Territory and Electoral Rules in Post-Communist Democracies

Territory and Electoral Rules in Post-Communist Democracies This page intentionally left blank Territory and Electo...

Author: Daniel Bochsler

68 downloads 482 Views 2MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Territory and Electoral Rules in Post-Communist Democracies

This page intentionally left blank

Territory and Electoral Rules in Post-Communist Democracies Daniel Bochsler Research Fellow, Centre for the Study of Imperfections in Democracies (DISC), Central European University, Hungary

© Daniel Bochsler 2010 All rights reserved. No reproduction, copy or transmission of this publication may be made without written permission. No portion of this publication may be reproduced, copied or transmitted save with written permission or in accordance with the provisions of the Copyright, Designs and Patents Act 1988, or under the terms of any licence permitting limited copying issued by the Copyright Licensing Agency, Saffron House, 6-10 Kirby Street, London EC1N 8TS. Any person who does any unauthorized act in relation to this publication may be liable to criminal prosecution and civil claims for damages. The author has asserted his right to be identified as the author of this work in accordance with the Copyright, Designs and Patents Act 1988. First published 2010 by PALGRAVE MACMILLAN Palgrave Macmillan in the UK is an imprint of Macmillan Publishers Limited, registered in England, company number 785998, of Houndmills, Basingstoke, Hampshire RG21 6XS. Palgrave Macmillan in the US is a division of St Martin’s Press LLC, 175 Fifth Avenue, New York, NY 10010. Palgrave Macmillan is the global academic imprint of the above companies and has companies and representatives throughout the world. Palgrave® and Macmillan® are registered trademarks in the United States, the United Kingdom, Europe and other countries. ISBN: 978–0–230–24827–4 hardback This book is printed on paper suitable for recycling and made from fully managed and sustained forest sources. Logging, pulping and manufacturing processes are expected to conform to the environmental regulations of the country of origin. A catalogue record for this book is available from the British Library. Library of Congress Cataloging-in-Publication Data Bochsler, Daniel, 1978– Territory and electoral rules in post-communist democracies / Daniel Bochsler. p. cm. ISBN 978–0–230–24827–4 (hardback) 1. Political parties – Europe, Central. 2. Political parties – Europe, Eastern. 3. Europe, Central – Politics and government – 1989– 4. Europe, Eastern – Politics and government – 1989– 5. Post-communism – Europe, Central. 6. Post-communism – Europe, Eastern. I. Title. JN96.A979B63 2010 324.0947—dc22 10 9 8 7 6 5 4 3 2 1 19 18 17 16 15 14 13 12 11 10 Printed and bound in Great Britain by CPI Antony Rowe, Chippenham and Eastbourne

2010007405

Contents List of Figures

viii

List of Tables

x

Notation of Variables

xii

Preface 1

2

3

4

xiii

Introduction: Electoral Systems and Party Systems in the New European Democracies 1.1 Previous research 1.2 Relevance and innovations of this study 1.3 A short guide through this book 1.4 Conclusions The Puzzle of Electoral Systems’ Impact in Post-Communist Democracies 2.1 Introduction 2.2 Electoral systems and party systems – the conventional view 2.3 The empirical puzzle: And it all looks like nonsense ... 2.4 Conclusions: Do electoral systems not work in post-communist countries? How to Measure Party Nationalisation 3.1 Introduction: The relevance of party nationalisation and the lack of a reliable measure 3.2 Previous measures of party nationalisation 3.3 The Gini coefficient to measure regional heterogeneity 3.4 Developments on the party nationalisation score: The impact of the size of territorial units 3.5 Testing the ‘standardised party nationalisation score’ on Central and Eastern European data 3.6 Conclusions Explaining the Nationalisation of Party Systems in Central and Eastern Europe 4.1 Introduction: Why study party nationalisation? v

1 2 6 11 13 15 15 17 30 36 37 37 39 45 47 52 56 58 58

vi Contents

4.2 4.3 4.4 4.5 5

6

7

Four explanations for party nationalisation Operationalisation Party nationalisation in Central and Eastern Europe Conclusions: Determinants of party nationalisation

How Party Systems Develop in Mixed Electoral Systems 5.1 Introduction 5.2 Mixed electoral systems: Terminology and application 5.3 How much compensation in mixed compensatory electoral systems? 5.4 Economics of votes: The run for better conversion rates 5.5 Party system stability and party system change 5.6 The consequences of mixed electoral systems – empirical study 5.7 Identifying party system equilibria 5.8 How to transform a compensatory into a non-compensatory system 5.9 The best of both worlds? Preliminary conclusions and suggestions for further research Counting Votes and Places: The Joint Effect of Electoral Rules and Territory 6.1 Introduction 6.2 About methodology: Going beyond directional models in political science 6.3 How to resolve the Central and Eastern European electoral system puzzle? 6.4 A model of district magnitude, number of districts and party nationalisation 6.5 The importance of national electoral thresholds 6.6 Umbrellas and rainbows – the early development of the new party systems 6.7 Testing the full party nationalisation model 6.8 A comprehensive model for the fragmentation of party systems 6.9 Conclusions Conclusions: An Institutional Model to Predict the Number of Parties in Central and Eastern Europe 7.1 A two-step model to investigate the effect of party nationalisation and electoral systems 7.2 Territorial ethnic divides, super-presidencies and party nationalisation

60 65 67 83 86 86 89 93 96 102 108 117 123 129 132 132 134 135 140 144 147 149 156 161 163 164 165

Contents vii

7.3 7.4 7.5 7.6 7.7 7.8

The joint impact of party nationalisation and district magnitude A comprehensive model for the fragmentation of party systems Mixed electoral systems – the surprise packet among the electoral systems Institutions work Relation to previous studies of electoral systems Party nationalisation and the size of party systems in different contexts

166 168 168 169 170 171

Appendices Appendix A Appendix B Appendix C

174 174 175 176

Notes

177

Bibliography

197

Index

213

Figures 1.1 2.1

2.2 2.3 2.4 2.5 2.6 2.7 3.1 4.1 4.2 4.3 4.4 5.1 5.2 5.3 5.4 5.5

5.6

Main model for this research project Effective number of parties in parliaments in Central and Eastern Europe, 1990–2007, time series for the number of the election Electoral systems and their impact on party system fractionalisation Effective threshold and the number of parties Balancing effects on the number of parties in single electoral districts: Determining the parameter k Three concepts to explain the number of parties in a party system Explaining the effective number of parties by district magnitude, 1990–2007 Explaining the effective number of parties through national legal thresholds, 1990–2007 The calculation of the party nationalisation score with weighted units in a fictitious country Party system nationalisation in twenty countries, development by country over time, 1990–2007 Budget centralisation and party nationalisation (PR tier), 1998–2001 Party nationalisation in PR elections, according to the OLS regression model Party nationalisation in the PR tier and in the single-seat district tier in mixed electoral systems Development of the effective number of parties in parliament under mixed electoral systems Semi- and fully compensatory systems and break-even point, where systems are just compensatory Contamination of the PR tier and the plurality/majority tier The relationship of volatility and level of coordination Development of the number of parties, according to the level of electoral coordination in previous elections, and according to the level of volatility Party concentration effect of low volatility

viii

8

20 21 25 27 29 32 34 46 68 74 80 82 88 94 110 114

115 117

Figures

5.7 5.8 6.1 6.2

6.3 6.4

7.1

Typology and dynamics of party system change under mixed semi- or non-compensatory electoral systems The impact of strategic vote splitting on the electoral success of small parties Two explanations for the high number of national parties in single-seat district systems Example of party nationalisation and the effective number of parties in a 200-seats parliament with 10-seats districts and k 1/3 Development of the number of parties over time Number of parties expected by partial functions of the model, according to legal thresholds, and according to the number of districts, district magnitude, and party nationalisation The two-step model to explain the number of parties in Central and Eastern European party system

ix

118 128 137

142 154

156 165

Tables

2.1

2.2

3.1 3.2 4.1 4.2 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 6.1 6.2 6.3 6.4

OLS regression models for the effective number of parliamentary parties (log), district magnitude and number of seats in parliament as explanatory variables OLS regression models for the effective number of parliamentary parties (log), with the legal threshold and a combined term as explanatory variables Indicators to measure the nationalisation of party systems and their discussed shortcomings Calculation of the relative deviation between the upper level and the lower level for both indicators Most important ethnic minority parties with a territorially concentrated structure of voters OLS regression models to estimate party system nationalisation Terminologies regarding mixed electoral systems Mixed electoral systems in Central and Eastern Europe, 1990–2007; types and application Overview of the variables used in the model of party system change OLS regression models for strategic coordination of candidate entry OLS regression models for the effective number of elective parties Types of outcomes of mixed electoral systems Four types of party systems under mixed non– or semi– compensatory systems Electoral results in Albania 2005 Two types of ‘failures’ of the concentrating effect of electoral systems’ characteristics OLS regression models for the effective number of parliamentary parties OLS Regression models for the effective number of parties OLS regression models for the effective number of parties, no intercept included x

33

35 40 54 71 77 90 92 109 112 116 119 122 126 136 144 152 153

Tables

6.5 6.6 A.1

OLS regression models for the effective number of parties, simplified model OLS regression model for the effective number of parties, global model Random–effects GLS regression for the effective number of parliamentary parties

xi

155 159 176

Notation of Variables c comp d e eC m N0 N2 NV p1 p1SSD . P1 . P1SSD pres q S SPR SSSD SSD t ∆t TD TE T ∆T

Share of compensatory mandates in mixed electoral systems [c 1 SSSD/S] Necessary share of compensatory mandates for a just compensatory system Number of electoral districts Ethnic fractionalisation (Hirschmann–Herfindahl index) Territorially based ethnic groups Average number of mandates by electoral district Absolute number of parties Effective number of parties (Laakso and Taagepera, 1979) Effective number of parties, based on their vote share Largest seat share of single party Largest seat share of single party in the plurality/majority tier of mixed electoral systems Absolute number of seats held by largest party in parliament . [P1 p1 · S] Absolute number . of seats held by largest party in plurality/ majority tier [P1SSD p1SSD · SSSD] Super-presidential systems (dummy variable) Parameter for the comparison of district-based PR with threshold-based PR systems Overall number of mandates in parliament Number of mandates in proportional or compensatory tier of mixed electoral systems Number of mandates in plurality/majority tier of mixed electoral systems [SSSD S · (1 c)] Single-seat district system (dummy variable) Legal threshold System with legal threshold (dummy variable for t 0) District threshold (1/(m 1)) Effective threshold – the higher of the district and the legal threshold Ordinal number of elections First multiparty elections (dummy variable for T 1)

xii

Preface This book is based on three years of research, which I started at the end of 2004 at the University of Lausanne, continued after 2005 at the University of Geneva, and completed in 2008. I have been lucky to find many people who gave me the support that I needed for my project. Foremost, I am deeply indebted to Pascal Sciarini and to Rein Taagepera. Pascal Sciarini has not only encouraged me to tackle this research, but, throughout the whole research and writing process, has given me full support as a very valuable advisor. His qualities as a critical commentator were very helpful not only for the development of my research design but also for the structuring and writing of the book. And his backing was crucial for the organisation of the research stays abroad. Thanks to this, I had the good fortune to work with Rein Taagepera, whom I followed to Estonia (Tartu Ülikool) and to the USA (University of California at Irvine). Rein Taagepera was not only always approachable and open for inspiring discussions, but he also made extensive in-depth comments which helped me to develop my project a great deal further and to sharpen my analysis. I am also indebted to Eugène Horber, Simon Hug and Jan-Erik Lane, who helped me with valuable comments and suggestions. I obtained crucial support from Alex Fischer, who was always extremely supportive for all aspects of my work – from conceptual discussions over the writing process, encouragement when I was becoming overwhelmed with dissatisfaction with my own work, up to organisational support for my research stays. During my research in Switzerland, the research stays in Estonia, California and Serbia and at the Central European University in Hungary, and at conferences and workshops, I could count on many colleagues, among whom I shall mention particularly Mirjam Allik, Aina Gallego, Sergiu Gherghina, Fabrizio Gilardi, Bernard Grofman, Ana Haro, Anthony McGann, Sebastian Maier, Evald Mikkel, Friedrich Pukelsheim, Carsten Schneider, Matthew Shugart and Sakura Yamasaki, but the list could certainly be longer. They all were involved in discussions about particular aspects of this study, or provided me with relevant comments on earlier papers. Besides this, I obtained valuable comments from further discussants and participants at conferences.

xiii

xiv Preface

A number of further members of the three host universities which I visited were helpful in organising my research stays. This book consists not only of ideas, but of numbers too. This required an extensive search for data, which was not always very easy, but many people have helped me to build an almost complete dataset. Among them are Marijana Bijelic´, academics such as Sarah Birch, Andrey Meleshevich and Tomas Zarycki, and officials of NGOs, Electoral Commissions and the OSCE/ODIHR, namely Zane Duze, Arben Idrizi, Diana Railean, Gilles Saphy and Zlatko Vujovic´. Important help was further provided by John Harbord and Kai Ostwald, who both proofread the manuscript. I thank Liz Blackmore and Alison Howson at Palgrave for their support, and the team at Newgen Imaging Systems for careful editing of the book. In addition, I want to express my particular gratitude to my family, who have supported me in my work. Finally, I am indebted to the Swiss and European taxpayers, who generously financed the work on this project partially with a scholarship of the Swiss National Foundation (grant nr. PBGE1–112936).

1 Introduction: Electoral Systems and Party Systems in the New European Democracies

The investigation of electoral systems’ effects in post-communist democracies has so far led to very surprising results. The effect of electoral systems in Central and Eastern Europe appears ‘not [to] be the same’ as in the rest of the world (Golder, 2002, p. 24); several studies conclude (Clark and Golder, 2006, p. 693; Grzymala-Busse, 2006, p. 421; Moraski and Loewenberg, 1999) that the common models that are valid in the rest of the world seem not to hold when they are tested on post-communist countries in Europe. Attempts to explain patterns of party systems through electoral systems often lead to results that fundamentally contradict common knowledge on electoral systems.1 Usually, elections in single-seat districts (or single-member districts) are associated with two-party systems, whereas elections with proportional representation (PR) allow the existence of large multiparty systems. But elections in post-communist Europe seem to invert this relationship, leading to either nonsensical or no results. Repeatedly, a much higher number of parties have resulted from single-seat districts than from PR elections. Overall, variables which are central in common studies on electoral systems are not conclusive when applied to post-communist countries in Europe. This result is more than a little puzzling. Are party systems in post-communist democracies non-institutionalised and just chaotic, so that one should avoid trying to make sense of them? Do they possibly confirm the limits of electoral system approaches? In contrast to many earlier conclusions, my study shows that, after an omitted variable is accounted for, electoral systems work even in post-communist countries in Europe. To put it briefly: it does not matter only how many votes you get, it matters where you get them from. If we want to understand how electoral rules affect party system fragmentation in post-communist countries, we need to analyse the territorial 1

2


structure of the vote, which has recently become a very popular aspect of party research, discussed under the slogan ‘party nationalisation’ (see Caramani, 2004; Chhibber and Kollman, 2004). Party nationalisation is understood as the territorial homogeneity of party support, or as the reverse of a regionalised system of parties, characterised by territorial splits. This is the first study that offers a description and an analytical discussion of party nationalisation in Central and Eastern Europe, and the first that employs this concept to explain the effect of electoral systems. My study is based on possibly the largest database on electoral results from Central and Eastern Europe, including results at the sub-national level (such as districts or municipalities) from twenty countries, and including data that was gathered from different Electoral Commissions and that was not previously publicly available. Altogether, I provide an almost complete dataset on post-communist elections in Europe, including ninety-five elections in the period 1990–2007. Here, I discuss the gaps in previous research, present the three main innovating ideas of this book, and give a short summary of each chapter.

1.1

Previous research

Why study electoral systems and party systems? The efficiency of electoral systems and party systems has widespread implications for democracy, politics and policies. Parties are one of the basic pillars of representative democracies, and elections and electoral systems are the foundation on which democracy is built. It is thus no wonder that the establishment of the electoral system is seen as one of the most crucial aspects of democratisation and democratic consolidation (Merkel, 2007, p. 416). The question of whether and how electoral systems affect party systems is too crucial to be left to paths of historical coincidence. While in a handful of micro-states modern democracy without political parties is thinkable and practised (Anckar and Anckar, 2000), in all other cases it is unthinkable. An important dimension of party systems is the contrast between representation and governability, or the contrast between large and small party systems. Large systems threaten to obstruct coalition-building (Nohlen, 2004, p. 158), cut the life expectancy of cabinets (Sanders and Herman, 1977; Taagepera and Sikk, 2010; Taylor and Herman, 1971) and over-emphasise particular group demands over national interests (e.g. Shepsle and Weingast, 1981, p. 109). Small

Introduction 3

party systems are said to increase government responsibility (Powell, 1989), but risk producing governments which lack broad popular support (Lijphart, 1999, pp. 167–8). While they might decrease the representation of extremist parties (Powell, 1989), they might also exclude minorities from politics (e.g. Rokkan, 1970, p. 157), and this might result in political violence (Powell, 1982) or the breakdown of democracy (Diskin, Diskin and Hazan, 2005). And those who care more about policies than about democratic coherence can find that party system size has consequences for welfare expenditures (Crepaz, 1998, pp. 74–5) or macroeconomic outcomes (Lijphart, 1994a; Roubini and Sachs, 1989). While party systems, and notably their fragmentation, have been shown to be one of the basic factors of the political system and to be crucial for many political outcomes, electoral systems are commonly treated as the key variable structuring party systems. Many further models include electoral systems as a proxy variable for the party system, acknowledging the central role of the party system, but avoiding problems of endogeneity that might emerge in some studies if it were included as an explanatory variable. Explaining party systems: Legacies, political cleavages and electoral systems The study of electoral systems and party systems in democratising countries has remained marginal so far, in spite of the voices of prominent scholars on democratisation, who stress that the efficiency of electoral systems and party systems has widespread implications for the consolidation of young democracies (Merkel, 2007, p. 416). This is valid for the young post-communist democracies too. Most cross-national studies on the electoral systems’ impact on party systems have either been written before Central and Eastern European democracies held several elections,2 or otherwise they have given a wide berth to young democracies, arguing that party systems in development might produce fuzzy results. The overwhelming view is that party constellations during early democratisation are no more than ‘kaleidoscopic configurations of individual politicians, devoid of anything akin to a system’ (Taagepera, 2007a, p. 6), where proto-parties arise and vanish in a short time. Twenty years after the opening of the wall, democracies in Central and Eastern Europe are approaching adulthood. Have their party systems reached political maturity? Many of the countries have experienced a series of four or five multiparty elections, and institutions now change much less often than they did in the early 1990s. Still, in many

4


of the countries the party systems are not highly institutionalised, or have not adopted a stabilised programmatic competition similar to Western democracies (Enyedi, 2006a; Innes, 2002). Many of the functions of political parties, such as working as a reliable mediator between the society and government (Sartori, 1976, p. ix), might suffer if the party system is ‘floating’ (Rose and Munro, 2003), if parties constantly merge, split, rename, and there is a high fluctuation of MPs between parties (Shabad and Slomczynski, 2004). However, for our understanding of political party development and of the electoral competition, a study of party configurations in young democracies is still worthwhile. Also, in this book, the term ‘party system’ is employed with a very broad understanding of party competition in general, not addressing the question to what extent this competition has been stabilised. Certainly, party competition is not ‘floating’ in all post-communist democracies, and at all times. Quite to the contrary, in many countries, two decades of free elections have brought about quite a clear pattern of party competition, which is based on policy issues (e.g. Tavits, 2008). In some cases, we find small party systems with low inter-election volatility, while in other countries large party systems prevail, with new parties emerging in almost every election. The results presented by previous studies mostly point to the problems of common approaches to understanding the party systems in postcommunist democracies. Three main approaches – based on historical legacies (Geddes, 1995; Grzymala-Busse, 2002, 2006; Ishiyama, 1998; Kitschelt et al., 1999; Tworzecki, 2003, pp. 192–5; Werning Rivera, 1996; see Kreuzer, 2004 for an overview), social cleavages (e.g. Johannsen, 2003; Rohrschneider and Whitefield, 2009; Werning Rivera, 1996; Whitefield, 2002; Zarycki, 2000) and institutional explanations (mainly based on electoral systems) – have been applied to explain the structure of party systems in post-communist democracies. So far, they have led only to partial results, and have shown some of the limitations of these approaches. In a nutshell, historical legacies have been important in explaining the roots of specific political parties, specifically the transformation of former communist parties (Grzymala-Busse, 2002, 2006; Ishiyama, 1998), but they are less powerful in explaining party system formats in a comparative approach. In their four-case study, Kitschelt et al. (1999) distinguish different types of legacies, but they fall short of explaining the different party system formats which have emerged.3 The social cleavage school has been reflected in studies on the social basis of political parties, but it has been noticed that parties along social

Introduction 5

divides in post-communist countries frequently change (Zarycki, 2000), and, apart from that, many parties that do not have strongly visible differences compete with each other in the social groups they appeal to.4 The post-communist context is not very favourable for the emergence of cleavage-based parties. The main divides along which political cleavages formed in Western Europe (namely the class, the rural–urban, the church–state and the centre–periphery cleavages) had been levelled down during communist rule, which changed the social and economic structure of the countries completely (Elster et al., 1998), so that the classical cleavage approach cannot be applied (Bielasiak, 1997; Miller et al., 2000; White et al., 1997). Lipset (1994, p. 13) argues that post-communist countries lack a strong enough civil society for the formation of a strong, cleavage-oriented party system. GrzymalaBusse (2006, p. 422) summarises the critique of the social cleavage approach in post-communist Europe, arguing that parties usually have neither the organisational wherewithal to organise voters, nor the willingness to seek out voter instructions, so that parties with true cleavage characters remain the exception rather than the rule. The common electoral system approaches seem not to lead to the expected results (Clark and Golder, 2006; Golder, 2002; GrzymalaBusse, 2006; Moraski and Loewenberg, 1999). Clark and Wittrock (2005) argue that presidentialism might be more important than electoral systems. It is, however, unsurprising that they do not find any electoral system impacts, because they only control for the difference between single-seat district systems and others – and in their database they have only two, furthermore mitigated, examples5 of these. When searching for reasons for the problems of the institutional approaches, Moser (1999a) and Birch (2003, pp. 115–18) identify former Soviet countries as the most important outliers. While Moser attributes this effect to low party institutionalisation, Birch (2003, p. 118) guesses that ‘high levels of regional fragmentation’ might play a role. The reasons for the different impacts are not entirely clear. If there are fairly simple formulae to explain party system size for given electoral system characteristics, and they work fairly well in established democracies, why do they fail in certain countries? The quantitative analytic models in both studies go only as far as to identify former Soviet countries as outliers, identified through mere dummy variables, but there is no accurate explanation of the classification. The simplified treatment of post-Soviet countries as deviant cases misses the fact that some of the post-Soviet party systems (such as Estonia, Ukraine or Moldova) became much more stable after a few

6


elections than others did (Bakke and Sitter, 2005; Lane and Ersson, 2007; Sikk, 2005; Tavits, 2005), and that some post-Soviet party systems are rather homogeneous across all regions (namely Moldova, Estonia and – if we look at the PR part of the elections – even Russia), while former Yugoslav countries (Bosnia, Macedonia) belong to the regionally most fragmented ones (Chapter 4). Despite these problems, the studies by Moser and Birch look promising and have shown a gap that needs to be filled with more accurate models, in order to obtain more stable results.

1.2 Relevance and innovations of this study The analytical study of elections in recently democratised countries is certainly a risky project. It is doubtful whether it is possible to get strong empirical findings in a region that is characterised by high electoral volatility, enduring changes of party organisations and labels, along with fissures and fusions, where parties lack ties to the society and are little appreciated and accepted in the population (cf. Kreuzer and Pettai, 2003; Mainwaring, 1998; Thames, 2007). In weakly institutionalised party systems, many of the properties needed for electoral systems to have the commonly expected effect are not met. It is commonly expected that, in weakly institutionalised party systems, institutional effects are less predictable. This, however, does not diminish the academic interest in party systems in such a context. Rather, there is major interest in tests of institutional effects in a different context. The robust findings of my study come even more as a surprise: electoral system effects in Central and Eastern Europe might be understood much more easily than commonly thought. Previous studies on political parties in post-communist Europe have been limited to a few countries. In contrast, my study is the first to include all post-communist countries in Europe, from Albania to Russia, excluding only Belarus.6 Not all elections in the region have been free and fair to the same extent as in established democracies (Bielasiak, 2006; Schedler, 2006). Certainly, elections that do not allow real competition, for instance due to the exclusion of important parties from elections or massive obstruction, are not helpful to understanding the development of a party system. However, many elections in democratising countries allow competition, even if they are not fully free and fair. It has been shown that competitive elections that do not meet all formal requirements are still important for democratisation and the development of a new political system

Introduction 7

(Howard, 2006). Based on this reasoning, I decide to apply a less restrictive definition of competitive elections, including both democratic and semi-democratic regimes, where elections are not fully free and fair, but reasonable conditions for competition are fulfilled (see below). It appears important to get a more profound knowledge of electoral processes and their formation in young democracies, risking that some elections might deviate from the expected pattern. If they nevertheless fit into the general model, this would be an even stronger confirmation than a dismissal. Yugoslav and Serbian elections before the year 2000, and all elections in Belarus, are excluded, because they did not qualify as even partially free. This study seeks to explain party system fragmentation in parliaments in the post-communist countries in Europe (first chamber)7 by looking at the common models of electoral system effects8 and further developing these. The core of the analysis is a model that shows that territory plays a crucial role for electoral system effects; based on the territorial structure of a party’s votes, the same electoral institutions can have very different effects. Party nationalisation is closely related to the territorial structure of social divides, and in turn affects the structure of party systems. Specifically, in countries with social and political conflicts that follow territorial lines, weakly nationalised party systems might emerge. At the same time, party nationalisation appears important to understanding electoral system effects. The study of interaction effects of electoral systems and social and political divides is not new. A few prominent articles in the last two decades (Amorim Neto and Cox, 1997; Ordeshook and Shvetsova, 1994; Taagepera, 1999b), relating both to the electoral system school and to the cleavage approach, have shown interactions of both variables, but they did not consider the territorial structure of social divides, or party nationalisation. This is why they came to very different conclusions than my theoretical model suggests. My study, however, focuses on divides with a territorial character. The analysis of the joint effect of the territorial structure of party systems and electoral systems on the party system format alone would, however, be very satisfying, given that party nationalisation is a measure that can emanate from party systems, and can only be established ex post after we dispose of electoral results. This is why party nationalisation cannot be used as an exogenous variable for this study. Instead, I proceed in a two-step design, first exploring the explanations for different levels of party nationalisation, and only in a second step showing the role that party nationalisation plays in the effect of electoral

8


Society

Party nationalisation

Institutions

Territorial social divides

Party system

systems on party systems. Party nationalisation thus has the role of an intermediate variable, first being explained, and then being used as an explanatory variable (Figure 1.1). Six innovations are particularly important in the analysis presented in this book. First, my novel theoretical model brings the dimension of territory into electoral system research. Usually, research on electoral system effects on party systems disregards the territorial structure of party systems. In my work, I prove that different levels of territorial heterogeneity in party support (party nationalisation) make a crucial difference. I develop a model that accurately predicts the number of parties, based on party nationalisation and the characteristics of the electoral systems. Common models explain the number of parties as a function of the size of electoral districts and of the number of seats in parliament (Lijphart, 1994b; Rae, 1967; Taagepera, 2007a; Taagepera and Shugart, 1989). I argue that district magnitude has the most immanent effect in countries where party support is rather homogeneous across the territory, what I call highly nationalised party systems. If, however, party strength varies substantially from region to region (low nationalisation), then the constraining effect of small electoral districts is limited (if at all) to local constituencies, but there will be many more parties in the national parties than at the local level. Elaborating on

Party system (number of parties)

Joint effect

Electoral system

First step: Intermediary model Explaining party nationalisation

Second step: Main model Explaining party system fragmentation

Figure 1.1 Main model for this research project. In the first step, party nationalisation is explained. In the second step, the joint effect of the electoral system and party nationalisation on the party system is shown

Introduction 9

an idea first proposed by Sartori (1986), and recent electoral system models by Taagepera (2007a), I formulate a model that helps to predict the number of parties at the national level, taking into consideration varying levels of party system nationalisation across countries. The methodology applied in this study goes far beyond previous vague operationalisations through general dummy variables on party institutionalisation or nationalisation: I measure party nationalisation properly. Second, I extend research on the territorial dimension of party systems (‘party nationalisation’) to Central and Eastern Europe. Previously, comparative party nationalisation research has been limited to Western Europe, North and South America, and a few countries in Asia. This opens a completely new perspective on party systems in post-communist democracies in Europe. The dominant explanation for variance is decentralisation of the state, which allows parties to compete for regional offices (Chhibber and Kollman, 2004). My study not only extends this research to post-communist countries – previously a blind spot for the research field – but also provides new explanations. I show in a systematic comparative way that the emergence of regional parties is closely connected to the existence of social divides which separate different regions from each other; in post-communist democracies in Europe, this is the impact of ethnic divides. This study first shows that the ethnic structure of the countries can almost perfectly explain the territorial structure of party systems, after controlling for the impact of national legal thresholds in the electoral systems and of super-presidential systems. Third, in order to conduct the study on party nationalisation, I have compiled what is possibly the largest database on electoral results from Central and Eastern Europe, including results at the sub-national level (districts, municipalities, etc.) from twenty countries. Two-thirds of the data was gathered from Electoral Commissions. This data had previously been either difficult to come by, or not available at all. My novel database includes an almost complete set of sub-national results for elections in post-communist Europe in the period 1990–2007, covering ninety-five elections. Fourth, I apply a new methodology, which promises important methodological advancement for political science. Commonly, models in political science only predict the directionality of effects. In contrast, I follow the school of quantitative predictive models.9 This allows me to predict the magnitude of effects and the precise way that several variables interact, going far beyond today’s standard in political science. I innovate

10


further using this methodology by testing the predictions with multivariate probabilistic models. My study is one of the first to bridge the gap between both methodological schools. Fifth, I question the previous enthusiastic view on mixed electoral systems. Mixed systems, combining PR elections with single-seat districts, had been a rare exception until 1990, but were then introduced in eight post-communist countries in Europe. They are applauded by scholars, political advisors and lawmakers as very promising solutions, in brief as the ‘best of both worlds’ (Shugart and Wattenberg, 2001a, pp. 592–6). However, researchers have had difficulties in comparing mixed systems with simple electoral systems, the main problem being that they are more complex than an addition of two simple electoral systems, because the combination creates conflicting incentives on the strategic behaviour of voters and parties (Cox and Schoppa, 2002; Ferrara et al., 2005). Their empirical outcomes deviate widely from the ideal expectations. This study develops a new approach to comparing the effect of mixed systems on party systems with that of simple (nonmixed) electoral systems. The large number of elections with mixed electoral systems, held in several countries of the region, offers an optimal laboratory for a comparative analysis. My conclusions have implications for institutional engineers: in opposition to the overwhelmingly positive view on mixed systems, my results suggest that they are not able to deliver on what they promise; they do not lead to moderate party systems precisely because some of them are very open to manipulation. Sixth, my book offers a new and extended view on party politics in postcommunist countries. Despite the importance of party systems on the consolidation of young democracies, most comparative studies of electoral systems give a wide berth to young democracies, and books on political parties in post-communist Europe have been limited to a few countries. In contrast, my study is the first to include all postcommunist countries in Europe (apart from Belarus), and it offers new suggestions about the development of party systems in democratising countries. I show why, after initial elections that are usually characterised by antagonism between two opposed united blocs, the party numbers mushroom in second elections, and thereafter the number of parties steadily declines, linked to the idea of the survival of the fittest. With this model, I unify two models that previously have been discussed as being contrary to one another, the party dispersion and shakedown hypothesis, which argue respectively for an increase and a decrease in party fragmentation (Reich, 2004).

Introduction 11

1.3 A short guide through this book In detail, these are the questions I shall address in the remaining chapters. The puzzle of electoral systems’ impact in post-communist democracies (Chapter 2) Do electoral systems make any difference? Several elections in newly democratised countries in Europe have contributed to political chaos and instability, due to the dominance of micro-parties and a lack of powerful parties. Often, the blame is shifted onto electoral systems: in Poland, the proportionality of the system was seen as the culprit, while in Russia and Ukraine it was the single-seat districts. The chapter provides an overview of the electoral systems in post-communist Europe. Using common electoral system models, mainly focused on district magnitude, it is shown that they do not help to account for party system size in the region. The following chapters show that the inclusion of party nationalisation does not only explain the surprising results in a few countries, but is furthermore the basis of a model that leads to an accurate account of the number of parties. Analysing the territorial structure of the vote: Methodological issues (Chapter 3) The first step of this model relies on the measure and explanation of party nationalisation. Therefore, I need to address methodological problems related to the measure of party nationalisation. Territorial heterogeneity is measured through electoral data from territorial units at the sub-national level, such as districts, municipalities or polling stations. The measurement has proven to be challenging, because indicators of heterogeneity depend on the size of political parties and varying size of the territorial units within a country, and across countries they can be biased when the number of territorial units varies from one country to the other. This chapter reviews the previously used measures and their shortcomings, and, elaborating on previous indicators by Gini (1921) and Jones and Mainwaring (2003), I propose a new standardised party nationalisation score which allows meaningful cross-country comparisons. Why do different regions vote differently? Explaining party nationalisation (Chapter 4) Using my new measure, I show the development in party nationalisation in twenty post-communist countries in Europe – from Albania to

12


Russia – over the period 1990–2007.10 After describing the nationalisation of parties and party systems, my model explains the differences that emerge from election to election, and most importantly between countries. Empirical evidence shows that the dominant but controversial hypothesis that party (non-)nationalisation is a consequence of the degree of government centralisation (Chhibber and Kollman, 2004)11 cannot be confirmed in this region. Instead, post-communist countries with a weakly nationalised party system undergo steps of government decentralisation, whereas party nationalisation in the region is driven by the absence of relevant territorial social divides and of super-presidentialism, and by nationalising electoral systems, that is, high national legal thresholds. The combination of these factors explains party nationalisation in post-communist Europe to a very high degree. The complexity of mixed electoral systems (Chapter 5) Despite their recent popularity, mixed electoral systems have been included into the common models in only rudimentary forms based on district magnitude, which is used to account for the effect of electoral systems. However, in the knowledge that mixed electoral systems are more than just the two halves they consist of, because there are interactions between both tiers, they cannot satisfactorily be included into comparative studies. The fifth chapter discusses the comparability of outcomes of mixed electoral systems with PR elections and with singleseat district systems. My theoretically derived formula shows that such systems are similar to PR only if a very substantial number of seats are allocated for the purposes of compensation. Otherwise, I suggest that, if political actors are highly sensitive about getting a maximal number of seats in parliament, and if there are no particularly strong cleavages, then it is likely that the effect of single-seat districts will predominate in mixed systems, and largely affect the overall outcome. The empirical numbers from post-communist Europe show that this indeed seems to be the case, even if only after a series of elections. These findings will allow for the operationalisation of mixed electoral systems in the chapter to follow. Given that mixed electoral systems do not only apply to post-communist countries, but are increasingly popular around the world, the tentative conclusions from my study are particularly interesting for the research on mixed systems. This is especially the case given the important pitfalls of mixed compensatory systems, which are revealed, for example,

Introduction 13

by a look at elections in Albania (where the system opened the door wide to manipulation by the major parties). Counting heads and places: The joint effect of electoral systems and party nationalisation (Chapter 6) Having measured and explained party nationalisation and discussed the consequences of a mixed electoral system, I can move to the comprehensive model which is the heart of this study in this last chapter. I build a quantitative model for the estimation of the legislative fragmentation of party systems. The core of the model is based on the idea that not only party strength, but also the territorial structure of a parties’ electorate, determines its electoral success. Therefore, I follow Sartori (1986), who pointed out that the impact of electoral systems depends on the nationalisation degree of party systems. Unlike the model I rely on (Taagepera, 2007a), I show how common electoral system models can be amended through the introduction of party nationalisation. Further attention is devoted to the development over time; specifically, I show why first competitive elections are usually quite particular in their character, and I explain how the number of parties in young democracies develops further. Conclusions (Chapter 7) While previous literature has discussed the impact of electoral systems on party systems predominately from a national perspective, my work includes the territory as a new dimension of electoral system research. It shows that there are factors – territorially based social divides – which lead to territorial heterogeneity in the party systems, and that they make a difference to electoral system effects. In post-communist Europe, two variables are strongly related to party nationalisation: the ethnic structure of the countries and the super-presidential regime type. The inclusion of party nationalisation resolves the puzzle of the apparent failure of electoral rules in the region. My model predicts the number of parties accurately.

1.4

Conclusions

Party systems have been shown to be among the most important elements of political systems, and electoral systems are considered as one of the most important institutional elements of constitutional orders (Sartori, 1994). Yet research on Central and Eastern Europe seems to show the contrary. Electoral systems are said either not to have any

14


impact on party system fragmentation, to provide for non-expected results, or to be effective only in some of the countries. Can the same institutional rules lead to different outcomes, depending on the region where they are applied? This study takes a closer look at the functioning of electoral systems in Central and Eastern Europe. With regard to common electoral system approaches, I innovate in three respects. First, I include the territorial structure of party systems, expressed through party nationalisation, as an additional explanatory variable. I argue that electoral systems have a joint effect with party nationalisation. Second, I extend the reach of electoral system models that have been developed for simple electoral systems to mixed electoral systems, frequently applied in the countries under study. Third, this project offers a new view of the party system changes that occur during the early development of democracy. Many previous studies on the impacts of electoral systems on party system fragmentation have steered clear of young democracies, on the grounds that they produce fuzzy outcomes. However, acknowledging that elections in the early stages of democracy are crucial for the establishment of party systems, it appears particularly important to devote specific attention to them. The goal of this book is to sharpen the analytic knowledge in this field, and to develop a model to explain party systems in Central and Eastern Europe, and possibly elsewhere too.

2 The Puzzle of Electoral Systems’ Impact in Post-Communist Democracies

2.1

Introduction

Electoral systems are a very powerful predictor of the party system format in Western democracies. Specifically, electoral systems with single-seat districts (plurality systems) are said to lead in many cases to the concentration of party system, whereas proportional elections with large districts are known to be inclusive electoral institutions, thus allowing a high fragmentation of the party system (Duverger, 1951; Norris, 1997; Rae, 1967; Taagepera and Shugart, 1989). Do these widely accepted results also hold for post-communist democracies in Europe? Several studies suggest that they do not (for instance, Moser, 1999a, p. 384; Moser and Scheiner, 2004, pp. 595–8). As outlined in the introduction to this book, many results of elections in Central and Eastern Europe deviated substantially from the outcomes which might have been expected from common electoral system models. It is thus little wonder that the effect of electoral systems in the young European democracies has been described as chaotic. In some of the countries, the fractionalisation of the party system is similar to established democracies with equal electoral systems. In others, it might appear that the results contradict our experience, turning electoral system theory upside down. Previous work has come to the conclusion that ‘proportional representation [ ... ] behaves more like single-member plurality systems’ (Moraski and Loewenberg, 1999),12 or that ‘the same 5 percent threshold produced very different competitive configurations’ and ‘district magnitude is not correlated to party competition’ (Grzymala-Busse, 2006, p. 421). Alternatively, ‘it could simply be the case that the effect of social heterogeneity and electoral institutions might not be the same in this region of the world’ (Golder, 2002, p. 24). 15

16


Clark and Wittrock (2005, p. 172) argue that ‘the assumptions about the strategic behavior of voters and parties contained in Duverger’s law do not operate with the same effect in different executive-legislative constitutional designs’. Varieties of party systems in Central and Eastern Europe After the fall of the communist regimes, the new democracies in Central and Eastern Europe developed very different types of party systems. ●

●

●

Albania and Hungary both developed a party system with two major parties that hold the overwhelming share of votes and seats. Other countries developed highly fragmented party systems. For example, at the time of writing, no party holds more than 24% of the parliamentary seats in Bosnia and Herzegovina, or more than 26% in Latvia. In yet other cases, such as the Czech Republic or Croatia, rather moderate multiparty systems with two larger and some smaller parties developed, with a fractionalisation level that is comparable to many Western European counterparts.

Even after many of the Central and Eastern European democracies have held up to five or six elections, there are countries where party systems have become small and stable and others where a highly fragmented pattern prevails. Interestingly, the patterns do not coincide with key aspects of political transition, such as the quality of democracy, or membership in the European Union. In an attempt to understand better the party systems in new European democracies, I apply conventional electoral system approaches to this set of countries. This chapter reviews theories on the impact of electoral systems on party system size, showing that those approaches do not help in the understanding of post-communist party systems. The main focus of these conventional electoral system approaches is district magnitude, legal thresholds, the size of parliaments, and the electoral formula. This chapter outlines the puzzle for the further study of the phenomenon. First, it focuses on conventional electoral system models (Section 2.2), before testing them on countries in Central and Eastern Europe. The analysis of elections in twenty democracies in the period 1990–2007 (altogether about ninety single elections) shows that electoral systems in post-communist countries have led to completely unexpected outcomes.

Electoral Systems in Post-Communist Democracies

17

2.2 Electoral systems and party systems – the conventional view Scholars of political institutions have stressed the importance of electoral systems for the development of party systems. The question of the size of party systems has been confronted from several directions. The common institutional view deals with the form of party systems as a product of institutional rules, namely of electoral systems. Its critics have often emphasised the problem of endogeneity when studying the relation between electoral systems and party systems, arguing that a country’s socio-political heterogeneity and resulting political conflicts determine the size of party systems. In this perspective, the electoral system is often a consequence of social and political conflicts and of a pre-existing party system. Both views will be incorporated here. After discussing the variable under study, namely the number of parties, I discuss the importance of the most common electoral system variables, such as district magnitude, legal thresholds and the size of parliament, before moving to the social cleavage view. 2.2.1

The number of parties as dependent variable

Representation and concentration are two key dimensions that matter for elections and party systems. The number of parties captures both of them. Representation concerns the proportionality of electoral systems: do they represent all relevant social and political groups in an approximately proportional way? Disproportionality indices (Gallagher, 1991), A-ratios (Taagepera and Laakso, 1980) and electoral responsiveness measures (Kendall and Stuart, 1950; King, 1990) measure the deviation of seat shares from vote shares, and thus help to establish the mechanical impact of electoral systems. But they are blind to the psychological effect of electoral systems: many parties that would be created and win votes under more proportional electoral systems do not even compete, once they realise that they would not be represented according to their vote share. If this problem is taken seriously, the fractionalisation of the party system in parliament turns out to be a more accurate expression of the proportionality of an electoral system than the disproportionality between seats and votes. Concentration is an important aspect for the governability of a party system. In small party systems, it is much easier to build a minimalwinning, one-party cabinet than in large party systems (Lijphart, 1999,

18


pp. 112–13), and this helps governments to survive longer (Taagepera and Sikk, 2010) because in one-party cabinets the potential for change, dissent, and conflict is lower. Fractionalisation of party systems has been established as the most important variable for the study of electoral system impacts. It touches both of the aspects mentioned, and both dimensions have immanent impact on macro-political outcomes (as described in the Introduction). Although the number of parties is a macro variable, it helps to understand what happens at the micro level: mainly it reflects whether small political interest groups are able to create political parties and have success with them in elections. The impact of electoral systems on the number of parties has direct implications for the chances to form new political parties (cf. Hug, 2001; Sikk, 2006), and salient political importance in countries with ethnic minority groups that demand the possibility of being represented by their own parties. There, the electoral system is decisive for the possibility of minority groups forming their own political parties and succeeding with them in parliamentary elections. There are different ways of measuring the number of political parties. One might look at party registers, which exist in some countries, but most of the registered parties are politically irrelevant, and do not even participate in elections. Commonly, the number of parties is measured based on parties that compete in elections or are represented in parliament. This, however, imposes a different problem: what if parties that are registered as distinct entities compete with a joint list in elections, or if parties form joint parliamentary groups? How should we count political groups that participate similarly to parties in elections, win seats in parliament, and possibly might even form parliamentary groups, but are not registered as political parties? A very central aspect to characterise a political party is its eligibility in elections. If not specified differently, I speak of political parties as organisations that submit their own electoral list and compete independently in elections,13 which guarantees also that they are fully independent in their legislative behaviour.14 Possibly the most natural indicator for the counting of political parties, the absolute number of parties N0 (the simple counting of all parties that hold seats in parliament15), does not differentiate small and large parties and consequently grows very much with the presence of several micro-parties. However, one or few micro-parties that are present with one or few seats in parliament affect the absolute number of parties massively, so that the measure is not really indicative to measure how many relevant parties exist. Setting a threshold in seats, beyond which a party is counted as relevant, would, however, be arbitrary, and create problems of


19

steadiness if small parties split up. For this reason, the Laakso–Taagepera index (1979) takes small parties into account, but weighs them less than large ones. Also called the effective number of parties (noted as N2), it has become widely accepted as the standard indicator for party fractionalisation. It is part of a whole family of related indicators, some of which are closer to the absolute number of parties, and others are rather sensitive to the share of the largest party (Taagepera, 1999c), or they are combinations of both (Molinar, 1991). The Laakso–Taagepera measure appears to be a happy medium that satisfies different criteria. The effective number of parties can be calculated either on the basis of the seats held by parties in parliament (effective number of parliamentary parties) or on the basis of the votes that parties win in elections (effective number of elective parties). If not stated differently, I refer to the indicator that is calculated from the seats held by parties in parliament, but where it is relevant for the discussion of the theory (in Chapter 5), and in the conclusive model (Chapter 6), I include the number of elective parties too.16 Independent candidates are a challenge for the calculation of the effective number of parties. In order to avoid the problem of the huge number of independent candidates that run in elections in Russia and in Ukraine, Moser (1999a) disregards candidates’ affiliation to political parties during the elections in these two countries, and instead looks at their membership in parliamentary groups. Such a measure is, however, not necessarily connected to the elections, but rather the result of group formation processes in parliament, which might also rely on non-electoral strategies, and it is unstable if members of parliament switch groups quite often (as in the case of the Russian Duma). It would also be inconsistent compared with other countries where we identify parties based on their electoral lists. Other authors simply exclude independent candidates from their count (e.g. Birch, 2003, pp. 109–10, 2007, p. 13), or, similarly, count only entities with more than one seat in parliament (Lewis, 2000, p. 88). This might produce paradoxes, however: if a small party with four seats splits into two new two-seat parties, the number of parties increases, but if it splits into four, and (some) MPs become independent, the number of parties decreases.17 Some studies instead merge independents together into a single ‘party of independents’ (Moser, 2001c, p. 48), which neglects the fact that independent MPs might have big political differences. In some cases, such a measure would also increase party system fractionalisation when an independent MP joins a political party, despite the fact that parliament is becoming more structured. Henceforth, I decide to treat every independent candidate as a separate unit – as if they were a one-MP political party (as suggested by many studies in the field (Chhibber and

20


Kollman, 1998; Diwakar, 2007, p. 544; Taagepera, 1997, 1999c, pp. 498–9, 2007a, pp. 58–9; Taagepera and Grofman, 2003)). Indeed, similarly to one-seat parties, independent MPs are not formally attached to any larger entity (as done by Clark and Wittrock, 2005, pp. 180–1). The impact of an independent deputy on the measure should not differ from the impact of a one-seat party. Indeed, often differences between one-seat parties and independent candidates with informally organised supporters are only of a formal nature. Specifically, in some countries, there are one- or two-seat micro-parties, along with non-partisan candidates (for Ukraine: Birch and Wilson, 1999, p. 281; for Lithuania: Clark, 1994, p. 47).18 However, it is worth noting that the common index that is used to measure the fractionalisation of party system, the effective number of parties (Laakso and Taagepera, 1979), is made on purpose not to overweight the relevance of small parties or of independent candidates. A party or any other organisation with only one or very few seats in parliament counts much less than a large party, and accordingly the impact of an independent candidate is only marginal, compared with a political party with dozens of seats (Taagepera, 2007a, p. 49). Still, parliaments that are dominated by MPs who are elected as independent candidates, or who belong to parties with only one or a few seats in parliament, are heavily fractionalised, and accordingly – as in the case of Russia or Ukraine – it seems correct that the effective number of parties will be larger than elsewhere. Figure 2.1 shows the effective number of parties in elections in twenty post-communist countries in Europe, and the development over time.

Number of parties

15 13

Number of election:

1

2

3

4

5

6

11 9 7 5 3 Ukraine

Slovenia

Serbia

Slovakia

Russia

Poland

Romania

Moldova

Montenegro

Macedonia

Latvia

Lithuania

Kosovo

Estonia

Hungary

Croatia

Czech Republic

Bosnia

Bulgaria

Albania

1

Figure 2.1 Effective number of parties in parliaments in Central and Eastern Europe, 1990 –2007, time series for the number of the election. (Not all data available) Note: Czechoslovak elections (1990, 1992) are shown under Czech Republic.


21

2.2.2 From Duverger’s duality to a continuous axis of electoral system proportionality Having defined the dependent variable, I now turn to the explanatory variables. The most frequent and basic distinction in the world of electoral systems is that between PR and plurality vote. In this binary approach, PR allows a large number of parties to survive, whereas plurality vote leads in to two-party systems (Figure 2.2, top). Majority vote (where a candidate needs 50% of the votes to get elected) was considered, along with PR, to allow several parties (Duverger, 1951).19 The general view has since been amended in two directions. On the one hand, scholars have elaborated on the conditions – partly already mentioned by Duverger – that need to be fulfilled for plurality vote to really lead to two-party competition. On the other hand, PR systems have been differentiated, mainly according to the size of their districts, so that the binary view of PR versus plurality vote became a continuous axis of proportionality, with district magnitude as the main variable. In this metrically measured view, plurality vote, in most empirical cases held in single-seat districts, marks the lower, least proportional extreme along the proportionality axis, and has the most concentrating effect on party systems. Due to the special character of single-seat districts, only the candidate with the largest number of votes wins the seat. PR systems with large multi-member districts or even countrywide districts constitute the opposite, permissive extreme with high proportionality (see Figure 2.2, bottom). The continuous measure of electoral system permissiveness is attributed to Rae (1967), who integrated the size of the constituency and the electoral formula in his explanatory model, and to Taagepera and Shugart (1989), who claimed that the constituency size is the most important characteristic of electoral systems. While, commonly, the definition of the district magnitude is obvious, there are more complicated cases. For my purpose, it is relevant

Duverger

PR (or majority vote) Institution Plurality vote > Concentrating (not proportional) > Permissive (proportional) Impact > Two-party systems > Multi-party systems

District magnitude Taagepera/ Institution 1 Shugart; Least proportional Rae Impact (most concentrating)

large Most proportional (most permissive)

Figure 2.2 Electoral systems and their impact on party system fractionalisation, according to Duverger (top) and Taagepera/Shugart and Rae (bottom)

22


to focus on ‘the units within which voting returns are translated into distributions of parliamentary seats’ (Rae, 1967, p. 19), since this is the level where the number of mandates has an impact. This becomes important because there are different types of proportional systems and electoral system elements that allow the election in small districts, but nevertheless guarantee a proportional allocation of mandates at the level of upper supra-districts, sometimes at the level of the whole electoral area (here: at the national level). Frequently in Central and Eastern Europe, a first seat allocation on parties applies at the national level, followed by a district-wise regional allocation of the seats won by parties (or similar mechanisms, where remainder mandates are allocated nationally). When mandates are accorded to parties in supra-districts, then the magnitude of the supra-district is relevant for the electoral system’s concentrating impact on the party system (provided that the number of compensatory mandates at the national level is large enough or the compensatory mechanism is strong enough to compensate for disproportionalities in the districts20). The impact of the size of electoral districts on party systems has been shown in a number of comparative studies, mostly focussed on Western democracies (particularly Lijphart, 1994b; Rae, 1967). The continuous logic of the district-magnitude variable is straightforward. In (proportional) electoral systems with a small number of seats per districts, the number of seat-winning parties per district is limited by the district magnitude. Alternatively, in a five-seat district, each seat weighs 20% of the district votes. Due to the fact that mandates cannot be divided – on the contrary, only entire mandates can be accorded to a party – the threshold to win a mandate is even slightly lower. For most electoral formulae, 16.7% of the vote is sufficient to win one of five seats in parliament21 (natural threshold TD, calculated as TD 1/(m 1)). Votes for smaller parties risk being wasted, if those parties fail to win parliamentary seats. In very large electoral districts, the threshold is low enough to allow even very small parties to win mandates, whereas small district concentrates the party system on large players. In districts with only one seat (or in plurality or majority vote districts), only (locally) large parties win parliamentary seats. Fifty per cent of the votes guarantees a victory, but, if there are more than two viable competitors, even a slightly lower vote share might allow a party to win a seat. Typically, the winning parties need some 35%–50% of the votes (Lijphart, 1994b; Taagepera and Shugart, 1989), and two large parties emerge.


23

2.2.3 District magnitude legal thresholds effective thresholds It has been shown that the size of electoral districts in multi-member PR systems constitutes an entry threshold for political parties, the natural threshold. However, the logic of district magnitude does not necessarily need to apply in every election, since some electoral laws introduce some elements that are more important than the limits given through district magnitude. Other electoral systems, indeed, set additional barriers (legal thresholds) besides the district thresholds to limit the access of small parties to parliament, namely through the definition of a minimal vote share that parties require to be considered in the seat allocation process, often around 5% of the national vote. Even parties that have sufficient votes to win district seats, but are below this legal threshold, are excluded. Thus, these legal thresholds become important as soon as they are higher than the natural thresholds. In a nine-seat district, for instance, the natural threshold is 10%. A legal threshold (TL) of 5% of the votes would thus not make any difference, since parties with fewer than 10% of the votes cannot win mandates due to the number of seats available. On the other hand, such a 5% legal threshold might be important in a thirty-seat district, where the natural threshold amounts to only 3% of the vote. There, the legal 5% threshold is the more important obstacle to pass. Lijphart (1994b, p. 29) thus introduced the notion of the effective threshold (TE), which is the more relevant of both values or the larger of both thresholds (TE max(TD; TL) or TE max(1/(m 1); TL)). There is a difference, however, between district thresholds and legal thresholds, due to the fact that legal thresholds define a fixed vote share that parties need to reach in order to gain parliamentary representation, whereas the thresholds given through the number of seats in a district cannot precisely be estimated, except for an upper threshold of exclusion and a lower threshold of inclusion. The threshold of inclusion refers to the vote share that is necessary to win a seat, the threshold of exclusion the sufficient vote share.22 More precisely, the importance of the territorial entity where thresholds apply should not be overlooked. In the case of national legal thresholds, the threshold defines a certain percentage of the votes at the national level, while natural district thresholds concern the electoral district level, and, likewise, some countries apply legal thresholds at district level. Only in a few cases is the distinction between the electoral district and the national level irrelevant. This is the case either if nationwide districts are in use or if a party’s vote shares are perfectly homogeneous across a country, so that the relative vote share at the

24


district level and at the national level does not differ (perfect nationalisation; see Chapters 4 and 6). If party nationalisation is imperfect, then the difference between district thresholds and national thresholds matters. Supposing ignorance of party nationalisation degrees, and going in line with the conventional school on electoral systems (notably an assumption taken tacitly by Lijphart, 1994b), I assume that party nationalisation is very high. This means that there is no significant difference in a party’s vote share at the national and at the district level (see Section 2.2.5 for a different assumption), and the concept of the effective threshold works quite well. Other electoral system elements – most notably differing proportionality formulae – although politically often highly controversial, proved to be less important regarding their impact on party systems, and are thus commonly not included as explanatory factors in comparative studies.23 More precisely, the choice of one or the other PR formula might affect the seat distribution by a few seats, but does not have a major impact on the party system in large PR districts or in PR systems with relevant legal thresholds. 2.2.4

A district-based model to predict the number of parties

Average district magnitude (number of mandates per district) is the main common predictor of the party system size. Sometimes it is operationalised as effective threshold (for instance, Moraski and Loewenberg, 1999; Moser, 2001b), which is a function close to the inverted value of the first one, but allows the integration of legal thresholds too. Both variables are often used in linear regression models, even if the impact on the (absolute or effective) number of parties is not a linear one, thus violating an important premise of linear regressions, as shown by the following examples. A change of the district magnitude by two might affect the character of an electoral system and party system size in a substantial way if single-seat districts (m 1) are merged into a small multi-member district (m 3). While in single-seat district systems electoral competition at district level is likely to be focussed on two large parties, a three-seat district would allow a third political party to appear and to have success if it achieves the support of about 20% of the voters. If the same change occurs in a system with already large districts, a much lower impact might apply. For instance, an increase of district magnitude from twenty to twenty-two seats might not make a big difference, since the effective threshold is lowered only from 4.76% to 4.35%, and affects only the success of parties in this tiny range. Even if by coincidence a


25

party existed that won just this vote share (for instance 4.5% of the votes), and if it were to win access to parliament, it would not affect the number of effective parties in the same way as the third party in the previous example, because small parties do not have a large weight in the effective number of parties indicator. In other words, for larger districts, the marginal importance of a change declines. A possible strategy to deal with this property would be to transform district magnitude into its inverted form (x 1/x), or similarly into the district threshold (TD = 1/(m 1)). This appears to take into account the reservations on the use of district magnitude: if the variable is inverted, a marginal change of district magnitude is much more important when districts are small than if districts are large. The maximum threshold of exclusion is about 50% in single-seat districts, related to two-party systems. If this is taken as a fixed point, and the number of parties as a linear function, contradictions emerge. From the threshold of exclusions, the conceptually possible maximal number of parties in a system can be derived. Essentially, no more parties than the inverted value of the threshold (1/TE) can exist (plotted by the dotted line in Figure 2.3). For instance, at TE 50% no more than two seat-winning parties can exist; at TE 25% the maximum number of parties allowed would be four parties with perfectly equal vote shares. Now linear functions of the number of parties should be fitted into this scheme of allowed values. Taking the point [0.5; 2] as given starting point, every line that

Number of parties

8

6 [0.25; 4] 4

2

[0.5; 2]

0 0

0.1

0.2 0.3 Effective threshold

0.4

0.5

Figure 2.3 Effective threshold and the number of parties. The dotted line describes the conceptual maximum of the number of parties, given a certain threshold. The bold line describes a possible OLS regression

26


respects the conceptual maximum and thus needs to remain below the dotted line would be very flat, and consequently estimate only very moderate numbers of parties even for extremely low thresholds slightly above 0%. Any more realistic linear function would either violate the conceptual maximum or not go through point [0.5; 2]. This is why I assume that the number of parties cannot be explained through a linear function, either of district magnitude or of the district threshold. Two approaches have been used to resolve the problem of non-linearity. The first is to take, based on empirical findings, the logarithm of average district magnitude (Ordeshook and Shvetsova, 1994, p. 107; Taagepera and Shugart, 1989, pp. 142–55), or the logarithm of the median district magnitude (Amorim Neto and Cox, 1997, p. 157). Second, and going beyond empirical thoughts, Taagepera and Shugart (1993, p. 457) propose a model that respects the conceptual limits of the function (Taagepera, 2007b). For the prediction of the effective number of parties, Taagepera and Shugart orient upon the upper conceptual limit N0 m. In an electoral district with m seats, the number of seat-winning parties can never be larger than m. The same is valid for the effective number of parties N2 . The lower conceptual limit for the effective number of parties is at N2 1. No matter how large the electoral district is, always at least one party wins seats. Neither of these two is a common case, however; rather, in larger multi-member districts, some parties might win several mandates. Accordingly, the function must fit into the space of the possible solutions, limited by the condition 1 N2 m. The larger the district, the better the chances of smaller parties to win mandates, so that the number of parties should increase monotonically with district magnitude. And, last, as discussed before, the marginal impact of an increase in district magnitude will decrease with larger districts. I formulate a first model, which operates at the district level, on which I shall elaborate further, considering the number of districts in parliament (in this chapter, following Taagepera and Shugart, 1993), and party nationalisation (in Chapter 6). The function N2 mk, with 0 k 1, lies within this conceptually allowed field, with k as a parameter to be defined. If k were equal to 0, the equation N2 m0 1 would apply, which corresponds to the conceptual minimum, where the number of parties is constant at 1. If k were 1, the equation N2 m1 m would apply, or the number of parties would always be equal to the number of seats in a district. However, it is very unlikely that the function takes either of those extreme values.


Conceptual maximum N2 = m k=1

27

log(N2)

Concentration effect

N2 = 3√m k = 1/3

Competition effect Conceptual minimum, N2 = 1, k = 0 log(m) Figure 2.4 Balancing effects on the number of parties in single electoral districts: Determining the parameter k

Rather, k is determined by effects that push it away from the conceptual limits. A party system at the lower limit (k 0, and N2 1; cf. Figure 2.4) would not hold very long, because, if all the seats were held by only one party, competitors would try to step into the electoral market and into parliament. The desire of new competitors to get into the electoral market, the competition effect, pushes the number of parties away from the lower limit. However, the conceptual maximum is unlikely too: voters and politicians join large parties because they are more visible, better known and perceived as more influential, and there is less risk that they will fail to pass the electoral threshold; this provides for the concentration effect. The resulting number of parties and the parameter k is a balance of both effects, and might be somewhere in the middle between 0 and 1, but the exact position might depend on exogenous factors, and vary for different socio-political environments.24 On theoretical and empirical grounds, and for Western democracies, Taagepera (2007a) proposes choosing k 1/3 respectively N2 3√m as best predictor, while, according to the same work, k 1/2 is the parameter to predict the absolute number of parties [N0 √m].25 2.2.5

The parliament – district-magnitude model

The most conventional view on electoral systems presented so far looks at mechanisms at the level of single electoral districts. However, it is mostly used to explain the size of the party system at national level. What works well in many empirical cases is theoretically odd: why should the outcome at a similar number of seats in a district be the

28


same, regardless of whether there is only one electoral district or a plurality of districts? As mentioned, this would be the case if the whole electoral area consisted of a single district, or if variation of the party system across districts were so small that it could be neglected. Based on the tacit assumption that there is a stable, but not perfect, degree of party system variation across districts, Taagepera and Shugart (1993) have developed a model that controls for both district magnitude and the size of a parliament. There are two conceptual limits: the smallest number of parties results if in each district exactly the same parties are elected with the same seat share, so that there would still be 3√m parties in parliament. Usually, however, in the case of several districts, the number of parties will be higher. Maximally, for a given number of seats in a parliament, the number of parties should not increase if the parliament is divided into districts, compared with a single nationwide district (Taagepera, 2007a).26 If a parliament with S seats consists of a single countrywide district, then parliament size S is equal to district magnitude m, and the number of parties can be estimated directly as N2 3√S. This is the upper limit. The geometrical mean of both measures amounts to N2 √ (3√m · 3√S) 6√mS. 2.2.6 Electoral systems, social cleavages and the endogeneity problem The institutional school was early confronted with the critique that electoral systems are ‘a result rather than a cause of the party system in a given country’(Grumm, 1958)27 (but see also Boix, 1999; Colomer, 2007; Rokkan, 1970). The varying shape of party systems, however, might be best explained by the cleavage structure of a political system (Lipset and Rokkan, 1967). This relation has been quantified by Lijphart (1984, pp. 147–9) and by Taagepera and Shugart (1989, p. 94): the number of parties N being the number of salient issue dimensions that are present in a democracy I plus one (N I 1). In old Western democracies, the (four) social cleavages in the industrial countries were frozen after the introduction of universal suffrage in the 1920s, and have shaped party systems (see Figure 2.5, sociological approach). Kitschelt et al. (1999) find different cleavages in post-communist democracies than in the West. Often, however, post-communist societies lack of strong and organised social groups on which political divides and parties are based. Such belong to the central characteristics of social cleavages (as defined by Bartolini and Mair, 1990). The social cleavage approach does not help to explain party systems in post-communist Europe. There are remnants of social cleavages that

Electoral Systems in Post-Communist Democracies Institutional approach Sociological approach Independent variable(s)

Dependent variable

Electoral system

Social cleavages

Party system

Party system

29

Integrated approach Electoral system

Social cleavages

Party system

Electoral system

Figure 2.5 Three concepts to explain the number of parties in a party system

shape the party systems (see Chapter 4), but too many phenomena remain unexplained, and many parties lack strong cleavage-based ties to their voters. While the social cleavage hypothesis thus does not lead to very powerful results, the critique on the electoral system school has two important shortcomings when it regards post-communist democracies in Europe. At first glance, it seems that the empirical reality in these countries confirms the criticism on electoral system studies, but two aspects show that the problem might not be salient. First, the electoral systems are made by political parties that hold a parliamentary majority, so that the relationship might be endogenous. Parties choose the electoral system with the most favourable expected impact. Parties move to a PR system in a changing environment, where their dominant position is challenged (Boix, 1999). Thus, even if the electoral system choice is not made at random, electoral systems are used as a mechanism to shape party systems. Considering that electoral systems are used as an institutional constraint to force or hinder party system change, the view of electoral systems shaping the party systems is even reinforced. Second, strategic considerations of parties might have a rather predictable impact in the short run, but the long-term development appears too uncertain to allow an effective purposeful institution creation in favour of a certain party. Particularly in newly established democracies with quickly changing political landscapes and high volatility, the parties’ intentions often fail even in the short run. Accordingly, Andrews and Jackman (2005) show that purposeful electoral system choice in Central and Eastern Europe suffered from informational shortcomings. Shvetsova (2003) found strategic choices of electoral systems in Central and Eastern Europe that were adopted in a self-preserving way, but they often failed to fulfil this goal. This diminishes the circular

30


dependency between electoral systems and party systems and makes it possible to consider the electoral system as almost an independent variable (Moraski and Loewenberg, 1999, p. 159). Instead of separating the institutional and the cleavage approach, a few studies combined them in one model (Figure 2.5, integrated approach). Social cleavages are the motor of party development, but electoral systems limit the number of groups that have access to the politic arena. The joint application of both institutional and sociological explanations allowed for a better account of the variety in party systems (Amorim Neto and Cox, 1997; Ordeshook and Shvetsova, 1994). Similarly to these studies, Mozaffar et al. (2003) and Brambor et al. (2007) have looked at the interaction of social cleavages with the electoral systems in Africa’s emerging democracies, and further included a measure of the territorial group concentration, pointing at the importance of territorial aspects of cleavages. However, since their results seem to contradict their hypothesis, this approach would profit from further study (see page 133 of this volume for details).

2.3 The empirical puzzle: And it all looks like nonsense ... This section assesses whether the conventional electoral system approach, as discussed in the previous section, works with Central and Eastern European data. After discussing the state of the literature, I shall test the district-magnitude model and the district-parliament size model. Due to non-results or statistical results in the opposite direction to that expected (Golder, 2002, p. 24; Grzymala-Busse, 2006, p. 421, footnote 5; Moraski and Loewenberg, 1999, pp. 163–4; Moser, 2001b, 2001c),28 several scholars have questioned the validity of the electoral system variable in this group of countries. Surprising effects of electoral institutions might suggest that non-institutional explanations such as history and social cleavages are more appropriate than electoral rules as an explanation of party formation in those countries (Beichelt, 2001; Kitschelt et al., 1999; Nohlen and Kasapovic´, 1996). Others have explained the surprising results instead through the political context, namely the lack of party system institutionalisation (Moser, 1999a) or low party nationalisation (Birch, 2003) in post-Soviet countries. Finally, presidential elections might explain a higher degree of party system fragmentation (Clark and Wittrock, 2005; Filippov et al., 1999). For several reasons it might be desirable to retest conventional electoral system models on my database before developing my own model


31

in the forthcoming chapters. Previous studies with the same goal, as discussed above, did not lead to any conclusive results. My tests are made on a broader range of elections, including elections up to 2007, and including several democracies that were often excluded from such analyses due to a lack of sufficiently long democratic experience or due to a lack of data.29 Later, I will use this dataset and comparable variables to compare the conventional model, which is outlined in this chapter, with the party nationalisation model, which will be developed later in this study. Using the same data and similar variables makes sure that the analyses are comparable. Previous studies restricted the cases of investigation or did not include the same variables.30 By now, all countries of the region have held at least four multiparty elections in a row, and Kosovo two. The empirical tests in this study are based on a database covering ninety-three elections, each included as a single case. In sixty-nine of these cases, simple electoral systems were used, while twenty-four times a mixed electoral system was used, combining PR and plurality or majority vote in the same electoral contest. Mixed systems are a genuine challenge. The estimation of the overall outcome is less straightforward than in the case of simple electoral systems because the vote distribution in each of the tiers of mandate allocation of mixed systems might be contaminated by the other tier(s). However, when it comes to an empirical test of hypotheses, mixed electoral systems cannot be ignored; due to their frequent application they have become an essential part of the electoral system study in post-communist Europe. For mixed electoral systems, I consider the number of parties and the electoral system characteristics of the single-seat district tier.31 My first models only include elections without legal thresholds, and further models include all elections. The usual impact of district magnitude on the number of parties seems not to work for Central and Eastern Europe: the scatter plot (Figure 2.6) shows no clear relationship between district magnitude and number of parties, but strong heteroskedasticity: the outcomes of electoral systems with smaller districts vary much more than those of systems with larger districts. Results of OLS regressions confirm the visually observed non-result. I tested both the pure district model and a model including district magnitude and size of parliament. District magnitude m does not help to estimate the effective number of parties. Logarithmisation of the district magnitude model (N2 3√m, or log(N2) 1/3 log(m)) permits testing it with an OLS regression. I would expect the coefficient to be close to 1/3, and the constant close to 0. In the second and third specification, I include both

32

Territory and Electoral Rules in Post-Communist Democracies 40

N2

30

20

10

0 0

5

10

15 m

20

25

30

Figure 2.6 Explaining the effective number of parties (N2, y-axis) by district magnitude, 1990–2007 (x-axis); only electoral systems without any legal threshold included Note: Forty-one cases, excluding Romania 1990 with a 328-seat district and Kosovo 2001, 2004, 2007 with a 100-seat district.

district magnitude m and the size of the parliament S, according to the Taagepera model (N2 6√mS, or log(N2) 1/6 · log(m) 1/6 log(S)). Two control variables are included. The inverted ordinal number of the election (1/T) controls for the role of the duration of democratic experience in the formation of party systems (see Section 6.6). The inclusion of ethnic fractionalisation (effective number of ethnic groups, e) allows control in a rudimentary way for the social cleavage hypothesis (Amorim Neto and Cox, 1997; Golder, 2002; Kim et al., 1992; Kostadinova, 2002; Ordeshook and Shvetsova, 1994, etc.). Critic of the ethnic fractionalisation measure has not addressed the calculation of fractionalisation itself rather than the need to assign each member of society to a certain ethnic group, being aware that ethnic groups are constructed and, often, persons might have multiple identities


33

(Chandra and Wilkinson, 2008; Fearon, 2003, etc.). In the Central and Eastern European context, however, ethnic identities appear to be fairly stable and clear, and very different sources mostly report the same ethnic composition (see Chapter 4). Due to the small number of cases, the short democratic experience – which might lead to some deviating cases – and the limited width of the approach, low statistical significance must not be of concern. What is rather astonishing is the almost complete absence of explanatory power (R2) for models that if applied on Western Democracies explain a reasonable part of the variance (Ordeshook and Shvetsova, 1994, p. 111; Taagepera, 2007a), and the very low coefficient for district magnitude. Its coefficients should be about 0.33 in the first, and 0.17 in the second specification. Similarly to previous studies on Central and Eastern European elections, this model thus confirms neither the conventional theory on electoral systems nor the Taagepera approach. The inclusion of typical control variables (ethnic fragmentation and democratic experience) seems to increase the importance of the size of parliament in the model, Table 2.1 OLS regression models for the effective number of parliamentary parties (log), district magnitude and number of seats in parliament as explanatory variables; robust standard errors. Elections with a national legal threshold excluded (1)

(2)

(3)

Regression model

B

Constant a

1.30

0.17

–

–0.23

District magnitude [log m]

0.06

0.06

–

0.07

0.07 1.03

0.06

Size of parliament [log S]

–

–

–

0.32

0.20 1.03

0.55** 0.19

Ethnic fragmentation [log e]

–

–

–

–

–

–

0.84*

0.39 1.28

Democratic experience [1/T]

–

–

–

–

–

–

–0.67*

0.32 1.09

–

–

–

1.03

–

–

–

–

N R

2

45 0.015

S.E. VIF

B

45 0.080

S.E.

VIF

B

S.E.

VIF

0.92

–

–1.35

0.93

–

45 0.199

Note: **significant at p 0.01; *significant at p 0.05; (*) significant at p 0.1.

0.06 1.07

1.33

–

1.19

–

–

34


Eff number of parties/PR tier

Eff number of parties/PR tier

but does not change the impact of district magnitude, the main variable of interest. Ethnic fragmentation seems to have a positive effect, according to the social cleavages hypothesis, but contradicting previous findings (Moser, 2001b). Electoral systems with legal thresholds are included in a second step. In order to test the impact of the legal thresholds, it appears sound to include the PR tier of mixed electoral systems.32 A visual and statistical analysis of the impact of national legal thresholds shows that the level of these thresholds does not help to explain the effective number of parties. The scatter plot shows the level of legal thresholds and the effective number of parties (Figure 2.7). Most legal thresholds are at 5% of the national vote, but their outcomes vary widely, from two up to almost eight effective parties. There is no correlation between the level of the legal threshold and the effective number of parties. No matter whether mixed electoral systems are included (left graph) or excluded (right graph), no differences are visible. For the statistical analysis, I have inverted the legal threshold, in order to make it comparable in the second and third specification to the number of seats in parliament and to district magnitude,

8

6

4

2 0.02

0.03 0.04 0.05 National legal threshold

0.06

8

6

4

2 0.02

0.03 0.04 0.05 National legal threshold

0.06

Figure 2.7 Explaining the effective number of parties (N2, y-axis) through national legal thresholds (t, x-axis), 1990–2007. In the left graph, all the cases with legal thresholds are included (seventy cases plotted); in mixed electoral systems, the proportional threshold and the PR tier are considered. In the right graph, mixed electoral systems are excluded (forty-seven cases). Xsimple systems, O mixed electoral systems Note: For mixed, semi-compensatory systems (cf. Chapter 5), the overall effective number of parties is used as dependent variable, because the number of mandates in the whole chamber is allocated according to PR vote. For mixed non-compensatory systems, only the number of mandates won in the PR tier is relevant, because only those mandates are affected by the threshold.


35

Table 2.2 OLS regression models for the effective number of parliamentary parties (log), with the legal threshold and a combined term as explanatory variables; robust standard errors (1) cases with legal threshold Regression model Constant

B 1.87

Legal threshold [log 1/t]

–0.18

Effective threshold [log (min(1/t, m))]

–

S.E.

VIF

0.60

–

0.19 1.13 –

–

(2) all cases B

S.E.

VIF

1.30

0.27

–

0.02

0.08 1.24

–

–

–

(3) all cases B

S.E.

VIF

1.28

0.11

–

–

–

–

0.05(*) 0.03 1.04

Ethnic –0.11 fractionalisation [log e]

0.17

1.07 –0.06

0.18 1.01 –0.11

0.18 1.04

Democratic experience [1/T]

0.29 1.07 –0.08

0.22 1.25 –0.03

0.24 1.01

N R

2

0.13

70 0.026

– –

1.09 90 –

0.004

– –

1.17 90 –

0.040

–

1.03

–

–

Note: **significant at p < 0.01; *significant at p < 0.05; (*) significant at p < 0.1.

and logarithmised (since it is always a positive number).33 The OLS regression confirms that the level of the legal thresholds has no explanatory power for variance in the number of parties (Table 2.2). Legal thresholds seem not to have any impact on the number of parties. In the first specification, I test the model only for the seventy elections where a national legal threshold has been applied. The second model looks at all elections (in cases with no legal threshold, the inverted threshold measure is set equal to the number of seats in parliament34). The third model employs the concept of the effective threshold (Lijphart, 1994b), comparing the legal threshold with district magnitude. The effective threshold is equal to the lower of the values inverted legal threshold and district magnitude. The lower of the two values is the more restrictive barrier against the entry of small parties. 35 Neither of the operationalisations, nor the control variables, explains substantively the number of parties.

36


2.4 Conclusions: Do electoral systems not work in post-communist countries? This chapter attempted to explain the number of parties in parliaments in post-communist democracies through the impact of electoral systems. The approaches, which work for established democracies, seem to fail in the post-communist world. These non-results allow several conclusions. Electoral systems cannot work differently in different regions of the world. They are well-defined rules, which, regardless of the cultural, political or economic background, always convert votes into seats in the same way. The direct, mechanical impact of electoral systems cannot vary by region. The failure to explain the variance in the number of parties through conventional electoral system theories thus cannot rely on a different functioning of systems themselves. The variety of systems employed in post-communist Europe suggests further that the failure cannot be attributed to a lack of relevant variance in the independent variable. Rather, the way electoral systems affect party systems might rely on further aspects, which are commonly discussed only briefly, taken for granted, or sometimes neglected completely, such as actors’ behaviour (discussed as psychological effect of electoral systems), the way parties link across electoral districts (party nationalisation), electoral coalitions, the type of linkage of voters to political parties (social cleavages or alternative ways of party orientation). This study attempts to resolve the electoral systems’ puzzle in postcommunist Europe with a focus on some of the aspects mentioned here. This should help to identify important aspects related to the functioning of electoral systems and to find a more expressive model which helps to explain the variance in the effective number of parties.

3 How to Measure Party Nationalisation

3.1 Introduction: The relevance of party nationalisation and the lack of a reliable measure How national is politics? The question becomes salient when a country is regionally divided, and is threatened with breaking up, when the struggle over political options goes along territorial lines. The ‘American Civil War’ might be the best-known example of national politics going regional; the struggles about the Ukrainian political direction, which became most pertinent in the 2004 elections, or the ethno-regional conflict in Bosnia and Herzegovina are the most prominent of many recent examples from post-communist Europe. Such territorialised conflicts could have completely different implications for politics than non-territorialised ones. Autonomy for a certain part of a country is a viable solution for territorial conflicts, as is separation. The question of how national or how territorialised is politics, and more particularly party politics, has attracted increasing scholarly interest in recent years. The subject has a particular importance because territorially based parties might ‘demand distinctive treatment for that region, at levels from the co-operative sharing of powers by federalism up to and including “national” independence’ (Rose and Urwin, 1975, p. 46). The term party nationalisation describes the territorial homogeneity of party support (Caramani, 2004; Claggett, Flanigan, and Zingale, 1984; Ishiyama, 2002; Jones and Mainwaring, 2003; Stokes, 1967), and party system nationalisation the territorial homogeneity of a party system.36 The empirical reality that is compared through the measuring of party nationalisation is the level of support of a party across territorial units. The nationalisation of the party system as a 37

38


whole is an aggregated measure for the territorial homogeneity of support of all the parties included in the system. A whole range of indices has been developed to measure party nationalisation, but all of them have certain characteristics that make their universal application to cross-country comparisons problematic. Many simple measures appear useful for the comparison of the nationalisation of parties of similar sizes, assuming the data on their support comes from similarly sized and numbered territorial units. However, in many empirical studies, such laboratory conditions are seldom fulfilled. Rather, we have differently sized parties, differently sized territorial units across the country where our electoral data stems from, or different numbers of territorial units from country to country. Since party nationalisation measures the territorial heterogeneity of party support, it should only vary as a function of the territorial variance of the votes and be influenced as little as possible by other, external factors, such as the drawing of the administrative units of a country, which are often the basis of the electoral statistics which are needed for the calculation of party nationalisation. As will be shown in this article, the number of administrative units into which a country is divided can completely distort the results of measures of heterogeneity. Comparative research faces the risk of being biased by the problems of these measures, either if countries with differently-sized units are compared, or if there is a change over time in the number of units within a country. For instance, several studies have investigated the question of whether party nationalisation is affected by electoral systems (Cox and Knoll, 2003; Nikolenyi, 2008). If, however, our party nationalisation measure relies on electoral results at the level of electoral districts, we do not know whether the measured effect of the number of districts is real, or whether it is an artefact emanating from possible distortions of our measure by the number of units. This is why careful examination for such biases and the standardisation effort are needed. My standardised party nationalisation score takes better account of varying district sizes inside a country, and it considers the varying number of territorial units between countries that are to be compared. My measure of party nationalisation is based on a Gini coefficient on inequalities in distribution. Tests on electoral data from Central and Eastern Europe show that it is more reliable than previous measures. 37 Before developing and testing my own measure of party nationalisation, I provide a systematic overview of the previous measures for the phenomenon, and explain the possible biases that might occur.

How to Measure Party Nationalisation

3.2

39

Previous measures of party nationalisation

Measures of party nationalisation basically investigate the diversity of party systems across territorial units. Four families of statistical indicators have been used to describe party nationalisation: frequencies, variances, distribution measures and inflation measures. In this section, I describe the development of indicators previously used to measure party nationalisation, and relate them to a number of possible biases which might impede comparability (see Table 3.1 for an overview). 38 Some indices are calculated for party systems as a whole, others for single parties. (Formulae for the discussed indices are listed in Appendix A.) Indices of frequency (share of contested districts) The first scholars who measured party nationalisation used competition indices. Especially in electoral systems with small constituencies, certain parties do not compete nationwide, but only in some districts. To capture this, Rose and Urwin (1975, p. 19 et seq.) counted the number of uncontested seats. This has been extended to the territorial coverage index (Caramani, 2004, p. 61), measuring the percentage of territorial units where a political party runs in elections.39 These measures are easy to calculate and to understand. Where applications address the supply side of the electoral market the offer of parties to voters they capture the problem best. However, the understanding of party nationalisation remains rather rudimentary: measuring the share of contested districts gives little information about regional heterogeneity of party support (first problem). Parties with a very regionally based electorate, but many candidates or own lists in most electoral districts, would appear to be highly nationalised, even if the regional support of their voters proved this not to be the case. In electoral systems with a single national constituency, frequency indices would by default indicate values of 100% for all parties. Indices of variance (variation of the district results) A second family of indicators is based on the variance in the vote share of political parties across territorial units. The index of variation developed by Rose and Urwin (1975, p. 28) is the mean absolute deviation of regional electoral results from the national ones, in analogy to the Rae index of disproportionality (Rae, 1967). Similar indices were proposed by Lee (1988) and Rose and Urwin (1975, p. 24).40 These indices can also be computed for single territorial units, comparing them with the national mean.

40 Territory and Electoral Rules in Post-Communist Democracies

The second to fifth problems can be discussed in relation to this group of indices. The second problem concerns the fact that these indices lack upper limits, which has been criticised in the literature (Caramani, 2004, p. 62). Indices that vary in a standardised range are more intelligible. Table 3.1 Indicators to measure the nationalisation of party systems and their discussed shortcomings

–

x

–

x

x

–

single party

x

–

–

–

x

x

–

single party or party sys.

–

x

x

x

x

x

–

–

–

x

–

–

x

–

–

–

x

–

x

x

–

–

–

–

–

(x)c

x

–

7. Large local party system bias

5. Insensitivity to transfers

x

6. Insensitivity to the number of territorial units

4. Lacking scale invariance (‘party size problem’)

party sys.

Level

3. Does not take different sizes of territorial units into account

2. Lacking upper limit

Index

1. No consideration of party support level

Reported problems (1–7)

a) Competition indices Number of uncontested seats (Rose and Urwin, 1975)a Territorial coverage index (Caramani, 2004)b b) Indices of variance Index of variation (Rose and Urwin, 1975) mean absolute deviation Standardised and weighted variability coefficient (Ersson et al., 1985) Index adjusted for party size and number of regions (Caramani, 2004) Cumulative regional inequality (Rose and Urwin, 1975)/ index of territorial distribution

Continued


41

Table 3.1 Continued

–

–

x

–

party sys.

–

x

x

–

–

x

x

–

–

x

–

–

x

x

–

x

–

–

–

x

x

Three-dimensional measure (Kasuya and Moenius, 2008)

–

x

–

–

–

x

–

Modified indicator of party aggregation (my modification)

–

(x)d

–

–

x

x

7. Large local party system bias

x

6. Insensitivity to the number of territorial units

–

5. Insensitivity to transfers

–

Level

4. Lacking scale invariance (‘party size problem’)

2. Lacking upper limit

single party or party sys.

Index

3. Does not take different sizes of territorial units into account

1. No consideration of party support level

Reported problems (1–7)

c) Distribution coefficients Party nationalisation score (complement of Gini index) (Jones and Mainwaring, 2003) d) Inflation measures Indicator of party aggregation (Chhibber and Kollman, 1998) Inflation score (Cox, 1999) Index of party aggregation (Allik, 2006) Inflation index (Moenius and Kasuya, 2004)

–

Notes: * Two of the reported shortcomings are discussed in detail in the text below. a Only for elections in single-seat district systems. b Not suitable for multi-tier systems. c Insensitive to transfers of party votes between districts. d Lacks upper limit, unless modified according to Cox (1999) or Allik (2006). For reasons of space only one version is reported. Source: Bochsler (2010).

42


Third, the index does not consider whether all territorial units within a country are equally large. In many countries, certain units are larger than others. In many empirical cases, not all of the territorial units have the same size; some of the units are much larger than others. Hence, the measures reflect to some extent the administrative structure, rather than the territorial structure of the vote. Reforms of administrative units would thus strongly and systematically affect the values of party nationalisation, depending on the question of whether they particularly target regions where a political party is strong or weak, or whether the most substantial reforms occur in regions where the electoral strength is close to the national mean. (For a more extensive discussion, see below.)41 Further, if party nationalisation measures do not consider the differences in the size of territorial units within a country, they lead to a potential paradox. One desirable property of variance measures is that they should indicate increase as we move votes from a weak stronghold to a more pronounced stronghold of a political party. However, if variance measures do not take into account the size of territorial units, then moving votes from a unit where the party is stronger than in the national mean to a unit where its electoral strength is higher might wrongly indicate lower variance, provided that the second unit is the larger one. The fourth problem that afflicts these indices regards party size, or the problem of scale invariance. Parties with a small vote share by their nature have smaller deviations than large parties.42 This problem can be illustrated if we imagine that a large party splits into two exactly equal successor parties, each of them taking half of the votes in each of the territorial units. At a given vote share, the variance of each of the successor parties is cut in half. Thus, for exactly the same pattern of vote distribution, smaller parties have smaller variances than large parties. All the measures of party nationalisation are empirically either positively or negatively correlated to party sizes (Caramani, 2004, pp. 646). We should, however, distinguish empirical correlations from measurement-induced biases. Empirically, small parties often rely for their support on local strongholds, and their strength is more heterogeneous. An unbiased measure should thus, when applied to empirical data, typically show more heterogeneity for small parties than for large parties. On the other hand, the absolute variation of vote shares among territorial units is often larger for large parties, solely on the grounds that for larger values variances are usually larger, as shown in the example of the party split. The mathematical concept of scale invariance appears thus as the only one relevant to the construction of nationalisation


43

measures; or, in other words, indicators should assure that, if a party is split in half, the measured values stay the same. And fifth, all of the measures which do not weight the deviations thus all except for the standard deviation measure contain the problem of ‘insensitivity to transfers’ (Firebaugh, 2003, pp. 7980; Monroe, 1994, p. 139; Smithson, 1982, p. 261). In certain situations, when votes are transferred from a district where a party is weaker to a district where the party is relatively stronger, the (non-weighted) indices are not affected, even if such transfers increase heterogeneity.43 All four problems occur frequently among measures of territorial heterogeneity, and are related to some other indicators too (see Table 3.1 for an overview). A number of standardisations have been proposed to tackle the problem of party size or scale invariance. Ersson et al. (1985, pp. 1756) employed the standardised and weighted variability coefficient, dividing the variance by the mean of the parties’ vote share. Their measure has been criticised due to its lack of an upper limit (Jones and Mainwaring, 2003) and because it is biased by the sample size in several regards (Smithson, 1982). The index adjusted for party size and number of regions divides the Lee index by the average vote share in the territorial units (Caramani, 2004, p. 62). The cumulative regional inequality measure takes into account the influences of electoral units’ size differences on the degree of homogeneity. It considers whether the distribution of a party’s total national vote in sub-national regions corresponds with the population (voter) share of those regions of the national population (Rose and Urwin, 1975, p. 30). Still, these measures are insensitive to transfers; the index adjusted for party size and number of regions also does not take into account different-sized territorial units. Distribution coefficients (Gini coefficients) The Gini coefficient of inequality is usually used to measure unequal wealth distribution among the population (Gini, 1921). Jones and Mainwaring (2003) use it in analogy to measure the uneven distribution of vote shares across territorial units, taking the complement of the Gini coefficient. The regional issue dimensions indicator suffers from the scale invariance problem (see above), and the two measures based on the Gini coefficient do not take differences in district sizes into account (Chhibber and Kollman, 2004, p. 177; Moenius and Kasuya, 2004, p. 522; Taagepera, 2005). The sixth problem concerns the level of aggregation of the electoral data. Generally, if a whole is divided into a larger number of units (lower level

44


of aggregation), and these units are characterised, more fine-grained differences among these units emerge than if the whole is divided into fewer units (higher aggregation). Party nationalisation might be calculated from electoral results from fine-grained units, such as precincts or municipalities. If these units are aggregated into larger units, such as provinces, the within-province variation is omitted, and only the across-province variation accounted for. Therefore, our indicators are affected by the level of aggregation of the electoral data. This might be a problem of comparability across countries (or within countries, across time), if we were to compare data that is based on a different number of territorial units. Inflation measures (difference between the size of the regional and the national party system) One class of measures is suited for the party system as a whole, but not for single parties. The inflation measures compare the number of parties at the national and at the local level. Chhibber and Kollman’s (1998, p. 331) indicator of party aggregation subtracts the average of the effective number of parties at the district level from the same number at the national level. Whereas the indicator might be useful for single-seat district systems, it has its limits for PR systems, when in each district many parties compete. Cox’s (1999, pp. 1556) inflation score considers this problem and measures the relative difference in numbers of parties between the district and the national level.44 The index of party aggregation (Allik, 2006) is the complement of the inflation score, so that 1 stands for high party aggregation, and low values signify low party aggregation. The inflation index (Moenius and Kasuya, 2004) considers the different size of constituencies. Inflation measures require slightly less information than other party nationalisation measures. If we had access to the electoral results by districts and overall results at the national level, but were not able to identify the parties correctly and to link the regional results to the national ones, inflation measures could still be calculated. However, inflation measures are not fully reliable, because in certain situations they suggest wrong interpretations. Given that inflation measures have been increasingly used and suggested in the literature, and that there is a general lack of a substantive critique of these measures, the following paragraph demonstrates the errors that might occur. The problems of inflation measures The basic idea behind inflation measures is that a territorially heterogeneous party system usually has fewer parties at the local level than at the national level. At the district level, there is Duvergerian two-party


45

competition, but, due to inter-regional differences, substantially more than two parties compete in the national parliament (examples in Chhibber and Kollman, 2004). Inflation measures are based on the difference in the number of parties at the local and at the national level, assuming that the number of parties at the local level is smaller than at the aggregated national level. This assumption leads to the seventh problem to be discussed, the bias through large local party systems. Local party systems might in some districts be larger than the national one. This typically occurs in urban areas, whereas in rural areas some parties are dominant and thus party system fractionalisation is lower (see Bochsler, 2010 for details). Kasuya and Moenius (2008) propose an alternative version of a party inflation measure: they add two new dimensions to their concept. They still take the difference between the average party system size at the district level and the national party system (party inflation) into account, but, besides this, they now also include the variance and kurtosis between districts. This new measure appears much more reliable than their previous one. However, their indicator remains an indirect measure of territorial heterogeneity, and the correction of previous biases might in some cases distort the results, so that new problems emerge (see Bochsler, 2010). It remains questionable whether any measure that is based on the effective number of parties in districts instead of party votes might lead to satisfying results. Which measure is the best suited? Table 3.1 provides an overview of indicators and their shortcomings. Not all the shortcomings are of the same seriousness and type. Whereas some might be just more difficult to understand (cf. lack of upper limit), others cast the applicability of the indicator into question (wrong results). And, finally, the characteristic of indicators taking high values for small parties is reported as a problem but, as will be argued below, it reflects instead the empirical reality that small parties are often less nationalised than large ones. Further, the best indicator is not necessarily the one with the smallest number of problems. What it is applied to, and whether problems are resolvable, matter more.

3.3 The Gini coefficient to measure regional heterogeneity Of the indicators described above, I choose the complement of the Gini coefficient, or the so-called party nationalisation score, as a basis for further consideration. It is better known than the other indices used and is based on a powerful measure of heterogeneity in distributions,

46


so that the shortcomings might be fixed. Below, I describe how the Gini indicator is constructed, before standardising it for both aspects which have been detected as problematic. The Gini coefficient is a measure of inequalities across the units, which are most frequently used to quantify heterogeneity in the distribution of wealth within a society (Jones and Mainwaring, 2003, p. 142). It has also been proposed as a measure of the disproportionality of electoral outcomes (Taagepera and Shugart, 1989, p. 263). The Gini coefficient can easily be converted into a measure of heterogeneity of the territorial vote distribution of a political party. In the case of a homogeneous distribution (high party nationalisation), every territorial unit will cast a number of votes for this political party which is approximately proportional to the unit’s size, or the party will win a similar vote share in every territorial unit. In the case of heterogeneous vote distributions, however, most of the votes are concentrated in a few territorial units. The calculation of the Gini coefficient can be explained geometrically (see Figure 3.1 below). The territorial units are drawn on the 2500

Calculated area = Gini coefficient Geographic units 1 to 6

2000

ct h

rfe

Pe

Party vote

ity

ne

ge

o om

1500 1000 500 0 0

1 2000

2

x

3

4

4000 6000 Population number

5 8000

6 10000

Figure 3.1 The calculation of the party nationalisation score with weighted units in a fictitious country with 10,000 electors and 6 territorial units. The party for which the score is calculated wins 25% of the total vote – from 2% in the first territorial unit up to 42% in the sixth unit. The six units are arranged in order of the relative party support (increasing gradient of the curve). On the x-axis, I reproduce the cumulative number of the population in the units, on the y-axis the cumulative number of party votes. Example: Point X at the upper right end of unit 2 marks the point with 3000 electors and 120 party votes. This means that 3000 electors live in the added units 1 and 2, where the party wins 120 votes in total Source: Bochsler (2010).


47

x-axis, ranked by their support share for party P, starting with the unit where party P gets its lowest share of the votes. The y-axis describes the cumulative function of votes across regions: thus for the first unit it just shows the number of votes party P scores in the first unit (with the lowest vote share). For the second unit, the votes scored in the second unit are added to the first, and so on. At the last unit, the y-axis shows the sum of votes scored in all territorial units, and thus the total number of votes party P scored in the whole country. From this, a convex line results. In the case of low party nationalisation, the area between the graph and the perfect homogeneity line (connecting the origin and the upper right angle) is large. With higher degrees of party nationalisation (homogeneous vote distribution), the graph approximates the perfect homogeneity line, and the area becomes smaller. The Gini coefficient is actually calculated as the area between the graph and the perfect homogeneity line. In the case of perfect homogeneity (perfect party nationalisation), this area is equal to 0, whereas in the case of extremely unequal distributions it is equal to 1. Jones and Mainwaring (2003) subtract this value from 1, because they do not intend to measure inequality, but rather equality or homogeneity, called party nationalisation. When party nationalisation is to be calculated for a whole party system (instead of single parties), the score is calculated first for each single party, and then the results are weighted by the number of voters per party in the system. PNS 1 Gi

¦ (1 Gi(P))p

N

1 ¦ (Gi(P))p N

[3.1]

3.4 Developments on the party nationalisation score: The impact of the size of territorial units For my standardised measure, I tackle two aspects differently from the score of Jones and Mainwaring (2003). Both concern the size of territorial units, on which the calculation of party nationalisation scores is based. Differently from previous measures, I take varying sizes of territorial units within countries and across countries into account. On the one hand, I weight exceptional voting behaviour more when it occurs in units with a large population than when it occurs in units with a small population. On the other hand, I take into account the fact that all measures will find more variance across small units of aggregation than across few larger units. Hence, I control for the number of units on which the calculation relies.

48


Correcting for unequal sizes of units in a country First, I take into account the share of each territorial unit out of the total number of votes in the country.45 By analogy to Firebaugh (2003, pp. 1267), it can be argued that, in certain situations, units should not be weighted by their size. For instance, this is the case when an empirical comparison of territorial units is used to analyse factors affecting voting behaviour, where each unit is a realisation of an electoral process. Differently, for an analysis of territorial homogeneity or inequality, inhabitants of large units should count equally to those of small units. I have conceptualised party nationalisation as a measure of territorial homogeneity of electoral support of political parties, and the territory is conceptualised regardless of politically decided administrative boundaries. Hence, my measure should be as insensitive as possible to the definition of territorial units through political instances or through statistical offices. An ideal measure of homogeneity should thus be affected as little as possible by changes of administrative boundaries.46 The size (in terms of population or area) of administrative units can vary considerably within countries. Using non-weighted measures of territorial homogeneity, electoral strongholds in a small and low-populated administrative unit affect our results equally to strongholds in a large and highly populated unit. For instance, taking the case of Canada, and calculating party nationalisation at the level of provinces, a different party landscape on the small province Prince Edwards Island (140,000 inhabitants) is as important as if the whole province of Quebec (7.7 million inhabitants) voted differently. Similarly, it might be questionable whether different voting behaviour on the Finnish island of Åland (26,000 inhabitants) should weight similarly to the whole province of Western Finland, which counts 70 times more inhabitants (Taagepera, 2005). For this reason, territorial units should be weighted by their size in order to measure inequalities (Firebaugh, 1999, 2003; Sala-i-Martin, 2002). Typically, one might rely on the geographic area, or on the population. Given that electoral representation is widely understood as a reflection of the will of the population,47 I use it as a baseline for the weighting of territorial units. Measures that weight territorial units are not sensitive to territorial reforms, such as the split or the merger of territorial units. Also, the relevance of the size of units is underlined by the fact that administrative units are often much more highly populated in urban areas than in rural areas. For instance, in Russia 50% of the electoral wards count just 19% of the population. Latvian electoral statistics (for the 2002 elections) count thirty-four counties, but the capital, Riga, counts 28%


49

of the countrywide voters, the same number of voters as in the eighteen smallest, often rural, districts combined.48 For the calculation of the weighted Gini coefficient, instead of just taking the number of units on the x-axis, I draw the cumulative number of voters by unit (see Figure 3.1). The resulting party nationalisation score with weighted units is a first step towards my standardised party nationalisation score. It allows the comparison of parties in a country with unequally large electoral units. The party nationalisation score with weighted units is the quotient of the area under the drawn cumulative function of votes across regions divided by the area under the perfect homogeneity line. In the case of perfect homogeneity (perfect party nationalisation), the score would take the value 1. In the case of extreme heterogeneity (very weak party nationalisation), the score would be close to zero.49 The following formula establishes the party nationalisation score with weighted units (PNSw) for a country with d territorial units [1; ... ; i;...; d], ordered according to the increasing vote share of party p. Each territorial unit i has vi voters, and pi of them vote for political party p. d





i

∑  v ⋅  ∑ p i

PNSw = 2 ⋅

1

j

−

1

d

pi   2  

d

∑ v ⋅∑ p i

1

[3.2]

i

1

Correcting for the unequal number of units across countries Second, my standardised party nationalisation score considers the number of territorial units in a country, on which it is calculated. This standardisation corrects, as far as possible, the effect of granularity of territorial data. Generally, if variance is calculated among the characteristics of n unequal parts of a territory, then variance will increase, as the number of parts n, into which the territory is divided, increases. Compared with data from many small units, in a few larger territorial units electoral strongholds will no longer be as pronounced, and will not necessarily be as recognisable.50 Therefore, for the same parties and exactly the same vote distribution across the territory, measures of homogeneity or variance can vary substantially, depending on the level of aggregation of the electoral data. Usually, studies of party nationalisation due to data availability rely on data from just one level of aggregation, such as electoral districts or provinces. I have constructed a few datasets that allow the comparison across different levels of aggregation. This allows a fully controlled

50


study of the effect of data aggregation on party nationalisation scores. For instance, for the 2001 elections in Poland, I rely on the full results from 2493 electoral precincts, which can be further aggregated into 377 districts, or into forty-one regions. For the same parties, with exactly the same vote distribution, measures of variance that are calculated at different levels of data aggregation lead to impressively different values. For the Self Defence party, the weighted nationalisation score amounts to 0.67 if data is aggregated at the level of precincts. It increases to 0.71 if moving to districts, or even to 0.80 in the case of provinces. For comparisons of party nationalisation across countries, this is a major problem. Often, regional electoral data is reported for electoral districts, administrative regions or municipalities. In some cases, only data from very large units exists, overall fewer than ten units per country. In other cases, the data is fine-grained from hundreds or thousands of electoral or counting districts. Since it is usually not possible to get datasets with a comparable number of units, I define a standard number of territorial units, and estimate party nationalisation at this level of granularity. If there were a single territorial unit, the measured party nationalisation score would be 1. With an increasing number of units, party nationalisation decreases. No function can precisely capture all the different imaginable patterns of vote distribution and aggregation effects. For certain parties or countries we might find strong electoral differences across regions, while in other cases electoral behaviour across regions is similar, but varies across municipalities or neighbourhoods. Currently, the common implicit assumption that users of party nationalisation measures need to make (as a lack of other measures) is that the level of aggregation (the number of units) has no effect on their results. This is so implausible that we might call the assumption misleading: it would require that there is no within-unit variation within the units where party nationalisation is measured. On the contrary, I assume that an increase in the number of units should lead to a steady increase in the heterogeneity that we observe. If we imagine n provinces, and each of them is further divided into n municipalities (so that in total there are n2 municipalities), then variation among the n provinces should be similar to variation among the n municipalities in each of these provinces. I assume that heterogeneity measured at the lower level, PNS(n2), corresponds to the squared heterogeneity measured at the upper level PNS(n), PNS(n2) PNS(n)2 . This idea is represented by the concept of the following exponential function, PNS(dx) PNS(d)x.


51

I choose ten as the standardised number of territorial units for which party nationalisation shall be estimated to be comparable across countries. Having calculated party nationalisation PNS(d) for any other number of territorial units d (depending on the level of granularity of the electoral data), the standardised level of party nationalisation PNS10 is estimated according to the formula: PNS10 PNS(d)1/log(d). The standardisation with log(d) ignores so far that not all the territorial units are of the same size. Since the units were weighted by their size (number of voters) for the calculation of the Gini coefficient, they should be weighted as well for the standardisation in this step. For instance, four units might give a very differently calibrated picture of the regional voter distribution. If each of the four units counts 25% of the countrywide voters, regional voting differences are captured much more precisely than if one unit counts 50% of the voters, and the others 16.7% each. The effective number of units E (according to Laakso and Taagepera, 1979) allows us to consider this aspect:

E=

 d   ∑ votersi  1 d

∑ voters

2

, and

2 i

1

d    i pi     ∑  vi ⋅  ∑ p −     2    1 PNS10 =  2 ⋅ 1 d d     vi ⋅ ∑ pi ∑   1 1

1 log( E )

[3.3]

Due to the assumptions I have made, that party nationalisation is a phenomenon that decreases steadily with every increase in the number of territorial units, my formula will provide less reliable estimations under particular circumstances. Amongst these might be a political ethnic conflict which is present in between regions, but where party support inside these regions is territorially very homogeneous. I expect my formula to provide good estimations in the absence of such extreme cases, if the measured parties are not too small, and if electoral data is disaggregated in a sufficiently large number of territorial units. Ten territorial units seem to be a usually sufficiently good basis, but an even larger number might be advantageous. Certainly, other ways of standardisation might be more appropriate for different conceptualisations of party nationalisation. For instance,

52


instead of taking a fixed number of units per country, the cross-country comparison of diversity might rely on territorial units of a fixed size, for instance comparing units of 100,000 inhabitants. The larger a country, the larger the number of territorial units it would be divided into.51 Such a measure would measure diversity across a territory, regardless of country borders, asking how large the territory is where voters are voting homogeneously, where a regional political party is present, or where a regional party system exists. It queries thus the perspective about the absolute spread of party support. Instead, my concept addresses the relative spread of party support within a polity. Starting from the territory of a country (or another given polity), it asks (at the party system level) how many different regional party systems exist, or how many territorial splits exist in a country’s party politics, or how homogeneously are the votes of a political party spread within a country.

3.5 Testing the ‘standardised party nationalisation score’ on Central and Eastern European data In order to test whether the standardised party nationalisation score solves the problem of granularity (number of territorial units taken into account), I test it on a series of elections for which data is available on several sub-national levels, such as provinces and municipalities. The most common datasets on sub-national electoral results, such as Caramani’s (2000) database on elections in Western Europe, or the Essex database on elections in Central and Eastern Europe, provide the results only on one level of aggregation, usually at the level of electoral districts or regions/provinces. Datasets that include several countries only include different levels of aggregation of electoral results for the same elections in some exceptional cases. This is why I employ my own database on post-communist elections in Central and Eastern Europe, which provides electoral results for the same elections on two or even three territorial levels for twenty-eight electoral contests, allowing the comparison of party nationalisation measures for different levels of aggregation. The selection of these elections is solely driven by the availability of data. For some elections, my database contains data, for instance, both at the level of counting circles and at the level of electoral districts.52 Those multiple datasets allow a controlled experiment using the party nationalisation measure. Since the data refers to exactly the same elections and the same vote distribution, the only aspect that varies is the number of territorial units


53

taken into account. These are ideal laboratory data for my examination of the standardised score. I calculated two scores for each of the twenty-eight elections. Thus, for each election, both score at the lower level (for instance municipalities) and at the upper level (for instance electoral districts). Ideally, between indicators that are calculated with the larger dataset (lower level, more units) and with the smaller dataset (upper level, fewer units), only random, non-systematic differences should remain. This should lead to a substantial reduction of the differences between both levels (apart from cases where the territorial differences between party strongholds and areas with weaker support strictly follow the boundaries between territorial units at the upper level). Without standardising, the results of the party nationalisation score range from 0.54 to 0.91, within a variation of 40%. For each pair of data, the score is on the average 12.2% higher if calculated at the upper level of aggregation (Table 3.2) over the twenty-eight pairs of comparison. Hence, the characterisation of a party system might often heavily rely on the number of units taken into account. For instance, the party nationalisation score for Hungary in 1998 at the lower level (176 constituencies, party nationalisation 0.72) would appear to belong to the group of countries with the lowest degree of nationalisation. Only countries with deep ethno-cultural divides, such as Macedonia or Ukraine, report such scores at the upper level. The same Hungarian elections, measured at the upper level of twenty regions, would result in rather a high level of party nationalisation (0.87). The huge difference between the lower and the upper level is eliminated to a high degree through the standardisation (Table 3.2). For the Hungarian elections of 1998, the two values vary by only 3.4% after standardisation (instead of 21%). In twenty-five out of twenty-eight analysed elections, the standardised score proves to be more reliable than the non-standardised one. The systematic deviation was reduced for the dataset under study by more than seven times, from 12.2% to 1.7% (means over all twenty-eight elections). A closer look at single cases shows that six cases remain where the standardised score is not very satisfactory. Specifically, there the difference between both scores is still larger than 5%. These are Macedonia (2006, 8.7%), Latvia (2006, second pair, 14.1%), Slovakia (1992, 9.9%), Moldova (1998, 9.9%), and Ukraine (2006, 18.7% 11.0%). This, however, should not be fully surprising. As pointed out above, if we have electoral data at one level of aggregation only, we cannot possibly conclude what party nationalisation would look like at a different level. The

Table 3.2 Calculation of the relative deviation between the upper level and the lower level for both indicators Upper level Country

Year

Units

Czech Republic Estonia

2006 2003

Estonia Hungary* Hungary* Hungry* Hungary* Hungary* Latvia Latvia Latvia Macedonia

2007 1990 1994 1998 2002 2006 2002 2006 2006 2006

Lower level

Relative difference

wPNS

sPNS

Units

wPNS

sPNS

wPNS (%)

sPNS (%)

14 (11.9) 12 (11.4)

0.894 0.814

0.901 0.823

91 (73.7) 248 (42.5)

0.870 0.733

0.927 0.821

2.8 11.1

2.9 0.3

12 (11.1) 20 (14.4) 20 (13.3) 20 (12.3) 20 (12.7) 20 (12.9) 5 (4.5) 5 (4.4) 34 (9.4) 6 (6.0)

0.828 0.725 0.845 0.874 0.909 0.907 0.817 0.811 0.747 0.710

0.834 0.738 0.855 0.883 0.917 0.915 0.742 0.726 0.741 0.651

402 (183.1) 176 (169.7) 176 (171.4) 176 (16906) 176 (170.2) 176 (168.9) 34 (10.2) 34 (9.4) 1005 (579.4) 88 (43.4)

0.735 0.606 0.757 0.723 0.821 0.841 0.756 0.747 0.671 0.593

0.870 0.755 0.863 0.854 0.914 0.920 0.758 0.741 0.863 0.713

12.6 19.6 11.6 20.8 10.8 7.9 8.1 8.5 11.3 19.8

4.1 2.2 1.0 3.4 0.3 0.5 2.1 2.1 14.1 8.7

Poland Poland Poland Poland Russia* Slovakia Slovakia Slovenia Slovenia Ukraine Ukraine

1991 1993 1997 2001 2003 1992 2002 1996 2004 2006 2006

37 (33.2) 52 (40.7) 52 (39.3) 41 (37.7) 225 (205.2) 4 (3.5) 79 (60.0) 8 (8.0) 8 (8.0) 27 (21.4) 27 (21.4)

0.684 0.759 0.809 0.810 0.793 0.831 0.695 0.889 0.901 0.605 0.605

0.766 0.839 0.873 0.873 0.904 0.715 0.806 0.879 0.891 0.685 0.685

2418 (209.8) 2465 (243.8) 2478 (226.7) 2493 (218.4) 2756 (1415.4) 42 (36.7) 2917 (224.5) 88 (83.9) 88 (83.7) 33,920 (22,995.3) 225 (192.2)

0.582 0.657 0.732 0.727 0.760 0.708 0.614 0.783 0.794 0.542 0.566

Average Notes: wPNS: party system nationalisation scores, with units weighted by their size. sPNS: standardised party system nationalisation scores, standardised by the number of territorial units. The effective number of units (according to the Laakso–Taagepera formula) is indicated in parentheses. * Hungary: single-seat districts; Russia: PR tier. Source: Bochsler (2010).

0.776 0.833 0.870 0.868 0.915 0.794 0.800 0.877 0.885 0.842 0.770

17.6 15.4 10.6 11.5 4.3 17.5 13.3 13.5 13.5 11.7 7.0

1.3 0.7 0.4 0.5 1.3 9.9 0.8 0.2 0.7 18.7 11.0

12.5

2.0


exact pattern might be different for certain parties and certain countries. All the problematic cases in my sample concern countries with important ethno-cultural divisions that follow (partly) the administrative regions or electoral districts. Then, moving from few territorial units to more fine-grained data from a lower level will report only slightly more heterogeneity, so that our standardisation might be overdone. This will be even worse when our data is aggregated into very few units, as in the case of Slovakia, with only four units at the upper level in 1992. In Slovenia, however, with eight electoral districts, the standardisation seems already accurate. However, these are limits that will occur in every attempt to account for the bias of the number of units: if data is aggregated differently, we never measure exactly the same. I can only provide a solution that is accurate for most, but not for all, cases. My measure eliminates the largest part of the bias that is contained in non-standardised measures, and that leads on the whole to reasonably similar results for different levels of aggregation.

3.6

Conclusions

There is a growing interest in the study and quantification of party nationalisation, defined as the homogeneity of the electoral strength of a political party across the territory. To measure party nationalisation, one relies on electoral results from territorial units, such as provinces, municipalities or precincts. In this study, I show that measures of territorial homogeneity are strongly influenced by the number of territorial units that are taken into account for their calculation. Accordingly, the administrative division of a country might heavily affect the results of cross-country comparisons of party nationalisation. This study proposes a new indicator that accounts for differently sized units within the same country, and considers the number of territorial units that were used to calculate party nationalisation. My new indicator allows a comparison of party nationalisation even if the number of territorial units, on the basis of which electoral data is taken, differs substantially across countries that are to be compared. When my new indicator is applied, the results are as far as possible not so heavily influenced by the number of units taken into account. This might be an important property for many studies, not least because often, for each country, electoral data is available only on one level of aggregation. Tests show that my score is quite robust when it is applied to data from different levels of aggregation. Thus, there is no systematic difference


57

between results if the measure is computed on a dataset that includes many territorial units (for instance, national electoral results at the municipal level) or few territorial units (regional level). If feasible, datasets with more than ten territorial units should be favoured, since the standardised party nationalisation score might be biased for a small number of units.

4 Explaining the Nationalisation of Party Systems in Central and Eastern Europe

4.1

Introduction: Why study party nationalisation?

In the previous chapter, I have shown how party nationalisation might be measured. Later, in Chapter 6, party nationalisation will be used as an independent variable to explain why electoral systems in Central and Eastern Europe in some cases have led to different outcomes from those we would expect when using the common models. However, before employing party nationalisation in order to explain electoral system effects, the phenomenon itself will be explained. Two explanatory approaches for party nationalisation will be addressed and tested. Two recent books discuss these concepts as possible determinants of party nationalisation. The first, by Chhibber and Kollman (2004), investigates the phenomenon in four countries with single-seat electoral districts, namely Canada, India, the UK, and the USA. Using cross-country and time-series analyses, they come to the conclusion that party nationalisation is linked to the power distribution in states. If economic and political power lies with the central government, parties will be nationalised; if regional governments are strong, however, parties are less nationalised. This view has dominated most of the studies of party nationalisation. The second book, by Caramani, describes a process of ‘territorial homogenization of electoral behaviour, both electoral participation and the support for the main party families’ (Caramani, 2004, p. 1) over the past two centuries in Western European politics. It proposes an alternative explanation, stressing the importance of the territorial structure of political cleavages for the nationalisation degree of a party system. Caramani describes the development of political parties along nationwide cleavages, ‘characterized by the progressive reduction of territorial diversity in the course of the nineteenth and twentieth centuries’ (2004, p. 5). 58

Explaining the Nationalisation of Party Systems 59

However, in post-communist Central and Eastern European countries, the process of party nationalisation might look different than in the West. Whereas many Western democracies went through parallel and interrelated processes of strengthening of the central state and of society, and of party nationalisation, the Central and Eastern European states were already heavily centralised when they started democratising in 1989/90, and other, federal countries broke apart. Further, their social and economic histories look fundamentally different. Social and economic integration and modernisation in Central and Eastern Europe preceded the formation of the post-communist party systems. On the other hand, the role of social cleavages in Central and Eastern European politics and party systems is heavily disputed, except for the salient ethnic cleavage. The study of party nationalisation in Central and Eastern Europe appears important not only because this appears to be a gap in previous research on party nationalisation, but because territory-related conflicts are particularly important in recent politics in these countries. The only previous study that employs an explanatory comparative approach (Tiemann, 2005) did not show very reliable results, due to problems with measuring party nationalisation.53 Many previous investigations of post-communist party systems in Europe have limited their focus to the most stable democracies, or to the new EU members. Furthermore, studies on party nationalisation have remained limited to one or very few cases (Ishiyama, 2002; Meleshevich, 2006). In contrast, this study is innovative in that it provides a broad inventory of party nationalisation in Central and Eastern Europe, covering the period 1990–2007 in twenty countries, and ninety-five elections altogether. The broad focus on all (reasonably) competitive elections in post-communist countries in Europe is motivated by several reasons. First, apart from Belarus, all countries and political systems in Central and Eastern Europe have experienced a series of several reasonably free and competitive elections. 54 Second, an exclusion of the less stable democracies in the Western Balkans, as in other analyses, would mainly exclude the least nationalised cases. However, since the regional fractionalisation of the party systems in some ex-Yugoslav countries might be a factor that contributes to political instability, an exclusion of these cases might be an indirect way of selecting the cases based on the dependent variable. Third, the increase of the number of cases from ten (new EU member states) to twenty (all new democracies) allows more reliable results from comparative analyses.

60


This basis allows me to assess different explanatory approaches systematically using standard regression models. In a first section of the chapter, I discuss previous research and explanatory approaches, followed by their development for Central and Eastern European democracies, and empirical tests.

4.2

Four explanations for party nationalisation

Why do parties adopt a national structure, instead of remaining local, atomised micro-units, restricted to one electoral district? Seats are won within electoral districts, so that – from a pure electoral system perspective – a party does not need to be organised across districts in order to win seats in parliament.55 Nevertheless, even in countries with many electoral districts, parties join in a national structure and organise across districts. In this section, I discuss several aspects that explain the formation of national party systems, and that might help to explain cross-country variance at the level of party nationalisation. 4.2.1 Centralised state, centralised parties – or the other way round? The prevalent school describes party nationalisation as a consequence of the centralisation of power, namely of the dominance of national executives over parliaments (Cox, 1997, pp. 186–93), of presidentialism (Cox, 1997, pp.187–90; Hicken, 2003; Kasuya, 2001; Samuels, 2002), and of a unitary and centralised state (Chhibber and Kollman, 2004; Harbers, 2010). One of the two mentioned institutional dimensions is legislative– executive relations. The arguments that are used in the literature are diametrically opposed to each other. Certain authors focus on strong presidential offices, arguing that presidential electoral campaigns might help in the formation of national political parties, by linking the party system to nationally competing presidential candidates (Hicken, 2003, p. 4; Samuels, 2002). Presidential candidates rely on party sections as local resources for their elections, while legislative candidates profit from linkages to the presidential campaign (Kasuya, 2001, pp. 15–18; Samuels, 2002, p. 468). In turn, powerful presidential offices deprive political parties of the function of government formation, notably if the president is not elected as a partisan candidate. Hence, nationalisation through presidential campaigns should only be expected if presidential candidates are closely linked to political parties, and if elections for


the executive and the legislature are held simultaneously (Clark and Wittrock, 2005; Hicken, 2003; Kasuya, 2001). In the super-presidential systems of Central and Eastern Europe, this is not the case. A different literature has, however, suggested a different link of party nationalisation to the executive–legislative relations. It highlights the function of parliamentary elections to indirectly elect the prime minister, and, if this view prevails, strong parliamentary executives might be responsible for high party nationalisation (Cox, 1997; Hicken, 2003). In countries where the parliament elects the government, voters and political actors attribute much more responsibility for national policies to the parliament than in presidential regimes (Tucker, 2002, p. 58). This, in turn, increases the importance of national political parties, while in (certain) presidential systems there might be no incentives for parties to organise nationally. The fully separate selection process of the executive and assembly and the lack of mutual dependence once in office permits voters and campaign contributors to demand different things of their executive and legislative candidates. As a result, voters need not demand that their legislators show loyalty to a party label. (Shugart and Haggard, 2001, p. 84) Certainly, under PR, there are other institutional incentives for political parties to nationalise – the electoral competition is structured around party lists, and often there are national compensation mechanisms that inhibit parties from competing countrywide. In single-seat district electoral systems, similar incentives are lacking: candidates are elected directly, and the electoral system does not encourage candidates to link to a political party or an electoral list. In the absence of any other systemic needs for nationwide party linkages, the form of government might play a particularly important role in electoral systems without partisan elements. Accordingly, I expect that party nationalisation is higher in PR elections and in parliamentary regimes.56 Second, centralisation of government has been employed as an institutional explanation of party nationalisation (Chhibber and Kollman, 2004; Cox and Knoll, 2003). In heavily centralised states, national political issues matter more to voters, because at this level of government the most substantial policies are decided. Political actors have little interest in organising on a local or regional level if there are no political responsibilities at this level, so that ‘local parties are abandoned altogether and disappear’ (Chhibber and Kollman, 2004, p. 222).

62


Decentralisation opens space for regional political parties (Blais and Carty, 1991, p. 85). While, for the study of certain countries, the centralisation of government dimension explains the variance fairly well (cf. Chhibber and Kollman, 2004), namely if these countries elect by plurality vote (the USA, Canada, India, and the UK), there might be reservations about this explanation and its causality in the Central and Eastern European case, for three reasons. First, the presented idea relies in large part on the indirect election of the prime minister, assuming single party majority governments and the existence of two main national parties, which compete against each other.57 Such cases are rare in Central and Eastern Europe, not least due to the widespread application of proportional representation and mixed electoral systems. If parliamentary majorities for a single party are rare, small parties, even regional ones, can gain a lucrative role as kingmaker, and in exchange obtain particular benefits.58 They might, for instance, press for increased regional autonomy. Hence, even in highly centralised states, there is no need for the voters to abandon small or regional parties. Second, even in a two-party competition, the two main competitors might as well be based on regional divides or interests (for instance in Ukraine, cf. Katchanovski, 2006). In such a case, party nationalisation will be low, even in a unitary state. To be precise, we might expect that countries with dominant regional cleavages might adopt a decentralised administration, so that the combination of strong centralisation and regional conflicts might be rare. Third, the centralisation hypothesis presumes that the institutional structure of the state gives incentives for the (non-)formation of regional parties. However, the contrary might be the case too: regionalist parties, when participating in governing coalitions, might demand a stronger political or financial decentralisation (Chhibber and Kollman, 2004, pp. 226–7; Heller, 2002).59 Federalist institutions might ‘reduce the expression of regional protest in the party system by opening up institutional channels of voice’ (Caramani, 2004, p. 292). They might be introduced in reaction to the emergence of regionalist parties. Party nationalisation might be not only the consequence, but also the cause of state centralisation. 4.2.2

The cleavage hypothesis

Alternatively, party nationalisation might be explained as a consequence of the territorial structure of social or socio-economic divides (Lipset and Rokkan, 1967). Caramani (2004, p. 15) addresses the


centre–periphery and the urban–rural cleavage as ‘territorial’ divides, connected to low levels of nationalisation. In contrast, ‘functional’ cleavages, such as the economic cleavage in Western Europe, do not have a territorial character, so that parties organising along such cleavages are highly nationalised (cf. Caramani, 2004; Cox, 1999, p. 159). The explanation of party system structures by cleavages has been criticised, though, because cleavages do not convert into parties as a matter of course, but this is produced by the political system itself (Zielinski, 2002, p. 187). In the case of Central and Eastern Europe, only a few scholars are looking for similarities of political divisions with social cleavages in Western democracies (Kitschelt et al., 1999). The overwhelming view is that cleavages, especially if they are narrowly defined, are of limited relevance in the region (Elster et al., 1998, pp. 247–70). However, one social divide60 appears to be an exception to the rule: the ethnic divide61 is salient in Central and Eastern Europe (Evans and Need, 2002) and helps many parties to mobilise their voters (Elster et al., 1998, p. 252). Ethnic minorities exist in almost all countries, and they vote in large numbers for their own parties. Furthermore, issues related to ethnicity also help nationalist parties of the titular nation to mobilise their voters. The investigation of ethnic divides in order to explain party nationalisation degrees may yield promising results, as many of the ethnic minorities in Central and Eastern European countries are territorially settled. If such ethnic divides become manifest in party politics, then the ethnic structure of a country will explain why the electoral strength of political parties varies across regions. Ethnic and national identities have been discussed as constructed categories (cf. Evans and Whitefield, 2000), and the unification of the nation state and a homogeneous ethnic structure have been identified as a two-way process (Caramani, 2004, p. 23). Ethnic awareness of (parts of) the population can be engineered (through education, media, use of language, etc.) or forced through violence (‘ethnic cleansing’, genocide). A prime example from the most recent history of the region is the wars in former Yugoslavia.62 However, in the post-1990 period, ethnic identities were substantially more stable than the party systems. Indeed, all current ethnic categories in Central and Eastern Europe have earlier roots, and, even if ethnic perceptions might undergo a slow change, all ethnic groups that can be identified today were already considered to be distinct ethnic groups before the recent transition.63Accordingly, the ethnic structure is almost an exogenous variable in the region. Still, in awareness of the

64


possibility of two-sided processes, particular attention is devoted to the ex-Yugoslav, and particularly to the Bosnian, case in the empirical part of this chapter. Ethnic mobilisation is constant at a very high level across the region, so that the results of such analysis do not stem from different levels of mobilisation. There are different motives for ethnic mobilisation (see, for instance, Brubaker, 1996; Kymlicka, 2002, p.20; Evans and Need, 2002, p. 656). However, mobilisation appears as a common constant for all countries under study, so that there should not be any problem of circular dependency. 4.2.3

The intervening character of electoral systems

Electoral systems might exert constraints against the formation of regional parties (Cox, 1999, p.159). My main focus relies on national legal thresholds, which exclude parties below a certain national vote share from representation in parliament (cf. Taagepera, 1998). The thresholds in Central and Eastern European democracies, reaching from 2.5% (Albania) up to 6% (Moldova) of the national vote, seem rather moderate. However, what is not particularly high for a national political party throws regional competitors out of the electoral race. A party which is only competitive in a region with 10% of the national voters would need to get half the votes in this region to pass a national threshold of 10%. Under substantial national legal thresholds, regional parties can thus only get access to parliament if their region is sufficiently large and if they have a fairly dominant position in their region, or, alternatively, they can anticipate that they have no chance of getting elected, and not even compete. In contrast to legal thresholds, small electoral districts do not have a special effect on party nationalisation. Since electoral districts are territorially defined, those with a small number of seats exclude locally small parties, but they do not particularly affect regional parties. Many of the minorities are rather small, so that legal thresholds might exclude minority parties from representation, and keep party nationalisation at high levels, even in countries with territorial ethnic divides.64 4.2.4

The development of party nationalisation over time

Lastly, time might play a role for party nationalisation, particularly in party systems in formation. In Western democracies, party nationalisation increased in line with the economic, social and cultural modernisation and integration of the society (Caramani, 2004). Unlike Western European countries during the time of enfranchisement, Central and Eastern European economies and societies were mainly modernised


and nationally integrated, when the countries were democratised. The starting point for the development of party systems and party nationalisation is thus not comparable in Central and Eastern Europe and in Western Europe. An initial lack of organisational strength of political parties (Olson, 1998, p. 434) might imply that party nationalisation was lower in a first period than in later elections. High inter-election volatility, as experienced in many countries in Central and Eastern Europe, might lead to the erosion of local party strongholds and to increasing party nationalisation.

4.3

Operationalisation

Despite the importance of party nationalisation for the analysis of party systems, it has only recently been measured and explained for a broader number of countries. So far, Central and Eastern Europe has not figured among these, even if party nationalisation varies considerably in this region, and is expected be a salient aspect in a region where a few countries have changed their state borders and face ethnic and regional conflicts. Measuring party nationalisation requires extensive databases covering electoral results at sub-national levels, including regions, districts, municipalities or counting circles. However, if common databases used in comparative politics contain electoral results, they are at the national level. I therefore built a new database on electoral results from Central and Eastern Europe at the regional and local levels, which included the elections held in the period 1990–2007. A part of the data could be taken from a previous project at the University of Essex, which collected district electoral results from ten countries altogether over the period 1990/91–2001.65 Further datasets were collected by the author. All in all, my database covers ninety-five elections, from twenty countries, with 66,623 sub-national units (such as constituencies, regions, municipalities, occasionally even polling stations). To calculate party nationalisation, I employ the standardised party nationalisation score that has been introduced in Chapter 3 of this book. Among the explanatory variables in my model, political decentralisation might be measured either through constitutional strength of regional or local governments, or through budget decentralisation. I choose the second option, due to the better availability of comparable data,66 and because local or regional governments exist, but in many cases lack any real autonomy due to insufficient transfer of government funds (Falleti, 2005). While all Central and Eastern European

66


democracies have a local level of administration, and in some cases a further intermediate level, these levels sometimes have almost no financial competences and funding. This is why a strong correlation of the centralisation of government expenses and the level of nationalisation of party systems might emerge. While party nationalisation data is provided for the whole period 1990–2007, the available data on financial centralisation in Central and Eastern Europe allows only a cross-country analysis, but no systematic analysis of the development over time.67 This should not be a major shortcoming, because, for the few countries where longitudinal quantitative data is available, the level of financial centralisation appears to be rather stable.68 When studying ethnic divides as a variable to explain the structure of a party system, it is relevant to identify all existing ethnic groups (as far as possible), independently of the politicisation of their ethnic identity. Otherwise, my indicators of the ethnic structure risk failing to measure ethnic diversity, but might rather measure the existence of a political player that mobilises on the ethnic divide, which would make the model circular. This is why common databases on ethnic minorities (such as Gurr, 2005) need to be handled with care, even if they are frequently used to identify the ethnic structure and ethnic conflicts in comparative studies (for instance Ishiyama, 2006), including studies on Central and Eastern Europe (for instance Alonso and Ruiz-Rufino, 2007). Existing databases look primarily at minorities in conflict or at risk, and might be not appropriate for the purposes of the study of party nationalisation, including only politicised ethnic conflicts.69 Other databases seem to avoid this problem (Alesina et al., 2003), but do not contain data on the regional structure of the ethnic groups. In order to avoid a potentially circular measure of ethnic divides, and following the critique of Campos and Kuzeyev (2007), I mainly use data from censuses held after the transition to democracy, and I crossvalidate the census results with other sources which appear to be complementary, such as the reports of human rights organisations. They help me to identify relevant ethnic groups, namely in cases where the national census does not consider all groups that are commonly understood as ethnic minorities, in order to achieve the most complete image possible of ethnic groups in the regions. Previous measures of ethnic divisions have considered the number of ethnic groups in a country, or the related concept of fractionalisation, namely the Hirschmann–Herfindahl index,70 expressing maximal


fractionalisation with 1, and 0 for perfect homogeneity.71 Unlike most other measures, I do not look at common ethnic divisions, but only at those with territorial implications. Only geographically concentrated ethnic minorities are expected to have a negative impact on party nationalisation. The indicator of territorial ethnic divisions (eC) is a measure of fractionalisation that counts all territorially concentrated minorities against a basis of groups consisting of the ethnic majority (or the largest ethnic group) and all non-concentrated minorities.72 If the country is divided into two equal-sized, concentrated ethnic groups, the index amounts to 0.5, in the case of three equal groups to 0.67, etc. The level of national legal thresholds is taken from my own database (see Appendix B). The impact of time is measured through the ordinal number of the election in a country after the first multiparty elections. Particularly in the initial elections, a part of the party system might consist of barely organised committees around a local base, so that the highest change rate is expected at the beginning of democratic consolidation. This is operationalised using the inverted ordinal number of the election (T 1/number of election).

4.4 Party nationalisation in Central and Eastern Europe There are substantial differences in party nationalisation among this group of countries. For some elections in the Czech Republic (in 2002, 2006), Hungary (2002, 2006), Russia (2003, PR tier) and Serbia (2000), the vote distribution is almost perfectly homogeneous across the country, with a level of party nationalisation of 0.90–0.94. At the other end of the scale, in single-seat district elections in Russia and Ukraine, party nationalisation scores in the 1990s, at often below 0.3, were by far lower than in any recent elections in Western Europe or in America. Besides these particular cases, due to the inflation of local, non-partisan candidates (see below), the party nationalisation scores of Bosnia and Macedonia appear as extremely low in the cross-country comparison with countries for which party nationalisation has been measured (Figure 4.1). The indicators of party system nationalisation give an aggregated impression of the territorial differences in electoral behaviour in countries in Central and Eastern Europe. Besides this, election studies on the basis of aggregated data have disclosed territorially based differences in party support in Central and Eastern Europe and allow a tracing back of the different electoral patterns to social differences between regions.73 Some

68



1 .8 .6 .4 .2 0

Albania

Bosnia

Bulgaria

Croatia

Czech Republic

Estonia

Hungary

Kosovo

Latvia

Lithuania

Macedonia

Moldova

Montenegro

Poland

Romania

Russia

Serbia

Slovakia

Slovenia

Ukraine

1 .8 .6 .4 .2 0 1 .8 .6 .4 .2 0 1 .8 .6 .4 .2 0

1 2 3 4 5 6

1 2 3 4 5 6 1 2 3 4 5 6 1 2 3 4 5 6 Number of election

1 2 3 4 5 6

Figure 4.1 Party system nationalisation in twenty countries, development by country over time, 1990–2007 Notes: X PR elections (and PR votes in mixed electoral systems) O Single-seat elections (and district tiers in mixed electoral systems) The graph ‘Serbia’ includes elections in the Federal Republic of Yugoslavia (1992, 1996), the graph ‘Czech Republic’ the Czechoslovak elections (1990, 1992).

have described the closely related political division between winners and losers of the transition, with rural voters suffering from the economic reforms and accordingly opting for ex-communist, socialist or left-wing parties (Tucker, 2006), but these divides are not very stable in electoral terms. In this section, I first describe two territorially based divides that are present in several countries of the region and discuss the development of party nationalisation in countries that split up during the period


of investigation, before passing to systematic analyses of the hypotheses. I first test the government centralisation and discuss reservations about the direction of the causality and then proceed to the cleavage hypothesis, tested on the elections in the period 1990–2007. I show how electoral systems might reinforce party nationalisation, and discuss the peculiarity of super-presidentialist systems in Russia and Ukraine. 4.4.1 The urban–rural divide in comparison Divides between agricultural regions and urban centres belong to the classic inventory of the cleavage theory as described by Lipset and Rokkan (1967). They play a role in party politics in Central and Eastern Europe too (Elster et al., 1998, pp. 248–9; Johannsen, 2003). On the one hand, a number of parties organised around agricultural interests after the breakdown of communism, notably in Poland (Zarycki, 2000) and to a lesser extent in Russia (Wegren and Konitzer, 2006). Urbanisation has been identified as the territorial variable that explains most of the variance between the electoral districts in Poland in 1993 (Wade et al., 1995, pp. 419–20). The urban–rural divide is further strengthened by the historical legacy: the politically rural element is stronger in the territory of former Russian rule in the East, while territories that were acquired from Germany after World War II stand politically closer to urban areas (Zarycki and Nowak, 2000, pp. 347–9). However, the Polish Peasant Party, despite organising along a classic cleavage dimension, failed to establish a stable electorate (Szczerbiak, 2001) and lost more than half of its electorate, so that nationalisation of the party system in Poland increased and stabilised at a rather high level. Some of the agrarian parties in post-communist countries were incorporated earlier in the communist system, or they stand politically close to ex-communist parties. Even more frequent than parties which exclusively address the agrarian electorate is the economically motivated rural vote for ex- communist, socialist or left-wing parties. Whereas urban and economically more successful areas vote for reform parties, farmers and citizens in more remote rural areas fear the new competition in the new open market, find themselves among the relative losers of the post-communist transition, and vote accordingly (Fidrmuc, 2000; Tucker, 2006). Similarly, in Romania, the initial urban–rural divide has declined, due to high levels of inter-bloc volatility, so that party nationalisation increased substantially over the years (Bochsler and Gherghina, 2008). In Lithuania, the urban–rural divide, cross-cut by the ethnic divide, has been an important determinant of party strength (Clark, 1994, pp. 55–7). Low party nationalisation scores are, however, mostly

70


registered in the single-seat district tier of the mixed electoral system, where they result due to strategic cooperation among the right-wing parties. Estonia is a similar case, where the urban–rural and the ethnic divide complement each other. 4.4.2 The ethnic or cultural divide Parties that mobilise along ethnic lines have been present across the whole of Central and Eastern Europe (Moser, 2005; Trifunovska, 1999). Similarly to the observation in the case of the urban–rural conflict, ethnic conflicts can be present in different forms in the party system. Most ethnic minorities are represented by an ethnic minority party74 which campaigns for voters of the ethnic minority (and often wins an overwhelming share of their votes), and otherwise one of the mainstream parties (non-ethnically defined) often has close ties to ethnic minorities. Since the presence of ethnic minorities might polarise, ethnic conflicts might also create identity fears among members of the titular nation and increase the vote share of nationalist parties of the majority population, which stress the dominance of the titular nation. Where ethnic groups live in close concentration, this might lead to the manifestation of territorial divisions in the party system. The most important post-communist party systems in Europe which are ethnically divided are those of Bosnia and Macedonia. In both countries, party systems are almost completely separated regionally along ethnic lines. Ethnically based parties with concentrated support can be found in several other countries. Some of the largest ethnic minority parties in the region are the Hungarian minority parties in Romania and in the Slovak Republic, the Turkish minority party in Bulgaria, Russian minority parties in Latvia, or Albanian minority parties in Macedonia. Table 4.1 gives an overview of the electorally more important cases. In post-Soviet countries (particularly in Estonia, Lithuania, Moldova and occasionally Latvia), mobilisation among ethnic minorities often looks slightly different. Whereas, in the previously discussed cases, parties addressing exclusively ethnic minorities were formed, in the postSoviet cases mainstream parties have often managed to mobilise ethnic minority voters. Members of ethnic minorities are more inclined to vote for left-wing parties in these countries (Moser, 2005, p. 129). In countries where ethnicity is the basis of a territorial divide, party nationalisation is expected to be rather stable, and determined by the ethnic population structure, provided that no further important territorial divides or external influences exist (such as changes to the electoral system). This is the case for a number of the countries mentioned in this

Explaining the Nationalisation of Party Systems 71 Table 4.1 Most important ethnic minority parties with a territorially concentrated structure of voters (only parties with more than 5% of the national votes; excluding Bosnia and Herzegovina). Where the ethnic affiliation is not obvious from the party name, it is indicated

Country

Party name

Bulgaria

Движение за права и свободи (Movement for Rights and Freedom DPS), Turkish minority

Estonia

Eestimaa Ühendatud Rahvapartei (Estonian United People’s Party) (1999; 1995 in coalition as Meie Kodu on Eestimaa! (Our Home is Estonia!)), Russian minority Par cilvçka tiesîbâm vienotâ Latvijâ (For Human Rights in a United Latvia; PCTVL) (1993–2006; in 1995 as Tautas Saskaņas Partija (National Harmony Party)), Russian minority Saskaņas Centrs (Harmony Centre) (only in 2006 elections), Russian minority Līdztiesība (Equal Rights) (in 1993) Latvijas Sociālistiskā Partija (Latvian Socialist Party) (in 1995), Russian minority Демократска партија на Албанците (Democratic Party of Albanians) (2002–6)

Latvia

Latvia

Latvia

Macedonia

Area of concentration


Several regions with strong Turkish minority (Kurdzali, Razgrad) and voters from abroad Tallinn (capital), Ida–Viru (Eastern Estonia)

0.59 (2005)

Riga (capital) and Latgale (Eastern Latvia)

0.63 (2006)

Riga (capital) and Latgale (Eastern Latvia) Riga (capital) and Latgale (Eastern Latvia)

0.56 (2006)

Western Macedonia, Skopje (capital)

0.40 (2002) 0.43 (2006)

0.54 (1995) 0.50 (1999)

0.38 (1993) 0.36 (1995)

Continued

72


Table 4.1 Continued

Country

Party name

Macedonia

Демократска унија за интеграција (Democratic Union for Integration DUI) (2002; 2006 in coalition with PDP), Albanian minority Партија за демократски просперитет (Party for Democratic Prosperity, PDP) (1994–8; 2002 below 5%), Albanian minority Uniunea Democratica Maghiara din România (Hungarian Democratic Union of Romania; UDMR) (since 1990)

Macedonia

Romania

Slovakia

Strana maďarskej koalície – Magyar Koalíció Pártja (Party of the Hungarian Coalition, SMK–MKP) (result of a merger of several Hungarian minority parties in 1998)

Area of concentration



0.34 (2002) 0.46 (2006)


0.24 (1994) 0.36 (1998)

Transylvania (Northwest; mostly in Harghita, Covasna, Mures, Satu Mare) Trnavský and Nitrianský kraj

0.33 (2004)

0.40 (2006)

listing, namely Bosnia, Bulgaria, Estonia, Latvia, Lithuania, Macedonia and Slovakia. Moldova shows a less steady development, with a major change in the 2001 elections when the national threshold was raised from 4% to 6%, reducing the chances of regional parties being elected (see discussion of the Gagauz case, p.81). 4.4.3 Brussels versus Moscow: Overlapping East–West divides in Ukraine The pattern of territorial divides in Ukraine is more complex, and not easy to relate to the previous distinction of urban–rural and ethnic conflicts. Due probably to the deep territorial division in the party system, and its changing and not very trivial shape, the largest body of literature on territorial voting patterns in Central and Eastern Europe deals with Ukraine.75 From the first multiparty elections in 1990 onwards,


the Ukrainian political landscape has been characterised by a strong division between the West and the East (Birch, 2000a). The division is characterised by at least four correlated differences, related to the ethno-linguistic political–historical identity of the voters and to an economic divide. Ethnic and linguistic differences in Ukraine are not as clear-cut as in other cases. While Russians are an ethnic minority, mostly in the South and the East, Russian is frequently used among ethnic Ukrainians, and some of them declare Russian as their mother tongue. The Western region around is the stronghold of pro-independence Ukrainian nationalists (namely Rukh) and parties in favour of integration in the European Union and independence from Moscow. Lviv, an agricultural region, is mostly settled by ethnic Ukrainians, the Ukrainian language is in use, and the Greek Orthodox and the Roman Catholic churches are strong (different from the rest of Ukraine, where Russian Orthodox believers dominate). Western Ukraine has experienced Austro-Hungarian, Polish, Romanian, and/or Czechoslovak rule, and has a much longer history of enfranchisement than other Ukrainian regions that used to be part of the Russian empire (Birch, 1995, p. 1,146). Eastern Ukraine appears as the pure opposite: it belonged to the Russian empire, has a high share of ethnic Russians, and mainly uses the Russian language. The region is the stronghold of the communists and of pro-Moscow parties. The political affiliations of the West and the East are clear, but the orientation of other regions, namely the South and the Left bank, is more subtly nuanced and varies from election to election and regarding the question posed (Birch, 1995, 2000a, pp. 27–8). The voting behaviour is similar to the Russophile East, but there are minor differences in the intra-bloc voting behaviour and there is some regional volatility from election to election. Nationalisation of the Ukrainian party system appears even lower in elections where single-seat districts were in use (or partly in use in mixed systems), because in many districts local non-partisan candidates run instead of political parties. The weakness of political parties is reflected by the legislative behaviour of MPs, for whom powerful economic or political elites have more importance than parties (Thames, 2005b). In sum, I have found several territorially based political conflicts that affect the nationalisation of party systems in single countries. To some extent, voters’ orientation in the post-communist democracies follows social divisions (Evans, 2006; Whitefield, 2002), many of which have a strong territorial component. The following section will establish whether these territorial divides and/or institutional aspects help to

74


explain the differing levels of party nationalisation across countries in Central and Eastern Europe. 4.4.4 The centralisation approach and the direction of causality Lacking time-series data for decentralisation, my analysis of the government centralisation hypothesis is limited to a cross-sectional comparison of twenty countries for the period around 2000. (Due to the limited number of observations, I rely on a bivariate analysis.) Figure 4.2 reports how party nationalisation is related to the centralisation of government expenses (PR elections considered). Most countries in the region are heavily centralised. There is a positive trend, in the sense of the hypothesis: decentralised countries more often have regional party systems, centralised countries nationalised ones. The correlation relies, however, solely on the case of Bosnia and Herzegovina, with extreme

1 0.9

RU99

Party nationalisation (PR tier)

0.8 0.7

MK02

0.6 0.5 BIH00 0.4 0.3 0.2 0.1 0 0

0.2

0.4

0.6

0.8

1

Budget centralisation Figure 4.2 Budget centralisation and party nationalisation (PR tier), 1998–2001 Notes: O Bosnia and Herzegovina, X other countries For countries with several elections in the period, the closest elections to 2000 were taken (Bosnia: 2000, Moldova: 2001). In the case of Macedonia and Ukraine, the 2002 elections were included, because no sub-national data is available for the PR tier in the 1998 elections (in both countries under a mixed electoral system). Seventeen cases with available data.


decentralisation and very low party nationalisation. After the exclusion of Bosnia, the correlation vanishes.76 Bosnia is thus the possible key case for a relationship of weak party nationalisation and extreme decentralisation, and ‘central government institutions exist largely on paper’ (International Crisis Group, 1999, p. 3). However, in this case, the causal sequence of the regional disintegration of the party system and decentralisation (if there is such an immediate link at all) goes the other way (Bochsler, 2006). In a nutshell, in the initial democratising elections in 1990, when Bosnia was still centralised and a Republic of former Yugoslavia, a new party system emerged, consisting mainly of three ethnically oriented parties, representing the main ethnic communities. The ethnic separation of the political parties has not changed to date. However, in the following five years, Bosnia experienced war, genocide and ethnic cleansing, which were only stopped by the Dayton peace agreement, which provided an extremely decentralised structure of the state. The historical sequence reveals that the ethnically based and disintegrated party system of 1990, already split along the same ethnic lines as after the war, was the basis for the extreme decentralisation of 1995. It might be too early to assess whether low party nationalisation has similar consequences all across the region, although there are other countries with weak nationalised party systems that have recently discussed or even decided steps to decentralise the administration.77 Following the ‘Ohrid framework agreement’ of 2001 in Macedonia, municipalities gained substantial autonomy. Municipal borders were redrawn, in order to allow the self-government of the Albanian-speaking minority, and to settle the ethnic conflict. Hence, in this case too, decentralisation was anticipated by the split of the party system into ethno-regional parties, more than a decade earlier. In brief, in Central and Eastern Europe, the correlation between government centralisation and party nationalisation relies on a single case, for which the often assumed causality that centralisation of the government leads to party nationalisation can be dismissed. More recent development in Macedonia reinforces the view that, in Central and Eastern Europe, changes in decentralisation are anticipated by the emerging of regionally based parties. 4.4.5

The ethnic divide and the electoral system model

Instead of the decentralisation approach, I explain party nationalisation through social divides. Following a broad number of studies, I operationalise this approach relying on ethnic diversity (Mozaffar

76


et al., 2003; Ordeshook and Shvetsova, 1994, to name a few). I expect that stronger territorial ethnic divisions eC will be related to lower party system nationalisation n. n a b1 eC « For countries without any concentrated ethnic groups (eC 0), party nationalisation might be high, but still not perfect. Other territorial divisions might impede party nationalisation from being perfect, and, to some extent, random variation of district results contributes to imperfect nationalisation. The intercept a, estimating party nationalisation in the absence of any territorial ethnic divides, will thus be slightly lower than 1, while the slope b1 is expected to be negative. Since both eC and n are measured on a scale from 0 to 1, and assuming a linear relationship, the absolute value of the slope needs to be smaller than the intercept, |b1| < a. Otherwise, our model would predict that, in countries with many territorial ethnic divides, party nationalisation is negative. The slope b1 depends on the strength of the translation of territorial ethnic divisions into the party system, and, if parties are a mirror of the ethnic structure of the country, |b1| will be close to a. In Central and Eastern Europe, where ethnic divides are fairly important for party competition, |b1| will be reasonably close to a. High national legal thresholds might have a moderating impact on the translation of territorial ethnic divisions into party nationalisation, while, in the absence of national legal thresholds, territorial ethnic divisions have a non-moderated impact on party nationalisation.78 The two super-presidential systems in the region, Russia and Ukraine, are identified through a dummy variable.79 In both cases, presidential and parliamentary elections were held non-concurrent for the elections under study,80 and presidents were absent from party campaigns for legislative elections (Moser, 2001c, p. 102). The presidential regime in Russia is considered to have subverted party formation (Hale, 2006, pp. 205–9). An interaction term pres SSD identifies elections in superpresidential regimes and in single-seat districts. Further control variables are included for single-seat district elections (SSD), for elections in the single-seat district tier of mixed compensatory electoral systems (SSD comp) and for the inverted ordinal number of elections (1/T). The model is tested in an OLS regression, with a first specifications including only elections under PR, and overall eighty-one cases,81 and four further models extending the view on elections in single-seat districts, or mixed electoral systems, counting up to 103 cases (Table 4.2). All five models reveal a substantial impact of concentrated ethnic

Table 4.2 OLS regression models to estimate party system nationalisation, robust standard errors, cases clustered by country Only PR elections Regression model Constant Territorial ethnic divisions (eC) National legal threshold (t) Interaction ethnic divisions/ threshold (eC t) Number of election (inverted) (1/T) Single–seat district elections (SSD) Single–seat district tier in compensatory systems (SSD comp) Interaction ethnic divisions/ SSD in non–compensatory systems (eC SSD (1comp)) Interaction number of elections (inverted)/SSD in non–compensatory systems (SSD (1comp)/T) Super–presidential systems (pres)

All elections (1)

B

S.E.

VIF

0.90 0.69**

0.03 0.07

0.33 9.67** 0.04

All elections (2)

B

S.E.

2.15

0.88 0.66**

0.05 0.08

2.49

0.54 2.71

2.99 2.78

0.10 9.22**

0.57 2.79

0.03

1.14

0.01

S.E.

VIF

0.92 0.71**

0.03 0.06

2.21

4.84 3.36

0.28 9.28**

0.52 2.59

4.69 3.23

0.07

1.13

0.07*

0.03

1.28

0.35*

0.17

6.15

0.05(*)

0.03

8.79

0.36*

0.16

2.40

0.05

0.03

4.15

0.94(*)

0.50

3.60

0.07

0.06

5.19

0.05

0.07

1.77

VIF

B

Continued

Table 4.2 Continued Only PR elections Regression model Interaction super– presidential/SSD in non– compensatory systems (pres SSD (1comp)) N Countries (cluster) R2

B

All elections (1)

S.E.

VIF

81 20

2.27

0.768

B

S.E.

VIF

103 20

3.42

0.600

All elections (3)

Constant Territorial ethnic divisions (eC) National legal threshold (t) Interaction ethnic divisions/ threshold (eC t) Number of election (inverted) (1/T)

All elections (4)

B

S.E.

VIF

0.89 0.64**

0.04 0.10

0.05 8.59** 0.05

All elections (2)

B

S.E.

2.20

0.92 0.72**

0.03 0.05

2.23

0.52

4.60

0.33

0.51

4.71

2.56

3.17

2.52

3.22

0.04

1.15

0.03

1.16

9.52** 0.07*

VIF

B

S.E.

VIF

0.45**

0.07

2.32

103 20

0.895

Single–seat district elections (SSD)

–0.01

0.04

3.37

–0.01

0.02

3.37

Single–seat district tier in compensatory systems (SSD comp) Interaction ethnic divisions/SSD in non– compensatory systems (eC SSD (1comp)) Interaction plurality/ SSD in non– compensatory systems (SSD (1comp) plurality) Super–presidential systems (pres) Interaction super– presidential/SSD in non– compensatory systems (pres SSD (1comp)) N Countries (cluster) R2

0.02

0.04

2.12

0.00

0.03

2.13

–

–

–

–

–

–

–0.34*

0.13

2.09

–0.05*

0.02

3.56

–

–

–

–0.05

0.07

1.77

–

–

–

–0.43**

0.06

3.34

103 20 0.543

– – –

– – –

103 20 0.896

– – –

2.83 – –

Note: **significant at p < 0.01; *significant at p < 0.05; (*) significant at p < 0.1.

80


groups, moderated by national legal thresholds. While in countries with no legal threshold, or a moderate one, a clear negative relationship of territorial ethnic divisions and party nationalisation emerges, this changes under high national legal thresholds – in the countries under study the highest threshold amounts to 6%. There, territorial ethnic divisions do not have any impact on party nationalisation. The effect is statistically significant in all models. In the absence of territorially based ethnic divides, party nationalisation is very high, and it slightly rises after increasing democratic experience, almost to 0.9. Where territorial ethnic divisions exist, party system nationalisation is much lower. The absolute value of the coefficient b1 ≈ −0.7 is almost as high as the negative value of a, from which it follows that, in Central and Eastern Europe, territorial ethnic divisions are strongly reflected in low party nationalisation. In four models, I include all elections, both in PR and in single-seat districts. For double-ballot elections under mixed electoral systems, both the PR and the single-seat tier are included separately, since, both theoretically and empirically, substantial differences between both tiers might be expected and are observed (Figure 4.4). The most pronounced differences between both tiers can be observed in the mixed electoral systems of Russia and Ukraine (and to a lower extent in Lithuania), where national parties could hardly establish successful candidates in 1


0.9 0.8 0.7 0.6 0.5 0.4 0.3

0% 2% 4% 6%

0.2 0.1

0.6

0.56

0.52

0.48

0.44

0.4

0.36

0.32

0.28

0.24

0.2

0.16

0.12

0.08

0.04

0

0

Territorial ethnic divides ec Figure 4.3 Party nationalisation (y-axis) in PR elections, according to the OLS regression model. Estimated for varying levels of territorial ethnic fractionalisation (x-axis) and varying levels of national legal thresholds (lines).


single-seat districts, and in many districts did not even compete (Birch, 1998, p. 98; Moser, 2001c; White et al., 1995, p. 199).82 The regression results reveal an impact that can be observed in a few examples in the region that underline the importance of national legal thresholds (see also Figure 4.3). ●

●

●

In Serbia, ethnic minorities such as Hungarians (4% of the population), Bosniaks (2%), or Albanians (1%) are well organised and represented with their own political parties in local and regional politics in the regions where they are strong (Bašic´ and Crnjanski, 2006). However, in the 2000 and 2003 elections in Serbia, a threshold of 5% hindered parties of ethnic minorities from winning mandates in parliament. Even with a unified list of nine ethnic minority parties and small local parties, they failed to get 5% of the vote in 2003. Four years later, the threshold was lifted, and six ethnic parties ran with their own lists, getting support mainly in the areas with a high concentration of the relative ethnic minorities (Bochsler, 2008a). Since 1994, Moldova has applied a national legal threshold, fixed first at 4% and later at 6% of the national vote. The Gagauz (4.4%) and the Bulgarian minorities (1.9%) that are concentrated in Southern Moldova cannot (or can only with great difficulty) pass these national thresholds, which might hinder the creation of regionally based parties (OSCE ODIHR 2001b). Latvia’s 5% threshold has caused the failure of the ‘Latgale Light’ party (‘Latgales Gaisma’, in 2002), a party with a highly regional voter structure and a very substantial share of the votes in Eastern Latvia (where many ethnic minorities live) and particularly in the largely Russian-populated town of Daugavpils.83 The party’s vote share would have been worth two out of seventeen seats in the Latgale region, but the party did not obtain enough votes nationally to pass the threshold.

In many other elections with national legal thresholds, there are no similar experiences of ethno-regional or regional parties. Many locally strong parties might anticipate the difficulty of the national threshold and as a result might not compete, or might join a larger party that operates nationwide or merge with a regional party in another region, all of which increase the level of party nationalisation. Analogous to the psychological effect of electoral systems on the number of parties, there might be such a psychological effect of a nationwide legal threshold on regional parties.

82


Party nationalisation (SSD)

1

0.8

0.6

0.4

0.2

0 0

0.2

0.4 0.6 0.8 Party nationalisation (PR)

1

Figure 4.4 Party nationalisation in the PR tier and in the single-seat district tier in mixed electoral systems. The line shows the equality line, where party nationalisation is equal in both tiers.

Other variables, such as the national legal threshold (as a non- conjunctional term) or the democratic experience, remain non-significant. In the four models covering all elections, additional variables and interaction terms test several competing explanations for the lower party nationalisation in the single-seat district tier of some mixed electoral systems. The variables control for differences between mixed compensatory and non-compensatory systems (model 1–4), for a special effect of ethnic divisions in mixed non-compensatory systems (model 1), for a slower effect of political learning in mixed non-compensatory systems (model 2), for differences between the majority and the plurality vote (models 3, 4), and for the impact of super-presidentialism, combined with non-compensatory electoral systems (models 2, 4) (see Bochsler, 2008b, pp. 81-8 for details). Only the hypothesis about superpresidentialism can satisfactorily explain the low party nationalisation in the single-seat district tier in elections in Russia and in Ukraine.

4.4.6 Party nationalisation in single-seat district systems: Russia and Ukraine Super-presidential regimes, jointly with elections in single-seat districts, lead to much lower nationalisation of party systems. While, both under


PR and in (semi-)parliamentary regimes, there are institutional reasons for candidates and voters to align with national parties,84 no similar institutional incentives exist in single-seat districts or in super-presidential systems (where presidential and parliamentary elections are not simultaneous). The logic of the electoral system does not encourage candidates to link over district borders, nor do the parliamentary elections have the character of nationwide elections of the prime minister, so that party nationalisation remains low. This mainly explains the peculiarities of single-seat district elections in Russia and in Ukraine. Other variables are not statistically significant at the conventional levels. Besides the role of super-presidentialism, political legacies of the post-Soviet space and socio-economic aspects have also been discussed to explain the low party institutionalisation in Russia and Ukraine. 85 Hale (2006, pp. 4–5) lists ten competing ones for the Russian case. However, many of them have not been tested systematically in crosscountry analyses. While most of these explanations are plausible for the explanation of the outcomes that are observed in Russia or Ukraine, they would similarly apply to other post- communist countries, or to all electoral systems, there is no explanation why they should only apply to the single-seat district part of the Russian and Ukrainian electoral system.86

4.5

Conclusions: Determinants of party nationalisation

The comparative analysis of party nationalisation has shown that ethnic structure is an important explanatory variable for party nationalisation in post-communist Europe. Unlike other social divides, in many countries ethnic identities are still present after the communist period, and now appear as crucial for politics and party systems. The impact of territorial ethnic divides on party nationalisation is not a one-way process, however. In the long run, what people accept as their identity depends on the political situation and on the influence of politics on societies. Central and Eastern Europe has not been an exception in this regard; in a continuity of languages and regional identities, common aspects have been stressed when it was important for the creation of a common national identity, and regional differences when this appeared opportune. Distinct ethnic identities stabilise political differences across population groups. The electoral system matters when it comes to the translation of regional differences into political parties. National legal thresholds manage in many cases to exclude regional parties from competition,

84


thus increasing party nationalisation and hindering territorial divides from being manifested. Not only can this impact can be shown in countries that changed their threshold over time (such as Moldova), but it furthermore appears from a regression analysis as well. My study is the first to show the impact of territorial ethnic divisions on party nationalisation in a quantitative analysis.87 It comes to different conclusions, however, from studies on other regions of the world with regard to the relationship between party nationalisation and state centralisation. Rather than party nationalisation being a product of centralised states, in key countries the impact appears to work the other way round. Regional political parties, which gain their support on ethnic grounds, pressure for the decentralisation of the state. Low party nationalisation appears to affect decentralisation more strongly than vice versa. It would be worth studying the reasons for these different findings, namely whether there are reasons for the exceptionality of Central and Eastern European countries, or whether other factors might account for differences.88 On the grounds of the ethnic structure and the level of national legal thresholds, party nationalisation can be predicted fairly well. The picture is less straightforward when it comes to elections under single-seat district rules. While many elections do not differ to a major extent from the results found for PR elections, two of the cases that applied mixed electoral systems do show very diverging levels of party nationalisation in both parts of the system. In Russia and Ukraine, party nationalisation is similar to other countries in the PR part of the elections, but substantially lower in the single-seat district tier. In other countries that apply mixed electoral systems, differences are less pronounced. The analysis of party nationalisation in single-seat district systems is thus influenced by the two outliers, Russia and Ukraine, and relies on some twenty elections altogether. Often, the peculiarity of electoral outcomes in the countries of the former Soviet space is related to a lack of party institutionalisation. This is, however, a fairly wide and fuzzy term, which includes many different dimensions, and is defined differently by different scholars (cf. Thames, 2007), sometimes even including party nationalisation (Moser, 1999a, p. 363), which in Russia and Ukraine is linked to further aspects of party institutionalisation. Both in Russia and in Ukraine, district candidates often do not even join national political parties, and instead compete as non-partisan candidates, so that party nationalisation is much lower than in the PR part of the elections. In a comparative analytic perspective, it would thus be interesting to specify more


precisely the reasons that explain why national parties fail to control many of the single-seat district elections within the mixed electoral systems in Russia and Ukraine. Many explanations have been provided for this phenomenon. Because many factors might empirically explain a Russian and Ukrainian peculiarity, methodologically it is not possible to separate one from the other in a cross-country study, even more because a cumulative and/or conjunctional effect of several factors might apply. Under the given political and institutional circumstances in Russia and in Ukraine, the net costs of forming or joining a national political party for single-seat elections might be higher than the benefits that candidates can expect from aligning with a national party. Supposedly, and supported by my cross-country analysis, the super-presidential system with non-concurrent elections and non-partisan candidates for the presidency might have a major impact on the low party nationalisation: the presidency has established itself as a non-partisan position, which deprives national parties of their function as the electoral college of the country’s executive. When parties no longer perform the important function of being responsible for the formation of national governments, there is even less need for strong national parties. An extension of this study to other regions might deepen the explanations further.

5 How Party Systems Develop in Mixed Electoral Systems

5.1

Introduction

German-style electoral systems have become a bestseller in the charts of European and worldwide electoral reforms of the 1990s and the early twenty-first century. Combining PR with elections by majoritarian rules, usually in SSD, the German electoral system unifies two different worlds of electoral system logics. Since Central and Eastern Europe is a region where mixed electoral systems are frequently applied, it is crucial for this study to look at their functioning. Under catchphrases such as ‘the best of both worlds’ (Shugart and Wattenberg, 2001b, pp. 592–6), mixed electoral systems have been discussed as combining the advantages of two opposing electoral system logics.89 Mixed electoral systems, as defined here, have several overlapping sets of electoral districts for the same elected body, in which each voter gives his or her vote according to different electoral formulae. One part of the seats, the ‘PR tier’, is elected by PR (often with national districts), whereas the ‘district tier’ consists of mandates elected in single-seat districts (see Section 5.2 for details). This combination of plurality/majority vote and PR is seen to have a moderating impact in a number of dimensions. The ‘best of both worlds’ school praises mixed electoral systems as striking a happy medium between fractionalisation and concentration of the party system and providing a compromise between both extremes with regards to proportionality (Grotz, 2000, pp. 707–8; Kostadinova, 2002; Shugart and Wattenberg, 2001b, p. 592). Similar advantages are praised for the combination of a party element and a personality element (Shugart and Wattenberg, 2001a, p. 582), local and national representation (Shugart, 2001; Stratmann and Baur, 2002; Thames, 2005a; see Crisp, 2007 for a critical view), 86

How Party Systems Develop in Mixed Electoral Systems

87

and, finally, they can allow compromises between different actors’ preferences in constitution-building processes (Schiemann, 2001 for an example). Recently, after they were discussed and introduced in many countries from the 1990s onwards, scholars have increasingly been interested in how mixed systems affect party strategies and voting behaviour (Cox and Schoppa, 2002; Ferrara et al., 2005; Gschwend, 2007; Herron and Nishikawa, 2001; Moser and Scheiner, 2004), legislative behaviour (Ferrara, 2004b; Smith and Remington, 2001; Thames, 2005a) or representation of women and minorities (Kostadinova, 2007; Moser, 2001a). Few previous studies have focussed on the fractionalisation of the party systems under mixed electoral systems in systematic crosscountry studies. The social laboratory school argues that each applied formula has its effect independent of others (see below). In contrast, research on contamination has found that voting behaviour in each tier is affected by the other (Ferrara et al., 2005; Herron and Nishikawa, 2001). Accordingly, the vote distribution in both tiers is interdependent, so that it cannot be directly compared to simple electoral systems with no contamination effect. So far, no model has been established that would make it possible to predict the number of parties in mixed electoral systems. After a series of elections under mixed electoral systems, some of the countries in Central and Eastern Europe have developed rather stable and small party systems (Croatia, Hungary and Albania, apart from the 2005 elections). In these countries, if no dominant party emerged in an election, disproportionality became rather low, similar to single-seat district systems with a stable two-party system. In Russia and Lithuania, however, the level of party system fragmentation has remained high even after repeated elections under a mixed electoral system. These differences persist after a number of elections, so that they do not seem to be simply random outliers occurring in the first few elections (Figure 5.1). Not all mixed electoral systems in the region are equal. Most of them have a roughly even share of seats allocated by PR and in single-seat districts, but they differ in the linkages between both tiers. In many cases, there is no linkage at all, but both Albania and Hungary employ (in certain elections) mixed compensatory systems with a partial linkage mechanism that should make the seat allocation more proportional. Ironically, however, in all cases where the element of proportionality is stronger, either due to such a (partial) linkage mechanism, because the PR mandates clearly outnumber the single-seat districts (Croatia in


Number of effective parties

88

10

7

Albania Croatia Lithuania Russia Hungary Ukraine

4

1 Figure 5.1 Development of the effective number of parties in parliament under mixed electoral systems Note: x-axis – Number of election under a mixed system (Only countries with at least two elections under a mixed system).

1995), the party systems were rather small, quite contrary to common expectations, while in Lithuania, in the absence of such PR dominance, not only is the party system large, but its fractionalisation has even increased over time. It might thus look more appealing to explain the differences based on historical legacies, such as the post-Soviet context in Lithuania, Russia and Ukraine, which hampers party institutionalisation and leads to fragmentation of party systems under mixed electoral systems (Ferrara et al., 2005, p. 47; Moser, 1999a, 2001c). However, what might be understood as institutionalisation is different in both cases: Lithuania is characterised by high volatility and strategic coordination, while Russia used to have an atomised party system. A legacy-based explanation of this kind might possibly explain the different starting points of party system development. However, this does not explain why the party systems have since developed in very different directions. The party system development in Croatia, Hungary and Lithuania was similar in the beginning, but the countries had fairly different outcomes later on. While elsewhere I have applied a model that explains why outcomes under mixed electoral systems vary widely, much more than under PR (Bochsler, 2009a), in this chapter I describe typical steps of the party system development under mixed electoral systems in Central and Eastern Europe, show how these steps are related to each other, and describe different types of party systems that emerge under mixed electoral systems. Previous research on party sizes under simple electoral systems is not directly applicable in efforts to explain the functioning of mixed


89

electoral systems. On the one hand, this chapter will show how the dynamic of mixed electoral systems can be compared with simple electoral systems. On the other hand, it shows how institutional differences within the family of mixed electoral systems – namely the distinction between mixed compensatory (with a possibly proportional seat allocation) systems and mixed non-compensatory systems – is crucial when they are compared with simple electoral systems. This chapter proceeds as follows. The next section defines mixed electoral systems and their elements. Against the view that sees mixed compensatory systems as PR-like, the third section shows that their proportional character relies on the share of compensation mandates that are available in a mixed compensatory system. Sections 5.4 and 5.5 discuss the effect of non-compensatory and partially compensatory electoral systems, and the range of possible outcomes and the conditions under which they might occur. In Sections 5.6 and 5.7, this is applied to the empirical reality in Central and Eastern Europe and compared with Japan. Special attention is finally given to the Albanian 2005 elections, which appear as an outlier throughout my study. Strategic voting allowed the largest party to gain more seats than intended by the electoral rules, creating an artificial parliamentary majority at the expense of the small parties.

5.2 Mixed electoral systems: Terminology and application The mixed systems in Central and Eastern Europe vary in the configuration of their institutional elements. Due to a lack of commonly accepted scholarly terminology for mixed electoral systems, it is essential to introduce a terminology and to clearly define mixed electoral systems and their elements, which can be divided into several subtypes, and to relate those definitions to the numerous definitions and terms already in use. Readers who are confident of understanding my terminology might either read on and be pleased by the lack of capacity of scholars to harmonise their notations, or skip this section. Confusion might arise when looking at the definitions of mixed systems. Often, they are defined as systems that include several electoral formulae (Ferrara et al., 2005, p. 17; Massicotte and Blais, 1999). This even includes cases where different formulae are used in different parts of the country or for different parts of the electorate. I chose a narrower definition, including only systems for which we expect a contamination effect on voting behaviour, because every voter votes according to several formulae. Hence, I only define an electoral system as mixed

Subtype of the system

Tiers of allocated mandates

Table 5.1

Terminologies regarding mixed electoral systems My notation

Definition

Other notations (selection)

Mixed electoral system

Two or more tiers, elected with different electoral systems; every voter votes according to two or more rules. Part of the mandates that is allocated by plurality or majority vote, mostly in singleseat districts. Part that is allocated by a proportional rule. The proportional tier, in cases where the system is compensatory (cf. ‘mixed compensatory systems’). Mandates in both tiers are allocated independently.

Mixed-member systems (Shugart and Wattenberg, 2001b), composite systems (Taagepera, 2007a)

Plurality/majority tier

Proportional tier Compensatory tier

Mixed noncompensatory system (Herron and Nishikawa, 2001, p. 65)

Nominal tier (D’Alimonte, 2001); SMD tier.

Party tier; list tier. –

Mixed-member plurality, mixed-member majoritarian system (Ferrara and Herron, 2005; Shugart and Wattenberg, 2001b); unlinked system (Moser and Scheiner, 2004); parallel (mixed) system (Reynolds et al., 2005; Taagepera, 2007a); additional member system (Budge et al., 1997, pp. 235–6); independent/superposition system (Massicotte and Blais, 1999, pp. 349–52); segmented electoral system (Nohlen, 2004, pp. 188–9); hybrid system (Christensen, 1996).

Subtypes of compensatory systems

Mixed compensatory system (Herron and Nishikawa, 2001, p. 65)

Mandates in the compensatory (proportional) tier are allocated dependent on the seats a party got in the plurality/majority tier; in such a way as to produce overall proportionality.

Mixed-member proportional system (Ferrara and Herron, 2005; Reynolds et al., 2005; Shugart and Wattenberg, 2001b); linked system/seatlinkage (Moser and Scheiner, 2004); correction systems (Massicotte and Blais, 1999, pp. 353–6); personalised PR (Kreuzer, 2004, p. 223; Moser, 1995, p. 383; Nohlen, 2004, pp. 188–90).

Positive/negative vote transfer systems

Special form of compensation: votes cast for non-winning candidates in the plurality/ majority tier are transferred to the compensatory tier (Hungary: positive vote transfer; Italy: negative vote transfer). Do not have sufficient compensation seats to allow proportional outcomes.

Partially compensatory systems with negative/ positive vote transfers (Ferrara and Herron, 2005); compensatory systems (Nohlen, 2004).

Mixed semicompensatory systems

92


where the same voters can vote in several tiers according to several formulae for the election of the same electoral body, either with one vote that counts for all formulae, or with several votes.90 Usually, such systems combine PR and single-seat district seat elections. The most important distinction between mixed electoral systems is that between non-compensatory systems (sometimes unlinked, segmented, superposition or parallel systems) and compensatory systems (or correction or linked systems) (Table 5.1). Non-compensatory systems allocate seats in non-related tiers, according to different formulae that are not linked to each other.91 In mixed compensatory systems, the proportional or compensatory mandates are used to correct the disproportionalities resulting from the seat allocation in single-seat districts. The overall seat distribution should be as close to proportionality as possible. The aim is that every party gets (under certain conditions; see Sections 5.3 and 5.8) the number of seats it is entitled to based on its overall vote share. The number of compensation mandates won is calculated as the difference between this overall number of seats and the number of district mandates for a certain party. Such a compensatory system, with the described classical compensation mechanism, was used in Albania, while Hungary applies a related, Table 5.2 Mixed electoral systems in Central and Eastern Europe, 1990–2007; types and application Non-compensatory systems

Compensatory systems

Albania 1996, 1997 Bulgaria 1990 Croatia 1992, 1995 Macedonia 1998 Russia 1993, 1995, 1999, 2003 Ukraine 1998, 2002 (Federal Rep. of Yugoslavia May 1992a) fully compensatory semi-compensatoryb partial positive vote transfer

Albania 1992, 2001, 2005 Hungary 1990, 1994, 1998, 2002, 2006

Notes: a The Yugoslav elections of 1992 were not free enough to be included in this study. b For the distinction between fully compensatory and semi-compensatory systems, see the following section.


93

special case, which uses a slightly different allocation mechanism termed the positive vote transfer formula. The most common type of mixed electoral system in post-communist democracies in Europe is non-compensatory systems (see Table 5.2). For my purposes, the distinction between compensatory and noncompensatory systems is crucial, because non-compensatory systems might provide a seat bonus for the largest parties when the results of the single-seat district tier are not compensated for, similarly to plurality/majority vote systems. In compensatory systems with fully proportional results, one might expect mechanisms affecting the party system that are similar to those in a PR system.

5.3 How much compensation in mixed compensatory electoral systems? The idea of compensatory systems is to arrive at an overall number of parliamentary seats that for a given party, as far as possible, is proportional to its vote share. This has led some scholars to speak of systems of ‘personalised PR’ (Table 5.1). What follows from this term is the widespread view that single-seat districts do not have any important impact on the overall proportional character of the compensatory systems (among many others, Herron and Nishikawa, 2001, p. 65; Nohlen, 2004, p. 190; Schoen, 1999, p. 475; Shugart and Wattenberg, 2001a, p. 584; Tiefenbach, 2006, p. 124). From this viewpoint, the question of whether compensatory systems can generally be considered as equally proportional to PR systems arises. More specifically, the question of whether this is the case for the compensatory systems in Hungary and Albania is also raised. If in a mixed compensatory system almost all of the seats were allocated by plurality or majority rule in single-seat constituencies, and only one or very few mandates as compensatory, these would almost certainly not be enough to compensate for the disproportionalities created in the plurality or majority tier. Rather, a system with a very low share of compensatory mandates would have a very similar effect to a pure plurality or majority vote system. How many compensation seats are needed in order to achieve an effect equal to a PR system? ‘Roughly,’ Moser and Scheiner (2004, p. 580) guess, with 50% of compensatory seats ‘the result is a distribution of seats almost fully controlled by the PR vote,’ but 25% would not be enough. This estimation, along with others (Cox and Schoppa, 2002, pp. 1029–30), is based on the observed degree of disproportionality in

94


elections in Hungary, Italy and Germany. Behnke (2003) and Behnke et al. (2003) have shown that in the German context 50% of compensation seats usually allow a proportional seat allocation, but not always. However, observations of disproportionality disregard the possible psychological effects of electoral systems (the strategic behaviour of voters and parties, anticipating electoral rules) and the possible preventive concentration of the party system to a size under which the results will be proportional. Indisputably, not every mixed compensatory system can provide fully proportional outcomes. The share of compensatory mandates c might help with the classification. If the share of compensatory mandates is low, they might not be enough to compensate completely for disproportionalities arising in the district tier, so that the systems are not fully proportional or cannot be taken as equal to PR systems (cf. Figure 5.2, left part). Despite allowing some compensation for under-represented parties, they remain semi-compensatory. Systems with a sufficiently high share of compensation seats, however, are fully compensatory: their outcome is the same as under PR; seat shares and vote shares are fully proportional (cf. Figure 5.2, right part). Both subtypes of compensatory electoral systems are separated by the point c comp, where the share of compensatory mandates c is just large enough to provide full compensation. Since there are many different PR systems with varying district magnitude or thresholds, the relevant basis for comparison is the rules applied in the compensatory tier of the mixed system. The position where a system is just compensatory (c comp) might be not universal, as it may depend on the nationalisation of the party system, the size of the parliament, the PR formula applied in the PR tier, and most of all on the vote distribution on political parties. Political actors do anticipate the impact of the electoral system (Duverger’s psychological Pure plurality/ majority vote

Semicompensatory systems

Just compensatory (c = comp)

Fully compensatory systems

Pure PR

share of compensatory mandates (0 < c < 1)

100%

comp 0%

Outcome: semi-proportional

Outcome: proportional

Figure 5.2 Semi- and fully compensatory systems and break-even point (c comp), where systems are just compensatory


95

effect), so that in certain occurrences of semi-compensatory systems only those parties might compete that get an (almost) proportional seat share. Henceforth, empirically observed measures of disproportionality might not be very helpful for a conceptual distinction of compensatory and semi-compensatory systems. Instead, a distinction should be based on the theoretical, whether an electoral system would allow as many parties to be represented in parliament as a proportional electoral system, without producing higher degrees of disproportionality. Taagepera and Ensch (2006) have established accurate estimations of the size of the largest party in a given simple electoral system,92 which are very useful for our purpose. Building on their work, I can estimate how many mandates the largest party (which is usually the most overrepresented party) wins in the plurality/majority tier, and how many it would win if all the seats of the parliament were to be accorded by PR. This allows me to establish a theoretically derived formula that indicates at what level a system will be just compensatory, so that a further increase of the compensatory tier would neither decrease disproportionality, nor allow an increase in the number of parties in parliament (Bochsler, 2007a).93 For a given number of seats in parliament (S) and a given legal threshold applied in the PR tier (t), and a parameter which allows comparison of legal threshold systems with district-based systems (q; for Central and Eastern Europe, estimated as q ≈ 1), the share of compensatory mandates where a system is just compensatory (comp) is given as follows: 1 q  comp = 1 − S 7 ⋅  − 1  t  

(

0.43 ⋅ −6

7

) [5.1, taken from Bochsler (2007a)]

According to this formula, both an increase in the number of seats in parliament and an increase in the legal threshold of the PR tier decrease the number of necessary mandates for full compensation. 5.3.1

Albania and Hungary as semi-compensatory systems

This model allows us to classify the two mixed electoral systems with compensatory elements in post-communist democracies in Europe, those of Albania and Hungary. The Albanian parliament (140 mandates, national 2.5% threshold in 2001 and 2005) would require 47.5% compensatory seats (sixty-seven in absolute numbers) in order to be comparable to a PR system, but there are only forty. In 1992, when the threshold was at 4%, fifty-three compensatory seats (37.2%) instead

96


of the forty would have been needed. This is why I conclude that the Albanian electoral system is semi-compensatory, and the 2.5% threshold is not the only barrier in the Albanian system. If actors only behaved according to the 2.5% threshold, too few compensation mandates would be available, and the results would become disproportional. Nevertheless, the Albanian system was characterised by low disproportionality until 2001, because there were only two or three effective parties, substantially fewer than might be expected for a pure PR system with a 2.5% threshold. If fewer parties compete, a small number of compensation mandates can still allow proportional outcomes. The 2005 elections, with much higher party system fragmentation and huge disproportionality, are discussed in Section 5.8. Hungary’s three-tier electoral system (386 seats; 5% threshold since 1994, 4% in 1990) is semi-compensatory too. With its three tiers, it is more complex, and its classification is disputed (Bochsler, 2009b for an overview): 152 mandates are allocated in non-compensatory PR districts, 176 mandates in single-seat districts, and fifty-eight as compensatory seats by positive vote transfer. To these fifty-eight compensation seats, a usually small number of remaining seats from the PR tier are added. The 152 PR mandates do not influence the compensation character of the Hungarian system, and for the other 234 seats (single-seat districts plus compensatory seats) with a 5% threshold, sixty-two mandates would be required for a fully compensatory system; slightly more than there are. Even more important, however, is that the mechanism used in Hungary to compensate for disproportionalities, the positive vote transfer system, is not as efficient in providing proportionality as common mechanisms, on which my estimation is based (Bochsler, 2009b). Hence, Hungary can be classified as a semi-proportional system too.

5.4 Economics of votes: The run for better conversion rates Mixed non-compensatory systems and semi-compensatory systems, as defined in the previous section, resemble in some respects a milder, more moderate form of plurality or majority vote systems. Lacking a mechanism that would provide full proportionality, they offer large parties an advantage over small parties in the form of a seat bonus. When it comes to the question of how many parties can win access to parliament, these systems rather resemble PR systems. Parties need only pass the threshold applied in the PR tier (or, alternatively, have a stronghold in one of the districts) in order to get represented in parliament. In contrast to PR systems, small parties, even if elected, risk remaining under-represented,


97

whereas large parties might benefit from seat bonuses in the single-seat district tier, and artificial majorities are possible. This creates contradictory incentives, which make it difficult to apply common electoral system models (assuming that there is a strategic adaptation of electoral rules on actors’ behaviour) to mixed systems (Bochsler, 2009a). This section discusses the implications of the contradictory incentives for political behaviour and the consequences for party system development in mixed non- or semi-compensatory systems. 5.4.1 The problem of ‘contamination’ for the study of outcomes of mixed electoral systems Electoral systems give incentives to political players, such as party leaders, party supporters and voters, to behave in a certain way, such as to turn away from small parties because they have little chance of getting elected, and instead support large parties. Duverger (1951) called this the ‘psychological effect’ of electoral systems. In simple electoral systems, these incentives are clear. In single-seat districts, for instance, it makes no sense supporting a third- or less-ranked candidate; and parties with no chances do not even compete. In mixed electoral systems, these incentives are mixed. The proportional tier gives small parties the opportunity to win representation, while the plurality/ majority tier restricts the competition to the strongest candidates in the districts only. Simply put, under mixed incentives, political players do not know which rule to follow. The proportional tier allows many parties to be represented; the plurality/majority tier, however, threatens to severely under-represent small parties that hardly win any of the district mandates. Certainly, in mixed electoral systems with two votes, voters can follow different strategies in each of both tiers (provided that their strategies are sufficiently sophisticated, which might be unrealistic under the complex electoral rules of mixed systems; see Karp et al., 2002). The same cannot be said for parties, candidates or party supporters. A party that runs and attracts votes in the PR tier is likely, however, also to present candidates and voters in the district tier. On the one hand existence and success in the PR tier already facilitates candidatures in the plurality/majority tier. It might be difficult for the national party office to deny a local party section the right to compete with its own candidate in an election, and accordingly it is possible for candidates to run even if they have no chance. On the other hand, this can increase the party’s visibility in public and in the media and in this way increase its overall result (Ferrara et al., 2005, pp. 65–79). Such interactive effects


between both tiers are discussed as ‘contamination’ (Cox and Schoppa, 2002; Ferrara et al., 2005). Contamination might have an inflationary effect if the mixed incentives increase the number of vote-winning parties, according to the described example. It might, however, also lead to a restrictive effect whereby parties which repeatedly fail to win seats in the plurality/ majority tier and remain severely under-represented in parliament (or for which such a development is expected) might lose attraction among voters and political actors (such as donors, party supporters or potential candidates). As a consequence, the concentrating impact of the plurality/majority tier might lead such parties to vanish or to merge with other parties (cf. Section 5.5.2). Although previous studies have impressively shown how contamination works, it remains unclear in which direction its impact goes, and what the implications are for the party system as a whole. Open questions include how strong contamination is, that is, whether it involves only a small part of the votes while generally the different voting patterns of plurality/majority systems as opposed to PR are still visible, and in which direction it occurs, that is, whether it leads to an inflationary or to a restrictive effect. The implication of this uncertainty is that we do not know which outcomes of mixed electoral systems are to be expected at the macro-level and thus how many parties typically compete – and how many get elected – in a mixed electoral system. 5.4.2 Models of the number of parties under mixed non-compensatory electoral systems Before introducing factors that might account for contamination, I shall outline the implications of different levels and directions of contamination on the development of party systems under mixed electoral systems. The most pragmatic logic that allows direct analogies to simple electoral systems is the concept of mixed electoral systems as social laboratories (Moser, 1999a; Moser and Scheiner, 2004). In this logic, we assume that contamination between both tiers is negligible, so that mixed systems overall produce the same results as if we combine two simple electoral systems and assume that actors behave as under simple systems.94 However, due to contamination, the model might not be accurate. Interestingly, the model is not very different from a model that starts from an assumption of strong contamination, which I call the average incentives model (Kostadinova, 2002, pp. 25–6). In this view, whatever strategy a party adopts, it takes into account both tiers at the same


99

time, and the result is in between both simple formulae. If one tier overwhelms, the overall result will be closer to this formula, in the opposite case closer to the other one. In systems with an equal number of mandates, voters will follow half the incentives provided by PR, and half the incentives provided by the SSD tier. This model, however, is not confirmed by results from Central and Eastern Europe: while mixed electoral systems are often related to fewer parties than pure PR, a statistically significant difference between mixed electoral systems and plurality or majority systems is not visible with regard to the effective number of parties (Kostadinova, 2002, p. 30, and Appendix C).95 Differently from both views, I propose a third model, which includes a parameter to measure the direction of the contamination effect. In environments where (almost) as many parties are tempted to compete as can win a seat in parliament, the logic of the PR tier prevails. In such a case, the contamination effect is strongly inflationary, and the contamination parameter is positive and high. In the opposite case, parties are very sensitive to obtaining an outcome that is proportional to their vote share, and accordingly they orient their behaviour towards the singleseat district tier. In such cases, the contamination parameter would be negative, indicating a strongly restrictive effect of contamination. It will be argued later in this section that a restrictive contamination effect might prevail in Central and Eastern European party systems. In both cases, the effect has implications first of all on the number of elective parties (or ‘vote-winning parties’, a measure based on the votes that are cast for political parties), rather than on the number of parliamentary parties (based on the number of seats by party). This is because the number of elective parties measures the psychological effect of electoral systems, or the strategic behaviour of political actors – to which contamination refers. The number of parliamentary (seat-winning) parties, however, is further based on the mechanical effect of electoral systems, and this mechanical effect is not linked to any contamination. The difference between the number of elective and the number of parliamentary parties is expressed through measures of disproportionality. In the case of a restrictive effect through contamination, only parties that can win seats in both tiers compete, so that disproportionality is low, and the numbers of elective and of parliamentary parties are similar. If an inflationary effect through contamination exists, then large disproportionality would be created in the plurality/majority tier, which means that the number of parliamentary parties would be substantially lower than the number of elective parties.

100


In this model, the format of the party system still depends on both tiers, but the importance of each tier is weighted by a parameter of contamination. In the case of a strongly restrictive effect of contamination, the party system develops mainly according to the constraints of the district tier, and the number of parties running in the PR tier would be very similar. In the case of an inflationary effect of contamination, the mechanical effect of the electoral system in the district tier would be important. The outcome would depend heavily on the exact vote distribution among parties (Kendall and Stuart, 1950), on the territorial distribution of the votes (see Chapter 6), and on possible strategies adopted by small parties in order to proportionalise the outcomes and to reduce the mechanical effect (see Section 5.5.1). Rational political actors will behave in such a way that their political power will have a maximal impact according to their preferences. This means that rational voters in parliamentary elections want to maximise their impact on the composition of parliament, so that their own voice is represented to a maximal degree. Similarly to voters, there is a whole range of further political players in elections, such as party activists, donors and candidates, who decide to support one or another party according to their preferences and strategic considerations. All these actors bring votes to a political party, which later are converted into seats. When the electoral system gives little chance to their favoured candidate, rational voters would switch and vote for the candidate closest to their own political stand among those candidates with chances of getting elected (Downs, 1957); other political players who might have an impact on the election might do likewise. Since the chances of getting elected are related to the ratio at which votes are converted into seats (A-ratios; see Taagepera and Laakso, 1980), rational players would, given two similar political options, favour parties with higher conversion ratios over those with lower ones. Votes for parties with higher A-ratios get translated to a proportional or over-proportional degree into parliamentary seats; the lower the A-ratio, however, the fewer seats are created per vote. In simple electoral systems, the rational choice is often clear. Parties either lose, or they get an (almost) proportional or even over-proportional seat share. Intermediate cases, where they win seats, but at a lower rate than other parties, are, however, rare. In mixed electoral systems, the effect needs to be looked at more closely. First, entry is easy, so that even small parties win a seat in one or the other tier, but only large parties win seats in both tiers, and can thus convert all votes into seats. If only


101

a part of the votes can be converted into seats, intermediate conversion might become much more common than under simple electoral systems. Second, mixed electoral systems require us to differentiate between mechanisms on the demand and the supply side of the electoral market, between voters and other political players, such as parties, candidates and party supporters. Often, the psychological effect is discussed for voters, and the role of actors on the supply side of politics and that of the elite remains neglected (Blais and Indridason, 2007, p. 193). In two-ballot mixed electoral systems, strategic voters can apply different strategies for each of both tiers. In contrast, actors on the supply side of the electoral market usually do not have such a choice. They invest their resources (such as ties to potential voters, time, money, expertise or reputation) in a political party, and they need to select a party that makes their effort worthwhile (see also Boix, 1999, p. 610). Assuming that the effort of a political player can generate an (approximately) constant amount of voters for whatever party he or she stands for, then the political player can choose among parties based on their political programme and their conversion rate of votes into seats, which depends on the party size and the electoral system. In the case of parties with equally attractive programmes, acting rationally, he or she will choose the one that guarantees the largest political power – or the largest number of seats, which might lead as well to an overall better representation of the specific interests of the political player. Certainly, political positions might matter: mobilising for a like-minded party might be easier and bring more utility than mobilising for a party with different stands. In cases of strong programmatic (or personal) differences, the vote–seat ratio is less important, but if parties were perfectly interchangeable it might become a dominant determinant of party choice. The substitutability of political parties might be much more pronounced in post-communist countries than in cleavage-based Western European election markets. In post-communist democracies, parties often rely much more on popular leaders, cadre personnel or patrimonial ties, rather than on programmatic cleavages. Even if a few strong conflict dimensions exist, a large share of voters seems to vote mainly on the basis of personality for parties with less clear programmatic differences. What is true for voters seems to be the case at the level of the elite too, with parties as exchangeable vehicles, with many candidates switching parties from one election to the other (Shabad and Slomczynski, 2004). Central and Eastern European party systems thus seem weak, volatile, and fluid enough to be competitive for better conversion rates.

102


Certainly, not every single actor evaluates parties based on their conversion rate of votes into seats, but an example from Hungary shows that success rates play an important role in political considerations: Due to the disbelief in the legitimacy of the opponent, the minor parties have been accused of playing with fire: by missing the threshold they are seen as serving the interests of the opposite bloc, and therefore risking the democratic or national character of the country. Following this logic even the Catholic Church names the size of the parties (that is, below or above the threshold) as one of the major criteria voters should base their choice on. (Enyedi, 2006b, p. 198) In sum, I have argued that the impact of mixed electoral systems on party systems might be very diverse. If there were no information about the importance of programmatic differences and substitutability, one would expect mixed non- or semi-compensatory electoral systems to lead to very varied results; in some cases inclining towards the more permissive tier, and in others towards the more concentrating tier. This would imply that there should be substantial heterogeneity and low predictability in outcomes, and, due to the mixed incentives, the consolidation of stable party systems (which is based on clear and repeated incentives) might last longer in simple systems. For the rather fluid party systems in Central and Eastern Europe, I expect high sensitivity to the conversion of votes into seats. In the long term, only parties that are competitive in both tiers might survive. The next section looks more closely at the consequences for a number of relevant factors of party system change under mixed electoral systems.

5.5

Party system stability and party system change

Central and Eastern European parties might be sensitive to electoral systems’ incentives. On the other hand, it is reasonable to assume– and we know from previous studies – that there is strong contamination between both tiers; thus, the number of elective parties is similar in both the plurality/majority and in the PR tier. This has consequences for party system change and stabilisation. Change occurs namely when parties are aiming at scoring better outcomes. Party system stability is given when the current situation gives few opportunities to ameliorate the outcome for single parties. Before discussing processes of party system change, I first shed light on equilibrium situations.


103

5.5.1 Equilibria Party system equilibria are situations where the electoral systems do not exert incentives that would lead to party system change. I have argued above that unequal vote–seat conversion rates are the main incentive for party system concentration (see paragraph 5.5.2 below). Following this, I now discuss the characteristics of party systems under mixed electoral systems that provide similar rates for all parties, or that due to other aspects of the party system prevent its change. 5.5.1.1 Two-party equilibrium through a restrictive plurality/majority vote tier Usually, the plurality/majority tier of mixed electoral systems is seen as the more restrictive (allowing fewer parties to win parliamentary seats). It is typically related to a competition between two large parties. Strong contamination might signify that the number of parties in the whole system is restricted to the two large competitors in the single-seat districts, lowering the number of parties in the PR tier. This would be the case of a restrictive contamination effect. Still, either some smaller parties might persist on the basis of regional votes, or some minor parties with strong ideological grounds might survive, but they might be represented only in the PR tier. In such an equilibrium, entry for newcomers is difficult. New parties need to become one of the two largest parties very soon, and win seats in the plurality/majority tier, or they will be severely under-represented. In such a situation it is not attractive for new parties to enter. A further party concentration is virtually impossible, because there will always be incentives for the formation of an opposition party to the ruling party. This is why two-party equilibria are expected to be stable. If a system of two-party competition is symmetric, that is, if both parties have similar vote shares, then disproportionality is low. If the system with two-party competition becomes asymmetric, with one party getting a much larger vote share, then there will be reasonable disproportionality, and the effective number of parties might be lower than two. 5.5.1.2 Multiparty equilibrium through an inflationary impact of contamination In other situations, mixed electoral systems might provide for large party systems, in line with many studies of contamination, which have shown that in the single-seat district tier of mixed electoral systems often many more than only two candidates compete (Ferrara et al., 2005, pp. 32–48). Despite the large number of competing parties, the

104


electoral results might show low disproportionality. This is the case if contamination between the two tiers creates situations of proportionalisation.96 There are a number of situations where plurality or majority vote does not have a majority-building impact (Cox, 1997; Riker, 1982; Sartori, 1986), but instead the outcome might be a proportionalisation of the plurality/majority tier (D’Alimonte, 2001, p. 327; Ferrara, 2004a). High inter-election volatility is a factor which keeps party systems large, despite the effect of the single-seat tiers. As long as the party system is large, and the vote shares change from election to election, it is difficult for large dominant parties to emerge. High volatility thus hampers party system concentration due to low predictability and the possibility of voting strategically, because vanishing parties are replaced by new parties, and because high volatility in mixed electoral systems gives small and medium-sized parties a better chance of survival. The literature on mixed electoral systems has also discussed strategic coordination of candidate entries in the single-seat district tier as a form of proportionalisation. Sartori (1994, pp. 74–5) expected that, if a mixed electoral system is introduced in a country with a pre-existing multiparty system, then parties will attempt to adopt strategies that still allow them to win proportional shares of seats. This can be achieved if parties compete with their own PR lists, but reach strategic agreements for the plurality/majority tier, and coordinate their candidates in a part of the country or in all the districts (Ferrara and Herron, 2005; for the Italian case, Ferrara, 2004a). If parties engage in strategic coordination, each of them remains independent and runs candidates under its own label, but some candidates are withdrawn from district races by mutual agreement. Parties which withdraw their candidates in a district ask their supporters to vote for the ally, and benefit from the inverted agreement in other districts. This helps all cooperating parties, even small ones, to win seats which correspond approximately to their vote share. In the absence of a dominant party in the party system, strategic coordination might proportionalise the electoral outcome. Mixed electoral systems are important for such coordination, because the PR tier allows every party still to compete nationwide (and not only in the selected constituencies), which might make cooperation less costly.97 5.5.1.3

Atomised party systems

Finally, there is a different type of constellation allowing the survival of a multiparty system under a mixed electoral system: the case of party system atomisation with very low party nationalisation degrees. Party system atomisation is characterised by mainly locally competing parties


105

or candidates, and, lacking strong national parties, local parties and local independent candidates have a good chances of winning mandates in their local or regional strongholds. National parties rely mostly on the seats won in the PR tier. In this system, the electoral system creates no dominance of one or a few successful parties that might attract more voters. Unlike the previous multiparty equilibria, this one is not necessarily based on the contamination effect, but is linked to low party nationalisation (Barkan, 1995; Sartori, 1986). Results might in the end resemble proportionality, because the vote share is mainly linked to the share of the territory controlled by a party.98 Different types of equilibria might in parts be combined. For instance, some atomised local parties might be found in a system with high volatility. Both aspects help to preserve an equilibrium with a highly fractionalised party system. An overview of the types of party system equilibria will be presented in the empirical section (5.6), including the quantitative characteristics of these equilibria. 5.5.2 Party system change In situations of party equilibrium, there is no pressure for party system change. There are, however, factors that might pressure for changes in the party system when none of these equilibria occur: namely, high differences in A-ratios between the parties might induce change in the party system. I propose three key variables which are important for processes of party system change – disproportionality, coordination and volatility. Disproportionality is measured as the aggregated difference between vote shares and seat shares of political parties (Gallagher, 1991), or differences in A-ratios. It is often understood as the location of an electoral system on the continuum from perfectly proportional electoral systems to very concentrating electoral systems, but this is not the case. Disproportionality is the motor of party system change. It either leads to electoral cooperation among parties, or it reduces the number of parties. In stable party systems, votes are cast strategically, based on experience and on predictions of electoral outcomes, so that in the end successful parties win most of the votes, and disproportionality is low, even in very concentrating electoral systems. If, however, disproportionality reaches high levels, then there is the potential for some parties and political actors to change their strategic behaviour in order to get more seats for the same number of votes. This goal can be achieved either through a concentration process towards fewer parties (politicians abandon small and unsuccessful parties for larger ones, formation of electoral alliances, party mergers, etc.), or through coordination of candidatures

106


in the plurality/majority tier. This does not hold in two-party competition with a dominant party (N2 < 2), where disproportionality emerges, but the party system does not contract any more. And, in situations of high inter-election volatility, disappearing parties are quickly replaced by new ones (see below), so that no concentration process is going on. Electoral coordination means that parties agree not to put up candidates in some single-seat districts in order to enable the election of the ally (Benoit, 2001). Unlike the formation of joint lists and common labels (which happens predominately in the PR tier), in the case of electoral coordination actors with their own candidates in a select portion of the districts and with their own lists compete independently and remain clearly identifiable. But, in the long run, cooperation of parties might ‘compromise their visibility, weaken their electoral base, and erode their capacity to extract policy concessions from their allies’ (Ferrara, 2004a, p. 398). While the need for cooperation corresponds to the short-term necessities of the logic of mixed non-compensatory electoral systems, in the long run it might lead to a re-identification of voters, de facto loss of independence of cooperating parties, and party mergers (Enyedi, 2007, pp. 121–2), so that electoral coordination appears only as an intermediate step towards the disappearance of the previous parties, except for the case of high inter-election volatility. Inter-election volatility is not a factor of party system change; volatility holds large party systems stable. Certainly, in the complete absence of volatility, hardly any change can occur; this would then mean that the original vote distribution and size distribution among parties would not change at all. In order to change the number of parties or their structure, a certain level of volatility is needed, but a moderate level is enough for this purpose. At high levels of volatility, however, new parties can easily be established, and this helps to keep large party systems alive, even if the electoral system appears not very favourable to this. When, as suggested in this study, small parties that repeatedly win only very few seats in elections, even if they get a higher vote share, drop out of the political game, this leads to a reduction in the number of parties. Larger parties that repeatedly gain a dominant position, however, become more and more attractive for voters and the political personnel, and grow even larger. High volatility together with the application of a mixed electoral system, however, disturbs these processes, because it does not allow repeated dominance for several reasons: ●

Small parties might wait for better times. The mixed electoral system allows small parties which fail in the single-seat districts to win some


●

●

107

PR mandates, so that the disproportional disadvantages of small parties are not so important; volatility gives them the hope that they might belong to the winners in the subsequent elections. The continuous renewal of political pluralism means that, under high volatility, disappearing parties are simply replaced by new competitors. As long as there is no dominant party, new emerging parties can easily win a substantial number of parliamentary seats. This supply of new parties makes a lasting concentration of the party system impossible. Blocking strategic behaviour: under high volatility, electoral outcomes are less predictable, particularly because surveys are often not accurate enough to predict the results at the district level, and experience in previous elections is scarcely helpful if the party system is volatile. This makes strategic decisions for voters and parties more difficult. In addition to this, if many parties compete in the district tier, the district threshold diminishes, which allows medium-sized parties to win a considerable number of district mandates. This is why volatile party systems remain large.

5.5.3 Hypotheses about contamination and the effect of mixed electoral systems From the previous discussion, a number of hypotheses about party system change in mixed electoral systems can be drawn out. 1. The main, and general, consequence of contamination is that parties in mixed electoral systems usually compete in both tiers and win similar vote shares in each tier, except for regional groups (or special cases where parties form joint lists for the PR elections, and run separately but in coordination in the single-seat districts). This is why the number of effective elective parties, calculated countrywide, should be similar in both tiers, if party nationalisation is high. I define ‘similar’ as a difference between both tiers of one at most. In cases of low party nationalisation, the number of parties might be higher in the plurality/majority tier than in the proportional tier. 2. Disproportionality leads to a coordination process among parties to get better representation, or to a reduction in the number of elective parties. 3. After a period of strategic coordination in elections, the number of parties is likely to be further concentrated. Only high volatility can avoid such a party concentration effect. 4. A party system under a mixed non-compensatory system will over time reach equilibrium, becoming a system with two major parties

108


and possibly some smaller parties after a process of transformation. The transformation process is frozen if the party system can stay fragmented due to high volatility or low nationalisation.

5.6 The consequences of mixed electoral systems – empirical study I test the model empirically, based on the elections in Central and Eastern Europe held under mixed electoral systems. This region has accumulated more experience with mixed electoral systems than any other part of the world. Eight countries have applied mixed electoral systems for twenty-four national parliamentary elections.99 Four out of these eight countries (Albania, Hungary, Lithuania and Russia) have applied mixed electoral systems over (almost) all the post-communist elections, and each of them offers us a series of four or five elections, which allows an investigation of their developments over time. The other half of the countries applied the mixed electoral systems only in one (Bulgaria, Macedonia) or two elections (Croatia, Ukraine). All the cases are non-compensatory or semi-compensatory systems (as defined in Section 5.3). Still, Central and Eastern Europe counts only very few cases, particularly since most of the hypotheses highlight dynamic processes that require several subsequent elections to be tested. A broader basis of cases would thus be advantageous, as would a comparison with similar developments in other mixed electoral systems. This is why I will briefly examine Japan, the only case outside the region that has applied a mixed non-compensatory system over a longer period (1996– 2005), but consequently highlight to what extent results rely on the Japanese case.100 As seen in the introduction to this chapter, party systems have developed very differently even under similar mixed electoral systems in Central and Eastern Europe. Due to the limited number of cases, conclusions that might be drawn from the empirical data need to be handled with caution, and results remain below any standard level of statistical significance. However, the following discussion of party system change might help to provide some preliminary suggestions about the impact of mixed electoral systems in Central and Eastern Europe, and might build a basis for discussion. A growing empirical number of cases will allow for more sensitive tests.101 First, I perform simple analysis of the stated hypothesis, before proceeding to a categorisation of the electoral outcomes, based on the previously defined equilibria.


109

5.6.1 Comparing party system change under mixed electoral systems I analyse step by step the party system dynamics as expected in my theoretical model using correlation analyses. Table 5.3 shows the key variables that are analysed in the model. First, I test whether contamination works in the electoral systems under study. Hypothesis 1: Vote distribution is similar in both tiers (strong contamination). According to the hypothesis of contamination, the number of seat-winning candidatures and the vote distribution are linked in both tiers, except for small regional parties. This hypothesis is opposed to the social laboratory view, suggesting that the electoral process is different between both tiers, so that a different number of parties will emerge. Since contamination captures the psychological effect of electoral systems or similar strategic behaviour of political actors in both tiers, it is best measured by an indicator that captures the voting behaviour rather than the resulting seat allocation. This is why I compare the number of Table 5.3 Overview of the variables used in the model of party system change Variable name

Variable description

Classification

Number of parties

Effective number of parliamentary parties (whole parliament). Effective number of elective parties in the PR tier. Seat share (whole parliament). Vote share (PR tier). Least–square index (Gallagher, 1991); comparing the overall number of seats won to the PR vote share. Ratio of effective number of elective parties in the plurality/majority tier by the effective number of elective candidates per district (Only applicable if party nationalisation is reasonably high). Lee index (PR tier).

low: 3.5; high: 3.5

Largest party

Disproportionality

Strategic coordination

Volatility

low: 40%; high: 40%

low: 0.10; high: 0.10

low: 1.4; high: 1.4

low: 25%; high: 25%


10 9 8 7 6 5 4 3 2 1

JP96 JP03

JP00

JP05

1

2

3 4 5 6 7 8 9 10 Elective parties (PR)

Parliamentary parties (SSD)

Elective parties (SSD)

elective (vote-winning) parties between both tiers. In the case of strong contamination, we will find the same strategic behaviour and thus an equal number of elective parties in both tiers, despite the different incentives they give. (Here, equality means that differences are smaller than 1.) In contrast, the social laboratory view states that the mechanical effect of each tier has a different impact on voting behaviour in each tier, so that the number of elective parties is substantially lower in the singleseat districts compared to the PR tier. The social laboratory hypothesis is scarcely plausible for twenty-three out of the twenty-four Central and Eastern European cases for which data is available, as shown in Figure 5.3 (left graph). The number of parties is equal in both tiers, and in nine cases even substantially higher in the single-seat district tier than under PR rules. Most clearly, the cases of party atomisation (Russia, Ukraine) fall into this category, together with party systems with strategic coordination and high volatility (three elections in Lithuania, 1998 elections in Hungary), and the Croatian 1992 elections.102 Only the Japanese 2000 elections lie slightly below the lower dotted line, with more parties competing in the PR tier than in the districts. The Albanian elections in 2005, with more than ten elective parties in the PR tier but fewer than three in the single-seat districts, are outliers due to specific strategic behaviour (see Section 5.8). Whereas the laboratory view can be rejected for the vote distribution, there is a substantial difference between the two tiers when looking at the effective number of parliamentary (seat-winning) parties. The right graph 10 9 8 7 6 5 4 3 2 1

JP03

JP96

JP00

JP05

1

2 3 4 5 6 7 8 9 10 Parliamentary parties (PR)

Figure 5.3 Contamination of the PR tier (x-axis) and the plurality/majority tier (y-axis), effective number of elective parties (left graph) and of parliamentary parties (right graph) (y-axis; censored at y 10). The upper dotted line shows the equality line, y x. The lower dotted line shows the line y x 1 Note:

X

Central and Eastern Europe; O Japan.


111

shows a number of data points beyond the lower diagonal line, showing elections where the number of parties that won seats through the PR tier is substantially higher than through the single-seat district tier. While political behaviour (elective parties) differs only marginally between tiers, or in the direction opposite to that expected, the mechanical effect of the electoral systems produces clear differences in seat allocation. The figures draw a picture of strong contamination of the voting behaviour, so that the number of elective parties is very similar in both tiers. Further tests will investigate the directionality of contamination: is there an inflationary effect (in the direction of the more permissive tier) or a restrictive effect through contamination (in the direction of the more concentrating tier)? Hypothesis 2: Disproportionality is the motor of party system development. It either leads to electoral cooperation among parties or reduces the number of parties. Disproportionality signifies that some parties get a higher seat–vote ratio (A-ratio) than others. When voters and political actors are sensitive to A-ratios, disproportionality might lead to changes in the party system, or to cooperation among parties in order to get a better outcome in subsequent elections. I use Gallagher’s least-square index (1991) to measure disproportionality. In two-vote systems, where voters cast one vote for their party and a second vote for their preferred district candidates, the party vote appears to be a more reasonable choice to measure disproportionality, because the personality vote might be affected by coordination strategies, for instance if several parties coordinate the race in certain districts, in order not to compete against an ally and to help each other win a better share of district seats. Strategic coordination might rise in the aftermath of elections with high disproportionality, since it might help previously under-represented parties to get better A-ratios. My index of district coordination measures the share of coordination of candidatures in the single-seat district elections (in two-round systems in the first round).103 The index is positive if there are fewer candidates in the districts than vote-winning parties at the national level in the single-seat district tier, which indicates that some parties decided not to compete in certain districts due to explicit party agreements over candidate withdrawal or candidate strategies in order to support candidates of another party. We need to be careful, however, not to confuse coordination with low party nationalisation, and accordingly exclude cases with low party nationalisation (taking 0.5 as the lower level, thus excluding elections in Russia and Ukraine).


The mechanism is well illustrated by Croatia in the period of 1992 to 1995. Plans among the opposition parties to coordinate candidates in the single-seat district elections in 1992 failed (Commission on Security and Cooperation in Europe, 1992, p. 12), and, as a result, the governing HDZ nationalists were largely over-represented. But by 1995 the opposition parties had learned from this experience, and moved towards a more inclusive form of alliance, with seven parties competing with joint candidates in the plurality districts104 and five on a joint list in the PR tier. This allowed the opposition to improve its result, and the parliament became more plural (2.6 effective parties in 1995). Disproportionality dropped substantially, but still not enough to break the HDZ dominance. The data supports a learning effect from disproportionality in previous elections: parties react with better coordination in order to achieve better proportionality. In order to control rudimentarily for the distortion of the measure through low party nationalisation, coordination in previous elections is included as a control variable.105 The results show that the sequence of disproportionality and strategic coordination in subsequent elections is significant (Table 5.4). What is more, strategic coordination is not a repeated game: when disproportionality is included in the model, the lagged variable does not exert an influence anymore.106 Coefficients and significance change only marginally if the three Japanese cases are excluded. Certain forms of strategic coordination are not captured by our measure. Further, parties might also react in a different way from disproportionality, for instance by merging or forming of joint electoral lists. Strategic coordination in the second round of two-round elections cannot be captured with the measured variables. It appears particularly Table 5.4 OLS regression models for strategic coordination of candidate entry; robust standard errors. Elections with low party nationalisation (

Territory and Electoral Rules in Post-Communist Democracies

Territory

Territory

Territory

Territory

Territory

Territory

Territory

Territory

Voter Turnout and the Dynamics of Electoral Competition in Established Democracies since 1945

Democracy and Authoritarianism in the Postcommunist World

In Milton Lumky Territory

In Milton Lumky Territory

Ethnicity and Electoral Politics

In Milton Lumky Territory

In Milton Lumky Territory

In Milton Lumky Territory

Federal Democracies

Slovenia (Postcommunist States and Nations)

Nationalism and Territory

The Politics of Electoral Reform: Changing the Rules of Democracy

New Territory

Dangerous Territory

Reckless Territory

Pack Territory

Dominant Territory

Pack Territory

New Territory

Reckless Territory

Territory and Electoral Rules in Post-Communist Democracies

Territory

Territory

Territory

Territory

Territory

Territory

Territory

Territory

Voter Turnout and the Dynamics of Electoral Competition in Established Democracies since 1945

Democracy and Authoritarianism in the Postcommunist World

In Milton Lumky Territory

In Milton Lumky Territory

Ethnicity and Electoral Politics

In Milton Lumky Territory

In Milton Lumky Territory

In Milton Lumky Territory

Federal Democracies

Slovenia (Postcommunist States and Nations)

Nationalism and Territory

The Politics of Electoral Reform: Changing the Rules of Democracy

New Territory

Dangerous Territory

Reckless Territory

Pack Territory

Dominant Territory

Pack Territory

New Territory

Reckless Territory

Recommend Documents