well as schlarship n vting behavir and party plitics. Hwever, while being a meaningful cnstruct, as we have shwn, left and right are flexible labels. Instead f being assciated with a fixed set f plitical views, in new demcracies left and right can becme linked with values that are the plar ppsite t thse established in the advanced-industrial demcracies f the West. In ther wrds, the left-right idelgical spectrum des nt lse meaning in these demcracies, their meaning is simply different. 67 This hetergeneity expses a need fr schlars t incrprate mre flexible measures f idelgy int crss-natinal research. 67 See Zechmeister (2006, p. 170) fr a similar warning abut treating the left-right labels hmgeneusly acrss cases. 103

119 Chapter 4: The Cheap Seats: Party Develpment and Lcal Electral Refrm in Pland Intrductin Stable plitical parties with rbust linkages t sciety are widely seen t be a critical cmpnent f successful, lng-lasting, and representative demcratic gvernance (Huntingtn 1968; Kitschelt, Mansfeldva, et al. 1999; Innes 2002; Mainwaring and Scully 1995). But building a successful plitical party in a mdern demcracy is difficult. In Western demcracies, parties were able t c-pt lng-standing scial structures the extensive linkages rted in class and cnfessinal memberships (Lipset and Rkkan 1967) which helped defray the rganizatinal csts assciated with party building, and made the mass party a viable rganizatinal mdel in many Western demcracies (Katz and Mair 1995). This mdel was nt, hwever, a viable ptin in pst-cmmunist demcracies, where the requisite scial structures were largely undernurished and inchate due t decades f cmmunist rule (Mair 1997, p. 177) and fr which demcratizatin was ccurring in an verall envirnment f 68 I wuld like t thank Paul DeBell, Luke Keele, William Minzzi, Frank Thames, and Peter Tunkis fr their helpful suggestins. This prject was supprted by the Ohi Supercmputer Center. An earlier versin was presented at the Annual Meeting f the Midwest Plitical Science Assciatin (Chicag, 2012). 104

120 dealignment (Daltn and Weldn 2007). In the absence f a strng scial fundatin, party leaders in new demcracies were frced t lk t alternative strategies fr building lasting party rganizatins. Eschewing strategies requiring extensive party building effrts, pst-cmmunist party leaders have instead turned t alternative rganizatinal mdels that leverage the mass media and charismatic party leaders in rder t attract vtes (Kpecký 1995). While nt ging s far as being cmpletely cartelized (Innes 2002, p. 91), this strategy reflects the cartel-like reliance f parties n the state, which ften finance party activities and grants airtime n media netwrks (Kpecký 2006, p. 256). 69 In a sense, the availability f the mass media prvides parties the lwest marginal cst fr each vte btained. Hwever, by many measures, the reliance n the mass media and tp-dwn rganizatinal structures has nt prduced the cnsistently stable and representative parties ne may hpe fr in a healthy demcracy. Instead, as is ften remarked in the literature, party and party system cnslidatin in pst-cmmunist cuntries has been slw t materialize (Lewis 2000; Bielasiak 2002; Bakke and Sitter 2005; Epperly 2011). This chapter investigates the rle f lcal electral institutins in the prcess f party building. Recent research has demnstrated the imprtance f party rganizatins, including lcal party branches, t party building in the yung demcracies f East Central Eurpe (Tavits 2012, 2011). While there is a lng and influential literature discussing the imprtance f natinal electral institutins n party system develpment (Duverger 1954; Cx 1997; Clark and M. Glder 2006), the imprtance 69 Of the 10 pst-cmmunist Eastern Eurpean cuntries they survey, van Biezen and Kpecký (2007) find that in 8 parties get direct funding frm the state. 105

121 f subnatinal electral institutins in this prcess is unexplred. 70 The mtivating hypthesis f this chapter is that lcal electral institutins can play an imprtant rle in natinal party develpment in new demcracies. This is fr tw reasns: (1) Different lcal electral institutins prvide different incentives fr lcal pliticians t jin natinal parties. (2) Lcal pliticians jining natinal parties creates incentives fr natinal parties t becme invlved in lcal plitics. I test the abve hyptheses by leveraging an unique lcal electral system discntinuity in Pland. Chice f lcal electral institutins and verall natinal party system develpment are likely highly endgenusly determined. As a cnsequence, the effects f lcal institutins n natinal party develpment is difficult t identify in stable electral and party system cntexts. A careful analysis f the discntinuity in lcal electral institutins in Pland helps me avid this endgeneity. In 1998, Pland s newly elected Slidarity calitin passed a brad package f lcal gvernment refrms. These refrms changed hw a certain class f municipalities elect their cuncil members. Fr a subset f municipalities, cuncil members wuld n lnger be elected in single-member districts and wuld instead switch t an pen-list prprtinal representatin system. Taking advantage f this discntinuity, I use a regressin discntinuity (RD) design apprach t determine if the switch frm SMD t PR had an effect n natinal party participatin and perfrmance in lcal electins as well as in natinal electins. I shw that in thse municipalities that experienced the change frm SMD t PR, the perfrmance f natinal parties imprves significantly in bth lcal and natinal electins. In ther wrds, the natinalizatin f the party system 70 Chhibber and Kllman (2004) prvide an excellent study f the imprtance f subnatinal plitics in the natinalizatin f party systems; hwever, their study fcuses n the nature f issue salience whether issues are natinal r subnatinal in scpe nt n electral institutins. 106

122 was aided by the change in the electral system. This highlights the imprtance f lcal plitics in the verall party and party system develpment. 4.2 Lcal Plitics and Natinal Parties in Pst-Cmmunist Cuntries Party develpment in pst-cmmunist Eurpe has been largely a tp-dwn affair (Kpecký 1995, 2006). Natinal elites have priritized parliamentary plitics, while simultaneusly using brad, shrt-term strategies t capture as many vtes as pssible. This is in cntrast t the develpment f party plitics in Western Eurpe, where cmpetitin was fcused n turning ut the vte f particular grups r classes f vters. There are several reasns why elites in pst-cmmunist Eurpe adpted this apprach. First, unlike in Western Eurpe, the flattened scial envirnment f pstcmmunist sciety did nt encurage the rganizatin f party cmpetitin arund a set f deep scial cleavages. Cnsequently, there were few natural divisins that culd be explited by party leaders fr the purpse f defining plitical cmpetitin and mbilizing vters (Mair 1997). Secnd, the cmmunist era created a great distrust f parties and plitics in general (Jwitt 1992, p. 215). Instead f representing the ideals f demcratic plitical cmpetitin, because they were assciated with the prir authritarian regime, parties and plitics mre generally were viewed quite cynically. As O Dwyer (2006, p. 123) puts it, tw defining features f pstcmmunist plitics are their dembilized scieties and delegitimized states. One manifestatin f this, was the ppular steretype that invlvement with plitical parties was a shameful activity t be frwned upn (Szczerbiak 2001b, p. 166). Third, parties clse ties t the state have meant that grassrts supprt was nt needed in rder t effectively 107

123 cmpete in electins. State funding and state mandated access t the media prvided the material resurces party leaders needed t cmpete. In ther wrds, in a very real sense, it was mre cst effective, frm party leaders perspective, t resist investing in party building effrts (Kpecký 1995). While natinal parties have nt spent a great deal f effrt n party building, there are very real electral and rganizatin advantages t develping strng party rganizatins and this has nt been lst n many pliticians. Tavits, fr instance, ntes that party fficials in Estnia have cited the imprtance f having active party membership as well as a netwrk f visible party ffices in develping ties t sciety (Tavits 2012, pp ). And the regeneratin f many ex-cmmunist parties was in great measure facilitated by the legacy f cmmunist-era party rganizatins (Grzymała-Busse 2002). Hwever, while there are certainly exceptins, verall initial incentives favred tp-dwn rather that bttm-up parties. The argument being made in this chapter is that lcal electral institutins can play an imprtant rle in shaping the develpment f natinal parties and party systems. Tw mechanisms individual candidate decisins t enter plitics and jin natinal parties and party incentives t invest in lcal party peratins link lcal electral institutins t the develpment f natinal parties. These are discussed in the fllwing sectins Lcal Elite Incentives The first mechanism tying lcal electral institutins t natinal party develpment is the way in which electral institutins shape the incentives f lcal pliticians t jin natinal parties. It is a fundamental finding in the study f party plitics that 108

124 electral institutins play a vital rle in shaping the character f parties, party systems, and the brader demcratic cntext. Electral institutins have been fund t affect everything frm the number f parties (Duverger 1954; Clark and M. Glder 2006; Amrim Net and Cx 1997), t turnut (Cx 1999; Blais 2006), t redistributin (Iversen and Sskice 2006). An imprtant mechanism thrugh which electral systems affect party systems is by shaping the incentives fr prspective pliticians t participate in electins. Pliticians will enter plitics when they believe they have a decent prbability f winning a seat, and this prbability is a functin f the vte share required t win. In the simplest case f single-member districts with first-past-the-pst vting, the vte share required t win a seat is the highest (50% + 1 in the case f a 2 persn race). In prprtinal representatin systems, where multiple candidates are selected fr each district, the share f the vte needed t win a seat will be lwer. Cmparing these tw examples, a prspective candidate wuld clearly be mre likely t enter the race in a PR system with a high district magnitude than in they wuld in a SMD system with plurality vting (Cx 1997). Anther key decisin a candidate faces is whether t run as an independent, jin an existing party, r t start ne f her wn. Of curse, while running as an independent in a candidate-centered electral system may be viable, this is nt ften pssible in the electral systems that use party lists. In such circumstances, the candidate will have n chice but t chse between jining an existing party r starting her wn. There are csts and benefits t bth ptins. On the ne hand, if the candidate jins an existing party, she will pssibly benefit frm financial supprt frm the party and the party label will be a knwn entity t vters. On the ther hand, jining a party 109

125 means being frced t sme degree t adpt that party s platfrm and t tw the party line n issues that may arise during the electins. 71 By starting her wn party, a candidate will have much mre flexibility t run n a platfrm she sees as mre cmpatible with the electrate and her wn views. Hwever, the csts f starting a party are high. Nt nly wuld the candidate need t pass any legal hurdles necessary t be recgnized as a party, she wuld als face the substantial barrier f building familiarity with the electrate (Aldrich 1995). 72 A candidate fr lcal ffice faces these same incentives. As is the case fr a prspective plitician at the natinal level, a candidate fr lcal ffice will base her decisin t enter a race by weighing her prspects fr winning against the csts f running. She will likewise face the decisin f whether t jin an existing party r t start her wn. Hw will this affect natinal party develpment? Simply put, lcal electins that rely n party lists and multi-member districts are mre likely than candidate-fcused electral institutins t (1) encurage prspective candidates t run fr ffice and (2) t see these candidates jin natinal parties. The first hypthesis fllws frm the abve argument that PR systems create mre pprtunities fr prspective pliticians, thus attracting a greater number f candidates. 73 Furthermre, in new demcracies, the effect f electral systems n new candidate entry is particularly strng. Lcal ffices in new demcracies, such as 71 This is even the case in East-Central Eurpe. As Szczerbiak (2001b, p. 57) ntes that while the lcal party fficials in Pland were given a great deal f autnmy n purely lcal issues,... [m]st party statutes cntained clauses that empwered the central ffices t intervene in the activities f any lcal branch when it felt that the party prgram r statute was being vilated r when the party was suffering damage natinally. 72 As Aldrich (1995) shws, there are als significant legislative incentives t jin existing parties. Thse are beynd the fcus f this argument. 73 This is knwn as the strategic r psychlgical effect f electral institutins n plitician and party participatin (Clark and M. Glder 2006; Duverger 1954). 110

126 thse f pst-cmmunist Eurpe, are ften captured by lcal elites (e.g., see O Dwyer 2006, ch. 5). In such situatins, there is little space fr the entry f new pliticians. Incumbent advantage keeps prspective pliticians ut f lcal races, particularly when the electral system is candidate-centered. PR in this case will prvide greater space fr new candidates t cmpete fr ffice. The secnd hypthesis that multi-member districts will encurage lcal candidates t jin natinal parties fllws frm the simple requirement fr candidates in PR systems t belng t a party. Every new candidate entry represents a ptential recruit fr natinal parties. If starting a party is cstly, relative t jining an existing ne, then natinal parties are a natural chice fr new candidates. This is particularly the case when natinal parties are well develped, as they wuld prvide the mst benefit in terms f visibility, rganizatin, and resurces; hwever, the attractiveness f natinal parties t candidates fr lcal ffice is als likely t be significant Natinal Party Incentives Once they have successfully attracted lcal pliticians t their banner, natinal parties have an incentive t expend resurces n lcal party develpment. This is the secnd mechanism linking lcal electral institutins t natinal party develpment. The mtivatins fr getting invlved stems frm parties need t build and maintain their brand. Parties will nt want t accept just any lcal candidates t run n their labels, and they will nt want these candidates t run n plicies that cntradict the values and psitins f the natinal party. Cnsequently, natinal parties will be mtivated t exert cntrl ver candidates, either thrugh their cntrl lcal party lists, as in PR, r by restricting membership under SMD. This will require that parties 111

127 develp the means f evaluating ptential lcal candidates and drawing up, r at least apprving, lcal lists. While cstly, there are significant spillver benefits t getting invlved in lcal electins, which have the ptential t strengthen natinal parties ver the lng term. First, as Geser ntes, establishing a lcal presence allws parties t maximize their chances f generating a large reservir f experienced yung adherents frm which future candidates fr higher rles r public psitins can be recruited (Geser 1999, p. 6). Secnd, investing in lcal rganizatin and participating in lcal plitics prvides the pprtunity fr natinal parties t test new, alternative campaign strategies. In the case f electral defeat at the natinal level, lcal plitics prvides a venue fr a party t regrup (Geser 1999, pp. 7 8). 74 Recent empirical research supprts this argument that lcal party rganizatin matters t natinal party develpment (Tavits 2012). Finally, a lcal presence prvides the means thrugh which lines f cmmunicatin frm the lcal t the natinal level and back can be established. Issues at the lcal level can be mnitred at the natinal level. Thus, this wuld make natinal plitics mre relevant t the electrate Observable Implicatins In the previus sectins, I have argued that different lcal electral institutins shuld affect natinal party develpment in different ways. Specifically, PR at the lcal level will prvide the incentives fr prspective lcal pliticians t jin natinal parties. Faced with the need t integrate new members, parties will be frced t develp rganizatins t manage lcal lists. Hwever, while cstly t set up, these 74 Als see (Tavits 2012, p. 86). 112

128 rganizatins will generate psitive spillver effects fr the party. They will (1) prvide the means thrugh which the party can the attract and train new members; (2) prvide a venue fr new electral and plicy strategies t be develped; and (3) develp int cnduits thrugh which infrmatin can pass between the lcal t the natinal levels. In sum, relative t plurality electral rules, prprtinal representatin in lcal electins shuld benefit natinal parties and, thus, shuld cntribute t the develpment f rbust party systems. There are tw immediate bservable implicatins assciated with this argument. First, if PR creates greater incentives fr lcal pliticians t jin natinal parties, verall natinal party perfrmance in lcal electins shuld be better in PR electral systems relative t their perfrmance in plurality systems. This wuld be particularly the case if higher quality candidates thse mre likely t win electins as independents in plurality electins are induced t jin natinal parties in PR systems. Secnd, if lcal plitics matter at all t natinal party and party system develpment, we shuld bserve sme effect f the presence and perfrmance f natinal parties at the lcal level n their perfrmance at the natinal level. In ther wrds, the increased expsure f natinal parties at the lcal level shuld translate int a greater familiarity n the part f vters and imprved party rganizatin at the lcal level, resulting in imprved perfrmance in natinal electins. The fllwing empirical investigatin will fcus n the first f these implicatins that PR leads t imprved perfrmance f natinal parties in lcal electins by taking advantage f a unique discntinuity in electral laws in Pland. 113

129 4.3 The Test Case: Lcal Electral Refrm in Pland The relatinship between lcal electral institutins and party develpment is highly endgenus. This makes a clear test f the prpsed hyptheses difficult. Fr instance, in established and stable plitical systems the stated hypthesis that sme lcal electral institutins facilitate natinal party develpment while thers inhibit such develpment wuld be empirically indistinguishable frm the inverse hypthesis that highly develped natinal parties in particular electral cntexts chse particular lcal electral systems. In ther wrds, des PR at the lcal level fster natinal party develpment r d highly develped natinal parties instead chse PR? While I have argued fr the frmer, there are gd reasns t believe the latter culd als be at play. Simply put, strng parties may institute PR fr lcal electins fr reasns that have nthing t party develpment. As a cnsequence f this prblem f endgeneity, nly a research design that takes advantage f sme surce f randmizatin can help us parse ut the effects f lcal electral institutins n natinal party develpment. A change in lcal electral institutins in Pland prvides a great pprtunity t d just that. In June 1998, the Plish parliament enacted a wide-ranging lcal gvernment refrm prgram. Cmprehensive lcal gvernment refrm had lng been a gal f the plitical right, and had becme an imprtant cmpnent f Slidarity Electral Actin s (AWS) platfrm in the natinal parliamentary campaign f Once AWS had wn the electin and established a calitin gvernment with Freedm Unin (UW), passing lcal gvernment refrm became a key pririty fr the gvernment. Balcerwicz, the architect f shck therapy and UW party leader, was particularly supprtive f refrms, while many far-right members f the calitin were ppsed 114

130 Figure 4.1: Plish Municipalities Affected by the Change in Electral System, The figure shws the gegraphic bundaries f all 2,478 Plish municipalities. The 194 municipalities between 20,000 and 40,000 residents affected by the change in electral law are highlighted in red. (Szczerbiak 1999). 75 The legislatin that ultimately passed parliament included significant changes at all levels f lcal gvernment, including the reductin f the number f prvinces (wjewództwa) frm 42 t 16 and the establishment f a new cunty (pwiat) level f gvernment, f which 308 were initially created (Regulski 1999) The natinalist and religius right was adamantly ppsed t the refrms and, in prtest t the legislatin, 15 MPs defected frm the calitin, 7 frm the Cnfederatin fr an Independent Pland Patritic Camp (KPN-OP) and 8 frm Plish Family Assciatin (SRP) (Szczerbiak 1999, p. 86). While being the prject f the Slidarity calitin, the ppsitin cmmunist successr parties were clsely invlved in the negtiatins that shaped the legislatin. As O Dwyer (2006, PAGE) pints ut, patrnage was critical t gaining supprt. 76 Of the 308 cunties created, 65 were larger municipalities that were als granted cunty status. 115