Economic voting and the Great Recession in Europe · 2020. 2. 5. · Abstract The Great Recession of 2007–09 was the worst global economic crisis since the Great Depres-sion of

Economic voting and the Great

Recession in Europe

A comparative study of twenty-five countries

Troy Alexander Cruickshank

August 2016

A thesis submitted for the degree of Doctor of Philosophy of

The Australian National University

© Copyright by Troy Alexander Cruickshank 2016

Declaration

This thesis is entirely my own work and has not previously been submitted for a degree or

diploma in any university. To the best of my knowledge and belief, the thesis contains no

material previously published or written by another person except where due reference is

made in the text.

Troy Cruickshank

i

Acknowledgements

I would like to thank first and foremost my supervisor Ian McAllister. Your advice to me

has always proven sound, whether relating to my thesis specifically or to academic life more

generally. I much appreciate your willingness to meet with me or read my work on short notice

throughout my candidature. I would also like to thank the other members of my advisory

panel, Juliet Pietsch, who frequently made herself available to discuss my work and whose

advice is also much appreciated, and Andrew Banfield, who stepped up at the last minute to

fill a formal vacancy.

I would also like to thank the other people who have taken the time to discuss my work with

me at various points throughout my thesis. These include, in no particular order, Hans-Dieter

Klingemann, Jeffrey Karp, Jill Sheppard, Richard Johnston, Shawn Treier, Timothy Hellwig

and Yusaku Horiuchi. Finally, I would like to thank my friends and family who have supported

me in countless ways throughout this process. You know who you are.

iii

Abstract

The Great Recession of 2007–09 was the worst global economic crisis since the Great Depres-

sion of the 1930s. The effects were felt across most of the developed world and Europe was

no exception. In many European countries, austerity programmes were implemented in re-

sponse to the recession, which were often deeply unpopular. Many governments lost power

in the years following the recession, with sometimes strikingly harsh swings against them.

One notable example was the Irish election of 2011, in which the incumbent Fianna Fáil was

reduced from 71 to 20 seats, by far its worst result at any general election since independence

in 1922. This is congruent with the theory of economic voting, according to which voters will

remove from office governments that fail to deliver economic prosperity. Although there is an

enormous empirical literature supporting this theory, almost all of this evidence pertains to the

typical boom and bust cycle of individual countries and little is known about economic voting

during a severe global recession. The Irish result could have been indicative of the usual eco-

nomic vote, a bolstered economic vote due to the unusual scale and severity of the crisis, or of

dissatisfaction with the government’s handling of the crisis. This thesis investigates whether

the usual economic vote in European countries was altered during the Great Recession.

This thesis uses survey data from the 2004, 2009 and 2014 waves of the European Election

Studies (EES) to compare the economic vote in 25 European countries before, during and after

the Great Recession. Multilevel methods are used to model voters’ support for the parties they

could vote for at general elections in their own countries. Using this method, the results show

that the economic vote was weaker during the crisis than it was either before or after. In order

to explain these results, I analyse which parties voters tended to prefer after the crisis and

how attitudes towards the European Union evolved over time. The results find that there was

a shift away from centrist and pro-European parties towards radical and Eurosceptic parties

following the crisis. In addition, support for the EU fell over the same time period and voters

were increasingly likely to hold the EU responsible for economic conditions. Given the timing

v

vi

of these shifts as well as the association between European institutions and austerity policies,

these findings suggest that the austerity programmes implemented in the wake of the crisis

may have been a stronger catalyst for economic voting in Europe than the Great Recession

itself.

Contents

Introduction 1

1 Theory of economic voting: how economic conditions shape the vote 9

1.1 What is economic voting? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.2 Perceptions or indicators: the link between the economy and the vote . . . . . 13

1.3 Sociotropic voting: do voters only care about themselves? . . . . . . . . . . . . . 16

1.4 Prospective and retrospective voting . . . . . . . . . . . . . . . . . . . . . . . . . . 18

1.5 The vote choice process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

1.6 Reward and punishment: the logic of sanction . . . . . . . . . . . . . . . . . . . . 24

1.7 Competent government: the logic of selection . . . . . . . . . . . . . . . . . . . . 26

1.8 Political context and the instability problem . . . . . . . . . . . . . . . . . . . . . . 29

1.9 The Great Recession . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

1.10 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

2 Measuring the economic vote 41

2.1 Data sources and timing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

2.2 Measurement and variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

2.3 Method of analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

3 Voting in a time of crisis: how the Great Recession affected the economic vote 63

3.1 Party support theory of economic voting . . . . . . . . . . . . . . . . . . . . . . . . 65

3.2 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

3.3 Measuring the economic vote . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

3.4 A spatial model of party support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

3.5 The prime minister’s party and the economic vote . . . . . . . . . . . . . . . . . . 73

vii

viii CONTENTS

3.6 A multiparty model of the economic vote . . . . . . . . . . . . . . . . . . . . . . . 78

3.7 Influence of the economy on mean party support . . . . . . . . . . . . . . . . . . 84

3.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

4 Clarity of responsibility during a global recession 89

4.1 Clarity of responsibility: economic voting in different contexts . . . . . . . . . . 90

4.2 The dimensions of clarity of responsibility . . . . . . . . . . . . . . . . . . . . . . . 94

4.3 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

4.4 How clarity affects the prime minister’s party . . . . . . . . . . . . . . . . . . . . . 99

4.5 The effect of clarity on other parties . . . . . . . . . . . . . . . . . . . . . . . . . . 105

4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

5 Extreme and Eurosceptic parties: the changing policy preferences of European

voters 113

5.1 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

5.2 Classifying parties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

5.3 How party position relates to voter support . . . . . . . . . . . . . . . . . . . . . . 124

5.4 The changing fortunes of pro-European integration parties . . . . . . . . . . . . 127

5.5 Left–right position: a shift to the extremes . . . . . . . . . . . . . . . . . . . . . . 131

5.6 Beyond left and right: social and economic dimensions . . . . . . . . . . . . . . 134

5.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

6 Economic abstention: turnout intention in the face of economic pessimism 143

6.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

6.2 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

6.3 Measuring turnout intention . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

6.4 An economic model of turnout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

6.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

7 Attitudes towards European integration and institutions 161

7.1 Austerity and the European Union . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

7.2 Measuring attitudes towards the European Union . . . . . . . . . . . . . . . . . . 164

7.3 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168

7.4 Attitudes towards further European integration . . . . . . . . . . . . . . . . . . . 170

7.5 Popular support for EU membership . . . . . . . . . . . . . . . . . . . . . . . . . . 177

CONTENTS ix

7.6 Attribution of responsibility for the economy . . . . . . . . . . . . . . . . . . . . . 183

7.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187

8 Conclusion: revising theories of economic voting 191

A Countries and parties 199

B Coefficient tables 213

References 237

List of figures

1.1 Basic principle of economic voting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.2 Outline of economic voting theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

1.3 Vote choice process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

2.1 GDP growth in the European Union, 2000–2015 . . . . . . . . . . . . . . . . . . . . . 43

2.2 Distribution of raw and centred party support . . . . . . . . . . . . . . . . . . . . . . 50

2.3 Distribution of left–right distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

2.4 Distribution of prospective and retrospective assessments . . . . . . . . . . . . . . . 54

2.5 Relationship between prospective and retrospective assessments . . . . . . . . . . . 55

2.6 Data stacking process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

3.1 Support for incumbent prime minister’s party by economic assessment . . . . . . . 69

3.2 Relationship between party support and left–distance . . . . . . . . . . . . . . . . . . 71

3.3 Predicted support for prime minister’s party by left–right distance . . . . . . . . . . 75

3.4 Predicted support for prime minister’s party by economic assessment . . . . . . . . 77

3.5 Data classification structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

3.6 Economic vote by year and incumbency status . . . . . . . . . . . . . . . . . . . . . . 81

3.7 Predicted support by year and incumbency status . . . . . . . . . . . . . . . . . . . . 82

4.1 Economic vote for dominant government party by time in office . . . . . . . . . . . 103

4.2 Economic vote for dominant government party by ideological cohesion . . . . . . . 104

4.3 Economic vote for dominant government party by institutional clarity . . . . . . . . 105

4.4 Government and opposition economic vote by time in office . . . . . . . . . . . . . . 107

4.5 Government and opposition economic vote by ideological cohesion . . . . . . . . . 108

4.6 Government and opposition economic vote by institutional clarity . . . . . . . . . . 109

5.1 Relationship between party support and position . . . . . . . . . . . . . . . . . . . . . 124

5.2 Relationship between integration position and salience . . . . . . . . . . . . . . . . . 126

x

List of figures xi

5.3 Party support by position on European integration . . . . . . . . . . . . . . . . . . . . 129

5.4 Relative support for pro-integration parties . . . . . . . . . . . . . . . . . . . . . . . . 130

5.5 Party support by general left–right position . . . . . . . . . . . . . . . . . . . . . . . . 132

5.6 Party support by economic left–right position . . . . . . . . . . . . . . . . . . . . . . . 135

5.7 Party preference by social libertarian–authoritarian position . . . . . . . . . . . . . . 138

6.1 Average turnout at national elections in EU countries . . . . . . . . . . . . . . . . . . 151

6.2 Proportion intending to vote by year and economic assessment . . . . . . . . . . . . 152

6.3 Predicted probability of voting by past behaviour . . . . . . . . . . . . . . . . . . . . 155

6.4 Predicted probability of voting by economic assessment and party identification . 156

7.1 Support for European integration by prospective economic assessment . . . . . . . 173

7.2 Support for European integration by left–right position . . . . . . . . . . . . . . . . . 174

7.3 Evaluation of EU membership over time . . . . . . . . . . . . . . . . . . . . . . . . . . 179

7.4 Positive evaluation of EU membership by prospective economic assessment . . . . 180

7.5 Positive evaluation of EU membership by left–right position . . . . . . . . . . . . . . 181

7.6 Attribution of economic responsibility by prospective economic assessment . . . . 184

7.7 Attribution of economic responsibility by left–right position . . . . . . . . . . . . . . 185

List of tables

2.1 Sample size and interview mode of EES surveys . . . . . . . . . . . . . . . . . . . . . 44

2.2 Key variables used to measure the economic vote . . . . . . . . . . . . . . . . . . . . 48

3.1 Prime ministers’ parties model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

4.1 Components of institutional and government clarity . . . . . . . . . . . . . . . . . . . 95

4.2 Alternative government clarity models . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

5.1 Variation in party position over time . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

5.2 Regression model predicting general left–right position . . . . . . . . . . . . . . . . . 127

5.3 Left–right and extreme tendency by year and voter group . . . . . . . . . . . . . . . 133

5.4 Directional and extreme tendency along economic dimension . . . . . . . . . . . . . 136

5.5 Directional and extreme tendency along social dimension . . . . . . . . . . . . . . . 139

xii

Abbreviations

BNP British National Party

CDU Christian Democratic Union (Germany)

CHES Chapel Hill Expert Survey

CSU Christian Social Union (Germany)

ECB European Central Bank

EES European Election Studies

EU European Union

GDP Gross domestic product

IMF International Monetary Fund

MSZP Hungarian Socialist Party

PASOK Panhellenic Socialist Movement (Greece)

ParlGov Parliaments and Governments Database

PDY Political Data Yearbook (European Journal of Political Science)

SGP Stability and Growth Pact

Syriza Coalition of the Radical Left (Greece)

UK United Kingdom

xiii

Introduction

The Great Recession of 2007–09 was a deep global recession that severely harmed the world

economy. It has been described as the worst economic crisis since the Great Depression of the

1930s, both in general (Reinhart and Rogoff 2009, 208) and in Europe specifically (European

Commission 2009, iii). In many European countries, the period following this recession has

been characterised by political upheaval and many incumbent governments have lost power

as a result (Kriesi 2014). This can seemingly be explained by the theory of economic voting,

which predicts that voters’ support for incumbent governments is linked to good economic

conditions (Key 1961; Fiorina 1981; van der Brug, van der Eijk and Franklin 2007; Duch

and Stevenson 2008). This simple idea is so well established that it has been described as

‘virtually a social science law’ (Duch 2007, 805). Nonetheless, there are still open questions

about economic voting. As the theory was developed after the Great Depression, it is not yet

known whether the economic voting effect operates in the same way during a severe global

recession than it does at more normal times. The main question motivating this thesis is: was

the economic vote during the Great Recession different from that at other times? Since many

of the elections following the crisis coincided with deeply unpopular austerity programmes

(Kriesi 2014), the thesis also examines whether this austerity played any role in influencing

voters’ attitudes and intentions.

The Great Recession began with the financial crisis in the United States sparked by the

subprime mortgage lending practices of US banks (European Commission 2009, 1). The

European Union was strongly affected, with the whole EU entering a fifteen-month recession

from early 2008. Although this period ended in mid-2009, it was succeeded by a further re-

cession in late 2011 and yet another in 2012. The Eurozone remained in recession throughout

the entire period from late 2011 until early 2013.1 In several of these countries, government

funds were used to bail out banking sectors at risk of collapse (62). Austerity programmes were

1These figures are based on OECD seasonally adjusted quarterly GDP growth rate data, using the commondefinition of a recession as two or more consecutive quarters of negative growth.

1

2 INTRODUCTION

widely implemented. In the following years, a number of governments suffered memorable

defeats at the ballot box (Kriesi 2014). Decades of economic voting theory have established

that voters punish governments electorally when economic conditions are poor but the Great

Recession is of a much greater scale and severity than the typical national recessions that have

informed much of this theory. The last comparable event was the Great Depression of the

1930s, and although there have been studies of that time (e.g. Lindvall 2013, 2014), they

are fundamentally limited by the lack of any data beyond the electoral results and primitive

economic indicators of the time. The purpose of this thesis is to study electoral behaviour in

the European Union during the Great Recession and to determine whether voters responded

to economic dissatisfaction the same way then as they have at other times.

The Great Recession and its after-effects have played out differently in every country. It is

worth taking a closer look at some of the countries that were most deeply affected in order to

gain an understanding of the ways that governments and voters responded to the unfolding

situation. One of the most striking examples of an electoral backlash took place in Ireland.

Ireland was the first European country to enter recession, beginning in the second quarter of

2007 and, but for a short reprieve in late 2007, remaining in recession until the end of 2009.2

Unemployment more than doubled in the space of a few years, the rate rising from 6.4 percent

in 2008 to 14.7 percent in 2011.3 In 2009, 100 000 workers took to the streets of Dublin to

protest the government’s plan to levy the pensions of public sector employees to make up

for declining government revenue (BBC News 2009). Irish banks were heavily damaged by

the global financial crisis and later in 2009, the Irish government was forced to nationalise

the Anglo-Irish Bank to prevent its collapse, after other banks had already been bailed out

(Connor, Flavin and O’Kelly 2012). The first opportunity that Irish voters had to respond to

these events was the general election of 25 February 2011. This election proved catastrophic

for Fianna Fáil, the leading party of the coalition government at the time and the party which

had held the Irish premiership for all but eighteen of the 79 years between 1932 and 2011. At

the 2011 election, Fianna Fáil lost more than half of its vote compared to the previous election

and was reduced from 78 to a mere twenty seats in the Dáil Éireann (O’Malley 2012). Although

they have increased their support from that low at the 2016 election, they still received less

than one quarter of the vote (Irish Times 2016), showing that they have still not recovered

from most of the damage.

2According to OECD quarterly GDP data.3According to OECD annual unemployment rate data.

3

One of the countries that has been most severely affected by the crisis is Greece. The Greek

crisis has been characterised by high levels of sovereign debt and the inability of the Greek

government to repay or service that debt. Greece suffered an exceptionally long period of

recession, with negative or zero quarterly GDP growth rates in all but two quarters of the six

years from mid-2007 to mid-2013, according to OECD figures. In light of increasing concern

that Greece would be unable to repay its sovereign debt, the so-called Troika of the European

Commission, European Central Bank (ECB) and International Monetary Fund (IMF) offered

Greece a series of bailout loans (European Commission 2010, 2012). These loans were condi-

tional on the implementation of a package of austerity measures, including privatisation and

structural reforms. The measures were deeply unpopular with the Greek public, leading to

mass protests. One survey found that 29 percent of the Greek population had participated in

anti-austerity protests (Rüdig and Karyotis 2014, 488).

The economic instability in Greece has been matched by political instability, with five gen-

eral elections having taken place in the years 2009–15. The first election took place on 4 Octo-

ber 2009, during the recession but before the debt crisis took hold. The centre-left Panhellenic

Socialist Movement (PASOK) was able to secure an absolute majority of votes and took gov-

ernment from the centre-right New Democracy party (Mavrogordatos and Marantzidis 2010).

Two elections were held in 2012, after the bailout loans had been accepted and the austerity

programme had begun to be implemented. The first election took place on 6 May but as no

government could be formed, fresh elections were held on 17 June. The result of this election

was that a grand coalition was formed in which New Democracy and PASOK shared power

along with a third party, Democratic Left. This coalition was was explicitly pro-European and

had the objective of keeping the anti-austerity Coalition of the Radical Left (Syriza) out of

power (Mylonas 2013). This effort was ultimately futile, as Syriza gained enough seats in the

election of 25 January 2015 to form a new government with the support of the right-wing

populist Independent Greeks. This new coalition was re-elected in the September elections of

the same year (Aslanidis and Kaltwasser 2016). In effect, Greek voters have rejected not just

the party in power at the time the crisis began but the entire political mainstream.

Not all countries that were deeply affected by the Great Recession experienced such tur-

moil. The United Kingdom was also hit very hard by the crisis but the political effects were less

severe than in Ireland and Greece. The UK was in recession for eighteen months between early

2008 and late 2009.4 The unemployment rate rose from 5.3 percent in 2007 to 8.0 percent in

4According to OECD quarterly GDP data.

4 INTRODUCTION

2011, before slowly falling again to its pre-crisis level by 2015; youth unemployment peaked

at 21.3 percent in 2015.5 Two general elections have been held in the UK since the begin-

ning of the crisis. The first took place on 6 May 2010. In this election, the incumbent Labour

Party lost a considerable portion of its support, with its vote share reduced to 29.0 percent, a

6.2 point decline since the preceding election (Whitaker 2011). Despite the use of a majorit-

arian electoral system, no party received a majority of seats but the Conservatives were able

to form a centre-right coalition with the centrist Liberal Democrats. At the second election,

which took place on 7 May 2015, the Conservatives were able to form a government in their

own right and the Labour Party recovered some of its vote, with a 1.5 point increase in support

(BBC News 2015). Thus although there was a change of government, this was nothing like

the severe collapse in support faced by the incumbent party in Ireland or the reconfiguration

of the party system in Greece.

In all of these cases it may seem obvious that the incumbent governments suffered defeats

because they were in power at a time when economic conditions worsened considerably. This

would be the classic economic voting response, in which voters sanction governments for

failing to prevent an economic crisis. This is a reasonable interpretation but there are other

interpretations that ought to be considered as well. In the case of the United Kingdom, the

Labour Party had already been in power for thirteen years by the time of the 2010 election,

an unusually long tenure for a Labour government as no Labour government had previously

served more than six years and a few months. Furthermore, this was the first election faced by

Gordon Brown, who had succeeded Tony Blair as prime minister and who was less charismatic

than his predecessor. In other words, had there been no recession, Labour may well have lost

the election anyway. The sheer magnitude of Fianna Fáil’s defeat makes it difficult to deny that

the Irish result was related to the crisis. Similarly, the fact that the grand coalition of Greek

mainstream parties has failed to hold on to office is strongly suggestive of a crisis response in

Greece. Even so, an economic vote would normally be expected to benefit the main opposition

party, which in this case was also clearly rejected. Furthermore, voters may well have been

reacting to the unpopular austerity measures as much as or more than the recession itself.

This thesis aims to shed new light on economic voting theory by testing these alternative

explanations. Although there is an enormous economic voting literature showing that voters

do tend to vote against incumbent parties when economic conditions are poor,6 these results

have not been thoroughly tested in the context of a global economic crisis. This means that5According to OECD annual unemployment rate and youth unemployment rate data.6This literature is discussed in Chapter 1.

5

it is not yet known whether the economic voting response is stronger in such a context or

whether there are spillover effects, such as a loss of support for mainstream non-government

parties. It has previously been hypothesised that the economic vote may operate differently

in different economic contexts (Stevenson 2002, 45) and the Great Recession offers an ideal

opportunity to test this hypothesis. Furthermore, economic voting theory has little to say

about the electoral response to the austerity measures that played such a large role in the

political response to both the Great Depression and the Great Recession. These are the key

questions motivating to this thesis. To what degree, if any, does the electoral response to the

Great Recession differ from the ordinary economic voting response and how much of this can

be attributed to austerity measures rather than the economic conditions themselves?

In order to examine these questions, this thesis uses data from several sources. The most

important of these is the European Election Studies (EES) voter surveys. These are surveys of

approximately one thousand respondents in each European Union member state. A wave of

the EES surveys has been collected shortly after each set of elections to the European Parlia-

ment since 1979. Three of these waves are used in this thesis. The first, the 2004 wave, took

place a few years before the beginning of the Great Recession. The 2009 wave was collected

at a time when most of the countries had entered recession. The final wave, collected in 2014,

took place in the aftermath of the crisis, when many countries were still in recession and once

the effects of the austerity programmes had been widely felt. In other words, these three time

points represent the before, during and after periods of the Great Recession. The EES data

includes a number of questions that make it suitable for measuring the level of economic vot-

ing propensity. The EES survey data is also supplemented with aggregate data, most notably

from the Parliaments and Governments Database (ParlGov). Multi-level modelling is used to

estimate economic voting effects across the 25 countries that had joined the EU in 2004 or

earlier.

This thesis consists of eight chapters. Chapter 1 reviews the economic voting literature

and explains the theoretical model informing this study. This chapter argues for a sociotropic

model of economic voting and for an individual-level study using party support rather than

vote choice as the dependent variable. It argues that both the prospective and retrospective

theories of economic voting have merit but that a prospective approach is appropriate for this

thesis because asking people in 2009 how they felt the economy had changed over he past

twelve months is not very informative, since there is broad agreement in most of the studied

countries that the economy had become worse in that period. This chapter also outlines the

6 INTRODUCTION

two-stage model of vote choice according to which voters form an assessment about the eco-

nomic conditions in their country, which in turn influences their level of support for each of the

parties in their country, this preference finally determining their vote choice. This theoretical

model informs the empirical model developed in later chapters to measure the economic vote.

Chapter 2 explains in depth the data and methods used throughout the rest of the thesis.

The chapter discusses how the economic vote is measured using multilevel analysis to predict

an individual’s level of support for the parties in his or her country. The term ‘economic vote’

is used throughout this thesis to mean the degree to which an individual’s current support for

government parties relative to opposition parties is affected by his or her economic perceptions.

This means that there need not be an actual election in order to speak of the economic vote

in a given year. The chapter also discusses the choice of datasets in more depth as well as the

details of how the key variables have been measured.

The remaining chapters discuss the results of the analysis. Chapter 3 compares the eco-

nomic vote in 2004, 2009 and 2014. In this chapter, a multilevel model is developed to meas-

ure the effect of a change in prospective economic assessment on a voter’s levels of support

for each of the parties in their country. This model is used to show that support for incumbent

parties is positively associated and support for opposition parties negatively associated with

an individual’s prospective economic assessment, as economic voting theory predicts. It goes

on to show that this effect was stronger in 2004, before the Great Recession, than it was in

2009, when the recession was at its peak. It further shows that the economic voting effect was

stronger in 2014 in the aftermath of the recession than in 2009, although not as strong as in

2004. These results are contrary to expectations and some of the implications are discussed.

Potential explanations for this result are explored throughout the remainder of the thesis.

Chapter 4 is concerned with explaining why some countries experienced a stronger eco-

nomic voting effect than others. There is a body of evidence that certain political institutions

afford greater clarity of responsibility than others, making it easier for voters to apportion

blame and so strengthening the economic voting effect in those countries. For example, in the

United Kingdom, the majoritarian electoral system, the weak upper house and the procedural

norms together have the effect that prime ministers are almost always in a position to imple-

ment their preferred legislation without having to negotiate with a number of other actors.

In Germany, by contrast, federal chancellors usually have to contend with coalition partners,

a powerful upper house and a parliamentary committee system not necessarily controlled by

their party. As a result, it is not surprising that British voters are more willing than German

7

voters to hold their governments to account for poor economic conditions. This chapter in-

vestigates whether clarity of responsibility can also explain the variation between countries in

their response to the Great Recession. The only clarity effect that is found is that the economic

vote is stronger in countries where the government is more ideologically cohesive. Even this

clarity effect was depressed during the recession.

Chapter 5 examines which parties benefited from the Great Recession. Earlier chapters

have established the presence of economic voting during that time, meaning that incumbent

parties suffered a loss of support when voters believed the economy would worsen. This raises

the question of which parties those voters turned to and it is this question that this chapter

addresses. It is variously claimed that extreme parties, left-wing parties and Eurosceptic parties

ought to be the beneficiaries of any post-crisis change in party support. This chapter puts

these claims to the test by classifying each of the parties in the study and examining whether

parties in any particular group are more likely than other parties to gain support from voters

pessimistic about the economy. It is found that the biggest beneficiaries of the recession were

Eurosceptic parties and parties further from the centre than the mainstream.

As well as the economic voting behaviours discussed up to this point, there is also the

possibility that voters have chosen to respond to the dire economic situation by withdraw-

ing from voting altogether. Chapter 6 investigates the possibility of an economic abstention

effect. There is an existing body of evidence suggesting that poor economic conditions do

depress turnout but, as with economic voting, there have been very few studies of this effect

in the context of a severe transnational crisis. This chapter tests the hypotheses that there was

an economic abstention effect in all three years of the study and that the size of this effect

was greater during the crisis than it was at other times. Although the economic abstention

effect was found to exist, it was also found that this effect was weaker during the crisis before

becoming stronger than even previously in the years following.

The preceding chapters found that the strongest changes in attitudes and intentions mostly

took place too late to be attributed to the crisis alone, suggesting that the austerity introduced

in response to the crisis may also have played an important role. Chapter 7 tests this hypothesis

by examining attitudes towards the EU and European integration as well as how responsible

voters believe the EU is for the economy. Given the highly visible role of European institutions

in negotiating austerity programmes with several countries as a condition of bailout loans, it

is expected that any anti-austerity sentiment would influence this attitude. It is shown that

8 INTRODUCTION

support for European institutions and further integration fell in the years following the crisis

and voters became increasingly likely to believe that the EU is responsible for the economy.

Finally, Chapter 8 draws these various threads together to offer a coherent explanation of

European electoral behaviour before, during and after the Great Recession. Several possible

explanations for these results are discussed and the merits of each examined. It is argued that

the best explanation for these results is that the electoral response to the austerity measures

imposed in many countries was stronger than the response to the recession itself. It seems

that voters were able to be relatively forgiving of a crisis that spread to Europe from overseas

but they were less forgiving of their governments’ management of the crisis and unwilling

to accept the austerity programmes that many of those governments implemented. This has

implications for economic voting theories, which usually assume that specific policy measures

are of little import to voters as it is only the outcomes that interest them.

Chapter 1

Theory of economic voting: how economic

conditions shape the vote

Few theories in political science have been as successful as the theory of economic voting.

Economic voting effects have been observed time and time again, at numerous times and in

numerous countries. As Raymond Duch puts it, ‘it has now become virtually a social science

law that the economy is one of the most important influences on how individuals vote’ (Duch

2007, 805). Despite, or perhaps because of, this success, there are numerous variant theories

of economic voting. Some argue that economic voters are trying to select the best possible

party to govern them; others that they are simply trying to reward or punish the incumbent

government for its performance. Some theories posit that voters look to the future condition of

the economy; others that they simply react to its past performance. It has been theorised that

economic voters are only concerned with their own situation but it has also been theorised that

they also take into account the conditions of others. Moreover, economic voting suffers from

an instability problem (Lewis-Beck and Paldam 2000). What this means is that measurements

of the strength of the economic vote have a tendency to be very sensitive to both the country

and the time of the measurement, although there is no obvious reason why this should be

the case. Attempts to solve this problem have led to even more formulations of economic

voting theory, such as the idea that certain institutional arrangements are more conducive to

economic voting than others. This chapter reviews these competing theories in the light of

the existing empirical evidence in order to construct the specific economic voting theory that

informs this thesis.

This chapter begins with an overview of economic voting, explaining what it is and briefly

discussing the history of the theory. The next sections examine some of the key disputes in the

literature. The first of these is whether the economic vote should be modelled using objective

economic indicators or voters’ perceptions of the current economic conditions. The second

9

10 CHAPTER 1. THEORY OF ECONOMIC VOTING

Figure 1.1: Basic principle of economic voting

The central idea of economic voting theory is that a strong economy benefits the incumbentelectorally and a weak economy benefits the opposition. This implies that a person’s votechoice is affected differently by the prevailing economic conditions depending on which partiesare currently in power.

is whether economic voters are egocentric, that is to say entirely concerned with their own

situation, or sociotropic, which means at least partially concerned with the well-being of oth-

ers. The next debate is whether economic voting is prospective or retrospective in orientation,

in other words, whether voters are more concerned with the likely future trajectory of the

economy or with its history. Following that, there is a discussion of the process of vote choice

formation, where it is argued that economic voting should be studied in terms of party support

rather than final vote choice. Then there is a discussion of the two basic logics of economic

voting that have been proposed. These are the sanctioning models and the selection models,

which offer different explanations as to what individual economic voters intend to achieve

through their actions. Following this, there is a review of the literature about economic vot-

ing during the Great Recession. It will be shown that there is a need for a cross-national

individual-level study that explicitly compares economic voting during the Great Recession to

that at other times. Finally, it will be shown how these threads fit together to form the cohesive

theory that underpins this work.

1.1 What is economic voting?

The fundamental claim of economic voting theory is that voters are more likely to support

the incumbent government when the economy is performing well and more likely to support

opposition parties when the economy is performing badly. This idea is illustrated in Figure 1.1,

which is intended to be general enough to describe any variant of the theory suitable for a

multiparty system. As the diagram shows, a person’s vote choice is influenced by the current

1.1. WHAT IS ECONOMIC VOTING? 11

economic conditions and, crucially, the precise nature of this influence depends on which

party or parties are currently incumbent. Without this interaction, a strong economy would

always benefit the same party but the theory predicts that a strong economy benefits whichever

party happens to be in power. The term ‘economic vote’ refers to this interaction throughout

this thesis. In a strictly two-party context, such as the United States, this model can be, and

frequently is, simplified further so that the interaction can be eliminated entirely and the model

simply becomes support for the government relative to support for the opposition depends on

the prevailing economic conditions. It will be shown later in this chapter that this is not an

appropriate model for the multiparty systems analysed in this thesis.

Before proceeding to discuss specific aspects of this theory, it is worth briefly discussing

the historical development of the idea of economic voting. Economic voting has a long history

of being studied. Some of the earliest studies were published during the Great Depression,

although the term ‘economic voting’ was not used until much later. Tibbitts (1931), building

upon an even earlier study (Rice 1928, cited in Tibbitts 1931), found evidence that changes

in incumbent vote share at US congressional elections can be partly explained by fluctuations

in economic conditions. This study is unusual for such an early paper in that the ideas it

presents are very similar to economic voting as it understood today. Another early study of

economic voting was conducted by Pearson and Myers (1948), who found that price indices

compared favourably to opinion polls as a predictor of US presidential elections, with high

inflation aiding the challenger and low inflation the incumbent. On the other hand, Wilkinson

and Hart (1950) found almost no correlation between incumbent vote share at presidential

elections and an index of economic activity.

There was little agreement at this stage not only about results but also about which parties

should benefit from good economic conditions. Many studies of the relationship between

electoral behaviour and economic conditions were looking for party-specific effects rather

than the incumbent effects that characterise economic voting and which have come to be

expected today. For example, during Franklin D. Roosevelt’s presidency and motivated by his

depression-era social and economic reforms, there were studies seeking to determine which

groups had the greatest support for Roosevelt, finding that poorer counties (Ogburn and

Coombs 1940) and counties that had been most severely affected by the depression (Gos-

nell and Colman 1940) were most likely to support the president. Other studies anticipated

that voters would turn to the left when economic conditions are poor and to the right when

conditions are good, finding some support for this hypothesis in US election results at US pres-


idential (Kerr 1944) and congressional (Rees et al. 1962) elections. There is little discussion

from this period of economic voting outside the United States and that is not well-developed.

The idea that the fate of the contemporary Conservative government in the United Kingdom

might depend on economic conditions is explored by Durant (1965), the then-director of the

UK Gallup Poll, who quotes newspapers as the source of this idea, but it is not clear whether

this is expected because of that party’s political orientation or because of its incumbency. Even

though these ideas have largely been supplanted by the incumbent-focused theory of eco-

nomic voting, studies of this kind still appear occasionally. Carlsen (2000), for example, ana-

lysed the effect of unemployment and inflation rates on quarterly opinion polls in the US, the

UK, Canada and Australia, arguing that left- and right-wing parties are affected differently by

different kinds of economic adversity.

This debate was reignited when Kramer (1971) published a study introducing multivariate

statistical analysis to this question. This marked a shift away from the basic cross-tabulation

and correlation analyses of earlier studies but Kramer’s work continued in the tradition of ana-

lysing aggregate data sources, namely incumbent vote share and various economic measures

such as unemployment, income and price indices. He found a positive relationship between

real income and incumbent vote share in elections to the US House of Representatives. Shortly

afterwards, Stigler (1973) published a refutation, using similar methods to argue that this re-

lationship actually does not exist and theorising that it ought not exist in any case, as in his

view the two major parties in the United States are equally driven to manage unemployment

and inflation. Arcelus and Meltzer (1975) also found only weak evidence of any link between

aggregate economic conditions and the vote shares of the major US parties. Tufte (1975) on

the other hand found results consistent with Kramer’s, in that US midterm election results

were influenced by economic conditions. In an effort to explain these inconsistent findings,

Bloom and Price (1975) hypothesised that economic conditions affect voter behaviour much

more during times of deprivation than during times of prosperity. Their findings seemed to

confirm this hypothesis and they argued that voters were inclined to punish governments for

poor conditions at US House elections but not to reward them for better conditions. Even this

did not settle the debate, and further research found no important relationship between US

presidential vote share and economic conditions (Fair 1978).

Economic voting soon became a major focus of study within the discipline. Not only did

the literature expand considerably but various competing theories of economic voting arose.

Kinder and Kiewiet (1981) proposed the sociotropic voting theory, which rejects the older

1.2. PERCEPTIONS OR INDICATORS 13

egocentric voting theory and argues that voters are concerned, to at least some degree, with

the conditions of others, not just themselves. Fiorina (1981) brought new attention to Downs’

(1957) idea that economic voting could be explained by a rational model of selection, which

offered an alternative to the then-dominant idea that economic voting was an act of pun-

ishment or reward (Key 1961, 1964, 1966). Since the theory developed in so many different

directions, rather than continuing chronologically, this chapter will proceed thematically. Each

of the key debates is discussed in turn, along with any empirical evidence that sheds light on

the competing approaches and a discussion of how these developments are relevant for this

thesis.

1.2 Perceptions or indicators: the link between the economy and

the vote

Economic voting theory naturally raises questions about the nature of the link between the

prevailing economic conditions and the individual’s vote choice. For example, what sorts

of economic privations are voters sensitive too? Are these all equal? Do voters see these as

separate issues or facets of a cohesive economy issue? The answers to these questions typically

fall into one of two categories. The first category posits that voters respond to material changes

in well-being, which can be measured by standard economic indicators, such as inflation or

unemployment rates. The second category proposes that voters form an overall impression of

the condition of the economy and it is this impression that influences their vote choice. For

example, even if all of the economic indicators for a given year are positive, a particular voter

might be under the impression that the economy has worsened over the past year, leading

that voter to turn against the government. This might be the case if that voter’s personal

experience of the economy was bad or if the sources of information relied upon by that voter

gave a distorted impression of the true situation. There are further related questions, such

as whether voters are exclusively concerned with their personal situation or whether they are

concerned with the economy more broadly and also whether voters are primarily influenced

by their assessment of past economic performance or their expectations of future economic

performance. These related questions are discussed in the following sections. This section

examines the relative merits of the economic indicators and individual perceptions theories of

economic voting and explains why the perceptions approach is preferred in this thesis.

The different approaches offer contrasting benefits. While it is naturally true that voters

can only respond to their own perceptions of the economic situation, if these perceptions are


highly accurate, then an unnecessary link can be eliminated from the theoretical model of

economic voting behaviour. Indicator-based theories also lend themselves well to macro-level

empirical studies, where individual evaluations are not available in any case. Furthermore,

since most countries collect data pertaining to multiple different aspects of the economy, an

indicator-based theory can make empirically testable claims about how voters react to different

kinds of economic privation. For example, it is sometimes argued that voters are more sensitive

to unemployment when the incumbent government is left-leaning and inflation when right-

leaning (Powell and Whitten 1993, 404–405; Whitten and Palmer 1999; van der Brug, van der

Eijk and Franklin 2007, 59). Empirical efforts to pin down these sorts of differences have not,

however, been overwhelmingly successful. For example, Chappell and Veiga (2000) examined

136 elections in thirteen European countries between 1960 and 1997 and were able to find

significant economic voting effects for inflation but not for other kinds of economic problems.

Another study examined quarterly opinion polls from Australia, Canada, the United Kingdom

and the United States, finding that voters were more tolerant of inflation than unemployment

when right-wing governments were in power, but there were no conclusive results for left-

wing governments (Carlsen 2000). A recent study of OECD elections between 1975 and 2013

has also found a difference between patterns of economic voting for left-wing and right-wing

governments but in this case it was found that voters were more tolerant of inflation when left-

wing governments were in power (Bouvet and King 2016, 76). These findings are obviously

contradictory and it may well be that there is no systematic pattern of differences in economic

voting according to the ideology of the governing parties.

On the other hand, if voters’ perceptions of the economy are substantially distorted, then

omitting the perception formation process from the theory will cause problems, which may

explain the inconsistent findings discussed above. Furthermore, given the notorious instability

problem of economic voting studies, it is worth reconsidering some traditional strong assump-

tions, as they may provide some helpful clues. If it is in fact the case that voters’ perceptions

are distorted, then both the link between objective conditions and voters’ perceptions of those

conditions and the link between perceptions and vote choice ought to be studied carefully

in order to understand economic voting. Much therefore hinges on whether voters accurately

perceive economic conditions. This was studied by Paldam and Nannestad (2000), who found

that Danish respondents mostly knew little about the state of the economy, in terms of indicat-

ors like unemployment, inflation and foreign date rates (Paldam and Nannestad 2000). This

is problematic for an indicator-based theory, since if voters do not know how those indicators

1.2. PERCEPTIONS OR INDICATORS 15

are moving, then it is difficult to see how they could condition their votes. In response to

these findings, Sanders (2000) looked at voters’ assessments of the overall condition of the

economy in the United Kingdom between 1974 and 1997. He found that, although voters typ-

ically have little awareness of specific economic facts, their assessments of the general state of

the economy tend to be remarkably accurate.

Taken together, these findings suggest that a perception-based theory of economic voting

is more promising than an indicator-based theory. On the other hand, it has also been found,

in an aggregate study of German electoral behaviour, that hidden unemployment as well as

official unemployment harms support for the governing parties (Feld and Kirchgässner 2000).

It is thus possible that the evident lack of knowledge of official statistics does not represent

mass ignorance of the true state of the economy. Nonetheless, even if it is true that voter

perceptions are in some respects a better reflection of the economic reality than the official

statistics, this just strengthens the argument that published economic indicators are a flawed

tool for understanding the economic vote. It has also been argued that, even though the

average voter may have no incentive to learn about the condition of the economy, there is a

subset of informed voters that does have such an incentive and this subset could be sufficient

to explain the economic voting effect (Aidt 2000). Only a theory that includes perceptions can

properly account for this possibility.

One difficulty with a perception-based theory is that it has been theorised that voters have

a tendency to filter the information that they pay attention to based on their prior beliefs about

the incumbent government. Government supporters are more likely to be receptive to news

that supports the belief that the economy is being managed well than to news challenging that

belief, while the opposite is true for opposition supporters. If this is the case then this produces

a potential endogeneity problem since, according to this model, a citizen’s vote choice both

influences and is influenced by his or her perceptions of the economy. This point has produced

considerable concern in the literature (e.g. Wlezien, Franklin and Twiggs 1997; Evans and

Andersen 2006; Evans and Pickup 2010) and has been particularly forcefully expressed by

van der Brug, van der Eijk and Franklin (2007, 26). The empirical evidence, however, suggests

that this is much less of a problem in practice than it might appear. As part of their detailed

study of the economic vote, Duch and Stevenson (2008, 123–126) compared their key models

to variants specifically designed to account for any endogeneity effects in voters’ economic

perceptions and found no evidence of systematic bias in the results of the naive models. Other

attempts to exogenise economic perceptions have similarly confirmed the existence of the


economic vote (Lewis-Beck, Nadeau and Elias 2008; Lewis-Beck, Stubager and Nadeau 2013;

Hansford and Gomez 2015; Stevenson and Duch 2013, 318).1

1.3 Sociotropic voting: do voters only care about themselves?

A related question is whether voters are concerned with the broader economy or merely their

own personal well-being when they vote economically. As discussed earlier in this chapter,

many of the early studies of economic voting used aggregate data to demonstrate a link

between economic conditions and voter support for the incumbent government. These early

studies typically gave no account of the individual behaviour that gave rise to these large-

scale patterns and it fell to later scholars to produce theories to explain these trends. A com-

mon assumption was that individuals respond to changes in their own personal conditions.

Thus, if many people become worse off, both the government’s vote share and the national

measures of economic well-being decline simultaneously. Conversely, if many people become

better off, these indicators improve together. This is what is meant by egocentric (or pock-

etbook) voting—voters consider only their own personal circumstances, or perhaps those of

their friends and family. Although these early studies rarely made explicit an assumption of

egocentric voting, it tends to be apparent in their models. For example, Tufte (1978, 127–134)

breaks down the vote at the US presidential elections of 1968, 1972 and 1976 by income and

family financial situation. Similarly, Bloom and Price (1975, 1243) produced a model in which

the short-term variations in voting behaviour are primarily explained by short-term changes in

real income. In their influential book The American Voter, Campbell et al. (1960) also make the

claim that political behaviour, at least in some spheres, can be explained in terms of ‘primitive

self-interest’ (205). Again, their analysis focuses on the way that voting behaviour changes in

response to changes in personal economic conditions (381–401).

Although egocentric voting was widely assumed, the theory was rarely put to the test.

When Fiorina (1978) did test the egocentric voting hypothesis using individual-level survey

data, he failed to find support for it. Similarly, in a study of voting in US congressional elec-

tions between 1956 and 1976, Kinder and Kiewiet (1979, 521) found little support for the

idea that an individual’s vote choice was materially affected by any economic troubles that

individual had personally experienced. In order to explain this, and in light of the evidence

1It should also be noted that it is not the purpose of this thesis to establish the absolute size of the economicvote but rather to determine whether economic voting was different during the Great Recession. The substantiveresults of this thesis should thus be unaffected by any inflation in the estimates of the size of the economic voteowing to an endogeneity effect, unless that effect was itself affected by the recession.

1.3. SOCIOTROPIC VOTING: DO VOTERS ONLY CARE ABOUT THEMSELVES? 17

that a government’s fortunes are tied to the economic conditions it presides over, they the-

orises that voters respond more to national economic conditions than to their own personal

conditions (524). That is, voters do hold governments to account for the state of the economy

but not necessarily for the individual economic events in their own lives. A later study found

further evidence for what they refer to as sociotropic voting (Kinder and Kiewiet 1981). Using

a panel study of voters at the 1972, 1974 and 1976 US elections (congressional and, other than

in 1974, presidential), they once again tested the egocentric voting hypothesis. Again, they

found that sociotropic voting better explains their observations than egocentric voting. Al-

though they note that presidential elections have often provided more support for egocentric

voting than congressional elections (147), they conclude that ‘for presidential voting even

more completely than for congressional voting, citizens’ assessments of national economic

conditions—their sociotropic judgements—overwhelm economic grievances encountered in

private life’ (148).

Although it is used here in contrast with the term egocentric, the term sociotropic does not

simply mean altruistic. That would suggest that sociotropic voters act purely in the interests

of others and pay no heed to their own interests. Rather, the term was coined to mean ‘taking

some account—we needn’t say exactly how much—of other persons’ interests or, if you like,

of the collective’s interest’ (Meehl 1977, 14). So, both theories accept that voters heed their

own interests to some degree. Sociotropic voters do not necessarily ignore their own interests

but they do give some consideration to the interests of others. The key difference between

egocentric and sociotropic theories of voting is how the individual voter behaves. According

to the egocentric theories, a voter will react to changes in his or her personal conditions. So a

voter who loses his or job, for example, will be more likely to vote against the government than

one who keeps his or her job, irrespective of the unemployment rate. The sociotropic theories,

on the other hand, predict that the voter who just lost his or her job is not much more likely

than anyone else to vote against the government but if the unemployment rate goes up then

this will mobilise voters, employed or otherwise, to vote against the government.

The sociotropic theory of economic voting was criticised by Kramer (1983), who argued

that the evidence offered in support of this theory was insufficient to reject the null hypothesis

of egocentric voting. In his view, this evidence is merely a ‘statistical artifact’ (93). Despite

these early objections, sociotropic voting has become the consensus position in the literature

as the evidence has mounted (Nannestad and Paldam 1994). This consensus appeared to

be under threat when Nannestad and Paldam (1995, 1997a) published two studies finding a


stronger egocentric than sociotropic economic voting effect in Denmark. This was so unex-

pected that this result has been described as ‘virtually unique in the literature’ (Lewis-Beck,

Stubager and Nadeau 2013, 501). Even these findings have been disputed on methodological

grounds (Hibbs 1993, 62–63) and several later attempts to replicate this results have reached

the contrary conclusion (Borre 1997; Lewis-Beck, Stubager and Nadeau 2013; Stubager et al.

2014). In light of this, it now appears that Denmark is no exception to the usual pattern of

sociotropic voting.

Finally, there is also some evidence that the centre of focus for the economic vote may

lie somewhere between the individual and the entire nation (Rogers 2014). This idea that

voters are ‘communotropic’—mostly concerned with the economic conditions of their own

communities—is interesting but has not yet been thoroughly tested in its own right, unlike

the sociotropic theory. Furthermore, this approach would pose measurement difficulties, since

communotropic economic assessment questions are not currently common in survey instru-

ments. Given the considerable body of evidence that has been accumulated in support of the

sociotropic hypothesis, this thesis uses a sociotropic theory of economic voting.

1.4 Prospective and retrospective voting

Retrospective voting and prospective voting are the ideas that voters base their vote choice

on an assessment of the incumbent government’s past performance or an expectation of the

government’s likely future performance respectively. In many respects, the prospective voting

hypothesis is a more natural one than the retrospective voting hypothesis. A rational voter,

after all, ought to be more concerned with the future, which can still be influenced, than the

past, which cannot. Most retrospective studies do not give an explicit rationale for the decision

to use that approach but among those that do, there appear to be two main justifications.

The first is the idea that voters are not trying to rationally select the party that will pro-

duce the most favourable outcomes but rather to simply reward or punish the incumbent for

its successes or failures. The relative merits of the selection and sanctioning models of eco-

nomic voting are discussed later in this chapter. For now, it suffices to say that the sanctioning

model is not accepted by this thesis. The second justification for a retrospective theory is that,

while the rational voter is indeed trying to select the party most likely to deliver the desired

outcomes, past performance is simply a cheaper and more accessible means of assessing this

than the alternatives (Downs 1957, 38–40). It is argued that, while past performance is only

a rough measure of a party’s likely future performance, it is not worth the time and effort

1.4. PROSPECTIVE AND RETROSPECTIVE VOTING 19

for most voters to find and synthesise other sources of information to produce a better estim-

ate. While this argument has merit, it is not actually necessary to enquire into the process

of prospective assessment formation in order to understand economic voting. Why use the

retrospective measure as a proxy for the prospective measure, when the latter could simply be

measured directly? It should also not be ignored that citizens’ retrospective and prospective

assessments of the economy are often not particularly strongly correlated (Kuklinski and West

1981; Conover, Feldman and Knight 1987).

Efforts have been made to compare the prospective and retrospective theories empirically.

MacKuen, Erikson and Stimson (1992) made a distinction between ‘peasants’—retrospective

voters whose expectations of the trajectory of the economy are simple extrapolations of its

recent performance—and ‘bankers’—prospective voters whose expectations are formed by a

more sophisticated analysis. Using survey data, they attempted to determine which arche-

type more accurately described US voters, finding that the banker model was more accurate

than the peasant model. In other words, their findings supported by the prospective theory of

economic voting. They confirmed these findings in a follow up study, further noting that retro-

spective assessments do seem to influence future expectations to a degree but it is ultimately

the future expectations that predict vote choice (Erikson, MacKuen and Stimson 2000). Recent

studies have continued to find empirical support for prospective economic voting. Michelitch

et al. (2012) found prospective economic assessment to be an important predictor of vote

choice at both the US and Ghanaian presidential elections of 2008. The fact that they were

able to make the same findings in both the US and Ghana, two very different countries, is

promising, as it makes it unlikely that prospective voting is merely a quirk of American polit-

ics.

A further problem for retrospective voting is that there is evidence that voters’ retrospect-

ive assessments lack some of the qualities that would be expected from rational retrospective

voters. Using survey data collected from 40 countries between 1996 and 2005, Stanig (2013)

has shown that several psychological biases are in play when voters form retrospective as-

sessments of the economy. For example, voters are more likely to change their assessments

in response to worsening conditions than to improving conditions (736–737). Furthermore,

these assessments are also affected by partisanship and ideology, with supporters of govern-

ment parties typically less likely to be critical than opponents (737–739). Strangely, however,

this difference between supporters and opponents of the incumbent government is consider-

ably weaker during a recession than at other times (739–740). It has to be acknowledged that


prospective assessments may also suffer from some of these problems as well. Unfortunately,

there is little evidence either way, as most published research has focused on the retrospective

assessment.

A related question that has attracted recent scholarly attention is: how far back do ret-

rospective voters look? Hellwig and Marinova (2015) surveyed voters during the 2012 US

presidential election, finding that short-term and long-term retrospective assessments of the

economy were approximately equally accurate, which is to say, not particularly accurate at

all. They also found that neither the short-term nor the long-term assessment was a clearly

superior predictor of vote choice over the other. Based on this finding, they argue that the

almost universal choice of a twelve-month time window in retrospective voting studies is no

worse than any other choice (884). Wlezien (2015), on the other hand, found that voters at

US presidential elections between 1952 and 2012 appeared to look back approximately two

years when forming the retrospective assessment that would influence their vote choice. This

may well be a quirk of US elections specifically, given that partial legislative elections take

place two years before every presidential election. These findings may indicate that voters are

simply reflecting back on the period since the last time a federal election of some kind was

held. Finally, Taniguchi (2016) studied retrospective voting during the 2013 Japanese upper

house election. Based on a survey of Japanese voters, he argues that voters actually look to

the long-term, slowly adjusting their perceptions over time. Unfortunately, no consensus has

emerged from these studies, nor do these results shed any light on the time frames relevant

to prospective voters. Nevertheless, this thesis is limited to the questions actually asked in

the surveys used for its analysis. As will be seen in the next chapter, this means that the time

window has to be either the year just past or the coming year, at the time of the survey.

This thesis is based on a prospective theory of economic voting. Given that retrospective

voting tends to be the default choice for economic voting studies, this may be a slightly sur-

prising choice. The theoretical motivations discussed here are genuine—even if voters do use

retrospective evaluations as a shortcut for forming prospective assessments, as Downs (1957)

claims, it is not at all obvious that this means that those retrospective evaluations should be

used as a proxy for the prospective assessments when the means are available to measure

the latter directly. Nonetheless, it must also be acknowledged that there are more practical

motivations for pursuing a prospective rather than a retrospective study. As will be shown

in Chapter 2, there was something of a consensus among European voters in 2009 that the

economy had become worse over the previous year. This lack of variance in a key independ-

1.5. THE VOTE CHOICE PROCESS 21

ent variable would have posed serious problems for statistical analysis, so the retrospective

approach was not viable. Voters did, however, have a variety of opinions about the future

course of the economy, so a prospective analysis was possible. In any event, the fact that

the retrospective approach broke down during the Great Recession, at least for the data used

by this thesis, while the prospective approach remained viable, offers empirical support for

the argument that economic voting is fundamentally a prospective phenomenon, regardless

of how convenient retrospective evaluations may often be as a means of quickly forming a

prospective assessment.

1.5 The vote choice process

The preceding sections have discussed the nature of the link between the economy and an

individual’s vote choice but the mechanism by which this vote choice is made was left un-

specified. Most early studies of economic voting tended to use aggregate data, so there was

no need for a specific theory of individual vote choice and no means of testing such a theory

anyway. These studies simply looked for relationships between the incumbent government’s

vote share2 and various economic indicators. A good example of this genre is Pearson and

Myers (1948), showed that the incumbent party tended to win US presidential elections if

and only if the economy was strong. Aggregate studies still take place today—Carlsen (2000),

for example, who showed that government vote share in four countries was linked to the

unemployment rate—but they cannot shed light on the vote choice mechanism.

The simplest plausible vote choice mechanism is one whereby a voter simply weighs up the

strengths and weaknesses of the incumbent government and votes to return the government

should its overall performance exceed some (possibly idiosyncratic) threshold and votes to

reject the government otherwise. This naive picture of the vote choice process seems perfectly

logical in the context of a strict two-party system, such as the United States. This approach

tends to be assumed by individual-level economic voting studies. Given that the vast bulk of

such studies have investigated US politics specifically, this may well explain the near univer-

sality of this approach. This approach is typically modelled using logistic or probit regression

on an individual’s reported vote choice or intended vote choice. This is not only methodolo-

gically straightforward but also a natural operationalisation of this theoretical model of vote2Some early US studies looked at the vote share for the Republicans or the Democrats rather than the in-

cumbent party as such. This is because the link between economic conditions and support for the incumbentgovernment had not yet been decisively established, so these papers were testing theories about certain economicoutcomes favouring specific parties. For example, Kerr (1944) found a correlation between the conservative voteand several economic indicators in the United States.


choice, which may further explain the enduring popularity of this approach. An example of

this approach is Nadeau and Lewis-Beck (2001), who predicted the likelihood of US voters

supporting the incumbent party’s presidential candidate under various scenarios. This theory

of vote choice works well in the US context.

Many countries are not two-party systems, however, and certainly most European coun-

tries have more than two parties that typically receive a non-trivial proportion of the vote. This

extra complexity can be handled in several ways. The simplest way to generalise the naive vote

choice model to a multiparty system is to ignore parties altogether and recast the vote choice

task as voters simply making a decision to support or reject the government. Thus a vote for

any coalition party is treated the same and a vote for any opposition party is treated the same.

One example of this approach is Lewis-Beck (1986), who used various economic variables to

predict the likelihood of voters supporting any of the incumbent government parties as op-

posed to any of the opposition parties at elections in France, Germany, Italy and the United

Kingdom. Another example is Nadeau, Niemi and Yoshinaka (2002), who modelled this di-

chotomous vote choice in eight European countries. Interestingly, they found different results

in the majoritarian and consensual systems that they studied (419). This could be an artefact

of the forced dichotomisation of the vote choice, since majoritarian systems tend to have fewer

relevant parties than consensual systems. This method heavily simplifies a complex decision.

If voters really are primarily concerned with the choice between supporting and rejecting the

government, then it is not clear why such complex party systems are sustainable. In that case,

it would be expected that voters would simply support the largest government or opposition

party and the remaining parties would be starved of votes. Yet this is not the case.

An explicit theory of vote choice formation was proposed by Downs (1957, 36–50) and

elaborated by van der Brug, van der Eijk and Franklin (2007, 31–53). According to this theory,

vote choice is a two-stage process. The first stage is one of assessing party utility. For each

viable party, voters estimate by some method the utility they would derive from the success

of that party. This estimated party utility will be referred to as party support throughout this

thesis.3 The second stage of this process is where voters actually select which party will receive

their votes. This is assumed to be done by the obvious method of selecting the party that has

the greatest estimated utility to the individual voter, or in other words, each voter selects the

party that he or she supports most. The interesting stage is thus the first stage, in which

these levels of support are established in the first place. It is important to note that a voter’s

3This follows the usage of van der Brug, van der Eijk and Franklin (2007).

1.5. THE VOTE CHOICE PROCESS 23

levels of support for the various parties do not merely form an ordering between the parties

but also indicate the degree to which one party is preferred over another. This is important

because one of the key implications of this theory is that a particular event might cause two

different voters to adjust their levels of support for a particular parties in the same direction

and by precisely the same amount and yet lead to the two voters changing their vote choices

in different ways or to only one voter changing his or her vote choice while the other does not.

The final outcome depends not just on the immediate effect of the event on party support but

also on each individual’s levels of support for the other parties in the system. It is argued that,

by failing to capture this complexity, the dichotomous approaches discussed above invite or at

least exacerbate the instability that has long plagued the study of economic voting (14).

A further method that is sometimes used to predict vote choice in empirical studies is to

generalise the two-party logistic regression model to multiple parties using techniques such

as pairwise or ordered logistic regression. For example, Duch and Stevenson (2008, 42–52)

measure what they call the ‘general economic vote’—the total variation in vote choice that

can be explained by the economy—by using a multinomial logistic regression model to predict

which of several parties would be the respondent’s vote choice. Although this is done without

giving a detailed theory of vote choice, it is compatible with the party support theory outlined

above. This method also has the advantage of taking into account the full complexity of the

party system, unlike the dichotomous method. Nonetheless, the full party support model

offers two key advantages over this approach. The first advantage is that, by modelling party

support directly, it is possible to use much richer data, since a full vector of party support

levels encodes much more information that a single categorical vote choice item. The second

advantage is that the party support model allows important information about the party, such

as its incumbency status, to be treated as independent variables, which is not possible using

the ordered logistic regression method. This is particularly important for a comparative study

such as this. These methodological details are the subject of the following chapter but they are

worth mention here because they highlight the importance of an explicit vote choice theory.

There is some empirical evidence that economic voters in multiparty systems do operate

according to a more complex logic than a simple government–opposition dichotomy could

describe. For example, it has been found that unionist voters in Northern Ireland in 2011

tended to base their party choice not just on the performance of the government but also on

the perceived influence of each party within the governing coalition (Garry 2014). This latter

aspect can be explained by the party support theory but not by the dichotomous vote choice


theory. It has also been shown that partisans of particular coalition parties tend to believe that

the political position of the entire coalition is closer to the position of their preferred party

than other voters do (Meyer and Strobl 2016). Furthermore, it has been found that economic

voting effects are stronger for the head of government’s party than for other coalition partners

(Fisher and Hobolt 2010; Debus, Stegmaier and Tosun 2014). It has also been established that

the vote choice decision becomes more difficult for voters to make as the number of parties

increases (Orriols and Martínez 2014). These results all suggest that voters think about parties

rather than blocs and that the party rather than the coalition is thus the natural unit of analysis.

Consequently, the party support theory of vote choice forms an important piece of the overall

economic voting theory tested by this thesis.

1.6 Reward and punishment: the logic of sanction

So far this chapter has discussed various individual aspects of economic voting but there has

not yet been a discussion of the underlying logic of the phenomenon. What motivates people

to vote economically and what do they hope to achieve in doing so? This section introduces

the sanctioning model of economic voting, which was the first serious effort to answer these

questions. It is not, however, the only possibility and the competing selection model is dis-

cussed in the following section. The sanctioning, or reward–punishment, model posits that

voters seek to reward governments for good performance and punish them for bad perform-

ance. This would explain why voters are more likely to support governments when economic

conditions are good than when they are bad, which is the central observation of economic

voting. This theory is most strongly associated with V. O. Key, who described it thus:

The patterns of flow of the major streams of shifting voters graphically reflect the

electorate in its great, and perhaps principal, role as an appraiser of past events,

past performance, and past actions. It judges retrospectively; it commands pro-

spectively only insofar as it expresses either approval or disapproval of that which

has happened before. Voters may reject what they have known; or they may ap-

prove what they have known. They are not likely to be attracted in great numbers

by promises of the novel or unknown. Once innovation has occurred they may

embrace it, even though they would have, earlier, hesitated to venture forth to

welcome it. (Key 1966, 61)

1.6. REWARD AND PUNISHMENT: THE LOGIC OF SANCTION 25

In effect, he argues that election results must not be read as an endorsement of the policies

of the winning party, but as a judgement of the competence of the previous government (Key

1961, 473–474; 1964, 543; 1966, 51–52).

The most striking characteristic of this model is that it asserts that voters are outcome-

driven rather than policy-driven. What this means is that citizens do not compare the policy

packages put forward by the various parties in order to ascertain which is the best offering. It is

assumed that they not have the resources to do this effectively (Fiorina 1981, 45). Accurately

weighing the different claims as to which course of action will most benefit the economy4 is

likely to be a demanding task for even highly educated voters. Even reading all of the party

manifestos consumes more time than many people are likely to be willing to spend. On the

other hand, most citizens are likely to have some idea of how well the economy has been

functioning. Recessions do not tend to go unnoticed after all. In other words, comparative

policy assessments are difficult to come by but assessments of government performance—

assuming a link between economic outcomes and government policy—are readily accessible

(9). The sanctioning voter, according to this line of thinking, thus makes the best use of the

available information.

A second key characteristic of the sanctioning model is that it is explicitly retrospective

in orientation. According to this theory, voters are not merely looking at past performance

in order to inform their future expectations. A vote is actually an expression of the voter’s

assessment of the government’s past performance. Key (1961, 473) describes the public as

‘speak[ing] in disapprobation of the past policy or performance of an administration’. Else-

where he claims that ‘[r]etrospective judgments by the electorate seem far more explicit than

do its instructions for future action’ (Key 1964, 643). Thus according to this model of eco-

nomic voting, citizens use their vote to signal which outcomes they deem acceptable and which

they do not. For example, even if a voter expected that the economy would improve from a

recent crisis under the incumbent government, the sanctioning model predicts that this voter

would vote against that incumbent anyway. Doing otherwise would send the wrong message.

As Duch and Stevenson (2008, 11) put it, politicians ‘anticipate that voters will sanction them

if they underperform. And, to maintain the credibility of this threat, voters punish incumbents

at the polls when retrospective economic performance is substandard.’

This illustrates the vote choice logic at the heart of the sanctioning model and it is a rational

one. Although unremarkable today, this was an unexpected position to take in the wake of4Voter utility is certainly not limited to economic issues and there is no reason why a sanctioning voter would

not treat failures in other areas equally harshly but non-economic issues are beyond the scope of this thesis, so thisanalysis will assume that the issues in question are economic ones.


highly influential work such as Campbell et al. (1960), which had argued that vote choice was

driven primarily by party identification and not rationality. Key (1966, 7) admitted as much

when he described his own argument as ‘perverse and unorthodox’ before adding: ‘To be sure,

many individual voters act in odd ways indeed; yet in the large the electorate behaves about as

rationally and responsibly as we should expect, given the clarity of the alternatives presented

to it and the character of the information available to it.’ Others have produced formal models

of rational individual behaviour which would produce an aggregate retrospective voting effect

of the sanctioning kind. The basic idea is that voters must punish unsatisfactory outcomes in

order to incentivise incumbents to act in the public interest (Barro 1973; Ferejohn 1986). It

is worth noting that, for this to be effective, voters must even punish governments for poor

economic outcomes even if they are personally unaffected. This has led Ferejohn (1986, 23)

to argue that sociotropic voting is an essential feature of retrospective voting (23).

The sanctioning model is appealing for its simplicity. It is also important to understand

because it underpins so much of the early economic voting research. Nonetheless, it is not

the model that will be used by this thesis. For one thing, as has already been discussed,

this thesis is testing a prospective theory of economic voting, while the sanctioning model

assumes a thoroughly retrospective orientation on the part of the voter. The empirical evidence

that prospective voting exists (for example Erikson, MacKuen and Stimson 2000; Michelitch

et al. 2012) cannot be explained by the sanctioning model. The next section introduces an

alternative model, the selection model.

1.7 Competent government: the logic of selection

The selection model offers an alternative explanation for the phenomenon of economic vot-

ing. The differences between the two models stem from differing assumptions about what

voters intend when casting their votes. Whereas the sanctioning model asserts that voters are

primarily casting a judgement upon the incumbent government, the selection model posits

that voters are actually trying to select the party most likely to maximise their utility. For

economic voters in particular, this means selecting the party which can most reasonably be

expected to deliver positive economic outcomes. This model stems from Downs’ An Economic

Theory of Democracy, in which he explains the basic challenge of voting as one of selection:

When a man votes, he is helping to select the government which will govern him

during the coming election period (i.e., period t + 1). Therefore as we have just

1.7. COMPETENT GOVERNMENT: THE LOGIC OF SELECTION 27

shown, he makes his decision by comparing future performances he expects from

the competing parties. But if he is rational, he knows that no party will be able to

do everything that it says it will do. Hence he cannot merely compare platforms;

instead he must estimate in his own mind what the parties would actually do were

they in power. (Downs 1957, 39)

The point being made here is that if voters are assumed to behave rationally and to be mo-

tivated by selection, rather than sanction, then voter behaviour can only be understood by

knowing how voters measure the expected utility of each political party. Furthermore, this is

not simply a question of comparing party manifestos because they are not reliable indicators of

behaviour in office. This differs greatly from the sanctioning model, which argues that voters

do not think about a party’s future actions but only about its past performance.

According to the selection model, voters determine the expected utility associated with

each party by looking at their past performance. It is argued that this is precisely what a

rational voter would do:

Since one of the competing parties is already in power, its performance in period

t gives him the best possible idea of what it will do in the future, assuming its

policies have some continuity. But it would be irrational to compare the current

performance of one party with the expected future performance of another. For a

valid comparison, both performances must take place under the same conditions,

i.e., in the same time period. Therefore the voter must weigh the performance

that the opposition party would have produced in period t if it had been in power.

(39–40)

In other words, voters compare the incumbent party’s current performance—using their retro-

spective knowledge—to the hypothetical performance of an opposition party. It is important

to note that these assessments are not based purely upon retrospective knowledge. Downs

(1957, 41–45) explains that these assessments are modified by what he calls the ‘trend factor’

and ‘performance ratings’. The trend factor refers to the consideration that voters give to

current events and is the device used to project retrospective knowledge into the future. For

example, if it is clear that the economy is on the verge of a recession, voters are not going

to ignore this simply because the economy had been otherwise well-managed up until this

point. Performance ratings are only used to discriminate between two parties proposing in-

distinguishable platforms. In this case, voters can only contrast the past effectiveness of the

parties.


The model just described was not introduced as a freestanding theory but rather formed

part of Downs’ comprehensive theory of political behaviour. It was Fiorina (1981) who de-

veloped these ideas further and presented them as a coherent alternative to the sanctioning

model. His description of the model is succinct: ‘In deciding how to vote, the rational citizen

would compare the performance of the incumbent administration to the platform promises

of the challenger rather than compare both sets of platform promises’ (46). One key differ-

ence between the sanctioning and selection models is that the latter explicitly assumes that

a party’s past behaviour is a reliable predictor of its future behaviour (Downs 1957, 96–113;

Fiorina 1981, 45), whereas the logic of sanctioning implies that voters expect to be able to

condition the parties to behave as intended. If the parties did not respond to such cues, then

there would be no reason to punish or reward them electorally. Another difference between

the models is that the sanctioning model assumes that voters typically cannot compare policy

platforms, whereas that selection model assumes that this can be done in principle but that

it is too expensive, so retrospective information is a cheap and effective substitute (Fiorina

1981, 45–47).

The selection model has been a popular choice for empirical research into economic voting

beginning with Kramer (1971, 134), who argued that the ‘past performance of the incumbent

party in particular gives some indication of what it would do if returned to office, and of the

effectiveness of its policies and personnel’. Even his critic Stigler (1973, 165) agreed that a

rational voter should predict future performance based on past performance. The selection

model continues to inform recent work, such as Duch and Stevenson (2008). One of the

strengths of this theory is that selection effects have proven capable of explaining why in-

cumbent legislators can be so difficult to unseat (Ashworth 2005; Ashworth and de Mesquita

2008; Gowrisankaran, Mitchell and Moro 2008). The argument is that voters tend to return

high quality legislators and replace low quality ones, so the median incumbent tends to be

of a higher quality than the median candidate. This effect cannot be explained by the sanc-

tioning model. It should be mentioned that there are some recent experiment-based studies

finding support for a sanctioning effect (Huber, Hill and Lenz 2012; Woon 2012). Nonethe-

less, the evidence for selection effects is strong. It is also possible that the two effects exist

simultaneously (Fearon 1999, 56–57).

This thesis is informed by the selection model. This may be surprising, since the selection

model is a retrospective theory and it is has been stated that the thesis is operating from a pro-

spective perspective. These approaches are not contradictory, however. The selection model

1.8. POLITICAL CONTEXT AND THE INSTABILITY PROBLEM 29

assumes that voters have a prospective orientation, in that they seek to select the party most

likely to produce the best outcomes. What makes it a retrospective theory is that it posits that

voters use retrospective knowledge as a shortcut for forming their view as to which party this

is. As mentioned earlier, Downs expects voters to augment their retrospective evaluations with

a trend factor, reflecting their assumptions about the future course of present events. Given

this framework, it is not unreasonable to interpret a voter’s reported prospective economic as-

sessment as a projection of his or her retrospective economic assessment using a trend factor.

The prospective interpretation of the selection model used in this thesis thus predicts that the

retrospective and prospective economic evaluations are typically moderately correlated but

that the prospective evaluations are the stronger predictor of party support.

1.8 Political context and the instability problem

An enduring problem for economic voting research has been the instability of its findings.

It has frequently been the case that promising efforts to measure the economic voting effect

have produced wildly different estimates in different countries and sometimes even the same

countries at different times. Explaining this instability has increasingly become a focus for

research in this area (Lewis-Beck and Paldam 2000, 113–114; Dorussen and Palmer 2002, 1–

5). This instability is particularly salient for a comparative study such as this one because it is

clearly not reasonable to assume that the economic vote is the same for all of the different

countries under study or even that it is stable across time. An understanding of political

context is thus important and this section reviews what is already known about the impact

of context on the economic vote. Non-contextual explanations for the instability problem

have also been proposed. For example, van der Brug, van der Eijk and Franklin (2007, 16)

argue that model misspecification is the principle cause of this instability5 but even if true,

this does not mean that context does not also play a role. Voter heterogeneity has also been

proposed as a possible cause of the instability problem (Dorussen and Palmer 2002, 7) but

these possibilities are beyond the scope of this thesis.6 One way to take into account contextual

differences is simply to model them without trying to explain them. For example, if the model

allows for the possibility that, say, Poland and the Czech Republic have different levels of

5In particular, they argue that the instability results from the failure to take into account party competition.Their recommendation is to model party support explicitly rather than just vote choice and this is the approachtaken in this thesis, as was discussed earlier in this chapter.

6As well as adding considerable extra complexity and consuming degrees of freedom, the required data is notreadily available: ‘Information about relevant sources of voter heterogeneity is much more sparse, basically onlyavailable for a few countries and measured at irregular intervals’ (Dorussen 2002, 309).


economic voting then this can be accounted for. This is the method used by Chapter 3. Of

course this is merely an empirical fix and does not solve the theoretical problem, as it sheds

no extra light on these differences.

A more sophisticated approach requires an understanding of the causes of the variation

between political contexts. The most influential theory of contextual variation in the economic

vote is the theory of clarity of responsibility, which was introduced to the literature by Powell

and Whitten (1993) and further developed by Whitten and Palmer (1999). According to this

theory, economic voting is more likely to take place in contexts where voters can reasonably

hold a particular incumbent responsible for the condition of the economy. While this can be

reasonably done in a country like the United Kingdom, where there is very little to hinder

prime ministers from taking whatever course of action they feel is required, this is not neces-

sarily the case everywhere. For example, the German chancellor frequently has to negotiate

policies with coalition partners as well as gain the support of the upper house if certain forms

of legislation are required. Furthermore, certain policies may be beyond the jurisdiction of the

federal government and require the support of the states to be implemented. This implies that

it is less clear in Germany which government and who within the government, if anyone at all,

ought to be held responsible for any unwelcome economic events. Accordingly, the argument

runs, German voters ought to be less inclined to vote economically than their British coun-

terparts. Much of the work in this area has focused on precisely how clarity of responsibility

should be measured. In its original conception, the key variables were the voting cohesion of

the incumbent parties, the strength and inclusiveness of the parliamentary committee system,

opposition in any constitutionally significant upper house, and the presence of a minority or

coalition government (Powell and Whitten 1993). Alternative measures have been proposed

but these are discussed in Chapter 4, where a clarity of responsibility model is constructed.

An alternative contextual theory of economic voting has been developed by Duch and

Stevenson (2008). Basing their model on the rational expectations literature, and particularly

Alesina and Rosenthal (1995), they distinguish between two types of economic growth shocks.

These are competency shocks, which derive from the actions of the incumbent government,

and exogenous or nonpolitical shocks, which do not (Duch and Stevenson 2008, 132–133).

The strength of the economic vote depends on the proportion of shocks which are compet-

ency shocks (138). Duch and Stevenson (2008, 147) argue that this proportion is related to

the mix of ‘electorally dependent’ and ‘nonelectorally dependent’ economic decision makers.

Electorally dependent decision makers include both elected officers themselves and also the

1.9. THE GREAT RECESSION 31

government agencies that answer to them. Nonelectorally dependent decision makers are

‘everyone else whose decisions might impact the economy, including individuals, firms, in-

terest groups, nonelectorally dependent (entrenched) bureaucrats, foreign leaders, the WTO,

and many more’ (139–140). The argument is that in contexts where economic decisions are

largely driven by elected officials or bureaucrats responsible to them, a high proportion of

economic shocks will be competency shocks and the economic vote will be correspondingly

high. By contrast, in contexts where many economic decisions are made by statutory bodies or

international agencies or are otherwise outside of the control of the incumbent government,

the economic vote will be considerably lower. There are some similarities with the clarity of

responsibility theory, in that both theories relate the strength of the economic vote to polit-

ical control but there are importance differences. Clarity of responsibility theory is concerned

with the degree to which the dominant government party has to negotiate with other parties

to implement government policy, whereas this theory is concerned with the degree to which

economic policy is influenced by non-political actors. Although this is a promising theory, it is

not used by this thesis, mainly because it is based on some of the opposite assumptions from

those made here. For example, theirs is a retrospective vote choice theory, whereas this thesis

is based on a prospective party support theory.

1.9 The Great Recession

The central question motivating this thesis is whether the Great Recession affected the eco-

nomic vote in the European Union. Of course, classic economic voting theory already predicts

a strong electoral response, since the recession was both widespread and severe. The ques-

tion is whether the strength of the economic vote exceeded this expectation. There are two

threads in the literature that suggest that this might be the case. The first thread is the idea of

grievance asymmetry, which claims that the response to a good event is not necessarily of the

same magnitude as the response to a bad event. It is typically argued that the response to a

negative stimulus is stronger than the response to a positive one. This argument appeared as

early as Campbell et al. (1960, 555), who wrote:

Compare the change wrought in party fortunes in either of these cases [the panic

of 1893 and the Great Depression]with the change occurring when a party already

in office witnesses a surge of economic prosperity. This prosperity clearly benefits

the administration party, but it has nothing like the magnitude of the effect that


would result from economic distress. A party already in power is rewarded much

less for good times than it is punished for bad times.

This idea was tested by Bloom and Price (1975), who extended previous economic voting

models to account for this possibility. In an aggregate study of US House of Representat-

ives elections, they found that recessions were associated with reduced incumbent support

but that the opposite was not true for economic recoveries. Similar results have occasionally

been found elsewhere. For example, a grievance asymmetry was also found at Danish elec-

tions between 1985–92 using an individual-level model (Nannestad and Paldam 1997b) and

a study of Hungarian voters in 1997 found a similar effect (Duch 2001). Looking beyond

studies of economic voting specifically, it has also been found that the media and public opin-

ion response to negative economic news is stronger than the response to positive economic

developments (Soroka 2006). What these findings suggest is that economic voting is weaker

than expected during times of prosperity. It is not therefore too much of a stretch to hypothes-

ise that economic voting might be stronger than expected during a recession of exceptional

severity.

The second relevant thread in the literature pertains to the salience of economic issues.

The argument is that voters are more likely to vote economically when the economy is a sa-

lient issue for some reason and that the salience of economic issues is linked to economic

performance. Consequently, economic voting is expected to recede when the economy is per-

forming satisfactorily. Duch and Stevenson (2008, 171) summarise the argument thus: ‘when

the economy is in equilibrium, or possibly even out of equilibrium but in a positive direc-

tion, economic evaluations are likely to play a much less important role in the vote decision.’

This idea has some empirical support, with survey data from the 2008 US presidential elec-

tion (Singer 2011a) and cross-national survey data from 38 countries in the years 2001–2006

(Singer 2011b) showing that the salience of the economy is greater in contexts where eco-

nomic conditions are poor and among individuals who are personally affected by economic

privations. A follow-up study using aggregated survey data from elections in 43 countries

in the years 2001–2011 found further evidence that economic issues tend to be most salient

when the economy is less healthy (Singer 2013). It is interesting that the implications of the

asymmetry and salience arguments are so similar despite the differences in logic. Whether it

results from qualitative differences in the way people respond to different kinds of informa-

tion or from the fluctuating salience of economic issues, these literatures predict a stronger

economic vote during bad times than good times. This leaves open the possibility that a severe


global recession ought to produce a still stronger economic vote, beyond that which would be

predicted by classical economic voting theory.

The handful of comparative studies that have been published are largely based on aggreg-

ate data, which is probably because comparable aggregate data is more quickly available than

cross-national survey data. Several studies investigated the relationship between government

vote share and economic indicators. One study of EU countries in 2008–2011, which also

included aggregated perceptions data, found evidence of an economic voting effect during

the crisis, noting that most incumbents lost support, but their study design does not include a

baseline for comparison so it is not clear whether this economic vote is stronger or weaker than

expected (LeDuc and Pammett 2013). Similarly, a study of elections in 28 OECD countries in

the years 2007–2011 found clear evidence of an economic voting effect during the crisis, but

once again there is no baseline to compare this to (Bartels 2014). Other studies did compare

Great Recession voting behaviour to an earlier baseline. Hernández and Kriesi (2015) looked

at the relationship between economic indicators and the vote share of the dominant governing

party at post-crisis elections in 30 European countries. They found different patterns in dif-

ferent parts of the continent, with an economic vote typically exceeding baseline expectations

in Western European countries but not in Central and Eastern European countries, where the

response appeared to be more moderate. Likewise, Bouvet and King (2016) investigated the

relationship between incumbent vote share and economic indicators in 32 OECD countries at

elections between 1975 and 2013, with particular attention paid to the Great Recession. They

found evidence that voters were more likely to punish right-wing parties than left-wing parties

during the recession. A further study looked at economic opinion in eleven Western European

countries in the period 2007–2011, finding that economic opinion typically reflected the eco-

nomic reality and that the perceptions of voters during the Great Recession were by and large

not an overreaction (Anderson and Hecht 2014).

In the few years since the Great Recession, a literature has emerged on economic voting

during that time. The bulk of these have been studies of individual countries. Many of these

studies seek to establish whether there was an economic voting effect in some country during

the Great Recession. For example, a crisis economic voting effect was found in the United King-

dom, as well as qualified support for austerity policies, with evidence that this support is likely

to wane unless the economy starts to improve (Borges et al. 2013). A study of British voters’

attributions of responsibility for the crisis found evidence that Conservative partisans were

more likely to blame the government for the crisis compared to Labour partisans, who tended


to blame financial institutions (Hellwig and Coffey 2011). There is also evidence that Labour

Party support in the years 2004–2009 was more related to unemployment among low income

earners and to inflation among high income earners (Palmer and Whitten 2011; Palmer, Whit-

ten and Williams 2013). On the other hand, Duch and Sagarzazu (2014) used panel studies

from the UK and Germany to study perceptions of the Great Recession and economic voting,

finding little difference in the economic vote of rich and poor voters, despite poorer voters

feeling the effects of the recession more strongly. There is thus agreement that economic vot-

ing occurred in the UK during the crisis but not as to whether rich and poor voters behaved

differently.

Countries that were particularly heavily affected by the recession have tended to attract

scholarly attention. Unsurprisingly, economic voting appears to have been a feature of elec-

tions in these countries as well. For example, a Greek study used a combination of individual

and aggregate data to examine the economic vote in Greece both before and during the Great

Recession (Nezi 2012), finding evidence of economic voting throughout the crisis, although a

large shift in economic perceptions is necessary to produce a change in government. Turning

to Spain, Fraile and Lewis-Beck (2012) use survey data from 1982 to 2008 to show that eco-

nomic voting does exist in that country, despite some contrary claims in the literature. In a later

study, they then look at the 2008 election specifically, finding a clear economic voting effect

during the economic crisis, although they unfortunately do not compare 2008 directly with

previous years (Fraile and Lewis-Beck 2013). A study of post-election surveys from Italian

elections between 1990 and 2008 found that instability is not an inherent characteristic of

Italian politics but rather the result of valence political behaviour such as economic voting

(Bellucci 2012). And Marsh and Mikhaylov (2012) examined the 2011 Irish election, which is

particularly notable for the extreme loss of support of the previously highly successful incum-

bent party Fianna Fáil. While they naturally place importance on the specific circumstances of

the crisis and its aftermath, they unsurprisingly find support for a link between the economic

and political events of that time.

Economic voting is certainly not limited to those countries that suffered exceptionally from

the recession. Germany is a particularly interesting case because it was governed by a grand

coalition of the major centre-left and centre-right parties at the time of the recession, so eco-

nomic voters potentially had a much more difficult to choice to make. In other words, despite

the severe economic crisis, there was no viable alternative to the incumbent government. An-

derson and Hecht (2012) used panel survey data from before and after the 2009 German


elections to examine the economic vote in this context, finding that voters personally affected

by the recession used several strategies to shift their support away from the parties of the

grand coalition but that sociotropic voters tended not to continue supporting the coalition.

Clarke and Whitten (2013) used survey data of Germans in 2009 to compare valence mod-

els, which are a generalisation of economic voting models, to spatial models, finding that the

valence model was best equipped to explain vote choice. These results suggest that, despite

being governed by a grand coalition, economic voting still occurred in Germany during the re-

cession. Economic voting effects during the Great Recession have also been found in Sweden

(Martinsson 2013; Lindvall, Martinsson and Oscarsson 2013), Portugal (Freire and Santana-

Pereira 2012), Turkey (Çarkoglu 2012) and Hungary (Stegmaier and Lewis-Beck 2011).

Although this literature clearly establishes that economic voting during the Great Reces-

sion took place across many and varied European countries, there are two things that are

missing. The first of these is a comparative, individual-level study of economic voting beha-

viour during the crisis. This is important because there is only so much that can be learnt from

single-country and aggregate studies. Single-country studies are limited because they cannot

possibly test or control for contextual effects, such as clarity of responsibility. Moreover, when

looking at only one country, there is always the possibility, however remote, that any effect

that is found is not a typical behaviour but one characteristic to the particular institutions

of that country. Even though economic voting effects have been found in so many individual

European countries, it is not yet clear whether these effects are similar in size, since each study

uses different data and different methods. Aggregate studies are also limited because there

are so many different individual behaviours that can explain any observable aggregate effect.

Therefore, in order to gain a deeper understanding of the economic vote during this crisis,

a comparative individual-level study is needed. The published study that comes closest to

meeting these criteria is Whiteley (2016), who used European Social Surveys data to compare

voters from 21 countries in 2006 and 2012, in other words shortly before and shortly after the

crisis. He modelled incumbent support according to spatial, valence and cleavages theories of

voter behaviour, finding more similarities than differences before and after the crisis in each

case. This was not, however, a study of economic voting specifically and leaves open questions

about the nature and level of economic voting during the crisis.

The second thing missing from this literature is a study specifically designed to contrast

the economic vote before, during and after the recession. This is important because such a

study would make it possible to test the hypothesis that the economic vote was stronger than


expected during the crisis. Without a time comparison, the possibility cannot be excluded that

the economic vote seen during the crisis was simply a continuation of the normal economic

voting response that is seen at other times. There have been a small number of single-country

studies that have specifically made this comparison but no comparative studies. For example,

Martinsson (2013) found that the economic vote in Sweden was unusually strong at the post-

crisis 2010 election and Lindvall, Martinsson and Oscarsson (2013) compared the economic

vote in Sweden during the Great Recession to that during an earlier crisis in 1991–1993,

finding that economic status was a stronger predictor of economic voting behaviour during

the Great Recession than it had been during the earlier crisis. On the other hand, Freire and

Santana-Pereira (2012) found the economic vote in Portugal to be slightly depressed at the

2009 elections compared to previous years and Çarkoglu (2012) found a stronger economic

vote in Turkey in 2007 than in 2011. Others still have argued that the immediate economic

voting response to the recession was muted and that it was not until the implementation of

unpopular austerity programmes that voters started to turn against their governments (Bermeo

and Bartels 2014, 3–4). In the face of such contradictory results, there is a need for a systematic

cross-national individual-level study designed to compare the economic vote before, during

and after the Great Recession, a need which this thesis fills.

1.10 Conclusion

The basic idea behind economic voting theory is that voters are more likely to support in-

cumbents when the economy is performing satisfactorily than they are when it is performing

poorly. This basic idea is simple but there are many different theoretical behaviour models that

could support this idea. This chapter has reviewed the key theoretical debates in the economic

voting literature and discussed the empirical evidence in support of the various positions. Al-

though some of these questions are more settled than others, it is not possible to proceed

without making some assumptions about the nature of the economic vote. This theoretical

model is a selection model, which means that it is assumed that voters use their votes in order

to select the best possible government rather than to send a reward or punishment signal to

the incumbent. The assumption that voters are not entirely egocentric but rather sociotropic

economic voters is probably the least controversial, in light of the large body of evidence that

sociotropic economic voting exists. This thesis is based on a prospective theory of economic

voting. This is partly for data reasons, as the next chapter will show. Although retrospective

1.10. CONCLUSION 37

Figure 1.2: Outline of economic voting theory

This illustrates the specific theory of economic voting used in this thesis. A voter’s level ofsupport for a particular party is influenced by his or her perceptions of the economy and thedirection of this influence depends on whether that party is incumbent or not. The strength ofthis economic vote is in turn influenced by contextual factors such as clarity of responsibilityand the Great Recession.

Figure 1.3: Vote choice process

Vote choice is not modelled as being directly influenced by the economic vote. Rather, theeconomic vote independently affects a voter’s level of support for each party and these levelsof support determine the final vote choice.


studies tend to predominate, the empirical evidence does not strongly favour either altern-

ative. This thesis prefers an economic perceptions model over an objective indicators model

because, despite the criticisms of perceptions models, individual voters are not necessarily ob-

jective and do filter information in idiosyncratic ways. In light of the fact that almost all of the

countries under study have multiparty systems, a party support model is preferred to a vote

choice model. Finally, it is theorised that the strength of the economic vote may be affected

by contextual features such as clarity of responsibility or the Great Recession, as an example

of a deep international economic crisis.

Taken together, these features describe a cohesive theory of the economic vote, which is

outlined in Figure 1.2. As the diagram shows, the level of support that a particular voter has

for a particular party is influenced by both the prospective economic assessment of that voter

and the incumbency status of the party—which may have more than two levels if importance

within the governing coalition is taken into account. These two influences interact with each

other, as the effect of a particular economic assessment on support for a government party

is expected to be different from the same effect on support for an opposition party. It is this

interaction of effects that is referred to as the economic vote. It is worth noting that, although

the basic principle of economic voting appears straightforward, even obvious, it necessarily

describes an interaction effect.7 This explains some of the complexity of economic voting

models, including those introduced later in this thesis. This economic voting effect is itself

affected by contextual factors, such as the clarity of responsibility in a particular country and

events such as the Great Recession. It is of course the latter that is the focus of this thesis,

the primary objective of which is to establish whether that recession did indeed have an effect

on the economic vote. Figure 1.3 illustrates that a citizen’s final vote choice is determined by

his or her level of support for each of the parties that could potentially receive the vote. It is

assumed that voters simply select the party that they currently support the most.

This chapter has also discussed the existing literature on economic voting in the Great

Recession. Although there have been a number of studies of economic voting in individual

countries during that time frame as well as a handful of aggregate-data cross-national studies,

there are limits to what can be learned from these. Even among these studies, few explicitly

contrast what was observed during the crisis to what was observed at other times. In order

to gain a fuller understanding of how the economic vote was affected by the Great Recession,7Except in the special case of a two-party system, where support for the government and support for the oppos-

ition can be reduced to a single difference in support variable and the incumbency effect can be eliminated entirely.This special case describes the United States but is otherwise rare, especially among the European countries studiedin this thesis.

1.10. CONCLUSION 39

there is a need for an individual-level cross-national study comparing the economic vote be-

fore, during and after the crisis. This thesis fills this need. The next chapter discusses the data

used to undertake this study as well as the specific methods used to analyse that data.

Chapter 2

Measuring the economic vote

The previous chapter introduced the theory of economic voting and explained the specific

theoretical framework that informs this thesis. The following chapters use this framework to

develop statistical models describing the economic vote, which are then used to test hypotheses

that shed light on the research questions driving this thesis. The purpose of this chapter is to

prepare the ground for these statistical models by describing the data sources that are used,

the key variables that are analysed and the methods that are used to perform this analysis.

This chapter focuses on the details that are relevant for the entire thesis. Other details that

are more specific to individual analyses are discussed as needed in the chapters where they

are relevant.

As the purpose of this thesis is to examine whether and how economic voting behaviour

was affected by the Great Recession, it is important to be able to measure the economic vote

both during the crisis and at other times. As will be shown, it happens that the European

Election Studies have conducted multinational voter surveys at ideal times for making such

a comparison. These surveys include questions that indicate respondents’ levels of support

for various parties as well as their economic assessments. When combined with incumbency

data from other sources, this makes it possible to model the economic vote according to the

framework introduced in the previous chapter. It will also be shown that there is a structure

to this data, which has to be taken into account by the statistical methods used. This structure

is accounted for by using multilevel analysis.

This chapter begins by introducing the data sources that have been used, and shows how

the timing of the three survey waves used coincides with the course of the Great Recession.

Following this, there is a discussion of the particular measurements used. This includes a

description of the key variables and the precise question wording where appropriate, along

with some comments on the distribution and level of measurement. Finally, the methods used

to analyse all of this data are introduced and the reasons behind these choices are given.

41

42 CHAPTER 2. MEASURING THE ECONOMIC VOTE

2.1 Data sources and timing

As the previous chapter explained, this study is intended to be an individual-level cross-

national study comparing the economic vote before, during and after the Great Recession.

In order to achieve these goals, there is a need for survey data satisfying three key criteria:

it must include the necessary questions to measure the economic vote, it must be compar-

able across nations and it must consist of appropriately timed survey waves. Fortunately, the

European Election Studies satisfy all three of the requirements.1 The EES surveys are a series of

post-election surveys, which are collected in each European Union member state shortly after

every European Parliament election. As will be seen later in this chapter, these surveys in-

clude the required questions. These surveys are specifically designed to be comparable across

countries. They are based on a single questionnaire which is translated into the appropriate

local language or languages. It should be noted that this is not a study of European Parliament

elections but rather of hypothetical national elections. The advantage of this approach is that

it makes it possible to measure the economic voting tendency across all of the countries at

the same point in time, whereas the actual national elections are invariably spread out across

many years, which makes it very difficult to compare countries. Although the EES surveys cor-

respond to European Parliament elections, they also include questions about national politics

and these are the questions that are used.

The EES surveys are also suitably timed. This thesis uses the 2004, 2009, and 2014 waves

(EES 2004, 2009, 2014; Schmitt et al. 2009; van Egmond et al. 2010; Popa et al. 2015), which

were all conducted shortly after the European Parliament elections in those years. Figure 2.1

shows the quarterly GDP growth rate across the current 28 member states of the EU. The

Great Recession began in the aftermath of the global financial crisis of 2007–2008 and by the

end of 2008 the EU was in recession, following the common definition of a recession as two

consecutive quarters of negative growth. The dashed vertical lines show when the European

Parliament elections occurred and correspondingly when the three EES surveys were collected.

As the figure shows, the 2004 election took place well before there was any sign of crisis,

while the 2009 election took place shortly after its peak. Although there was a brief apparent

recovery, conditions worsened again in 2011, with a further recovery in 2013. By the time

of the 2014 European Parliament elections, the EU had been out of recession for eighteen1The survey data used in this thesis was originally collected by the 2004, 2009 and 2014 EES research groups.

Those studies have been made possible by various grants. Neither the original collectors of the data nor theirsponsors bear any responsibility for the analyses or interpretations made here. The data is available from thehome page of the European Election Study (http://eeshomepage.net/) and from the Archive Department of GESIS–Leibniz Institute for the Social Sciences (http://www.gesis.org/).

2.1. DATA SOURCES AND TIMING 43

Figure 2.1: GDP growth in the European Union, 2000–2015

-2.5

-2.0

-1.5

-1.0

-0.5

0.0

0.5

1.0

2001 2003 2005 2007 2009 2011 2013 2015time

GD

P g

row

th (

%)

This shows the seasonally adjusted GDP growth rate across the entire 28-country EuropeanUnion by quarter, each compared to the previous quarter. The dashed vertical lines show whenthe European Parliament elections took place in 2004, 2009 and 2015. Source: OECD

months. These three time points thus offer a before, during and after snapshot of the Great

Recession. It should be noted that this is the overall picture, as the recession took a slightly

different course in each country. Nonetheless, the only two EU member states to avoid the

recession altogether were Poland and Slovakia.

The EES surveys typically encompass each member state of the EU. The EU was enlarged

during the time frame of this study, with Bulgaria and Romania joining in 2009 and Croatia

joining in 2014. Since it would be difficult to separate any Great Recession effect from that

of joining the EU, these countries have been excluded from this study. This leaves the 25

countries that have been member states since 2004.2 The political parties included in the study

are those for which there is a measure of the dependent variable, party support. These are the

parties that were chosen by the EES coordinators in each country as being the most important

parties in that country. These typically include all of the parties that could reasonably expect2Unless otherwise specified, the Swedish data from 2004 has been excluded from the analysis. This is because

some variables were measured on a different scale from that used elsewhere. Most problematically, this includesthe economic assessment questions, which were measured on a three-point scale instead of a five-point scale. Sincethese responses could not be reconciled satisfactorily with those from other countries, the decision was made toignore the Swedish survey in that year. Sweden is however still included in the other years. Also, whenever partysupport is the dependent variable, Belgium, Lithuania and Luxembourg in 2004 also had to be excluded becausethe party-specific questions were not asked in those countries.


Table 2.1: Sample size and interview mode of EES surveys

2004 2009 2014Country Sample Mode Sample Mode Sample Mode

Austria 1010 phone 1000 phone 1114 faceBelgium 889 mail 1002 phone 1084 faceCyprus 500 face 1000 phone 530 faceCzech Republic 889 face 301/720 phone/face 1177 faceDenmark 1317 phone 1000 phone 1085 faceEstonia 1606 face 300/707 phone/face 1087 faceFinland 900 phone 1000 phone 1096 faceFrance 1406 unknown 1000 phone 1074 faceGermany 596 phone 1004 phone 1648 faceGreece 500 phone 1000 phone 1085 faceHungary 1200 face 300/705 phone/face 1104 faceIreland 1154 mail 1001 phone 1081 faceItaly 1553 mail 1000 phone 1091 faceLatvia 1000 face 300/701 phone/face 1055 faceLithuania 1005 face 300/705 phone/face 1096 faceLuxembourg 1335 phone 1001 phone 538 faceMalta — — 1000 phone 544 faceNetherlands 1586 mail 1005 phone 1101 facePoland 960 face 302/700 phone/face 1223 facePortugal 1000 phone 1000 phone 1033 faceSlovakia 1063 face 301/715 phone/face 1095 faceSlovenia 1002 phone 1000 phone 1143 faceSpain 1208 face 1000 phone 1106 faceSweden 2100 face 1002 phone 1144 faceUnited Kingdom 1500 phone 1000 phone 1421 face

The 2004 wave used telephone, mailback and face-to-face interviews, depending on the coun-try. The 2009 wave used telephone interviews in every country, which were supplemented byface-to-face interviews in certain countries. All interviews in the 2014 wave were conductedface-to-face. Source: EES

to win a seat at a general election. Table 2.1 shows the sample size and interview mode of

each survey in each of the three waves. At least five hundred responses have been collected

in each country and one thousand responses is typical, except for some particularly small

countries. Malta was not surveyed at all in 2004. Various interview modes have been used,

with telephone and face-to-face modes the most common. A mix of modes was used in 2004,

including mailback surveys in four countries. It is not known which method was used in France

in that year.3 Interviews were predominately conducted by telephone in 2009 but these were

supplemented with face-to-face interviews in several countries where a representative sample

could not be otherwise obtained (van Egmond et al. 2010, 5). All interviews were face-to-

face in the 2014 wave. Although there are a mix of modes in this data, mode effects are not3The codebook simply states that ‘France did not report the technical implementation of its study’ (Schmitt

et al. 2009).


expected to influence the findings, since economic voting is not something that respondents

might be embarrassed to admit. In fact, it cannot be determined if any individual respondent

is an economic voter since these patterns are only observable in the aggregate.

Although the EES survey data forms the primary dataset for this thesis, some contextual

data has been drawn from other sources. The main contextual information required is the

incumbency status of each party at the time the surveys were collected. This information

was taken from the Parliaments and Governments Database (Döring and Manow 2015), also

known as ParlGov, which aggregates information about election results and parliament and

cabinet composition from EU and other countries. Since multiple governments may be in

power in a country during a single calendar year, it was necessary to decide precisely which

dates to use to measure incumbency. In most cases, the measurement date chosen was the first

day of the survey fieldwork, namely 14 June 2004, 8 June 2009 and 22 May 2014. Whichever

party held the post of head of government on this date is designated the prime minister’s party4

in that year. Parties holding other positions in the government are designated cabinet parties.5

All other parties are coded as opposition parties. This approach makes sense because survey

respondents were asked to reflect on the government currently in power at the time.

In two cases, however, slight adjustments were made in order to accommodate the peculiar

circumstances. On 7 May 2009, the Czech government was replaced by a nonpartisan care-

taker government, in advance of elections in October of that year (Linek and Lacina 2010).

Since this caretaker government was not officially associated with any particular parties, it is

treated in this thesis as an extension of the preceding coalition government led by the Civic

Democratic Party. Similarly, the Prime Minister of Hungary Ferenc Gyurcsány resigned from

the minority Hungarian Socialist Party (MSZP) government on 21 March 2009, being replaced

by Gordon Bajnai on 14 April (Várnagy 2010). Although he was an MSZP appointment and a

government minister, Bajnai was a compromise candidate proposed to appease other parties,

and he is recorded as a nonpartisan prime minister in the ParlGov database. In light of these

circumstances, this thesis treats the MSZP as the party holding the office of prime minister

at the time of the 2009 surveys. Other cases worth special mention are Luxembourg in 2004

and 2009 and Belgium in 2014. In these instances, the national elections coincided with the

4This also includes the party holding the presidency of Cyprus, which is the only EU country with a full pres-idential system. Since the phrase ‘party of the head of government’ is rather unwieldy, the term ‘prime minister’sparty’ is used throughout this thesis instead and should be understood to include the party of the President ofCyprus.

5Note that in some countries, the official cabinet does not include all government ministers. The term is notused here to refer to such an inner cabinet but rather to all government ministers, irrespective of seniority. Thisusage applies throughout this thesis.


European Parliament elections, so there was the potential for a change of government to have

occurred during the survey fieldwork, which would have been problematic. In practice this

was not an issue, as both Luxembourg elections led to only minor changes to the cabinet,

which in any case did not take place until after the fieldwork period (Dumont and Poirier

2005; Dumont, Kies and Poirier 2010), and the Belgian election did not result in any change

to the party composition of the cabinet (Rihoux et al. 2015). The full list of parties analysed

in their thesis along with their incumbency status in each year can be found in Appendix A.

The most important sources of data are those discussed already, the EES surveys and the

ParlGov database. These are used throughout the thesis. There are also some supplementary

data sources that have been used to supply additional variables that are only required by

certain chapters. Chapter 4 extends the basic economic voting model to test the clarity of

responsibility theory developed by Powell and Whitten (1993). This requires country-level

variables measuring various aspects of the political system. Some of these, such as whether

the country has a presidential or a parliamentary system, were generated from the literature

or common knowledge. Other variables, such as the composition of the parliament, were

complex enough to require a systematic approach to measurement and these were mainly

taken from the European Journal of Political Research Political Data Yearbook (PDY). Chapter 5

extends the basic model to examine how party support is influenced by the party’s spatial

position. The spatial position data was collected from the Chapel Hill Expert Survey (CHES).

Some of the remaining chapters also use extra variables from the EES surveys in addition to

what is described in this chapter. In all of these cases, the measurements are discussed in the

relevant chapters before they are used.

There are two further complexities that had to be considered when coding political parties.

The first of these is the phenomenon of regional and regionalist parties. Regional parties are

those that are only active in parts of a country, such as the Christian Social Union in Bav-

aria (CSU), which does not operate outside the German federal state of Bavaria. The CSU

is a special case, since it works very closely with the Christian Democratic Union of Ger-

many (CDU), which operates everywhere else in Germany. In practice, the CDU and CSU

tend to be seen as a unit at the national level—the CDU/CSU faction—and this is how they

are treated in the EES surveys as well. A more difficult example of regional parties are the

parties in Belgium, which are typically organised along linguistic lines, with French-speaking

parties and Flemish-speaking parties, rather than at the national level. These are not normally

part of a de facto greater unit, like the CDU/CSU faction. Regionalist parties are those that


seek independence or at least greater autonomy for their region, such as the Scottish National

Party in Scotland and Plaid Cymru in Wales. There are further examples in Flanders and many

regions of Spain, namely the Basque Country, Navarre, Galicia and Catalonia. The problem

posed by these parties is that the analysis takes place at the level of the nation-state. In the case

of Belgium, this is sometimes dealt with by treating Flanders and Wallonia as separate states

but this is not the approach used here, since they do share a national government, despite

their different party systems. Moreover, this approach does not scale to the other countries

mentioned. Instead, a record has been made of the regions where each party is active. For

most parties, these comprise the entire country. The EES surveys also include a variable not-

ing where each respondent was interviewed. When the stacked dataset was produced, any

rows matching voters to parties not active in their region were discarded. Thus the analysis is

affected by all British voters’ opinions of the Labour Party but only Welsh voters’ opinions of

Plaid Cymru, for example.

The second complexity encountered when coding political parties was that the party sys-

tems of the various countries in the EU feature varying degrees of fluidity. Northern European

countries, such as Germany, Ireland, Sweden and Denmark, tend to have stable party systems

whereas the party systems of Mediterranean countries, such as France, Spain and Italy, are

characterised by frequent mergers and splits, which still occur today despite some increased

stability in recent decades (Krouwel 2012, 49–78). This means that, in many cases, it is far

from straightforward to identify political parties in 2004 with the same parties in 2009 and

2014. The Irish party named Fianna Fáil in 2004 is obviously the same party as the Fianna

Fáil of 2014 but can the now defunct Democratic Movement of France be identified with the

Union for French Democracy, which succeeded it in 2007? The Italian party Forza Italia even-

tually merged with several other parties in 2009 to form The People of Freedom, which later

dissolved in 2013, reviving Forza Italia. Should these all be treated as the same party? This

question is even more difficult to answer in the case of a party split, in which case a party may

have multiple successors. To make matters worse, political parties frequently form short-lived

electoral alliances for the European Parliament elections and often the EES surveys ask voters

about these alliances, rather than the constituent parties. In order to avoid making arbitrary

decisions, each party is treated as a discrete unit in each year. This means, for example, that

responses to questions about the Fianna Fáil of 2004 are not pooled with those about the

Fianna Fáil of 2014. This only applies to the pooling of responses. Measurements of a party’s

time in office are unaffected by this decision.


Table 2.2: Key variables used to measure the economic vote

Variable Scale Level Source

Party support 0 (low) to 10 (high) measurement EESYear 2009 1= yes, 0= no measurement EESYear 2014 1= yes, 0= no measurement EESLeft–right distance 0 (close) to 10 (far) measurement EESParty ID 1= yes, 0= no measurement EESCabinet party 1= yes, 0= no party ParlGovPrime minister’s party 1= yes, 0= no party ParlGovProspective assessment −2 (worse) to +2 (better) individual EESFemale 1= yes, 0= no individual EESAge 18 to 101 years individual EESHigh education 1= yes, 0= no individual EESLow education 1= yes, 0= no individual EESUrban area 1= yes, 0= no individual EESRural area 1= yes, 0= no individual EESUnemployed 1= yes, 0= no individual EESNot in workforce 1= yes, 0= no individual EES

These are the variables used in every chapter. Most chapters include further variables, whichare introduced as used. The dependent variable is party support, unless otherwise specified.The measurement level is shown for each variable. This indicates whether the variable isspecific to the party or the individual. Variables that depend on both are said to be at themeasurement level. Although none are shown here, some chapters also include variablesmeasured at the country level. The source is either the European Election Studies or theParlGov database.

2.2 Measurement and variables

There are a number of key variables that are fundamental to this thesis and appear in every

chapter. This section introduces the main dependent variable, party support, and several inde-

pendent variables, including party identification, ideological distance, incumbency, economic

assessment and some control variables. For each variable there is a discussion of why it is im-

portant and how it is measured. These variables are listed in Table 2.2. There is an inherent

structure to the data used, which should not be ignored. Each EES survey wave consists of

a number of national surveys. Random sampling is used for each national survey but some

countries are more heavily sampled relative to their population sizes than others. Individual

responses are thus nested within countries. Furthermore, individuals are asked about each

of the important parties in their country, so these responses are nested within both individu-

als and parties because they relate a particular individual to a particular party. This will be

referred to as the measurement level. Thus measurements are nested within individuals and

parties, each of which is nested within with countries. The problem of structured data is ad-

2.2. MEASUREMENT AND VARIABLES 49

dressed later in this chapter. For now, it will simply be noted at which level a particular variable

is measured.

The dependent variable used for the economic voting models is party support. Party sup-

port relates a particular individual to a particular party and is thus measured at the measure-

ment level. This variable is intended to measure the voter’s current preference level for the

party. It is derived from EES questions asking voters the likelihood that they would vote for

that party at a future national election.6 The precise wording of the question in 2009 is:

We have a number of parties in [country] each of which would like to get your

vote. How probable is it that you will ever vote for the following parties? Please

specify your views on a scale where 0 means ‘not at all probable’ and 10 means

‘very probable’. If you think of [party]: what mark out of ten best describes how

probable it is that you will ever vote for [party]?

The wording is almost identical in the other years, except for that fact that the 2004 wave

used a scale starting at one instead of zero. These were transformed to a zero to ten scale

for comparability with the 2009 and 2014 data. In order to ensure that this did not bias

the results, several potential transformations were considered. These included variations on

scaling the data and recoding particular points. The transformation that was ultimately used

was to recode one on the ten-point scale as zero and to leave the other levels unchanged. This

approach was chosen because it produced a distribution very similar to the distribution of the

2009 variable. It seems that most scores correspond directly on the two scales except of course

that parties that are strongly disliked were given the score zero where that was available or

one where it was not. Using the alternative transformations yields similar results in any case.

The decision to use party support rather than vote choice as the dependent variable was

explained in the previous chapter. It is based on the theory that vote choice is a two-stage

process, the first stage determining party support levels based on economic and other con-

siderations and the second stage determining vote choice based on those party support levels

(van der Brug, van der Eijk and Franklin 2007, 31–53). In this model, it is assumed that the

second stage is deterministic, with voters always selecting the party that they have the highest

level of support for. In order to test whether this assumption is reasonable, the pattern of

support levels was examined among those voters who answered the question asking which

party they would vote for at a national election held the following day. Among these voters, a

6Q12 in 2004, Q39 in 2009 and QPP8 in 2014.


Figure 2.2: Distribution of raw and centred party support

raw centred

0.00

0.25

0.50

0.75

1.00

0 2 4 6 8 10 -8 -6 -4 -2 0 2 4 6 8 10party support

norm

alis

ed c

ount

The left panel is a histogram of raw party support and the right panel is a histogram of partysupport centred around the individual-level mean, from the three surveys combined. The binsize is 0.5 in both instances. Source: EES

consistently high proportion nominated the party they had the highest reported support for.7

The proportion of voters matching this criterion was 92.7 percent [92.3%,93.2%]8 in 2004,

93.7 percent [93.3%, 94.1%] in 2009 and 94.4 percent [94.0%,94.8%] in 2014. Although a

small number of respondents have not responded as expected, these results confirm that this

vote choice model is a reasonable approximation of actual behaviour.

The party support variable has been centred around each individual’s mean reported party

support. The distributions of the raw and centred variables are shown in Figure 2.2. As the

figure shows, the distribution of the raw variable is both multimodal and highly skewed. A

disproportionate number of reported party support levels were zero, corresponding to 40.9

percent of the total measurements. These are the instances in which a voter reported that

they would never vote for a particular party. There are secondary modes at five and ten.

These properties are problematic for regression analysis, which assumes that residuals are

normally distributed (Gelman and Hill 2007, 46).9 In fact, the raw level of support reported

by a particular individual depends on that individual’s interpretation of the eleven-point scale.

7In many cases, respondents assigned the same support level to multiple parties and in some of these instancesmultiple parties shared the distinction of being a particular respondent’s most preferred party. As long as theresponse to the party choice question was one of those parties, the statistics reported here treat these cases asmatches.

8Square brackets indicate 95% confidence intervals.9Just because the dependent variable is not normally distributed does not necessarily mean that the residuals

will not be. For this to be the case, however, the independent variables would need to be distributed in a veryparticular way, so as to compensate for the non-normality of the dependent variable, which they are not.


By centring the scores around the individual-level mean, the differences between the scores

are preserved and it is these differences that are of substantive interest, not the absolute levels.

This is because the theory predicts a change in vote choice whenever the voter’s support for a

new party exceeds their support for their previous choice, not when it exceeds some absolute

bar. As the figure shows, the distribution of the centred variable is somewhat positively skewed

and it appears to be bimodal, although the mode at zero is an artefact resulting from the fact

that some individuals give all the parties an equal score. Nonetheless, the distribution of the

centred variable is far more appropriate for regression analysis than that of the raw variable.

The timing of the survey is indicated by two dummy variables. The base case is the 2004

survey wave and there is a dummy variable indicating the 2009 wave and one indicating the

2014 wave. As discussed earlier, the 2009 survey wave took place when the initial crisis was

at its peak, so this dummy variable indicates the mid-crisis time point. Similarly, the 2014

wave took place after the end of the recession, so its dummy variable indicates the post-crisis

time point. The reference case is naturally the pre-crisis time point.

As well as predictors related to economic voting specifically, the economic voting model

developed in the next chapter also includes other variables known to be predictive of vote

choice. The first of these, left–right distance, is derived from the spatial theory of voting

(Downs 1957; Hotelling 1929; Davis, Hinich and Ordeshook 1970). According to this theory,

individual voters prefer the parties that are closest to themselves in some political space, which

is most frequently conceived of as a single dimension stretching from left to right. The left–

right distance variable measures this closeness between a voter and a party. This is known

to be a strong influence on vote choice (van der Eijk, Schmitt and Binder 2005; Kroh 2009).

Two survey questions were used to produce this measure. The first measures the individual’s

spatial position.10 The precise wording of this question in 2009 was:

In political matters people talk of ‘the left’ and ‘the right’. What is your position?

Please indicate your views using any number on a scale from 0 to 10, where 0

means ‘left’ and 10 means ‘right’. Which number best describes your position?

The second question measures the individual’s opinion of the spatial position of each party.11

This question immediately follows the self-location question and is worded: ‘And about where

would you place the following parties on this scale?’ The wording of these questions was very

similar in the other years, except for the fact that the 2004 survey used a 1–10 scale instead

10Q14 in 2004, Q46 in 2009 and QPP13 in 2014.11Q14_i in 2004, Q47 in 2009 and QPP14 in 2014.


Figure 2.3: Distribution of left–right distance

0

20000

40000

60000

0 1 2 3 4 5 6 7 8 9 10left–right distance

coun

t

This is a histogram of the left–right distance between each individual and party, from thethree surveys combined. A distance of zero indicates that the individual believes the partyto be perfectly aligned with them ideologically. A distance of ten indicates the belief that theindividual and party occupy the opposite extremes of the spectrum. Source: EES

of a 0–10 scale, as was the case for party support. These were adjusted in the same way as

party support, described above, placing all values on the 0–10 scale.

The left–right distance between a particular voter and a particular party is the absolute

difference between that voter’s reported left–right position and his or her assessment of that

party’s left–right position. This effectively measures the distance an individual perceives be-

tween him- or herself and the relevant party. This can be expected to be a reliable measure of

spatial distance, as there is evidence that most voters are able to place themselves on a left–

right spectrum (Mair 2007) and that in most countries there is a strong correlation between

left–right self-identification and views about the desirable level of government intervention in

the economy as well as certain social issues (Dalton, Farrell and McAllister 2011, 81–108).

Additionally, the relative positions of parties according to the aggregated placement of survey

respondents tends to coincide well with other measures of party position (109–141). The

measure of left–right distance described here ranges in value from zero, in the case of perfect

alignment, to ten, which means the voter is on the far left and the party the far right or vice

versa. The distribution of this variable is shown in Figure 2.3. The mean value is 3.13 and

the standard deviation is 2.66. Frequency falls off quickly as the distance increases past five,


which is to be expected because most voters place themselves near the centre and extreme

distances are only possible for voters placing themselves near the endpoints.

The other key predictor derived from non-economic voting theories of voter behaviour is

party identification. The theory of party identification originated with Campbell et al. (1960),

who argued that a key determinant of vote choice was an individual’s sense of identification

with a particular party, this identification being largely inherited from parents or social milieu.

Green, Palmquist and Schickler (2002, 24–51) have investigated some of the characteristics of

party identification. Identifying with a party is not the same as supporting it or voting for it,

but rather a deeper and longstanding attachment to a particular party. People may vote against

their identification on occasion but it changes only rarely. Not everyone identifies with a party

but those who do are more engaged with politics. The original analysis of party identification

in the United States has been replicated in recent years (Lewis-Beck et al. 2008). The theory of

party identification has its origin in studies of the United States and its application to European

politics has not been uncontroversial but there is evidence that party identification models do

have some applicability to European countries (Berglund et al. 2005).

Party identification is, like party support and left–right distance, measured at the measure-

ment level, that is it relates a particular individual to a particular party. It is a dummy variable,

taking the value one if the individual identifies with the party and zero otherwise. No indi-

vidual identifies with more than one party12 but some individuals identify with no party. The

question used to measure this variable13 was:

Do you consider yourself to be close to any particular [political]14 party? If so,

which party do you feel close to?

Overall, 55.4 percent of respondents reported identifying with a particular party. As a result,

the party identification variable takes the value one for 7.7 percent of the party–individual

pairings.

Further independent variables are needed to measure the economic vote. In the previous

chapter, a theoretical model of economic voting was developed. According to that model, party

support is influenced by both the party’s incumbency status and the individual’s prospective

economic assessment. Incumbency status is measured at the party level of course and consists

of two dummy variables, one indicating that the party held the office of prime minister (or

12There is some evidence that people can identify with multiple parties (Schmitt 2009) but the EES surveys didnot ask about secondary identifications.

13Q29a in 2004, Q87 in 2009 and QPP21 in 200414In the 2014 survey only.


Figure 2.4: Distribution of prospective and retrospective assessments

2004 2009 2014

0

3000

6000

9000

0

3000

6000

9000

prospectiveretrospective

-2 -1 0 1 2 -2 -1 0 1 2 -2 -1 0 1 2economic assessment

coun

t

This shows the distribution of prospective and retrospective economic assessments in eachyear. Both range from −2 (much worse) to +2 (much better). Source: EES

president in the case of Cyprus) and the other indicating that the party held at least one cabinet

post at the relevant time. These are not mutually exclusive categories and in particular the

prime minister’s party is always a cabinet party, since the prime ministership is itself a cabinet

post. The method used to determine a party’s incumbency status has already been explained

earlier in this chapter. Of the 504 parties in the pooled dataset, 69 (14%) were prime ministers’

parties, a further 96 (19%) were other cabinet parties and 339 (67%) were opposition parties

at the time the surveys were conducted.

Prospective economic assessment is measured at the individual level using the following

survey question:15

And over the next 12 months, how do you think the general economic situation

in this country will be? Will it get a lot better, a little better, stay the same, a little

worse or get a lot worse.

15Q21 in 2004, Q49 in 2009 and QPP16 in 2014.


Figure 2.5: Relationship between prospective and retrospective assessments

2004 2009 2014

-1.0

-0.5

0.0

0.5

1.0

-2 -1 0 1 2 -2 -1 0 1 2 -2 -1 0 1 2retrospective

prospective

Each panel shows the mean prospective economic assessment for a particular retrospectiveassessment in the given year. These are shown with their 99% confidence intervals, althoughowing to the high sample size many of these are not visible. Source: EES

The responses to these questions have been recoded so that they range from −2 to +2, with

negative numbers indicating a pessimistic assessment and positive numbers an optimistic one.

Zero means the economy is expected to stay the same. The surveys also include an equivalent

retrospective question asking the respondents to assess whether the economy has improved

or worsened over the past twelve months. It was explained in the previous chapter that this

thesis uses a prospective voting model for both theoretical and data-driven reasons. The lat-

ter can be seen from Figure 2.4, which shows the distribution of both the prospective and

retrospective variables. In both 2004 and 2014, the distributions of both variables resemble

the binomial distribution, but for the fact that extremely positive assessments are remarkably

rare. In 2009, however, this is not the case. Voters’ retrospective assessments are very skewed

in that year, with 79.2 percent of respondents reporting a negative assessment, 13.7 percent

a neutral assessment and only 7.1 percent a positive assessment. While such a negative retro-

spective assessment of the economy was certainly justified at that time, this sort of consensus

among respondents is at odds with the variance desired in a key independent variable. The

prospective assessment in 2009 is also a little more skewed than in other years but the skew-

ness is much more tolerable, with 39.9 percent reporting a negative assessment compared to

26.4 percent a neutral and 33.7 percent a positive assessment.

It is also interesting to look at the relationship between prospective and retrospective eco-

nomic assessments in each year. These are shown in Figure 2.5. It can be seen that the relation-


ship between these variables is approximately linear in 2004 and 2014 but once again 2009

is different. In that year, not only does the relationship appear to be somewhat less linear but

retrospective assessment is also a weaker predictor of prospective assessment. Another way

of seeing this is to look at the correlations between the two variables in each year. This is 0.56

in 2004, 0.28 in 2009 and 0.63 in 2014. In other words, there is a strong correlation between

the two before and after the crisis but only a weak correlation during the crisis. These obser-

vations about the pattern of responses to the economic perceptions questions, combined with

the theoretical reasons outlined in the previous chapter, have led to the decision to base the

economic voting analysis in the following chapters on prospective, rather than retrospective,

economic assessment.

In addition to these key independent variables, demographic variables are used as con-

trols. These are gender, age, education, urban density and workforce participation. Gender

is a dummy variable taking the value one for women and zero for men. Women make up

54.4 percent of the sample. Age is measured in years but this is sometimes expressed as dec-

ades in the models used in the coming chapters so that all of the estimated coefficients are

on similar scales, since the effect size of age in years is typically very small. Unfortunately,

the other demographic questions were often measured on very different scales in the different

survey years and sometimes even countries within a survey year. This makes it difficult to

find granular correspondences between the different years, so responses have been grouped

into coarse categories instead, aiming to keep the size of the categories roughly equal. Edu-

cation has been divided into low, medium and high groups. Low education respondents are

those who did not complete high school. Medium education respondents are those who have

completed high school but do not have a university degree. Current students are included in

this category. High education respondents are those who have a degree or higher. Medium

education is the most common category and also the reference case, covering 39.9 percent of

respondents, followed by high education, covering 36.9 percent, then low education, covering

23.3 percent.

Urban density indicates the density of the area where the respondent lives. This is also di-

vided into three groups. These groups are necessarily rather subjective, owing to the phrasing

of the questions. The reference case is a small or medium-sized town. The other groups are

rural, which indicates a village or rural area, and city, which also includes large towns. Town

was the most frequent response (35.1%) followed by city (34.0%) and finally rural (30.9%).

The final control variable is workforce participation. Once again this is divided into three

2.3. METHOD OF ANALYSIS 57

groups. The reference case is those who are employed, with the other groups being the un-

employed and those who are not in the workforce for any other reason. The latter group is

mostly retired people but also includes stay-at-home parents and disabled people, among oth-

ers. Unsurprisingly, a majority of respondents (50.2%) indicated that they were employed,

with only a small number (8.0%) claiming to be unemployed. The remaining 41.8 percent are

not in the workforce. All of these variables are coded as dummy variables indicating each of

the non-reference cases.

2.3 Method of analysis

This thesis uses a stacked dataset design, which was strongly influenced by van der Brug, van

der Eijk and Franklin (2007, 40-46), although the methods used are different. The motivation

for this design is that the key variables of interest exist at the measurement level, which relates

particular individuals to particular parties. For instance, the dependent variable, party support

is such a variable. The degree to which individual i supports party j cannot be said to belong

completely to either the individual level or the party level—it belongs to both. Since this

party × individual, level, which is referred to in this thesis as the measurement level, is where

the most important measurements occur, it is appropriate for the unit of observation to be a

party–individual pair. This objective was achieved by transforming the EES survey dataset,

which was measured at the individual level. Parties are represented by repeated questions—

that is, each party-related question is asked once for each party. The dataset was transformed

such that each individual appears in the dataset once for each party they were asked about,

with only the questions about the relevant party included in each new row. Figure 2.6 illus-

trates this process.

The key analytical method used in this thesis is multilevel modelling. Multi-level model-

ling is an extension of linear regression and generalised linear modelling that explicitly models

variation between groups. This is achieved by giving the model coefficients a probability model

of their own (Gelman and Hill 2007, 1). The decision to use multilevel modelling was mo-

tivated by the clustered nature of the data. One of the consequences of stacking the dataset

as described above is that individual observations have been repeated in the stacked dataset

many times. If linear regression were used this would artificially deflate the estimated stand-

ard errors but multilevel modelling can account for this sort of structure. In any event, the

data used in this thesis is inherently structured. For one thing, there are twenty-five countries

in the analysis and to ignore this would be to assume implicitly that there is minimal variance


Figure 2.6: Data stacking process

country resp_id age support_p1 support_p2 support_p3

UK 1 88 0 8 5UK 2 21 7 3 8UK 3 40 3 8 0

⇓country party_id resp_id age support

UK UK-Lab 1 88 0UK UK-Lab 2 21 7UK UK-Lab 3 40 3UK UK-Con 1 88 8UK UK-Con 2 21 3UK UK-Con 3 40 8UK UK-LD 1 88 5UK UK-LD 2 21 8UK UK-LD 3 40 0

In the original dataset, each observation corresponds to an individual and variables relatingto particular parties are repeated for each party. In the stacked dataset, each observationcorresponds to a party–individual pair, eliminating the need to repeat variables.

between these countries, which is a strong assumption to make. Furthermore, many of the

variables that are of interest, including party support, the dependent variable, involve opinion

about specific parties. Since there may be variation between a country’s parties as well as

variation between countries, there are clusters within clusters.

While ignoring this clustering would lead to deflated standard errors, it must be acknow-

ledged that there are other ways of addressing this particular problem. It has been shown

that robust standard errors and aggregation can produce the same results as multilevel mod-

els under the appropriate conditions (Arceneaux and Nickerson 2009). Nonetheless, each of

these other methods is limiting in some specific way. Aggregation cannot be used to estimate

individual-level effects and robust standard errors are not recommended for clusters of fewer

than about twenty observations (188). The number of parties in each country in this study

varies from five to fifteen. Furthermore, multilevel modelling is sufficiently flexible that is

possible to include predictors at multiple levels within the same model. This is particularly

important for this study, since economic voting theory relates variables that necessarily relate

to multiple levels. Incumbency is only meaningful at the party level. Economic performance is

measured at the national level. And it is individuals who make vote choices. A key advantage

of multilevel modelling is that these variables and their interactions can be included in a single


model (Gelman 2007a, 7-8). Of course, it would be possible to aggregate all of these together

and estimate everything at the national level. It would also be possible to model separate

equations for every political party of interest. Both approaches are problematic however. The

former approach, complete-pooling analysis, ignores all variation between groups and the lat-

ter, no-pooling analysis, is statistically inefficient. A key strength of multilevel modelling is

that it offers a compromise between these extremes (Gelman and Hill 2007, 256). Multilevel

models produce both fixed effect and random effect estimates. The fixed effects are those of

primary substantive interest given the research questions behind this thesis. The decision to

use multilevel modelling is motivated by the need to account for the structure of the data,

rather than interest in the random effects specifically. As a result, the discussion of random

effects is limited to discussions of variance and sometimes covariance. Individual countries

are not normally discussed, as the purpose of this thesis is to gain an understanding of large

cross-national trends.

The decision to use multilevel linear regression models distinguishes this study from pre-

vious survey-based research into economic voting behaviour. As discussed in the previous

section, the stacked dataset design of this thesis was influenced by van der Brug, van der

Eijk and Franklin (2007), but they chose not to use multilevel modelling in their study. They

reason that the chief advantage of multilevel models is that they avoid the biased standard

errors that would result from naive regression modelling and they observe that the this prob-

lem can be avoided equally well by using robust standard errors, which they do (47). They

also argue against multilevel modelling on the grounds that it is not well equipped to handle

the cross-classified data structure that results from the stacked dataset design (48). While

there is merit to these arguments, there is still much to be said for multilevel modelling in

this context. Although it is true that the use of robust standard errors can avoid the problem

of deflated standard error estimates, there are other advantages to using multilevel model-

ling, particularly in that they permit greater flexibility in the models. As for the problem of

cross-classification, it must be acknowledged that this does add complexity to the models but

modern computing power combined with Bayesian estimation techniques makes it possible to

estimate these complex models. Duch and Stevenson (2008), on the other hand, do use some

multilevel modelling. They adopt two different methods for their analysis, a one-stage and a

two-stage strategy. Their one-stage strategy uses multilevel models to estimate all of the para-

meters together, whereas their two-stage strategy involves estimating the level of economic

voting in national surveys and then using those estimates in cross-national models (94-100).


One of the key differences between their methods and the methods used in this thesis is that

they use multinomial logistic regression to predict vote choice, whereas this thesis uses linear

regression to predict party support, for the reasons given in the previous chapter.

Most of the independent variables in the models presented in the following chapters have

been centred around the grand mean. By contrast to the centring of the dependent variable,

the reasons for which have been given earlier, this centring of the independent variables has

been done for technical reasons. Specifically, centring in this way markedly increases the con-

vergence speed. It is also helpful for variables which are involved in quadratic and interaction

terms. One consequence of this centring is that some caution is required when interpreting

intercepts and interactions. Because of this and also owing to the complexity of some of the

models, model coefficients are typically not interpreted directly in this thesis. Instead, post-

estimation simulation is used to derive quantities of more direct interest. The method used is

that described in Gelman and Hill (2007, 140–143). Because the predictions resulting from

this method take into account the uncertainty of many predictors, the error bands around them

can be deceptively wide. In particular, it is very often the case that two predictive intervals

overlap even though there is a significant difference between the actual predictions. In order

to minimise confusion, predictive intervals are not normally shown in any plots but instead

the question of significant difference is tested directly and discussed in the text. Standard

errors are of course quoted in the text for most predictions as well. The full model coefficient

tables can be found in Appendix B. There are multiple methods for computing p-values for

multilevel models. The p-values shown in the coefficient tables are based on Satterthwaite

estimates of the degrees of freedom, although these values play no part in the analysis in this

thesis. The Pseudo R2 reported for these models is based on Xu (2003).

One challenge that arises with any use of survey data is that of missing data. This has

been dealt with using listwise deletion. Owing to the stacked dataset design, a respondent

does not have to be excluded using this method simply because he or she has not answered

a question about a particular party. Only the row corresponding to that party–individual pair

is removed, while rows are included for each party that the individual did answer questions

about. This means that the only individuals that had to be excluded completely were those who

refused to answer any party related question—and those voters leave very little basis to impute

missing values—and those who did not answer questions at the individual level. As people

typically did not refuse to answer the demographic questions posed, the only problematic

question was prospective economic assessment. The proportion of respondents who declined


to answer this question was 13.3 percent in 2004, 4.1 percent in 2009 and 5.4 percent in 2014.

Other techniques for managing missing values were also considered. Multiple imputation in

particular has much to be said for it (Rubin 1978; King et al. 2001, 50) but that approach did

not prove viable here because multiple imputation techniques and software have not yet been

developed for all of the multilevel models that are used in this thesis. Ultimately, the structure

of the data was considered a more important issue than missing data and that is what led to

the decision to use listwise deletion and multilevel modelling.

The survey response rate has also been given some consideration. Response rates in the

EES surveys typically ranged from 60–80 percent for the face-to-face mode but were lower for

the telephone mode, sometimes below 20 percent. One approach to the issue of potential bias

resulting from low response rates is post-stratification survey weighting. Survey weights have

been included in the EES survey data to correct for non-response bias. These weights were

computed using a raking procedure on the variables of age, sex, region, education and, in

some countries, whether or not the household has a fixed phone line. In principle, weighting

in this way allows for the more accurate estimation of population parameters from the survey

data by reducing the effect of non-response bias. These weights are not however used in

the multilevel models in this thesis. Instead, non-response bias is minimised by including all

relevant demographic variables in the models as controls. Since the weights are conditioned

on variables that are already being controlled for in the models, they add no extra information.

There has long been a controversy in the literature about the relative merits of weighted and

unweighted least squares estimators for linear regression (DuMouchel and Duncan 1983, 535)

and the relative merits of model-based and sampling-based approaches to the problem of unit

non-response continue to be debated today (Gelman 2007a, 2007b; Bell and Cohen 2007).

The decision to use a model-based approach is motivated by the fact that the parameters of

most interest are regression coefficients rather than simple means or proportions. There is

also the problem that weighted estimators simply have not been developed for certain cross-

classified multilevel models, so the sampling-based approach would severely limit the types of

analysis that could be undertaken using these methods.

The statistical analysis described in this thesis has been performed using the statistical pro-

gramming language R (R Core Team 2016). In addition to the core language and its libraries,

the analysis was supported by the lme4 (Bates et al. 2015), lmerTest (Kuznetsova, Brockhoff

and Christensen 2016) and ordinal (Christensen 2015) packages. Most of the plots in this

thesis were produced using the ggplot2 package (Wickham 2009). Finally, the arm (Gelman


and Su 2015), dplyr (Wickham and Francois 2015) and RSQLite (Wickham, James and Falcon

2014) packages were all used heavily for utility purposes.

2.4 Conclusion

This chapter has introduced the dataset and the methods that will be used throughout the rest

of the thesis, as well as discussing how the key variables have been measured. This study will

undertake a multilevel analysis of survey responses collected in twenty-five European Union

member states. The primary data source is the pooled responses from the 2004, 2009 and

2014 waves of the European Election Studies surveys. These waves correspond to time points

before, during and after the Great Recession. This survey data is supplemented with contextual

data from other sources. The key dependent variable is party support, the degree to which

a voter states that he or she is likely to vote for a particular party in the future. Important

independent variables are the survey year, spatial distance between party and voter, party

identification, incumbency and prospective economic assessment.

The next chapter will introduce the basic models used with various extensions throughout

this thesis. Owing to the complexity of the complete model, several simpler models are presen-

ted first, examining the first hypothesis from various angles. These simpler models are easier

to interpret but only the complete model takes advantage of all of the available data, which

provides a clearer overall picture. The intention behind this approach is to use the simpler

models to illustrate various aspects of economic voting behaviour and then use the complete

model to obtain definitive estimates. Later chapters of the thesis extend or alter these models

in various ways, so as to test the remaining hypotheses.

Chapter 3

Voting in a time of crisis: how the Great Recession

affected the economic vote

After the global financial crisis of 2007–08, most developed countries slipped into recession, in

an event that has become known as the Great Recession. This was the worst event of its kind

since the Great Depression of the 1930s and a number of governments suffered catastrophic

electoral defeats in the following years. The example of Ireland was introduced earlier, where

the formerly dominant party Fianna Fáil was reduced to approximately half of its previous

vote in 2012 (Marsh and Mikhaylov 2012, 478). Such results accord with the established

theory of economic voting, which predicts that poor economic conditions will lead to voters

turning against their governments at the ballot box. On the other hand, this theory was almost

entirely developed using evidence relating to less turbulent economic conditions, what might

be described as the ordinary boom and bust cycle of the economy. Whether or not voters

respond to a severe transnational crisis in the same way as a typical recession is not yet clear.

This chapter explores this question by comparing voters’ party suport levels in 2004, well

before the crisis, to those in 2009, at the height of the first wave of the Great Recession, and

in 2014, after the initial shock had subsided.

In the immediate aftermath of the Great Recession, some political commentators argued

that the situation would benefit the Left, since they are traditionally critics of the economic

system which produced the crisis (Bartels 2012, 44) but these expectations have not been

borne out. There is little evidence of an ideological shift in OECD countries as a result of the

crisis (Bartels 2014). Commenting on the United States, Bartels (2013, 70), observes that:

The truth of the matter is that ideological mandates are exceedingly rare in Amer-

ican politics, even in times of economic crisis. Indeed, what may be most striking

about the politics of the Great Recession is how ordinary they look. In bad times,

as in good times, ordinary citizens have a stubborn tendency to judge politicians

63

64 CHAPTER 3. VOTING IN A TIME OF CRISIS

and policies not on the basis of ideology or economic doctrine, but of perceived

success or failure.

Economic voting theory offers a more plausible account of electoral behaviour during the crisis.

In an analysis of aggregate data from twenty-eight OECD countries, Bartels (2014, 188–194)

shows that governments generally received an increased vote when GDP growth was positive

and a decreased vote when negative. Kriesi (2014, 305–315) similarly finds a relationship in

European countries between incumbent vote share and the economic indicators, particularly

inflation in Central and Eastern Europe and unemployment in Western Europe. Kenworthy

and Owens (2011, 212–216) examined US voter attitudes in survey data since the 1970s and

found that voters do tend to lose confidence in whichever party is governing at the time but

they also found that this effect was actually quite weak during the Great Recession. Others

have also found the electoral response to the crisis to be weaker than might be expected (for

example, Kriesi 2012).1

These results suggest that the electoral response to the Great Recession was an economic

voting one, rather than one motivated by a deep ideological shift among voters, but they still

leave questions unanswered. The key question concerning this chapter is: was the economic

vote stronger during the Great Recession than at other times? Given that the recession was

far deeper, this seems likely, but some of the studies just mentioned have found clues that the

opposite might actually be the case. On the other hand, such findings might also be artefacts

of the particular methods used by those studies, since there has not yet been enough research

done to know how different approaches affect the results found.

This chapter has two key purposes. First, it describes how a multilevel model was construc-

ted to measure the economic vote using the particular theoretical framework of this thesis, and

second, it uses this model to compare the economic vote before, during and after the reces-

sion by using multinational survey data from the 2004, 2009 and 2014 waves of the European

Election Studies in order to shed light on the relative strength of the economic vote during

the Great Recession. The chapter begins by reviewing some relevant theory and introducing

the hypotheses that will be tested. After a brief discussion of measurement, a series of mod-

els is constructed, starting with a simple spatial voting model and proceeding to models that

measure the economic vote using first prime ministers’ parties alone and eventually all parties.

Finally, the findings will be summarised and the implications discussed.

1See Chapter 1 for a more thorough review of the literature on economic voting during the Great Recession.

3.1. PARTY SUPPORT THEORY OF ECONOMIC VOTING 65

3.1 Party support theory of economic voting

The theoretical framework of economic voting underlying this thesis was described in depth

in Chapter 1. Following van der Brug, van der Eijk and Franklin (2007, ch. 2), the vote choice

decision is conceptualised as a process in which voters have a certain level of support for each

of the parties in their country and this support level is informed by such considerations as

their ideological congruence with each party or how competent they believe that party to be

in government. Individuals are assumed to vote for the party for whom their level of support

is greatest, irrespective of the absolute level of that support. According to this model, voters

form an assessment about the condition of the economy and this assessment affects their beliefs

about the relative competence of the current government and opposition, which in turn affects

their party support levels and potentially their vote choice. This process can thus be seen as

having three stages: economic assessment, party support adjustment and finally vote choice.

The focus of this thesis is on the second stage—how does a particular sociotropic economic

assessment affect an individual’s party support levels?

Although economic voting is conceptually straightforward, there is no consensus as to how

it should be measured. In order to make the claim that one election had more economic voting

than another election, the concept of economic voting has to be precisely defined and opera-

tionalised. Different scholars have used different approaches towards this end. For example,

earlier studies used aggregate data, such as parties’ national vote share as well as national pro-

duction or unemployment rates to make an argument about economic voting. For example,

Kramer (1971) examined incumbent vote share in US House of Representatives elections.

These were the only sorts of data available during the last comparable crisis, the Great De-

pression of the 1930s. These methods are limited, however, as it is not generally possible to

make inferences about individual behaviour based on aggregate data.

More recent studies tend to use individual-level data and most of these define economic

voting implicitly, using regression analyses with interacting predictors to determine which con-

textual variables influence economic voting. It is however possible to define an explicit meas-

ure of the economic vote so that a specific figure can be estimated for the level of economic

voting at a particular election. Duch and Stevenson (2008, 44–46) did just this, construct-

ing their measure of the economic vote from survey questions asking for a respondent’s vote

choice and retrospective economic assessment. An important criterion for their vote choice

question was that respondents had to be explicitly offered the option of declaring no intention

to vote, in which case they were excluded from the analysis. The criteria used for selecting


an economic assessment question was that it had to be retrospective, ask about the national

economy and ask how the economy had changed rather than its absolute condition. They

explain that the choice of retrospective rather than prospective analysis was largely dictated

to them by practical concerns, particularly in that the majority of surveys available to them

only included the retrospective question.

In order to produce a measure of economic voting from these questions, Duch and Steven-

son (2008) estimate a multinomial logistic regression model to predict the likelihood of a

change in vote choice as a result of a change in retrospective assessment, with relevant con-

trols included. For each respondent they construct a vector of the change in vote choice prob-

abilities corresponding to an arbitrary change in assessment. They define their measure of

the ‘general economic vote’ as the average size of this vector across the entire survey (49–52).

This is described as the ‘general’ economic vote because this approach requires no measure of

incumbency—any economically motivated change in vote choice is included in this measure.

Because this is more general than most current definitions of economic voting, which see eco-

nomic voting as an activity discriminating between government and opposition parties, they

define several refinements of their measure conditioned on various measures of incumbency

(55–59).

Others are sceptical of the widespread tendency to measure economic voting in terms of

vote choice, preferring to measure party support levels instead (van der Brug, van der Eijk and

Franklin 2007, 33–36). Party support questions ask voters to assess their level of support for

each of the relevant parties in their countries. They prefer a two-stage model of vote choice, in

which voters decide who to vote for by selecting the party that they have the greatest level of

support for at that moment, it being the party support levels rather than the vote choice directly

that are influenced by the usual variables, such as economic conditions. They argue that the

widespread use of vote choice as a dependent variable has contributed to the instability of

economic voting results (15–16). Their argument is that when party support levels are quite

close a small change in predictors can cause a change in vote choice but when those support

levels are far apart, it requires a large predictor change to see a change in vote choice. Hence,

according to them, it makes more sense to use party support as a dependent variable and

this should lead to more consistent results. Although their study used individual-level survey

data, they chose to measure economic conditions using national indicators of unemployment,

inflation and economic growth, arguing that these indicators best reflect the information that

voters use to form their own judgements (69–71).

3.2. HYPOTHESES 67

This study is influenced by both Duch and Stevenson (2008) and van der Brug, van der

Eijk and Franklin (2007), and combines features of both approaches. The latter’s argument

in favour of measuring party support levels rather than party choice is persuasive, particularly

in light of the fact that the majority of the countries examined in this thesis are multiparty

systems and therefore any analysis reducing the party system to a government–opposition di-

chotomy necessarily treats disparate significant parties as identical. Furthermore, using party

support instead of vote choice offers the possibility of measuring not just which party a voter

most prefers but also by how much. This difference in support between a voter’s first and

second choice is important because if that difference is large then even a strong economic

vote might not be sufficient to cause a change in vote choice but it will cause a change in party

support. This choice of dependent variable is combined with Duch and Stevenson’s method of

measuring the degree of economic voting at a particular election by simulating the effect of

an arbitrary adjustment in economic assessment. This study also uses an individual measure

of economic assessment, but a prospective measure is used, rather than their retrospective

measure. There are theoretical reasons behind this choice, which are discussed extensively in

Chapter 1, as well as data-driven reasons discussed in the next section.

3.2 Hypotheses

This chapter tests three specific economic voting hypotheses. The first is:

Hypothesis 3.1 Individuals’ party support levels were influenced by their prospective sociotropic

economic assessments in each survey year.

This hypothesis tests whether the specific framework of economic voting developed in this

thesis so far accurately describes the intended voting behaviour of the respondents in these

surveys. As this framework is built on very well-established principles of economic voting the-

ory, a failure to confirm this hypothesis ought to be seen as a sign of flawed operationalisation,

rather than a deficit in the theory. This hypothesis makes no reference to incumbency as it is

concerned with the most general form of economic voting, namely any change in party support

attributable to economic voting, irrespective of which parties benefit and which suffer.

The second hypothesis relates to the more specific and most usual conception of economic

voting, in which voters are supposed to treat government and opposition parties differently:


Hypothesis 3.2 There was an observable tendency for citizens to vote economically in each year.

In particular, voters holding an optimistic economic assessment are more inclined to support gov-

ernment parties and less inclined to support opposition parties than those holding a pessimistic

assessment.

This builds directly on the first hypothesis. If it can be shown that the general form of economic

voting does occur then the next step is to show that voters discriminate between these different

groups of parties.

One of the central claims of this thesis is that voter behaviour during times of severe global

recession cannot necessarily be predicted from theories developed to describe the typical boom

and bust cycle of a national economy. It has long been suspected that economic issues play

a larger role in voting behaviour as they become more severe (for example, Bloom and Price

1975, 1240). Although some recent studies found the crisis response to be weaker than ex-

pected (Kenworthy and Owens 2011; Kriesi 2012), these were studies of government vote

share, so they may be concealing movements that would be observable with a party support

measure. This chapter’s third and final hypothesis is thus:

Hypothesis 3.3 The economic voting effects were heightened during the Great Recession. Spe-

cifically, there was a greater observable tendency to vote economically in 2009 than in either 2004

or 2014.

As the 2009 survey took place at the peak of the first wave of the crisis, unlike the 2004 survey,

which was conducted during a time of normal economic activity, any form of crisis-specific

behaviour should be observable in a comparison between the two surveys. As the economy

was naturally a highly salient issue during the crisis, it is predicted that the inclination towards

economic voting became stronger at that time. Similarly, once the initial shock of the recession

had passed, it would be expected that normal patterns of behaviour would reassert themselves,

so it is predicted that the inclination to vote economically had receded to their pre-crisis levels

by 2014.

3.3 Measuring the economic vote

There are strong theoretical arguments for both retrospective and prospective models of vot-

ing, and it has even been argued that the two are in many respects equivalent, as the former

informs the latter (Downs 1957). Often the choice of which measure to use is driven by the

3.3. MEASURING THE ECONOMIC VOTE 69

Figure 3.1: Support for incumbent prime minister’s party by economic assessment

2004 2009 2014

2

3

4

5

6

7

2

3

4

5

6

7



mea

n su

ppor

t

Mean support for the incumbent prime minister’s party according to the respondent’s prospect-ive and retrospective economic assessment in each survey year, shown with 95% confidenceintervals and accounting for party-level variance. Both forms of economic assessment aremeasured on a five-point scale ranging from −2, indicating a very negative assessment, to +2,indicating a very positive assessment, with 0 indicating a neutral assessment. Source: EES

available data, and this is an important consideration here too. Although each survey includes

both retrospective and prospective economic assessment questions, descriptive analysis shows

that these questions are not equally useful, particularly in 2009 during the crisis. In both 2004

and 2014, responses to the prospective and retrospective questions are similarly distributed,

appearing approximately binomial besides a marked tendency for respondents to avoid the

most positive category. In 2009, the prospective assessment variable is similarly distributed

but the distribution of the retrospective assessment variable is heavily skewed towards the

negative responses, with fewer than eight percent of respondents giving a positive response.

This is not surprising, since the condition of the economy at that time had in fact worsened

badly, but it does mean that the retrospective evaluation in 2009 is not suitable for use in this

analysis, as there is too little variance.


Prospective assessment can be predicted from retrospective assessment by a simple linear

model with moderate success in both 2004 (R2 = 0.30) and 2014 (R2 = 0.39), which confirms

the Downsian view that the two measures can be seen as related. On the other hand, this

relationship breaks down in 2009, with the same model no longer able to predict prospect-

ive assessment with any accuracy (R2 = 0.08). It is interesting that, despite the consensus

among respondents about the worsening of the economy over the past year, there was still

a diversity in opinion regarding its future course, with a similar distribution to that of the

other years. Furthermore, since the dependent variable is prospective in orientation too, it is

more consistent to choose a prospective orientation for the economic assessment measure as

well. Finally, the relationship between support for the prime minister’s party in each country

and both economic assessment measures was examined. As Figure 3.1 shows, the relationship

between these measures is approximately linear, except in the case of retrospective assessment

in 2009. For these reasons, only the prospective measures are used in this thesis.

A measure of party support has been constructed from a sequence of questions included

in each of the EES surveys asking voters the likelihood that they would ever vote for each of

a number of parties in their countries, on a scale from zero to ten, where zero is described as

‘not at all probable’ and ten as ‘very probably’. Although the question does ask respondents if

they would ‘ever’ vote for a particular party, this interpretation treats the responses as current

support for that party instead, and there is some evidence that the respondents interpret the

question that way as well. For example, even people who stated that they identify with a

particular party do not overwhelmingly respond with ten for that party. Furthermore, the fact

that there is a relationship, as will be shown, between this measure and questions about recent

economic conditions does suggest that the current feeling interpretation is more appropriate

than a strict reading of the question as posed. As each individual responds to several party

support questions, corresponding to the key parties in their country, there is one measurement

for each party–individual pair, and these measurements have been centred around the group

mean for each individual.2

3.4 A spatial model of party support

Naturally the economy is not the only consideration that influences a person’s party support

levels. As outlined in Chapter 1, there is an enormous vote choice literature describing the vari-

ous influences that are known to affect the vote choice decision. In order to obtain unbiased2More details on the measurement of this and other variables can be found in Chapter 2.

3.4. A SPATIAL MODEL OF PARTY SUPPORT 71

Figure 3.2: Relationship between party support and left–distance

1

2

3

4

5

0 2 4 6 8 10left–right distance

mea

n su

ppor

t

Mean support for parties according to the left–right distance between the party and the re-spondent, shown with 95% confidence intervals and accounting for variance at the individual,party and country levels. Source: EES

estimates of the influence of economic considerations on party support, other key influences

need to be controlled for. There are two key predictors that will be taken into account, spatial

proximity and party identification.

In order to measure spatial proximity, survey questions asking respondents to place both

themselves and each major party in their country on the left–right spectrum are used. It is

expected that those voters who consider themselves close to a party on the left–right spectrum

will generally have a greater level of support for that party than those who consider them-

selves further away. There is however no reason to expect that relationship to be linear, as

the difference between zero and one point of distance, representing a change from a com-

plete alignment to a close alignment, is conceivably of greater significance to a voter than the

difference between nine and ten points of distance, both of which represent a fundamental

mismatch. Plotting the mean party support for each level of left–right distance confirms that

this relationship is indeed negative and non-linear, as Figure 3.2 shows.

As well as spatial proximity, party identification is known to be an important predictor of

vote choice. Accordingly, it is expected to be a strong predictor of party support as well and

cursory analysis of the data confirms that these two variables are closely correlated. Plotting


mean party support against left–right distance for party identifiers and non-identifiers separ-

ately shows that the effect of left–right distance is different for the two groups. In order to

model these characteristics accurately, the model needs to include both a linear and quadratic

term for the left–right distance variable as well as an interaction between left–right distance

and party identification. This leads to the following regression equation for a single party:

y = β0 + β1∆+ β2∆2 + β3 I + β4∆I + ε,

where y represents an individual’s support for that party,∆ represents the mean-centred left–

right distance between that individual and that party and I is a dummy variable indicating

that the individual identifies with that party.3

The dependent variable, party support, has been centred around the mean for each indi-

vidual. Most individuals surveyed recorded a level of support for each of several parties, so

the mean response for a particular individual was subtracted from that individual’s response

for each party. This was done because it is the difference between the levels of support for

each party that is of most interest in this models and the absolute levels may be misleading

if different respondents adopted different conceptual reference points when responding to

these questions. Another advantage of centring the dependent variable in this way is that

the centred variable is more approximately normal than the raw variable, which is heavily

skewed because respondents are much more likely to assign low scores than high scores. The

assumption of residual normality is therefore more plausible using the centred variable. The

means are examined separately later in the chapter. The predictors have also been centred in

order to improve the efficiency of the estimation process. This should not in principle change

the particular coefficient estimates but it does mean that the intercept estimate can no longer

be directly interpreted. The left–right distance variable has been centred around the mean

for each election, that is each country and year. The party identification dummy variable was

centred around the grand mean.

In addition to these predictors, a number of further predictors were included to control

for personal attributes generally known to influence voting behaviour. These predictors were

age, sex, education, population density and labour force status. Age is measured in decades,

so that the corresponding coefficient is not too small to be conveniently interpreted. Education

level is divided into three groups, the first being those with a university degree, the second

3An argument could be made for the inclusion of an interaction between the quadratic distance term and theparty identification dummy variable but empirical tests have shown that the models including that term are notsignificantly more accurate, so it has been omitted for the sake of parsimony.

3.5. THE PRIME MINISTER’S PARTY AND THE ECONOMIC VOTE 73

being those who have not completed high school and the reference group being those who

have completed high school but who do not possess a university degree. Population density

is also divided into three groups, those residing in cities, those residing in towns and those

in rural areas, the reference group being towns. Lastly, labour force status is divided into

working, unemployed and other groups. The other group consists of people who are not in

the workforce for any reason, such as retirement. The reference group is workers.4 All of

these control variables were centred around the grand mean. An interaction has also been

included between party identification and age, since it has been known for some time that

party identification tends to be stronger among older people (Campbell et al. 1960, 161–164;

Converse 1976; Berglund et al. 2005).

As was mentioned earlier, the dependent variable, party support, has been centred around

each individual’s mean response, so as to control for differing centres of support. Modelling

the uncentred variable with these control variables has shown that there are relationships

between some of the controls and an individual’s mean party support. For example, women

tend to report higher levels of party support than men do, irrespective of the party, although

the difference in levels of support between the parties does not appear to be related to these

variables. A naive analysis of an arbitrary party would thus be likely to conclude that women

are more likely to support that party, even when this is not the case. Centring party support in

this way avoids this problem, while preserving the differences in support between the parties.

3.5 The prime minister’s party and the economic vote

There are several ways this spatial model can be extended in order to test the hypotheses

posited at the beginning of this chapter. One of the most straightforward ways to do this is to

analyse support for only the prime minister’s party in each country. As this is the party that

could be expected to be most strongly affected by any economic voting effect, if the hypotheses

are true then this ought to be reflected in such a model. This does not mean that the other

parties in each country are ignored. Since the party support variable is centred around the

individual’s mean response, it is not measuring absolute support for the prime minister’s party

but rather the relative support for that party. Therefore, even situations where the absolute

level of support for the prime minister’s party remains steady while that for opposition parties

increases would be observed by such a model. Another advantage of this approach is that the

multilevel structure of the data can be simplified considerably. The data can be described by4Further details on how these variables were measured can be found in Chapter 2.


Table 3.1: Prime ministers’ parties model

Fixed effect Coeff. SE p

Intercept 0.067 (0.095) 0.492Year 2009 0.490 (0.183) 0.014Year 2014 0.159 (0.130) 0.235Left–right distance −0.428 (0.019) < 0.001Left–right distance2 0.023 (0.001) < 0.001Party ID 4.706 (0.129) < 0.001Prospective assessment 0.323 (0.053) < 0.001Female 0.054 (0.021) 0.011Age (decades) −0.007 (0.008) 0.353High education 0.084 (0.025) 0.001Low education 0.028 (0.031) 0.357Urban area −0.018 (0.026) 0.495Rural area 0.065 (0.027) 0.016Unemployed −0.083 (0.043) 0.055Not in workforce 0.052 (0.025) 0.038Distance × party ID 0.280 (0.029) < 0.001Age × party ID 0.179 (0.015) < 0.001Prosp. assess. × year 2009 −0.051 (0.048) 0.310Prosp. assess. × year 2014 −0.038 (0.065) 0.565

Fixed effect coefficient estimates from Model 3A. The dependent variable is support for theincumbent prime minister’s party in the respondent’s country. Several country-level randomeffects terms are also included in the model and the corresponding variance estimates canbe found in Appendix B. Sample size is 51962 individuals within 25 countries. Pseudo R2 is0.541. Source: EES

a two-level model, where individuals are grouped within countries. No party level is required

as each country only has one prime minister’s party.

Several things have to be done in order to extend the spatial party support model to meas-

ure economic voting intention for prime ministers’ parties in each of the studied countries.

The intention of Model 3A is to measure the effect of economic assessment on party support,

so a prospective economic assessment predictor was added to the model, centred around the

grand mean. The survey year is indicated by the addition of dummy variables for 2009 and

2014, using 2004 as the reference case. Interaction terms between the time and economic

assessment variables were also added, so that it can be ascertained whether the degree of eco-

nomic voting differed among the three years. As this is a multilevel model, a random intercept

for the country was added, along with several random slopes. The economic assessment and

time variables, as well as their interactions, were given random slopes so that it is not as-

sumed that the degree of economic voting intention was equal in every country. Similarly, the

party identification and left–right distance variables and their interaction were given random


Figure 3.3: Predicted support for prime minister’s party by left–right distance

2

4

6

8

0 2 4 6 8 10left–right distance

pred

icte

d su

ppor

t

groupparty identifiers

others

Predicted support for the incumbent prime minister’s party in 2004 according to the left–rightdistance between the respondent and party as well as whether the respondent identifies withthat party or not. Predictions are derived from Model 3A and are shown with 95% predictiveintervals. These predictions assume a neutral prospective economic assessment. Source: EES

slopes because single-party analysis showed that the shape of these relationships differed of-

ten considerably from party to party. Random slopes were considered for each of the control

variables as well but the estimated variances of these slopes were very small and the resulting

models did not fit the data significantly better, so these were not included in the final model.

The key results from this model are shown in Table 3.1. As most of the variables have been

centred, it is not trivial to interpret the intercept or the interaction coefficients. For this reason,

post-estimation simulation has been used to derive estimates of the quantities of interest.5

As expected, spatial proximity is an important predictor of party support. Figure 3.3 shows

the relationship between left–right distance and predicted party support for both the party

identifier and non-party identifier groups in 2004. In both groups, it is assumed that the

individual has a neutral prospective assessment of the economy.6 From this plot it can be seen

that party support falls as left–right distance increases but that the size of this drop diminishes

5The post-estimation simulation approach to interpreting complex models is used throughout this thesis. Sincethe regression coefficients are usually not interpreted directly, the coefficient tables for most of the models discussedare not included in the main text but can be found in Appendix B. See Chapter 2 for a more detailed discussion ofthis decision and the reasons behind it.

6Unless otherwise specified, these predictions are for a 40-year old male who has completed high school butnot university, who lives in a town and is employed.


with increasing distance. In other words, the difference between zero and one point of distance

is much more important than the difference between nine and ten points of distance. It can

also be seen that party identifiers have a much stronger level of support than non-identifiers at

any degree of proximity. In fact, the importance of distance is weaker for identifiers than non-

identifiers. There is one surprising feature of this plot, which is that after six points of distance,

it appears that further distance is associated with increased party support. This is most likely

an anomaly resulting from the fact that few respondents identify with a party on the other side

of the political spectrum from themselves. In fact, the error around the predictions is large

enough that a levelling off is perfectly plausible.

The inclusion of prospective economic assessment in the model makes it possible to es-

timate the economic vote.7 Because interactions were included between this variable and

the time dummy variables, the model gives a measure of the economic vote in each year. In

particular, the economic vote for a particular party will be defined as the difference in party

support between the optimists, those who believed the economy will be much better in twelve

months time, and the pessimists, those who believed the economy will be much worse in

twelve months time. In other words, this quantity is the total change in an individual’s sup-

port for a particular party that could be accounted for in the model by a change in economic

assessment. For example, the party support predicted by the prime minister’s party model for

an individual in 2004 who does not identify with that party but who occupies the same posi-

tion on the left–right spectrum is 5.21 (SE= 0.19, p < 0.001) if that individual is an optimist

but only 3.92 (SE = 0.12, p < 0.001) if that individual is a pessimist. The difference, and

hence the estimated economic vote in 2004, is 1.29 points (SE= 0.21, p < 0.001).

By contrast, the estimated economic vote for 2009 is only 1.09 points (SE = 0.13, p <

0.001). There was thus an apparent decline in the economic vote between the two years

of 0.20 points (SE = 0.19, p = 0.30) but this was not significant. Similarly, the estimated

economic vote for 2014 is 1.14 points (SE = 0.17, p < 0.001), which is not significantly

different from either 2009 (∆ = 0.06, SE = 0.21, p = 0.79) or 2014 (∆ = 0.14, SE = 0.26,

p = 0.58). In other words, this model suggests that the strength of the economic vote was

approximately constant over the entire period, at least when considering prime ministers’

parties only. Figure 3.4 shows the relationship between prospective economic assessment and

7The term ‘economic vote’ is used in this thesis to describe any change in voters’ support for the differentparties in their countries that can be attributed to their assessment of the economy. This might not necessarily bemanifested as an actual vote choice for several reasons, which are discussed in Chapter 1. The empirical focus ofthis thesis is on the stage of the voting process in which party support levels are formed, which precedes the votechoice stage.


Figure 3.4: Predicted support for prime minister’s party by economic assessment

4.0

4.5

5.0

5.5

6.0

-2 -1 0 1 2prospective economic assessment

pred

icte

d su

ppor

t

year2004

2009

2014

Predicted support for the prime minister’s party according to prospective economic assessmentfor each survey year, based on Model 3A. Economic assessment ranges from −2 (very pessim-istic) to +2 (very optimistic). These predictions are for an individual who does not identifywith the party and who occupies the same left–right position as the party. Source: EES

predicted support for the prime minister’s party in each year. The slope of each line indicates

the strength of the economic vote in the corresponding year, so the fact that the three lines are

approximately parallel reflects its stability over time. It is interesting to note that the lines do

not overlap, despite being approximately parallel. This suggests the level of support for the

prime minister’s party was greater in some years than in others, irrespective of the individual’s

economic assessment. In particular, it appears that support was greatest in 2009. For example,

a typical voter holding a neutral economic assessment was 0.48 points (SE = 0.18, p < 0.01)

more supportive of the prime minister’s party in 2009 and 0.32 points (SE = 0.18, p = 0.06)

more supportive in 2014 than in 2004. Thus, although the economic vote for prime ministers’

parties was stable over time, those parties did receive a small boost in support at the height

of the recession, irrespective of economic assessment.

In general, the control variables have little if any effect. Women are slightly more likely to

support the prime minister’s party than men, although the size of this effect is tiny, accounting

for an increase in party support of only 0.05 points (SE= 0.02, p = 0.01). For comparison, an

increase from zero to one point of distance accounts for a decrease of 0.57 points (SE= 0.02,


p < 0.001). Education also plays a modest role, with support 0.08 points (SE = 0.02, p <

0.001) greater among university educated voters than voters who have not attended university.

The effect of employment status is likewise small. The employed and unemployed groups are

not significantly different from each other (∆ = 0.08, SE = 0.04, p = 0.06) but the not in

the workforce group is slightly more likely to support the prime minister’s party. Once again

this effect is tiny, accounting for a mere 0.05 point (SE = 0.02, p = 0.04) increase in party

support over the employed group. Age has an even smaller effect, with each extra decade of

age associated with a 0.02 point (SE= 0.01, p < 0.01) decrease in support.8 There is also an

interaction between the effects of age and party identification, and this is more substantial. At

the age of twenty, party identification is associated with a 3.30 point (SE = 0.13, p < 0.001)

increase in party support and this is increased by a further 0.18 points (SE= 0.02, p < 0.001)

for each additional decade of age.

In summary, Model 3A offers evidence that support for the prime minister’s party is linked

to the individual’s prospective assessment of the economy, with optimistic voters being more

supportive than pessimistic voters. These findings support the first two hypotheses. The third

hypothesis predicted that the strength of the economic vote would be greater in 2009 but

this was not supported by the model, which found the economic vote to be stable over time.

It was however found that support for prime ministers’ parties was slightly higher in 2009

than at other times and this was true irrespective of economic assessment. This model has

the advantage of being relatively straightforward to interpret but it is also limited by the fact

that it only analyses support for a single party in each country. This is problematic because

a change in support for the prime minister’s party is of limited import if support for all of

the other parties moved in the same way.9 The next section discusses how this model can

be extended to take into consideration support for multiple parties in this country and thus

address this limitation.

3.6 A multiparty model of the economic vote

The economic voting model developed thus far can be generalised to analyse not only prime

ministers’ parties but also the other parties in each country. In principle, this requires a far8This may appear to contradict Table 3.1, which shows no significant age effect. This apparent contradiction

arises because the model includes an interaction between party identification and age. Since the party identificationvariable has been centred, it is not actually zero for non-party identifiers. This illustrates why post-estimationsimulation has been used to interpret model estimates rather than discussing regression coefficients directly.

9As the dependent variable is centred around the individual’s mean party support level, parallel movementhas already been eliminated, so this particular example would not be a problem in practice. Unfortunately, thereare similar situations that are not neutralised by centring the dependent variable, so these possibilities do requireserious consideration.

3.6. A MULTIPARTY MODEL OF THE ECONOMIC VOTE 79

Figure 3.5: Data classification structures

Country

Party Individual

Measurement

(a) Classification structure of the stacked dataset.Measurements are cross-classified into partiesand individuals, which are also independentlynested within countries.

Country

Party

Measurement

(b) Simpler, hierarchical classification structurein which measurements are nested within partiesand parties are nested within countries.

more complex model in order to be analysed accurately. This is because the same individuals

responded to questions about each party within a country, so measurements such as the left–

right distance between a party and an individual are cross-classified within both parties and

individuals. Furthermore, both of those groups are nested within countries. Because of this

data structure, which is illustrated in Figure 3.5a, it is important to be aware of the possibility

that observations within these groups are more alike than observations from different groups.

The degree of this within-group similarity can be measured by the intraclass correlation. The

intraclass correlation of the uncentred party support variable is 0.08 by individual, 0.13 by

party and 0.01 by country. This implies that measurements within a particular individual or

a particular party are more similar than between those groups but once these effects have

been accounted for there is very little further similarity between measurements within a given

country. Since the party support variable has actually been centred around the individual-level

mean, some of the within-group similarity can be expected to have been eliminated. In fact,

the intraclass correlation of the centred party support variable is 0.14 by party and zero by

both individual and country.

This suggests that random intercepts are only required for the party level, despite the

complex structure of the data. Even though the intraclass correlations of the other groups are

low, it is still possible that random slopes might be required for these groups, so a subset of the

data was used to explore various options but in every case the variance of the random slope

was either too small for the model to be estimated successfully or a chi-square difference test

showed the resulting model not to be a significant improvement over the equivalent model

without the random slope. In other words, the full complexity of the cross-classified model is

not required and the data can be modelled with the hierarchical structure shown in Figure 3.5b


instead. This means that the generalised model need only differ from the prime minister’s party

model in two respects. First, it can include a number of parties in each country, rather than

just the one. And second, whereas the random intercept and slopes are grouped by country

in the simpler model, they are grouped by party in the generalised model. Of course, this

difference is purely a question of interpretation, as with only one party per country in the

prime minister’s party model, the party and country groupings are identical. Model 3B was

formed in this way and differs from the prime minister’s party model in one further respect,

which is the inclusion of party-level random slopes for each of the control variables.10

The results from this model are similar in most respects to those from the prime minister’s

party model but there are some important differences. Since not just incumbent parties but

also opposition parties are included in this model, it would be expected that the fixed effect

for party support would no longer be significantly different from zero. That is, whereas a

unit increase in optimism about the economy is associated with a small increase in support

for the prime minister’s party, it is not expected to be associated with an increase in sup-

port for any arbitrary party. Using the same measure of economic voting intention as before,

namely the change in predicted party support associated with the maximum possible increase

in prospective economic assessment, the economic voting intention for an arbitrary party was

−0.13 points (SE= 0.09, p = 0.14) in 2004, −0.07 points (SE= 0.08, p = 0.34) in 2009 and

−0.13 points (SE= 0.08, p = 0.13). None of these is statistically significant, nor is any of the

differences between them. This confirms these expectations.

Although the economic voting intention for an arbitrary party is not significant, this is not

necessarily the case for a specific known party. It has already been shown that for prime minis-

ters’ parties this effect is positive. Furthermore, the theory predicts a negative economic voting

intention for opposition parties. Because the model includes random slopes for the economic

assessment and time variables as well as the interaction between them, it is possible to extract

separate economic voting intention estimates for each party. Figure 3.6 shows the estimated

economic voting intention for each party grouped by year and incumbency status. This shows

that in each year, government parties were usually subject to a positive economic voting inten-

tion and opposition parties were mostly subject to a negative economic voting effect, although

there are some exceptions, particularly among opposition parties. Note also that the points in

10These random slopes make sense from a theoretical perspective because different parties can be expected toappeal to different groups in society, so the control effects ought to take different values for different parties. Thisargument applies equally well to 3A but in practice the variances were low enough to cause convergence problemswith the smaller sample size, so they were omitted from that model.


Figure 3.6: Economic vote by year and incumbency status

2004 2009 2014

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●●

●●

●●

●

●●

●

●

●●

●

●

●

●

●

●●

●

●

●●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●●●

●

●

●

●

●

●

●

●

●●● ●●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

● ●

●

●

●

●

●

●

●

●●

●

●

●

●

●●

●●

●

●

●

●

●

●●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●●●

●

●

●

●

●

●●

●●

●

●

●

●

●●

●

●

●●●

●

●

●

●

●

●

●

●

● ●●

●

●

●

●

●

●

●

●

● ●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

● ●

●

●

●

●

●● ●

●

●

●●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●

●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●●

●

●

●

●●

●

●

●

●

●

●

●

●●●

●

●●●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

−3

−2

−1

0

1

2

3

PM

cabin

et

oppo

sition PM

cabin

et

oppo

sition PM

cabin

et

oppo

sition

econ

omic

vot

e

Economic vote for each party according to the survey year and whether the party was theprime minister’s party, a junior coalition partner or an opposition party. Economic vote is thepredicted increase in support for a party associated with the maximum possible increase inprospective economic assessment, based on Model 3B. Each point represents a single party.Source: EES & ParlGov

each group are less spread out in 2009 than they are in the other years. The standard devi-

ation of the party economic votes in 2009 was 0.77 points, compared to 1.20 points in 2004,

and 0.91 points in 2014. This suggests that the overall amount of economic voting decreased

during the recession and recovered partially in the following period.

The above analysis suggests that an optimistic economic assessment is typically associated

with an increased support for government parties and a decreased support for opposition

parties but it does not test this hypothesis. In order to do so, the model was extended so as

to include an incumbency term and the interactions between incumbency and the terms used

to measure the level of economic voting intention. The resulting Model 3C allows for the

possibility that economic voting intention differs for government and opposition parties. This

was done by including a dummy variable to distinguish between government parties, those

represented in cabinet, and other, opposition, parties. The model also includes a random

slope at the country level for the interaction between incumbency and prospective economic

assessment, which allows for the possibility that the amount of economic voting was different

in different countries. A random intercept for the country level was also added. Random slopes


Figure 3.7: Predicted support by year and incumbency status

2004 2009 2014

●

●

●

●

●

●

●

●

●

●

●

●

●

●

●

3.50

3.75

4.00

4.25

4.50

−2 −1 0 1 2 −2 −1 0 1 2 −2 −1 0 1 2prospective economic assessment

pred

icte

d su

ppor

t

status ● government opposition

Predicted support for a party according to the respondent’s prospective economic assessment,the party’s incumbency status and the survey year, based on Model 3C. Economic assessmentranges from −2 (very pessimistic) to +2 (very optimistic). These predictions are for an indi-vidual who does not identify with the party and who occupies the same left–right position asthe party. Source: EES & ParlGov

were considered for related terms, such as the three-way interactions between incumbency,

economic assessment and time, but these had a variance very close to zero and were therefore

omitted.

This extended model was used to estimate the net economic vote, that is, the total pre-

dicted change in support for governing parties relative to opposition parties that can be at-

tributed to economic assessment. For example, in 2004, the economic vote for government

parties was 1.02 points (SE = 0.15, p < 0.001) and the economic vote for opposition parties

was−0.70 points (SE= 0.10, p < 0.001). This means that the net economic vote for 2004 was

1.72 points (SE= 0.20, p < 0.001). As expected, the economic vote is significant and positive

for government parties in each year and significant and negative for opposition parties in each

year. The net economic vote in 2009 was only 1.18 points (SE = 0.19, p < 0.001), which is

0.55 points (SE = 0.20, p < 0.01) less than in 2004. By 2014, the net economic vote was

1.22 points (SE = 0.19, p < 0.001), which is not significantly greater (∆ = 0.05, SE = 0.19,

p = 0.81) than in 2009. In other words, the net economic vote became weaker during the

recession and did not recover in the following years.


The economic vote in the three different years can be seen graphically in Figure 3.7. The

lines show the relationship between economic assessment and party support in each year. The

solid lines show government support while the dashed lines show opposition support. The fact

that an optimistic assessment is good for government parties and bad for opposition parties

is reflected by the positive slope for each of the government lines and the negative slope for

the opposition lines. The steeper slopes in 2004 show the stronger economic vote in that year.

This figure also shows that support for government parties was greater in 2009 and 2014

than in 2004, irrespective of the voter’s economic assessment. As a result of this increase in

government support, the intersection point of the lines appears to have changed over time.

This would imply that a more strongly pessimistic assessment was required to cause a net

shift in support towards the government in those later years. In fact, it is only in 2004 that the

intersection occurs at a point corresponding to an economic assessment significantly different

from completely neutral, specifically −0.79 [−1.50,−0.22].11 The corresponding assessments

were −0.18 [−0.62,0.25], in 2004 and −0.39 [−1.01,0.16], in 2014. On the other hand,

none of the differences between these assessments was significantly different,12 so it cannot

be concluded that this apparent difference did not arise by chance.

So far this discussion has only contrasted government and opposition parties. It has been

suggested that the prime minister’s party ought to be more strongly affected by economic vot-

ing than junior coalition partners (van der Brug, van der Eijk and Franklin 2007, 56–58). The

argument is that these parties are often able to criticise the government during election cam-

paigns and make the case to voters that they ought to have more seats and so be in a position

to demand greater influence within a government. The prime minister’s party naturally lacks

this freedom to distance itself so easily from the government’s performance. In order to test

this hypothesis, Model 3D was formed by adding further terms to distinguish prime ministers’

parties from other cabinet parties. Based on this model, it appears that prime ministers’ parties

typically are subject to a somewhat stronger economic vote than other coalition parties. This

stronger economic vote amounts to an additional 0.44 points (SE = 0.23, p = 0.06) in 2004,

0.61 points (SE = 0.21, p < 0.01) in 2009 and 0.70 points (SE = 0.22, p < 0.01) in 2014.

These effect is not significantly different from year to year and the estimated variance of the

corresponding random slope is very close to zero. This suggests that the extra economic vote

that prime ministers’ parties receive is stable across both time and country. The results from

11Square brackets indicate 95% confidence intervals.12The difference between 2004 and 2009 was −0.61 [−1.42,0.09], the difference between 2009 and 2014 was

0.40 [−0.41, 1.27] and the difference between 2004 and 2014 was −0.22 [−0.96,0.50].


this model are otherwise similar to the results from the previous model, which only distin-

guishes government and opposition parties.

The multiparty approach to modelling the economic vote has confirmed that a positive

economic vote is not only associated with increased support for governing parties but also

decreased support for opposition parties. These results support for the first two hyptotheses.

It was also found that prime ministers’ parties in particular are more strongly exposed to eco-

nomic voting effects than other opposition parties. The net strength of the economic vote

became weaker during the recession, rather than stronger as the third hypothesis predicted.

This result also appears to contradict the earlier finding that the economic vote that prime

ministers’ parties were exposed to remained stable over time. This is actually not a contra-

diction as the net economic vote takes into account the impact of economic assessment on

both government and opposition parties. In fact, the results from Model 3D confirm both

that the economic vote for prime ministers’ parties remained stable over time and that the net

economic vote fell in 2009. This is interesting because it implies that the weakening in the

economic vote occurred not because voters were more forgiving of their governments than

usual but rather because they were less inclined to support opposition parties in their stead.

3.7 Influence of the economy on mean party support

Since the dependent variable has been centred around its individual-level means, the analysis

presented so far only takes into consideration relative changes in party support, not a shift in

mean party support. That is, any increase in support for one party at the expense of another

is observed but an increase in support for all parties would not be evident. Although it is

this relative change that is of primary interest, it is worth exploring the possibility of absolute

change in party support in order to gain a more complete understanding of voter behaviour.

This can be done by examining the means of the various party support scores given by each

respondent to the parties in his or her country. These are the same values that the scores

were centred around to form the centred party support used as the dependent variable in the

earlier models. Model 3E was constructed to predict mean party support from an individual’s

prospective economic assessment, their demographics and the year, as well as the interaction

between their economic assessment and the year. This corresponds to the basic economic

voting model introduced earlier except that none of the party-specific variables are included

as they are not meaningful here. As with earlier models, random slopes have been included

for the time and economic assessment variables and their interactions.

3.7. INFLUENCE OF THE ECONOMY ON MEAN PARTY SUPPORT 85

The results from the mean party support model are consistent with the findings made so

far. The predicted mean party support for an individual with a neutral economic assessment

was 3.62 points (SE = 0.11, p < 0.001) in 2004, 3.27 points (SE = 0.10, p < 0.001) in

2009 and 3.19 points (SE = 0.10, p < 0.001) in 2014. This means that voters were less

supportive (∆= 0.38, SE= 0.15, p = 0.02) of all parties in 2009 than in 2004. There was no

significant difference (∆= 0.08, SE= 0.11, p = 0.48) between 2009 and 2014. The predicted

increase in mean party support associated with a maximal increase in economic assessment

was 0.52 points (SE= 0.10, p < 0.001) in 2004. This means that optimistic voters were more

likely to to report higher support for all parties than pessimistic voters. By 2009, this had fallen

by 0.21 points (SE = 0.09, p = 0.02) to 0.31 points (SE = 0.09, p < 0.001) and by 2014 it

had risen by 0.49 points (SE = 0.13, p < 0.001) to 0.80 points (SE = 0.10, p < 0.001). This

is also significantly greater (SE = 0.14, p = 0.04) than the 2004 level. Taken together, this

means that voters became less supportive overall of their political parties during the recession

and their overall support was less strongly influenced by their economic assessment. This is

consistent with the earlier findings that economic voting effects generally were depressed in

2009.

Some of the control variables were also shown to have a significant effect on mean party

support. Each extra decade of age was associated with a 0.09 point (SE = 0.004, p < 0.001)

decrease in mean party support. Women had a slightly higher mean party support level than

men (0.10 points, SE= 0.01, p < 0.001). Those with a university education had a mean party

support level 0.03 points (SE = 0.02, p = 0.02) greater than those who had only finished

high school. The differences between the other education groups were not significant. There

were also no significant differences between people living in cities, towns or rural areas. Un-

employed voters had a mean support level 0.04 points (SE = 0.03, p = 0.10) greater than

unemployed voters and 0.07 points (SE= 0.03, p < 0.01) greater than those not in the work-

force. Although many of these differences are statistically significant, the effect sizes are very

small. Furthermore, this model has a pseudo R2 of only 0.12, meaning that almost all of the

variance in mean party support is unexplained by these variables.13 These results seem to

confirm the assumption of the earlier models that it is primarily relative party support and not

absolute party support that is affected by the key variables of interest.

13It is well known that feeling thermometer scales and survey attitude measurements suffer from low reliability(Wilcox, Sigelman and Cook 1989; Alwin and Krosnick 1991; Alwin 1997), so it should not be surprising that thereis so much residual variation. Individual variation in absolute scores need not affect the substantive results of thisstudy, so long as relative party support is consistent.


3.8 Conclusion

This chapter has tested three main hypotheses. The first hypothesis was that party support

was influenced by an optimistic economic assessment and the second was that this improved

support for government parties and reduced support for opposition parties respectively. The

evidence strongly supports both of these hypotheses. In all three years economic perceptions

had a clearly observable effect on party support and this is apparent in the relatively simple

prime minister’s party model as well as the more complex multiparty models. The former

model further suggests that this operates in the expected direction, inasmuch as prime minis-

ters’ parties benefit from a positive economic assessment. The multiparty model confirms that

this is the case and demonstrates that other cabinet parties also benefit from such an assess-

ment and that opposition parties benefit from a negative economic assessment. These results

are not very surprising as they are in agreement with a large economic voting literature. What

makes these results significant is that they demonstrate the validity of the multilevel party

support approach to measuring economic voting. Although party support models have been

pioneered by van der Brug, van der Eijk and Franklin (2007), the use of party support within

a multilevel model is novel, so it is important to show that this method can replicate what is

already known.

The third hypothesis is that these economic voting effects were heightened during the

Great Recession. This question has generated little prior comparative research, as these events

were quite recent. The expectation was that the increased salience of the economy would

have led voters to be more inclined than usual to hold their governments to account for the

exceptionally bad economic conditions at the time. In fact, these findings show the opposite to

have been the case. The size of the net economic voting effect was shown to be considerably

smaller in 2009 than in 2004 and it had not recovered to its pre-recession levels by 2014.

What this means is that a larger change in economic assessment was required to achieve a

unit change in party support in 2009 than in 2004. However, this may not necessarily have

been to the benefit of incumbents. The change in party support required for a change in vote

choice depends on how close the prior party support levels were, so if the party support levels

were close enough then even the smaller economic voting effect could have led to a change in

government. Although surprising, others have also found the electoral response to the crisis

to be weaker than expected (Kriesi 2012; Kenworthy and Owens 2011).

How do we explain this surprising result? One possibility is that during the Great Reces-

sion, the economy may have been seen as a less useful tool to discriminate between parties.

3.8. CONCLUSION 87

As many opposition parties have previously been in government, if the crisis were seen as a

structural issue then even opposition parties may have been held partly responsible for the

situation, depressing the importance of the economic vote. If this were the case, then it would

be expected to see a move away from the centre or that extreme or anti-system parties would

benefit from negative perceptions of the economy. This possibility is explored in detail in

Chapter 5. Another potential explanation is that voter dissatisfaction was not expressed by a

change in party support but by an increased likelihood of abstaining from voting. This idea

is explored in Chapter 6. A third potential explanation is that voters did not hold their own

governments responsible for the crisis. They could hold the financial sector responsible, for ex-

ample, or the European Union, which has centralised certain aspects of economic governance.

This possibility is explored in Chapter 7, which looks at how attitudes towards the European

Union have evolved over the same period studied in this chapter.

Chapter 4

Clarity of responsibility during a global recession

An enduring difficulty for the theory of economic voting has been the so-called ‘great instabil-

ity’ of the economic vote, which refers to the fact that empirical studies of certain countries at

particular times have found very strong evidence of an economic voting effect while similar

studies of different countries or different times have found little if any evidence of such an

effect (Paldam 1991, 26). One proposed explanation for this instability is that some countries

and times provide more favourable contexts for economic voting behaviour than others (Pow-

ell and Whitten 1993). In particular, it is argued that citizens are most likely to engage in

economic voting when there is a clearly recognisable authority who is seen to be responsible

for the state of the economy. This ‘clarity of responsibility’ is stronger in some countries and

at some times than others, depending upon both institutional design and the degree to which

a single party controls the institutions of government (C. J. Anderson 2000; Nadeau, Niemi

and Yoshinaka 2002). For example, institutional designs in which a single-party executive

effectively dominates the government ought to have a higher economic vote than a country

in which power is shared across multiple parties and multiple branches of government. Thus

parliamentary systems ought to be expected to experience a higher level of economic voting

than presidential systems, for example, since power is less divided. The economic vote should

also be higher during periods of majority government than periods of coalition government,

for the same reason.

An alternative explanation for the instability of economic voting findings is model mis-

specification. Van der Brug, van der Eijk and Franklin (2007, 16) argue that this instability

results from the widespread use of vote choice as a dependent variable and contend that party

support models, such as the models developed in the previous chapter, are better suited for

measuring the economic vote.1 It is worth noting that both the clarity of responsibility and

party support approaches would explain the weak level of empirical support for economic vot-

ing in multiparty systems. In the former case, this is because the coalition governments and1See Chapter 1 for a detailed explanation of this argument.

89

90 CHAPTER 4. CLARITY OF RESPONSIBILITY DURING A GLOBAL RECESSION

limited executive control that feature in many multiparty systems of government are among

the key characteristics of low-clarity contexts (Whitten and Palmer 1999). In the latter case,

one of the key intentions of the party support approach is to provide a better measure of the

economic vote in contexts where the vote choice cannot meaningfully be described as a binary

one (van der Brug, van der Eijk and Franklin 2007, 8–15). This overlap in intention between

the two approaches raises the question as to whether a clarity of responsibility effect can be

found when party support rather than vote choice is used to measure the economic vote. This

chapter answers this question by extending the economic voting models from the previous

chapter to take clarity of responsibility into account.

The chapter begins by discussing the theory of clarity of responsibility and some recent

relevant developments in that area. Following on from that is a discussion of how clarity

of responsibility will be measured for this analysis. Next, three specific hypotheses are in-

troduced, which test whether clarity of responsibility effects can be found and whether they

appear to be constant over time. The following analysis is broken into two parts. The first

part tests the chapter’s hypotheses by analysing citizens’ support for whichever party happens

to control the office of prime minister at the time. The second part expands the analysis to

take into account all of the parties for which support data is available. The results from these

different analyses are then compared so as to determine whether or not the evidence overall

supports each of the hypotheses. Finally, there is a discussion of the implications of these

findings.

4.1 Clarity of responsibility: economic voting in different

contexts

The notion of ‘clarity of responsibility’ was first introduced into the literature by Powell and

Whitten (1993). Motivated by the failure of earlier cross-national studies of economic voting to

replicate the success of economic voting studies in the United States and the United Kingdom,

they argued that these discrepancies could be explained by differences in the political systems

of the different countries. In particular, they argue that economic voting depends on the ability

of voters to attribute responsibility to the incumbent government:

We suggest that the critical linkage of the voter’s assignment of responsibility to

the government is not merely an individual-level idiosyncrasy or rationalization.

Rather, it will strongly reflect the nature of policymaking in the society and the

4.1. CLARITY OF RESPONSIBILITY: ECONOMIC VOTING IN DIFFERENT CONTEXTS 91

coherence and control the government can exert over that policy. The greater the

perceived unified control of policymaking by the incumbent government, the more

likely is the citizen to assign responsibility for economic and political outcomes to

the incumbents. (398)

In other words, economic voting should be expected to occur predominantly in countries where

power is visibly concentrated and less in countries where it is diffused among different parties

or political actors.

Powell and Whitten (1993, 399–402) nominated five features of a political system that

were thought to interfere with clarity of responsibility and so reduce the level of economic vot-

ing in a country. The first of these was a ‘lack of voting cohesion among the major governing

party or parties’, that is, a marked tendency for elected representatives of major parties to vote

in accordance with their own views irrespective of their party’s official position. Although it is

unusual for political parties to tolerate this, especially in parliamentary systems, four countries

were identified as lacking voting cohesion, namely Italy, Japan, the United States and Switzer-

land. The second feature was a legislative committee system which allocates real power to

non-government parties, for example, one that distributes committee chairs to parties pro-

portionally to their representation in the legislature. They found this to be the case in several

European countries at the time. The third feature they identified was the presence of an upper

house that is both powerful, such as the US and Australian Senates or the German Bundesrat

but unlike the British House of Lords, and controlled by opposition parties. The fourth and

fifth features identified as reducing clarity of responsibility were the presence of minority gov-

ernment and coalition government respectively. Coalition governments in particular tend to

be an enduring characteristic of most European political systems.

The presence or absence of each of these features was used to classify countries as low

or high clarity and it was then shown that the economic voting effects were stronger in the

high clarity group than the low clarity group (405–409). These results were later replicated

by Whitten and Palmer (1999) using a broader data set. Unlike the earlier study, they divided

countries into three distinct groups, described as ‘most clear’, ‘mixed clarity’ and ‘least clear’

(57). Their results showed that high clarity countries generally experienced a greater eco-

nomic voting effect than mixed clarity countries, with low clarity countries experiencing very

little measurable economic voting (57–63). These findings offered further support for the idea

that voters’ tendency to hold governments to account for economic conditions is mediated by

their ability to identify an incumbent party as the clearly responsible political actor.


Although groundbreaking, the Powell–Whitten index has been criticised for some details

of its operationalisation. The decision to reduce all of the variation to a simple dichotomy

between high and low clarity countries has been seen as limiting as is the fact that the index

is almost constant over time (Nadeau, Niemi and Yoshinaka 2002, 404; Royed, Leyden and

Borrelli 2000, 677). Scepticism has also been voiced about the idea that such highly technical

features of the political system as the allocation of committee chairs could influence voters’

perceptions of responsibility (Royed, Leyden and Borrelli 2000, 678). At least one aggregate-

level study has been unable to find any clarity of responsibility effect at all (Chappell and

Veiga 2000). Other explanations have also been suggested for the high variation in levels

of economic voting in different times and different countries, such as that the salience of

economic issues also varies across time and space (Singer 2011b). Nonetheless, the clarity of

responsibility model is the most widely accepted.

A number of scholars have since proposed revisions to the Powell–Whitten clarity index in

order to incorporate various factors that are conjectured to contribute to or detract from the

clarity of responsibility of a polity. Many of these proposals pertain to long-term institutional

characteristics. For example, Cutler (2004, 2008), argued that federal structures cloud per-

ceptions of responsibility. He found that, in Canada at least, federalism poses a real obstacle to

voters’ ability to hold the appropriate government responsible for policy failures. On the other

hand, Arceneaux (2006, 748) found that voters in federal states do hold politicans responsible

in the areas where they actually exercise power but only under certain favourable conditions.

C. D. Anderson (2006) extends this idea further, finding that multilevel governance in general

tends to weaken the economic vote, since voters have to determine which level of government

to hold to account. In an extensive comparative study of democratic systems, Hellwig and

Samuels (2008) looked more broadly at the effects of different institutional designs on elect-

oral accountability, arguing that certain designs impede voters’ capacity to hold governments

to account even when they accurately recognise which political actors are responsible. For

example, in parliamentary systems, members of parliament can replace the head of govern-

ment mid-term, depriving voters of the opportunity to express their dissatisfaction electorally

(69–70). Similarly, when legislative and executive elections are held separately in presid-

ential systems, voters are only able to hold to account the particular branch of government

that happens to be facing election on that occasion (Hellwig and Samuels 2008, 70; Samuels

2004). Thus parliamentary systems as opposed to presidential systems and concurrent elec-

tions within presidential systems are both linked to reduced clarity of responsibility.

4.1. CLARITY OF RESPONSIBILITY: ECONOMIC VOTING IN DIFFERENT CONTEXTS 93

Other proposed revisions to the Powell–Whitten clarity of responsibility index are con-

cerned with characteristics of the incumbent government. For example, it has been suggested

that divided government within presidential systems might impede clarity of responsibility,

although there is evidence that American voters, at least, respond to this problem by simply

holding the president accountable, irrespective of which branch is actually responsible (Nor-

poth 2001). C. J. Anderson (2000) has argued that both ‘governing party target size’—-the

degree to which a single party dominates the incumbent government (154–155)—and ‘clarity

of available alternatives’—the ease with which voters can identify a clear alternative govern-

ment from among the opposition parties (155–156)—ought to mediate the economic vote.

Using Eurobarometer data from 1994, he showed that both of these features were related to

the level of economic voting measured in a country (160–168). Nadeau, Niemi and Yoshinaka

(2002, 404) also proposed a revision to the Powell–Whitten clarity of responsibility index on

the grounds that the original index is mainly composed of relatively static items, so it offers

little scope to measure any changes over time. They extended the index to include measures

of four transient characteristics that are hypothesised to affect clarity of responsibility. The

first of these is Anderson’s notion of governing party target size. The second is the ‘ideological

cohesion of the governing coalition‘, that is, the proportion of government members of parlia-

ment who share the dominant party’s ideology, defined as simply left or right. The third item

is the length of time the current government has been in office. The final item is the number

of parties holding a minimum threshold (three percent) of seats in the parliament, which is

intended to measure Anderson’s clarity of available alternatives (409–411).

These various developments in the clarity of responsibility theory were brought together by

by Hobolt, Tilley and Banducci (2013), who argued that the concept of clarity of responsibility

ought to be seen as a two-dimensional construct, with one dimension measuring institutional

clarity and the other government clarity (168). Institutional clarity refers to those aspects of

clarity of responsibility resulting from stable institutional arrangements, whereas government

clarity refers to the clarity arising from the cohesiveness of the current national government.

Items forming part of the institutional clarity dimension include formal divisions of govern-

ment power such as a bicameral legislature, a federal state, clear separation between the

legislative and executive branches and a strong committee system in which opposition parties

routinely chair important committees (169). Items forming part of the government clarity di-

mension are those indicating the cohesion of the incumbent government, such as the number

of parties in the cabinet, the ideological cohesion of the cabinet as a whole and the relat-


ive strength of the head of government’s party in the cabinet (170). Using the 2009 European

Election Studies survey, they hypothesised that both clarity dimensions would impact the level

of economic voting measurable, with higher clarity countries experiencing a greater level of

economic voting. What they found, however, was that only government clarity was a strong

predictor of economic voting, with institutional clarity playing little if any role (175–178).

They argue that it is not so much the case that institutional clarity is of no importance but

rather that institutional clarity is likely to be a key determiner of government clarity, which in

turn mediates the economic vote (180).

One thing that remains unclear in this body of research is whether the clarity of respons-

ibility mechanisms tend to operate the same way under extreme economic conditions. As

with economic voting in general, most of the research that has been conducted has studied

behaviour during fairly typical economic conditions. Even when studies have examined the

Great Recession specifically, such as Hobolt, Tilley and Banducci (2013), who looked at the

European Union in 2009, they have not explicitly contrasted those times with more typical

periods. Given that the Great Recession was global in scale, it could be argued that individual

governments can hardly be responsible for its occurrence, since that responsibility is presum-

ably shared by the governments of the other affected countries. This is in fact a clarity of

responsibility argument and so it is worth exploring whether the clarity of responsibility effect

changed during the crisis or remained stable throughout. This chapter does just this, by com-

paring the impact of clarity of responsibility among EU countries at the height of the Great

Recession, in 2009, to that well before the recession, in 2009, and that following the recession

in 2014.

4.2 The dimensions of clarity of responsibility

The two-dimensional approach to measuring clarity of responsibility developed by Hobolt,

Tilley and Banducci (2013) forms the starting point for the measures used in this chapter.

Their approach is used because it consolidates the various predictors identified by previous

studies and because it groups these predictors into two cohesive dimensions, rather than com-

bining into a single index items related to institutional design with those relating to the in-

cumbent government. Each dimension is measured by an index composed of four variables,

which are shown in Table 4.1. The first of their two dimensions, institutional clarity, is made

up of three dummy variables and one scale variable. The three dummy variables indicate a

4.2. THE DIMENSIONS OF CLARITY OF RESPONSIBILITY 95

Table 4.1: Components of institutional and government clarity

Item Scale

Institutional clarityWeak committee system 1= weak, 0= strongParliamentary government 1= parliamentary, 0= semi-presidentialUnicameral parliament 1= unicameral or weak bicameral, 0= strong bicameralCentralisation of government From 0 (highly federal) to 4 (completely unitary)

Government claritySingle-party government 1= one party, 0= coalitionAbsence of cohabitation 1= no cohabitation, 0= cohabitationIdeological cohesion Proportion (1 means ideologically unified government)Dominance of main party Proportion (1 means all cabinet posts held by same party)

These are the component items of the government and institutional clarity of responsibilityindices, according to Hobolt, Tilley and Banducci (2013).

weak parliamentary committee system, a parliamentary as opposed to semi-presidential sys-

tem of government and a unicameral or weak bicameral system respectively. The scale variable

measures the centralisation of power at the level of central government, ranging from zero for

highly federal states to one for completely unitary states. These items are summed and scaled

so that the resulting measure ranges in principle from zero to one (174). Since these institu-

tional features are not expected to change over short time periods (169), this chapter uses the

institutional clarity index reported by Hobolt, Tilley and Banducci (2013, 181) for each of the

countries in this study.

The second dimension is government clarity, which they also measure by an index com-

posed of four variables. The first of these is a dummy variable indicating that there is currently

a single-party rather than a coalition government. The second is a dummy variable indicating,

for semi-presidential systems, that the prime minister and president are of the same party. In

other words, this variable indicates the absence of cohabitation. For parliamentary systems,

this variable is fixed at one, since there can never be cohabitation. The third variable repres-

ents the ‘ideological cohesion’ of the government and this is measured by the proportion of

government-held seats belonging to parties sharing an ideology with the head of government’s

party. The final component is the ‘dominance of the main governing party’, which is measured

by the proportion of cabinet positions held by the head of government’s party. Once again, the

final index is scaled so that it ranges in principle from zero to one (174).

Since the government clarity of a country can be expected to change frequently, it is not

sufficient to use the values of the index reported in the original paper (181) as these values are

specific to 2009. Instead, the index has been reconstructed from its components for each year


and each country under study. In order to determine whether there was a single-party govern-

ment and to find the dominance of the main governing party, information was needed about

the cabinet composition of each country at the appropriate time. This data was collected from

the European Journal of Political Research Political Data Yearbook.2 The absence of cohabita-

tion variable only needs to be measured in the few countries using a semi-presidential system.

Hobolt, Tilley and Banducci (2013, 184n16) follow Elgie’s definition of a semi-presidential

system, which treats as semi-presidential any system in which there exists both a ‘popularly

elected fixed-term president [and] a prime minister and cabinet who are responsible to par-

liament’, irrespective of the division of powers between those institutions (Elgie 1999, 13).

Following from this definition, he includes Austria, Finland, France, Ireland, Lithuania, Por-

tugal, Poland and Slovenia3 as examples of semi-presidential systems in Europe (14). This is a

rather broad definition and includes countries that others would classify as parliamentary. For

example, the description of modern Finland as semi-presidential is disputed,4 and in Ireland,

the few constitutional powers reserved to the president are exercised sparingly in practice (El-

gie 2012). On the other hand, Hellwig and Samuels (2008, 81) found that ‘the direct election

of a president—whether powerful or weak—introduces a special element into electoral politics

under semi-presidentialism not present in pure parliamentary systems’ (emphasis in original).

This is a strong argument for using the broader definition of semi-presidentialism, so that

practice has been continued here. Among all of these semi-presidential countries, the only

instances of cohabitation coinciding with the relevant surveys were in Finland and Poland in

2009 and Portugal in both 2004 and 20095 (Döring and Manow 2015).

Hobolt, Tilley and Banducci (2013, 174) operationalise ideological cohesion as ‘the pro-

portion of seats held by parties in government that are of the same ideology as the dominant

governing party’. In order to measure this, it is necessary to know which parties were in gov-

ernment at the appropriate times and how many seats each of them held in the parliament.

This data has been collected from the ParlGov database (Döring and Manow 2015). It is also

necessary to assign an ideology to each party. Following the original method, this was done2In many cases, this information was gathered from the ‘interactive’ interface to the PDY data (EJPR 2016).

At the time of writing, only a subset of the required data has been made available through that interface, so theremaining data was gathered from the detailed annual reports published in the yearbook editions of the journal(EJPR 2003–2015).

3Of the EU member states, Bulgaria, Croatia and Romania were also included in the list but they do not formpart of this study because they acceded to the EU after 2004. Cyprus is the only EU member state with a purepresidential system of government but cohabitation is impossible in a pure presidential system.

4Nousiainen (2001, 108) writes that, since the constitutional reforms of the 1990s, ‘there are hardly anygrounds for the epithet “semi-presidential” ’ with reference to Finland.

5There were also instances of an independent president in both Lithuania in 2009 and 2014 and Sloveniain 2009. These were not treated as instances of cohabitation, following the example set by Hobolt, Tilley andBanducci (2013).

4.3. HYPOTHESES 97

by dividing parties into ‘left’ and ‘right’ groups, according to their European Parliament party

groupings. In particular, parties belonging to the European United Left–Nordic Green Left,

the Progressive Alliance of Socialist and Democrats and the Greens–European Free Alliance

groupings and are treated as ‘left’ and all other parties are treated as ‘right’ (Hobolt, Tilley and

Banducci 2013, 184n18).6

In addition to government and institutional clarity, time in office is also treated in this

chapter as a clarity measure. The fact that the EES survey dates coincided with different

points in the electoral cycles of the surveyed countries means that some of the governments in

power at the time were elected much more recently than others. For example, some ‘incum-

bent’ governments had only been in power for six weeks at the time the survey took place,

whereas others had been in power for several terms. Time in government is only occasionally

identified in the literature as an aspect of clarity of responsibility—one example is Nadeau,

Niemi and Yoshinaka (2002, 410)—but it is not difficult to imagine that voters who had only

recently elected a new government might be less willing to judge that government harshly

for poor economic conditions—likely inherited from the previous government—than voters in

countries where the same party had been in power for a considerable period of time. There

are two types of models used in this chapter and time in government is measured slightly dif-

ferently according to the type. For those models that analyse exclusively the prime minister’s

party in each country, time in government is measured by the number of years that the party

has held the office of prime minister. When multiple parties in each country are analysed to-

gether, this definition cannot be used because many incumbent parties do not hold the office

of prime minister, so time in government is measured by the number of years that each cabinet

party has held any cabinet posts at all. In both cases, the number of years is expressed as a

fraction, accurate to the number of days in office.

4.3 Hypotheses

The purpose of this chapter is to test whether clarity of responsibility effects existed during

the period under study and whether the magnitude of this effect changed as a result of the

Great Recession. Three potential effects have been derived from the clarity of responsibility6This method leads to a handful of surprising codings, particularly as some centrist parties could be plausibly

regarded as belonging to either group. Given that some codings are effectively arbitrary, any consistent dichotom-ous coding will lead to some minor anomalies. An alternative operationalisation was developed, which amountedto the negated standard deviation of the left–right positions of each government member of parliament, wherethis position is assumed to be the same as that of the party that member belonged to, measured on an eleven-pointscale. Although this eliminated the minor anomalies identified, the resulting measure was in fact less predictivethan the original, so the analysis reported here uses the original dichotomous measure.


literature: a time in office effect; a government clarity effect; and an institutional clarity effect.

In each case, the chapter will test whether the effect in question exists and, if it does, whether

the magnitude of that effect changed over time. To this end, the chapter tests three specific

hypotheses. The first of these is:

Hypothesis 4.1 Governments that have recently been elected are subject to less economic voting

than governments that have been in power for some time.

As was stated in the previous section, there is good reason to believe that a government that

has been in power for a longer period of time, and therefore had more opportunity to influence

the economy, might be more exposed to economic voting than a government that is relatively

new to office.

The second hypothesis is that there is a traditional clarity of responsibility effect:

Hypothesis 4.2 Countries with high clarity of responsibility experience a greater level of eco-

nomic voting than countries with low clarity. This applies to both government and institutional

clarity.

Since this chapter measures clarity of responsibility by the two different dimensions of gov-

ernment and institutional clarity, this effectively has two sub-hypotheses, which are tested

separately. If this hypothesis is found to hold, then this would naturally be considered evid-

ence in support of the clarity of responsibility theory. If, on the other hand, this hypothesis is

rejected, then this would support the idea (van der Brug, van der Eijk and Franklin 2007, 16)

that the instability problem of economic voting theory results from model misspecification,

since the method used to measure the economic vote in this thesis is informed by suggestions

for a better approach (26–29).7

The final hypothesis concerns clarity of responsibility during the recession period:

Hypothesis 4.3 Clarity of responsibility had a smaller influence on the level of economic voting

during the Great Recession than it did before or after.

There are two reasons to expect that a global recession of such a scale might erode any eco-

nomic voting differences between high- and low-clarity countries. The first reason is that the

recession was presumably a highly salient issue for voters, so that even in low-clarity coun-

tries, voters might be more motivated than usual to hold someone accountable, even if it is

7These ideas have not been adopted in their entirety. See Chapter 2 for a detailed explanation of how economicvoting is measured in this thesis and the reasons behind this approach.

4.4. HOW CLARITY AFFECTS THE PRIME MINISTER’S PARTY 99

not necessarily as obvious who ought to be held to account as it is in other countries. The

second reason is that the Great Recession was not a mere local recession but an international

one that began outside of Europe. This being the case, it is possible that voters in high-clarity

countries might find it difficult to hold their own governments to account for a crisis that those

governments arguably had very little control over.

4.4 How clarity affects the prime minister’s party

The economic voting models developed in the previous chapter have been extended in order

to measure how clarity of responsibility impacts the economic vote. This analysis is divided

into two parts. The first part focuses on the head of government’s party in each country

and specifically those factors affecting the degree to which relative support for that party8

was influenced by economic perceptions. This single-party approach has the advantage of

producing relatively simple models which are straightforward to interpret but they do not

take advantage of all the data available. The second part of this analysis uses this extra data

by looking at all parties together, not just the dominant government parties. The resulting

model is inevitably more complex than the single-party model but it offers a different view

of the evidence. Taken together, these two approaches allow for greater insight than either

approach used alone.

The first clarity of responsibility model, Model 4A, is based on Model 3A from the previ-

ous chapter, which predicts support for the head of government’s party in each country based

primarily on the left–right distance between the voter and the party, whether that voter iden-

tifies with the party and, crucially, the voter’s prospective economic assessment. The effect of

this economic assessment on support for the prime minister’s party is a measure of the overall

level of economic voting. In order to test the hypothesis that clarity of responsibility influences

the economic vote, therefore, this model must be extended to include an interaction between

the economic assessment predictor and the chosen measure of clarity of responsibility. This al-

lows the model to estimate different levels of economic voting for different levels of clarity. As

the underlying model is a multilevel model, the fact that clarity of responsibility is a country-

level variable is accounted for. The new model includes predictor terms for both institutional

8As in the previous chapter, the dependent variable is a voter’s level of support for a particular party, centredaround that voter’s mean level of support for all parties. This means that even the models that focus on primeministers’ parties only still take some account of voter attitudes towards other parties, which is appropriate becausea voter is not expected to switch their vote unless their support for a new party exceeds their support for the partythey previously supported.


Table 4.2: Alternative government clarity models

Fixed effect Mod. 4A Mod. 4B Mod. 4C

Intercept 0.54 (0.24) 1.91 (0.41) 1.40 (0.32)Year 2009 0.33 (0.17) 0.31 (0.17) 0.34 (0.18)Year 2014 0.04 (0.13) −0.06 (0.14) −0.06 (0.14)Prospective assessment 0.13 (0.09) −0.13 (0.14) −0.11 (0.10)Time in office (PM) 0.02 (0.02) 0.01 (0.02) 0.01 (0.01)Government clarity −0.89 (0.29)Single-party government 0.21 (0.19)Absence of cohabitation −0.27 (0.21)Ideological cohesion −1.58 (0.32) −1.39 (0.30)Dominance of main party −0.14 (0.34)Institutional clarity 0.05 (0.26) −0.15 (0.26) −0.08 (0.24)Prosp. assess. × year 2009 −0.08 (0.03) −0.07 (0.03) −0.07 (0.03)Prosp. assess. × year 2014 −0.07 (0.03) −0.04 (0.03) −0.03 (0.03)Prosp. assess. × time in office 0.01 (0.00) 0.01 (0.00) 0.01 (0.00)Prosp. assess. × govt clarity 0.19 (0.09)Prosp. assess. × single party −0.05 (0.06)Prosp. assess. × no cohabitation −0.03 (0.07)Prosp. assess. × cohesion 0.39 (0.09) 0.38 (0.08)Prosp. assess. × dominance 0.04 (0.11)Prosp. assess. × inst. clarity 0.14 (0.12) 0.19 (0.12) 0.15 (0.11)

Comparison of clarity of responsibility models. The dependent variable is the individual’ssupport for the current head of government’s party. Only the key fixed effect coefficients areshown here, with standard errors in brackets. The full results can be found in Appendix B.Source: EES, ParlGov & PDY

and government responsibility, as well as time in office.9 These three variables account for

the different types of clarity discussed in this chapter. The model furthermore includes inter-

actions between each of these and the voter’s prospective economic assessment. This model

has been estimated and the key results are summarised in the first column of Table 4.2.

Before interpreting any figures in detail, it is worth discussing some alternative model spe-

cifications. The government clarity index consists of four component variables, as discussed

earlier. Model 4B replaces the index with its four components so that their relative impact can

be assessed.10 The key results from this model form the second column of Table 4.2. Com-

paring the two models, it is striking that the interaction between economic assessment and

ideological cohesion in the second model appears to be stronger than that between economic9This is specifically the number of years this party has held the office of prime minister, even if they held other

cabinet posts before that.10Multicollinearity might be a concern, as the components of an index can be expected to be correlated with

each other. In fact, the components are not so strongly correlated as to cause serious difficulties. The highestcorrelation is between dominance of the main governing party and single-party government variables (r = 0.70)but these variables are dependent since single-party government prevails if and only if the prime minister’s partycompletely dominates the government. The second highest correlation is between dominance and ideologicalcohesion (r = 0.31).


assessment and the government clarity index. Although it is not straightforward to compare

model coefficients directly, given the similar scale of the two variables, this difference is sug-

gestive. It is also surprising that the interactions involving the other three component variables

are not even close to significance. This evidence is hardly conclusive but it does suggest that

ideological cohesion alone may be a better predictor of economic voting than the complete

index. In order to test this hypothesis, Model 4C was estimated, in which ideological cohe-

sion alone replaces government clarity. The key results from this model make up the third

column of Table 4.2. Comparing this to the other models shows that this model fits the data

better (∆AIC = 26, ∆BIC = 27) than the government clarity model (Model 4A) and that the

components model (Model 4B) does not improve the model fit enough to justify the extra

complexity relative to this model (∆AIC= 7, ∆BIC= 61).

These results support the hypothesis that ideological cohesion alone is a better predictor

of economic voting levels than the complete index. Consequently, Model 4C forms the basis

of this analysis. As in the previous chapter, post-estimation simulation is used to derive key

quantities of interest rather than interpreting model coefficients directly. This chapter’s first

hypothesis is that governments that have been power for some time experience greater levels

of economic voting than governments which have been recently elected. The level of eco-

nomic voting predicted from the model can be measured by the difference in support for the

prime minister’s party between a highly optimistic individual and an otherwise similar11 highly

pessimistic individual. Using this definition, the predicted level of economic voting for a gov-

ernment that has just been elected is 1.23 (SE = 0.16, p < 0.001). A party that has been in

power for a full decade, on the other hand, has a predicted economic vote of 1.50 (SE= 0.18,

p < 0.001). This amounts to a difference of 0.26 points (SE = 0.15, p = 0.07), which is not

statistically significant. In other words, it is possible that a longer time in office is associated

with a higher level of economic voting but the effect is weak if it exists at all.

The second hypothesis is that clarity of responsibility increases the level of economic vot-

ing. This has two sub-hypotheses, relating to government and institutional clarity respect-

ively. Since government clarity is measured in this model by ideological cohesion alone, the

11Unless otherwise stated, all predictions in this chapter are made for a context in which ideological cohesion,institutional clarity and time in office are held at their means. This is 0.86 for cohesion and 0.61 for institutionalclarity. The mean time in office is 4.03 years where this refers to the time the party has held the office of primeminister, as in this case, or 4.92 years where this refers to the time the party has been part of the governingcoalition, which becomes relevant later in the chapter. As in the previous chapter, the individual is an employed40 year old male, living in a town, who has completed high school but not university.


first sub-hypothesis is tested by comparing high-cohesion to low-cohesion contexts.12 The pre-

dicted economic vote for a low-cohesion country is 0.79 (SE = 0.21, p < 0.001) whereas the

predicted economic vote for a high-cohesion country is 1.54 (SE = 0.15, p < 0.001). This

corresponds to a difference of 0.75 points (SE = 0.16, p < 0.001) between high- and low-

cohesion countries. In other words, the economic vote is almost twice (1.94, [1.39, 3.54]13)

as high when the ideological cohesion of the government is high than when it is low. The

second sub-hypothesis pertains to institutional clarity. The predicted economic vote in high-

clarity countries14 is 1.08 (SE = 0.24, p < 0.001) and the predicted economic vote in low-

clarity countries is 1.58 (SE = 0.23, p < 0.001). The difference of 0.50 points (SE = 0.38,

p = 0.19) is not significant. Based on this model then, it appears that government clarity, and

in particular ideological cohesion, is the strongest predictor of economic voting in a country.

The third hypothesis is that these clarity of responsibility effects were weaker during the

recession than at other times. As the models introduced so far assume that these effects are

static over time, a new model is needed to test this hypothesis. Model 4D was created by

extending Model 4C with interactions between the time dummy variables and the clarity of

responsibility and time in government measures, thus allowing each of these effects to vary

in strength over time. This model was used to predict the level of economic voting under

different clarity contexts in each year. Figure 4.1 compares the predicted economic vote for a

government that has just been elected with that of a government that has held office for a full

decade. This plot shows that countries whose governments have held power for some time

experienced a greater level of economic voting than countries who have experienced a recent

change in government. This effect changes over time, appearing to vanish in 2009 and return

much stronger in 2014. These impressions are mostly borne out by a numerical analysis except

that the effect falls just short of significance (∆ = 0.77, SE = 0.40, p = 0.06) in 2004. The

effect is clearly not significant (∆ = 0.12, SE = 0.19, p = 0.52) in 2009, and the difference

between the two years is also not significant (∆ = 0.64, SE = 0.42, p = 0.13). By 2014, on

the other hand, the decade in office accounts for a 1.98 point (SE= 0.43, p < 0.001) increase

in the economic vote and this effect is significantly stronger than both earlier years (compared

to 2004, ∆ = 1.22, SE = 0.58, p = 0.04). In other words, time in office was only a strong12High cohesion means a cohesion measure of 1, indicating no differences in ideology among government

parties. Low cohesion means a cohesion measure of 0.5, indicating that half of the government-held seats areheld by parties not sharing an ideology with the prime minister’s party. This value has been chosen because lowervalues are unlikely, since the prime minister’s party is typically to be the largest governing party. It is also thelowest value occurring in the dataset but it is by no means an outlier.

13Square brackets indicate 95% confidence intervals.14High clarity means an institutional clarity measure of 1 and low clarity a measure of 0.18, these being the

highest and lowest scores assigned to any country by Hobolt, Tilley and Banducci (2013).


Figure 4.1: Economic vote for dominant government party by time in office

0.5

1.0

1.5

2.0

2.5

2004 2009 2014year

econ

omic

vot

e

time in office10 years

just elected

Predicted level of economic voting for the head of government’s party according to the timethat the party has held office and the survey year. The economic vote is the difference in sup-port for the party from a highly optimistic and a highly pessimistic voter based on predictionsfrom Model 4D. Ideological cohesion and institutional clarity are held at their means. Source:EES, ParlGov & PDY

predictor of economic voting in 2014. The fact that this effect was so weak in earlier years

may explain why no effect was found when it was assumed not to vary over time.

Figure 4.2 shows the predicted level of economic voting for a high-cohesion and a low-

cohesion context in each year. A high-cohesion context is a country in which all of the gov-

erning parties are ideologically similar and a low-cohesion context is one in which half of

the government-held parliamentary seats belong to parties not sharing the ideology of the

prime minister’s party. This shows a similar pattern to the previous figure, in that there is

an apparent difference between the two groups in 2004, which closes in 2009 before widen-

ing again in 2014. These observations are supported by numerical analysis, which shows

that in 2004, the economic vote experienced in a high-cohesion context was 1.65 points

(SE = 0.41, p < 0.001) higher than in a low-cohesion context. By 2009, this difference

had fallen (∆= 1.37, SE= 0.47, p < 0.01) to 0.29 points (SE= 0.27, p = 0.28) and by 2014

had risen once again (∆ = 0.68, SE = 0.33, p = 0.04) to 0.97 points (SE = 0.24, p < 0.001).

In other words, high ideological cohesion was associated with more economic voting before


Figure 4.2: Economic vote for dominant government party by ideological cohesion

0.0

0.5

1.0

1.5

2004 2009 2014year

econ

omic

vot

e

cohesionhigh

low

Predicted level of economic voting according to the ideological cohesion of the incumbentgovernment and the survey year. The economic vote is the difference in support for the primeminister’s party from a highly optimistic and a highly pessimistic voter based on predictionsfrom Model 4D. Institutional clarity and time in office are held at their means. Source: EES,ParlGov & PDY

and after but not during the Great Recession. This supports the third hypothesis with respect

to government cohesion.

Similarly, Figure 4.3 shows the relationship between institutional clarity and economic

voting by comparing the difference between high and low scores on the institutional clarity

index. This figure shows little if any difference between high- and low-clarity contexts in

2004 and 2009 but a large difference in 2014. Numerical analysis confirms that there was

no significant effect in either 2004 (∆ = −0.19, SE = 0.51, p = 0.71) or 2009 (∆ = 0.43,

SE= 0.53, p = 0.42), nor is there a significant difference between these two years (∆= 0.62,

SE= 0.47, p = 0.18). In 2014 on the other hand, the economic vote was considerably stronger

among high-clarity countries than low-clarity countries (∆ = 1.67, SE = 0.50, p < 0.001),

which is a significant increase over both other years (compared to 2009,∆= 1.24, SE= 0.47,

p < 0.01). This means that high institutional clarity was not a predictor of economic voting

either before or during the Great Recession but it did become one in the aftermath of the

recession. This evidence provides only qualified support for the third hypothesis with respect

4.5. THE EFFECT OF CLARITY ON OTHER PARTIES 105

Figure 4.3: Economic vote for dominant government party by institutional clarity

0.5

1.0

1.5

2.0

2004 2009 2014year

econ

omic

vot

e

clarityhigh

low

Predicted level of economic voting according to the institutional clarity of the country andthe survey year. The economic vote is the difference in support for the prime minister’s partyfrom a highly optimistic and a highly pessimistic voter based on predictions from Model 4D.Ideological cohesion and time in government are held at their means. Source: EES, ParlGov &PDY

to institutional clarity, since it was predicted that the importance of institutional clarity would

be weaker during the recession than either before or afterwards.

4.5 The effect of clarity on other parties

The models discussed so far have focused on support for the current prime minister’s party.

This approach has the advantage of simplicity but it also has the disadvantage that it disregards

respondents’ levels of support for coalition partners and opposition parties, so it does not use

all of the data available. This means that while it is possible to determine the degree to which

prime ministers’ parties are supported relative to other parties in general15 it is not possible to

compare them to opposition parties specifically or to examine support for government parties

more generally using this approach. A further model (Model 4E) has been constructed using

the available data for every party so that the contrasting effects of economic voting between

15Since the dependent variable, party support, is centred around its individual-level mean, these models aremeasuring support relative to the other key parties in each country rather than absolute support. This centringhas both theoretical and data-driven motivations, which are discussed in detail in Chapter 2.


government and opposition parties in general can be examined in different clarity contexts.

This model is based on Model 3D, which predicts party support based on, among other things,

the individual’s prospective economic perceptions and whether the party is a government or

opposition party. This was extended to take into account clarity of responsibility by allowing

the key economic voting terms to interact with the clarity of responsibility predictors. In

the base model, economic voting is measured by a three-way interaction between economic

assessment, incumbency and time. This means that the model allows government and non-

government parties to be affected differently by an individual’s economic assessment, which is

the key claim of economic voting theory, and further allows the strength of this economic vote

to change over time. Clarity of responsibility theory further asserts that the level of economic

voting varies according to country-level clarity variables and if these effects are to be allowed

to vary over time, this necessitates a four-way interaction.16 This is admittedly a high level of

complexity but it reflects the complexity of the hypotheses.17 Without these interaction terms,

it would not be possible to determine whether clarity of responsibility had a stable effect on

economic voting propensity over time.

Once again, post-estimation simulation has been used to interpret this model. The method

used to ascertain the effect of the clarity of responsible variables on economic voting is to com-

pare the level of economic voting under different clarity scenarios in each year. In this instance,

the level of economic voting is defined in terms of relative government support. Relative gov-

ernment support is the difference between the individual’s support for a government party and

that same individual’s support for an opposition party. This is positive if that individual gen-

erally prefers government parties and negative if that individual generally prefers opposition

parties. The economic voting level is the difference between the relative government support

of a highly optimistic individual and that of a highly pessimistic individual. The higher the

economic voting level, the more influence economic conditions have on the decision as to

whether to support the government or the opposition. If this is zero then the economy plays

no role at all. A negative economic voting level would be surprising as this would indicate

that governments are harmed electorally by good economic conditions.

This model has been used to test the chapter’s hypotheses once again, as the different

approaches offer different perspectives. The first hypothesis is that the economic vote is greater16Ideological cohesion and institutional clarity are both allowed to interact with the three-way economic voting

interaction. Time in government is only meaningful for parties that are actually in government, so this variableonly interacts with time and economic assessment. As always, the coefficient estimates are shown in Appendix B.

17Van der Brug, van der Eijk and Franklin (2007, 210–212), who use the same dependent variable, also usefour-way interactions in some of their economic voting models. Although their method of measuring the economicvote is quite different, this illustrates that this level of complexity inevitably arises when studying the economicvote through the lens of party support.


Figure 4.4: Government and opposition economic vote by time in office

0.8

1.0

1.2

2004 2009 2014year

econ

omic

vot

e

time in officejust elected

10 years

Predicted level of economic voting in each survey year for a party that has been in governmentfor ten years compared to a party that has just been elected to government. The economic voteis the difference in relative support for the party compared to an opposition party between ahighly optimistic and a highly pessimistic voter based on predictions from Model 4E. Ideolo-gical cohesion and institutional clarity are held at their means. Source: EES, ParlGov & PDY

in countries where the government has been in power for a longer period of time. Figure 4.4

compares the predicted economic vote of a government that has been in power for ten years to

one that has only just been elected. This figure shows that there is an economic voting effect

favouring both short-term and long-term governments in all years. Surprisingly, governments

that have only been in power for a short period appear to have been subject to slightly stronger

economic voting than those that have been in power for a longer time in both 2004 and 2009.

Neither of these differences is significant, however, the larger of the two being the 0.15 point

(SE= 0.16, p = 0.35) difference in 2009. None of the three groups saw a significant difference

in the economic vote between 2004 and 2009. By 2014, long-term governments were exposed

to a greater level of economic voting than short-term governments (∆ = 0.35, SE = 0.18,

p = 0.04). The increase in the economic vote experienced by long-term governments from

2009 to 2014 was 0.58 points (SE = 0.25, p = 0.02). None of the other differences over

time were significant. Based on these results, it appears that time in office only played an

important role in 2014. One possible explanation for this is that those parties who held power

continuously since the Great Recession were seen differently by voters from parties that had


Figure 4.5: Government and opposition economic vote by ideological cohesion

0.0

0.5

1.0

1.5

2004 2009 2014year

econ

omic

vot

e

cohesionhigh

low

Predicted level of economic voting according to the ideological cohesion of the incumbentgovernment and the survey year. The economic vote is the difference in relative support fora governing party compared to an opposition party between a highly optimistic and a highlypessimistic voter based on predictions from Model 4E. Institutional clarity and time in gov-ernment are held at their means. Source: EES, ParlGov & PDY

been elected since that time. This would make sense as it means that parties who could be

seen as responsible for the recession experienced a stronger economic vote.

Figure 4.5 contrasts the predicted economic vote of highly ideologically cohesive govern-

ments with that of less cohesive governments. It can be seen that the economic vote exper-

ienced by highly cohesive governments was stronger than that experienced by less cohesive

governments. This difference amounts to 2.07 points (SE = 0.59, p < 0.001) in 2004, before

falling by 1.86 points (SE = 0.69, p < 0.01) before 2009, when the difference between the

two groups almost vanishes (∆ = 0.21, SE = 0.37, p = 0.57). The difference between 2009

and 2014 is not significant (∆ = 0.74, SE = 0.50, p = 0.14) but the size of the gap between

the two groups in 2014 is 0.95 points (SE = 0.33, p < 0.01). An anomaly arising from this

analysis is that the predicted economic vote for low cohesion governments in 2004 is actually

negative (∆= −0.36, SE= 0.54, p = 0.50) but this is not statistically significant. In summary,

it appears that the ideological cohesion of the government does mediate its exposure to the

economic vote but it also appears that this effect was weakened by the Great Recession. These


Figure 4.6: Government and opposition economic vote by institutional clarity

0.6

0.9

1.2

1.5

1.8

2004 2009 2014year

econ

omic

vot

e

clarityhigh

low

Predicted level of economic voting according to the institutional clarity of the country andthe survey year. The economic vote is the difference in relative support for a governing partycompared to an opposition party between a highly optimistic and a highly pessimistic voterbased on predictions from Model 4E. Ideological cohesion and time in government are heldat their means. Source: EES, ParlGov & PDY

findings support both the government clarity hypothesis and the hypothesis that this clarity

effect was weaker during the recession than at other times.

Similarly, Figure 4.6 compares the predicted economic vote of governments in countries

with high institutional clarity to that of governments in low-clarity countries. Unfortunately,

owing to the uncertainty in the model estimates, there is little that can be confidently said

about the effect of institutional clarity. The only year in which there is a statistically signi-

ficant difference between the groups is in 2004, when low clarity countries experienced a

1.18 point (SE = 0.53, p = 0.03) stronger economic voting effect than high clarity countries.

This runs opposite to the hypothesised effect, which was that high clarity countries should

have a stronger economic vote. The point estimate of the effect was in the opposite direction

in both 2009 (∆= 0.43, SE= 0.48, p = 0.37) and 2014 (∆= 0.61, SE= 0.49, p = 0.21) but

neither of these estimates is significant. There was significant difference in effect size between

2004 and 2009 (∆ = 1.60, SE = 0.72, p = 0.03) but not between 2009 and 2014 (∆ = 1.78,

SE = 0.69, p = 0.80). Given that the clarity effect is either not significant or significant but

opposite to the expected sign, the conclusion has to be drawn that this model does not support


the institutional clarity hypothesis. Although there appear to be some significant time effects,

in the absence of clear support for the basic hypothesis, it would be unwise to attempt to find

support for the hypothesis that the effect varies overs time. In summary, institutional clarity

does not appear to be a reliable predictor of economic voting, at least according to this model.

4.6 Conclusion

This chapter has examined the impact of clarity of responsibility as a mediator of the economic

vote before, during and after the Great Recession. Clarity of responsibility was divided into

three facets, each representing a distinct quality of a political context expected to affect the

ease with which voters can hold the incumbent government responsible for the prevailing

economic conditions. These facets were government clarity, institutional clarity and time in

government. Government clarity refers to those characteristics of the incumbent government,

such as consisting of a coalition of parties, that might make it difficult for voters to identify a

party to hold to account. Institutional clarity refers to institutional characteristics having the

same effect, such as the presence of a strong upper house or a federal system of government.

These concepts were measured using the indices proposed by Hobolt, Tilley and Banducci

(2013), except that it was found that the government clarity index was outperformed by one

of its component measures, the ideological cohesion of the government, which was therefore

used in its place as the measure of government clarity. Finally, time in government refers

to the amount of time that a party has held office, since a party that has only recently been

elected could plausibly claim not to have had an opportunity to influence the condition of the

economy.

Three hypotheses were tested in this chapter, the first of these being that a longer period

of time in government was associated with a stronger economic vote. This hypothesis was

not supported by the data. When the effect was modelled as constant over time, there was

no significant difference between parties that had been in power for a short period and those

that had been in power longer. When the size of the effect was allowed to vary over time, a

significant effect was only found in 2014. This was the case whether all parties were analysed

together or the analysis was restricted to prime ministers’ parties, even though time in office

was operationalised differently by the two approaches. This is certainly not strong enough

evidence to claim that time in government is a mediator of the economic vote but it is inter-

esting that there does appear to be some effect shortly after the Great Recession, since this

4.6. CONCLUSION 111

effectively means that time in government does matter when that time implies that the gov-

ernment held office during the recession. In other words, this suggests that voters were more

closely monitoring the economic performance of governments that survived the recession.

The second hypothesis is that greater clarity of responsibility is associated with a higher

level of economic voting. There are two sub-hypotheses, in that this is expected to apply both

to government clarity and institutional clarity. As mentioned earlier, government clarity is

measured in this chapter by the ideological cohesion of the incumbent government. In each

of the models, a positive relationship was found between government clarity and the level of

economic voting. The same cannot be said for institutional clarity. No significant institutional

clarity effect was found when the effect was assumed to be constant over time. Even when

the effect was assumed to vary with time, no significant effect was found apart from in 2014

when considering only prime ministers’ parties. Even worse, when examining all parties to-

gether, the effect was found to be in the opposite direction to that hypothesised in 2004 and

otherwise not significant. In other words, the data offers strong support for the hypothesis

with respect to government clarity but little if any support with respect to institutional clarity.

This mirrors the findings of Hobolt, Tilley and Banducci (2013, 177), who first made this dis-

tinction explicit and who also found that government clarity outperforms institutional clarity

as a predictor of economic voting levels. The fact that some clarity effects were found using

a contextual model of economic voting shows that existing clarity of responsibility findings

cannot be completely explained away as an artefact of model misspecification. On the other

hand, the lack of support for any clarity effect other than ideological cohesion supports the

argument that model misspecification plays at least some role in the instability of economic

voting results (van der Brug, van der Eijk and Franklin 2007, 16).

The final hypothesis is that clarity of responsibility was a weaker predictor of the economic

vote during the recession that it was at other times. With respect to government clarity, this

hypothesis has strong support. Whether all parties were analysed together or the analysis

focused solely on the parties of the heads of government, the findings were the same. The level

of economic voting was considerably higher in high cohesion countries than in low cohesion

countries in both 2004 and 2014. In 2009, during the recession, this effect was not nearly

as strong and, in fact, was not significantly different from zero, although the point estimates

were still positive. In other words, the recession temporarily eroded the distinction between

high- and low-clarity countries. In principle this hypothesis also applies to institutional clarity


but since institutional clarity was not shown to a predictor of the economic vote at all, this

sub-hypothesis is not meaningful.

In summary, it was found that there was a clarity of responsibility effect, in that voters in

countries with highly ideologically cohesive governments were more likely to vote economic-

ally than those in countries with ideologically fragmented governments. It was further found

that the level of economic voting was greater before and after the Great Recession than it was

during its peak. Considered in conjunction with the finding from the previous chapter that the

overall level of economic voting was reduced during the same period, this suggests that the

event itself had the effect of clouding economic responsibility. This makes sense as the reces-

sion was international in scale and had begun in the United States before spreading to Europe,

so governments could plausibly deny responsibility for it. By 2014, however, governments had

had plenty of time to react to the crisis and economic voting levels as well as the mediating

effect of government clarity had returned to their pre-crisis levels. Any governments that had

held power continuously since the crisis were held responsible to an even higher degree than

before for the condition of the economy.

This chapter and the preceding chapter have looked at the circumstances under which

poor economic conditions voters are likely to cause voters to abandon their support for their

incumbent governments. This raises the question of which party or parties are likely to benefit

from this shift in allegiances. The next chapter seeks to answer this question by examining

whether some types of parties are more likely to receive the support of economic voters than

others.

Chapter 5

Extreme and Eurosceptic parties: the changing

policy preferences of European voters

In the Greek parliamentary elections of 25 January 2015, Syriza won 149 of the 300 seats

in the Hellenic Parliament. This allowed the radical left party to seize power from the grand

coalition that had governed Greece since 2012 (Matakos and Xefteris 2016). The coalition

parties, which included the centre-right New Democracy, the centre-left PASOK and at various

times other parties, had desired to maintain Greece’s membership of the Eurozone, which

meant imposing the austerity measures demanded by their creditors (Rori 2016). Syriza,

on the other hand, stood on an anti-austerity platform.1 Much has been written about the

situation in Greece but it raises interesting questions about how voters respond to economic

crises. Why did Greek voters turn to Syriza specifically? Was it because they were leftist, or

because they were Eurosceptic, or simply because they promised to end austerity? Was this

event part of a greater Europe-wide trend?

This chapter explores these questions by examining which types of parties have benefited

from the events of the Great Recession and which have not. It will be seen that there are many

different ways that parties have been classified but for the purposes of this chapter they will be

primarily classified according to their positions on economic issues, social issues and European

integration. Previous chapters have shown that incumbent parties lost support from voters

with a pessimistic economic outlook during the recession, although not by as much as at other

times. The purpose of this chapter is to establish which parties gained support as a result. The

focus is not on individual parties so much as types of parties. Syriza, for example, is a radical

left party (Stavrakakis and Katsambekis 2014), so it could be argued that Greek voters shifted

to the left. On the other hand, it could also be argued that they voted for Syriza not because

1Despite this platform, the Syriza-led coalition government ultimately negotiated a new bailout programmeas well as further austerity. They were nonetheless re-elected at the snap September election in the same year(Matakos and Xefteris 2016; Rori 2016).

113

114 CHAPTER 5. EXTREME AND EUROSCEPTIC PARTIES

it is specifically left-wing but because it advocated a radical rejection of the status quo. In

other words, a right-wing party that promised to abandon austerity might have been equally

successful. Syriza has also been described as a ‘soft Eurosceptic’ party (Verney 2015, 280),

although their leader Alexis Tsipras is reported to have rejected all forms of Euroscepticism

(To Vima 2014). These are the three categories of parties that might be expected to benefit

from the crisis: parties on one side of the spectrum, parties far from the centre, and Eurosceptic

parties.

This chapter begins by discussing the specific hypotheses that will be tested. After that,

there is an explanation of how parties’ positions have been measured. Spatial position in

particular is measured in two different ways, both as a single-dimensional left–right spectrum

and as a two-dimensional space. The remaining sections discuss how parties’ positions towards

European integration and their spatial positions, in both one and two dimensions, affected

support for those parties before, during and after the crisis. This is done by extending the

multilevel model of party support introduced in Chapter 3 so as to control for the economic

voting effect that has already been observed. The results show that the parties who benefited

most from the recession are Eurosceptic parties and those holding more extreme2 economic

positions and conservative social positions. Since most of these changes took place well after

the recession, it is argued that the political response to the recession, rather than the recession

itself, has been the main driver of these changes.

5.1 Hypotheses

The purpose of this chapter is to determine which types of parties benefited from the Great Re-

cession. The previous chapter showed that there was an economic voting effect, that is, voters

who held a pessimistic economic assessment preferred opposition parties over government

parties. Not all opposition parties are equal, however. This chapter tests several hypotheses

about which groups of parties were most likely to receive new support from voters as a result

of the Great Recession. As in previous chapters, the primary sources of data are the 2004,

2009 and 2014 waves of the EES voter surveys. The first of these waves was collected well

before for the beginning of the crisis, the second set of surveys was collected at a time when

most of the EU countries were in recession as a result of the crisis and the third wave was

collected well after the initial crisis but at a time when many countries were suffering from

2The term ‘extreme’ in this chapter is used to mean positions that are relatively far from the political centre ofthe relevant country, irrespective of whether they lie to the left or the right.

5.1. HYPOTHESES 115

double-dip recessions or grappling with controversy resulting from severe austerity measures.

It is therefore expected that the voter response to the recession would have first arisen in

2009 and perhaps receded somewhat by 2014 and this is the time frame used in each of the

following hypotheses.

One possibility is that there was a voter backlash against those parties that were seen as

supportive of the institutions of the European Union. Given the international scale of the reces-

sion, domestic politicians may have been able to attribute some of the blame for the recession

to the EU (Hobolt and Tilley 2014), particularly since EU institutions do control important

aspects of European economic policy, notably including monetary policy for countries using

the shared currency. Furthermore, the EU was also seen to play a role in the often unpopular

austerity measures that were imposed in many countries as a result of the crisis. This be-

ing the case, it would not be surprising if parties advocating more national autonomy gained

some support at the expense of those parties advocating further integration within the EU.

This question has been addressed by Hobolt and Leblond (2014), who found that support for

remaining in the Eurozone has remained consistently high throughout the crisis and its after-

math, even in countries with severe economic problems or stringent austerity measures. They

argue that this is because citizens estimate the risks associated with leaving to be greater than

those of staying. As theirs was an aggregate study, it will be interesting to see whether their

results can be replicated with an individual-level model. The first hypothesis is thus:

Hypothesis 5.1 The Great Recession provoked a shift in support away from parties supporting

further European integration and towards those opposing it.

Differing convictions about how best to manage the economy are one of the fundamental

divisions underlying the political spectrum. This being the case, it is not difficult to imagine

that a severe economic shock could shake voters’ confidence in some of these positions and lead

them to prefer parties with somewhat different approaches. This naturally raises the question

of whether voters should be expected to move to the left or the right following an event like

the Great Recession. The popular expectation seems to be that voters should move towards the

left. For example, commenting on the results of the 2009 European Parliament elections, one

commentator described the left as having ‘failed to capitalise on an economic crisis tailor-made

for critics of the free market’ (The Economist 2009). This analysis is typical of the newspaper

commentary of those election results, with the left seen as the natural beneficiaries of the Great

Recession, but somehow failing to take advantage of their position (Lindvall 2012, 514).


Within the academic literature, on the other hand, the tendency is to find that poor eco-

nomic conditions are typically followed by a rightwards shift. For example, Durr (1993) at-

tempted to explain why the policy preferences of US voters were further to the left in some

periods, such as the 1960s, than in other periods, such as the 1980s. He proposed that there

was a link between support for redistributionary policies and economic conditions, namely

that voters were more in favour of these policies when they felt economically secure but saw

them as a form of discretionary spending to be reigned in when they felt the economy itself

was threatened. This has been described as a ‘luxury model’ (Anderson and Hecht 2014, 57).

It was found that support for leftist policies was indeed highest when optimism about the

economy was high and low when attitudes were more pessimistic (Durr 1993, 167). These

results were later replicated in a comparative study of fourteen European countries (Steven-

son 2001). An alternative to the luxury model is the idea that the different kinds of economic

troubles are associated with different policy preferences and there is some support for this idea

too. Erikson, MacKuen and Stimson (2002) studied aggregate trends in US politics in diverse

specifics between 1952 and 1996. One of their findings was that ‘policy mood’, a composite

measure of survey responses to specific policy questions, tends to the left when unemployment

is high and to the right when inflation is high (232–235).

There have been some observations of a rightwards shift during the Great Recession but

these are hardly decisive. Soroka and Wlezien (2014), for example, found that there has been

a long-term drop in support for redistribution among British Labour supporters in particular

and that the Great Recession has not arrested the long term trend. Bartels (2014) found that

there did appear to be a general move towards the right following the Great Recession but

also that the effect size is much smaller when controlling for economic conditions—the coun-

tries with left-wing incumbents were often those where the recession was worst (199). They

conclude that ‘the Great Recession produced surprisingly little overall change in the ideolo-

gical proclivities of voters—and that retrospective voting was a stronger and more consistent

factor than ideology in accounting for observed shifts in electoral behavior in OECD countries’

(200). Lindvall (2014) argues that the change in preference over time is more complex than a

simple leftwards or rightwards trajectory. By comparing the patterns of electoral results in the

aftermath of the Great Recession and the Great Depression, he found that in both cases voters

tended to move towards the right in the short term and back towards the centre or even the

left as the crisis continued, even when controlling for economic voting. He proposes that this

5.1. HYPOTHESES 117

could be explained as a result of the different social classes becoming exposed to the severity

of an economic downturn at different times (Lindvall 2013, 149).

These studies inform the second hypothesis:

Hypothesis 5.2 The Great Recession led to increased support for the parties of the right relative

to those of the left, particularly in the short term.

Alternatively, it may be the case that rather than turning to the parties of the left or the

right specifically, voters have turned away from the mainstream parties and towards extreme

or fringe parties. The argument behind this idea is that the mainstream parties suffered a

loss of credibility by failing to prevent the Great Recession from taking place. Since this event

had deep structural causes, even those mainstream parties that were in opposition at the time

could have been tainted by the crisis. In any event, the parties in power across Europe were not

uniformly of the left or the right, so neither the centre-left nor the centre-right could plausibly

claim to be blameless so long as any blame could be attributed to national governments. This

being the case, it might be expected that voters would turn to those parties who are critical of

the established political order or who seek to advance radical or anti-system policy platforms.

Aggregate cross-national election result data has already shown some evidence of an increase

in support for the populist parties of the radical left and radical right, especially in Western

Europe (Hernández and Kriesi 2015). This forms the basis of the third hypothesis:

Hypothesis 5.3 The Great Recession led to a shift in support from centrist parties to more extreme

parties.

There have already been studies on whether the recession has helped more extreme parties

but no consensus has been reached. For example, Allen (2015) studied support for extreme

right parties, defined as those occupying the right–authoritarian corner of the political space,

also using the 2004, 2009 and 2014 waves of EES survey data. He found some support for his

hypothesis that economic grievances were associated with greater support for far right parties

in the period after the crisis. Similarly, Mayer (2014) looked at whether the French working

class moved to the far right as a result of the Great Recession. It was found that although

the far-right National Front has increased its support, particularly among the working class,

a former stronghold of communist support, the magnitude of this increase was actually quite

small (288–290). Grittersova et al. (2015), on the other hand, argued that austerity actually

suppresses support for extreme parties. Using election-year panel data from sixteen countries


between 1978 and 2009, they found that austerity policies are associated with reduced support

for both radical left and radical right parties. Importantly, their data does not extend to the

period of austerity following the Great Recession, although the authors argue that there is

no obvious cause to believe that this would change their results (21). Given this diversity in

results, the hypothesis is worth testing here.

Lastly, it should not be assumed that all voters would respond to the recession in the same

way. One obvious question is whether poor and rich voters responded similarly. Duch and

Sagarzazu (2014, 225) use panel data to examine voter behaviour from the 2009 German and

2010 British parliamentary elections. They measured the economic vote and economic prefer-

ences of different groups of voters, finding that both rich and poor voters had similar levels of

economic voting but that the poor continued to prefer greater levels of redistribution than the

rich (253). This still leaves open the possibility of an individual’s economic assessment influ-

encing their policy preferences. It is also well known that voters are influenced by economic

conditions when deciding whether to vote for the incumbent government and it was shown

in Chapter 3 that this economic voting effect can be observed in the party support levels of

survey respondents before, during and after the recession. In particular, it was found that a

voter’s prospective economic assessment influenced that voter’s support for incumbent parties.

Similarly, if any of the predicted trends are observed then it is also expected that these effects

would be influenced by the voter’s economic assessment. For example, if support for extreme

parties did increase as a result of the recession, it is expected that the size of this increase

would be greatest among those voters who were pessimistic about the future of their national

economy and least among those who were optimistic. The final hypothesis for this chapter is

thus:

Hypothesis 5.4 These effects are strongest among voters believing that the economy was worsen-

ing and weakest among those believing it was improving.

5.2 Classifying parties

In order to test these hypotheses, the various parties need to be classified by some means.

There is a large literature on the classification of political parties, beginning with Duverger

(1959, 61–132), who drew a distinction between ‘mass’ and ‘cadre’ parties. This distinction

pertains to the basic structure of the party. Cadre parties are elite organisations, loose as-

sociations of like-minded politicians, with no true membership beyond their candidates and

5.2. CLASSIFYING PARTIES 119

office holders. Mass parties, on the other hand, have a large membership drawn from the gen-

eral population, who are committed to the party’s ideological project, pay dues to the party

and exercise some influence over the party’s elites. The cadre party model is the traditional

model of conservative parties, whereas the mass party model originated with European social-

ist parties and has been successful enough that some conservative parties have tried to emulate

it. Kirchheimer (1966) later introduced the notion of the ‘catch-all party’, observing that large

mass parties have a tendency to start trying to appeal to people outside of their traditional

support bases in order to improve their electoral performance. Building upon both earlier

works, Panebianco (1988, 262–274) argued that the key distinction among modern parties

is now between ‘mass bureaucratic’ and ‘electoral–professional’ parties. Mass bureaucratic

parties are still heavily membership-driven, depending upon financial dues and motivated by

ideology, whereas electoral–professional parties have only weak ties between their elites and

their membership, are often financed externally and seek primarily to appeal to the elector-

ate. Katz and Mair (1995, 2009) take this further, arguing that the major parties have evolved

into ‘cartel’ parties, which are effectively part of the state itself. On the other hand, they note

that these parties face new opposition from emerging radical parties who challenge their close

relationship with the state (Katz and Mair 1995, 24).

These ideas are helpful for understanding the evolution of the party system but it is not

clear that these terms are particularly useful for classifying individual parties (Koole 1996).

Moreover, these approaches focus on party structure, whereas this thesis is concerned with

individual behaviour. It may well be the case that party structure can explain why some parties

are more able than others to capitalise on the crisis but the question motivating this chapter is

which parties voters have turned to in response to the crisis. Given that party structure is not

highly visible to most voters, this is unlikely to be a motivating factor for them. As a result, this

chapter will instead focus on parties’ spatial location, which is also more congruent with the

hypotheses introduced in the previous section. There is evidence that voters do understand

the terms ‘left’ and ‘right’ and can accurately locate a party on the left–right spectrum (Busch

2016). It is thus reasonable to hypothesise that voters might treat parties differently according

to their left–right location when responding to the crisis. There is also evidence that party

positions tend to remain stable over time (Dalton and McAllister 2015), and that this continued

to be the case during the Great Recession (Dalton 2016).

This thesis focuses on those parties for which the EES surveys include party support ques-

tions. The set of these parties differs somewhat from wave to wave, as the researchers have


tried to include whichever parties were deemed to be most important at the time. The number

of parties in a particular wave varies from 147 to 195. In total, there are 274 unique parties

included for the relevant countries. In order to test this chapter’s hypotheses, it was necessary

to determine whether each of these parties was a left-wing or right-wing party, whether it was

an extreme or moderate party and whether it was a pro- or anti-European integration party.

Leftist parties are reasonably straightforward to identify, since the EES surveys ask respond-

ents where they would place each party on the left–right political spectrum. Other datasets

also include left–right placement data, such as the Manifesto Project, the Comparative Study

of Electoral Systems and the Chapel Hill Expert Survey. Those parties which are generally

agreed to be on the left by survey respondents or with leftist manifestos could be considered

leftist parties. Another possibility is to use a qualitative descriptor of the party’s family, in-

formation which is also included in the CHES dataset. Parties which are classified as socialist

or communist parties would then be considered leftist parties.

It is less straightforward to identify extreme parties, as there are different kinds of parties

that could be described as extreme. For example, extreme parties might be identified with

anti-system parties, which oppose the existing political order. This can be problematic when it

comes to regionalist parties who seek greater autonomy or even independence for a particular

region. These parties are common in Europe and are sometimes seen as outsider parties and,

as such, even subject to the cordon sanitaire, whereby mainstream parties informally agree to

minimise cooperation with them or exclude them from any coalition (McDonnell and Newell

2011, 444). On the other hand, within the affected regions, these parties often occupy a

mainstream position, making it inappropriate to categorise then as extreme. For example, the

Scottish National Party has become the dominant party in Scotland, managing to win major-

ity government in 2011 in a proportionally elected parliament (Johns, Mitchell and Carman

2013).3 The party does however argue for Scottish independence from the United Kingdom,

a position that could well be seen as extreme in other parts of the UK, yet the party is clearly

mainstream within Scotland. This illustrates the difficulty of classifying parties as centrist or

extreme by using qualitative family descriptors.

Another way to identify extreme parties could be to consider one of the same measures of

left–right position discussed above. Those parties which are held to be far to the left or far to

the right could be classified as extreme and the others as moderate. This too is not without

problems, as there are some parties where EES respondents do not agree on a position. The3They have since lost that majority at the 2016 Scottish Parliament election but they remain the dominant

party, having won 63 seats, more than twice the 31 seats won by the next biggest party (BBC News 2016b).


British National Party (BNP), for example, has been described as an extreme right party (Ford

and Goodwin 2010, 1), owing principally to its positions on race and immigration (5). The re-

spondents in the EES survey do not uniformly agree with this classification, however. In 2009,

the only wave in which respondents were asked about the BNP, 30.4 percent of respondents

assigned them a left–right position of ten, indicating a far-right position but 21.6 percent as-

signed a position of zero, indicating a far-left position. Some 16.3 percent of respondents

assigned no position and none of the remaining possibilities attracted more than 5.5 percent

of the response. Although the party’s positions on certain issues is right-wing, they also work

hard to win support from traditional Labour voters (5). This could be behind the pattern of

responses here, with some voters indicating that they would consider the party to be on the

left and that they find it to be extreme, even though the party is certainly not a communist

party, the usual meaning of far left.

The problem of inconsistent party placement is likely to affect all populist parties, whose

positions on various issues are not necessarily motivated by a coherent, or at least conven-

tional, ideology. One way of addressing this issue could be to measure the mean distance

from the centre reported by respondents, rather than aggregating the reported positions and

then measuring the distance from the centre. This means that a voter reporting that a party is

far left and one reporting that it is far right are considered to be in agreement that the party

is extreme. This is an improvement but it still leads to some surprising classifications, such

as the centre-right New Democracy party in Greece being classified as extreme—presumably

as a reaction to the austerity measures they introduced when they were in government. An

alternative to using the EES measures of left–right position is to use expert surveys of party

positions. The CHES dataset includes a similar measure of general left–right position as well as

separate measures of position on the left–right economic and authoritarian–libertarian scales.

Parties could be classified as extreme according to their positions on any of these scales. Being

an expert survey, the CHES dataset has highly reliable measures of party position and so does

not suffer the same problems described above. In any event, Dalton, Farrell and McAllister

(2011, 121) have shown that expert surveys, citizen surveys and manifesto assessments of

party position display a ‘striking consistency’.

Finally, parties need to be classified according to their position regarding European integ-

ration. The EES surveys include a question asking voters to assign parties a score from zero

to ten, where zero means unification ‘has already gone too far’ and ten means it ‘should be

pushed further’. This measure would be ideal except for the fact that there were problems


with this particular variable in the 2014 wave of the survey, and as a result, this measure is

simply unavailable in that wave at the time of writing of this thesis. Most of the parties asked

about in the 2014 wave were also asked about in either or both of the earlier waves, and the

data from those first two waves shows that party positions tended not to change by very much

if anything between 2004 and 2009. By assuming that this was also true between 2009 and

2014, the missing data could be imputed for about three quarters of the parties in the 2014

wave, although this solution is not highly satisfactory, as some of the party position data is ten

years out of date at that point. The CHES dataset also includes a question asking experts to

assign parties a score indicating the party’s orientation towards European integration, as well

as a question asking the relative salience of that position in the party’s public image. This is

useful because it makes it possible to distinguish between parties that are overtly Europhile

or Eurosceptic and those parties who may hold similar positions but are less strongly associ-

ated with them in the public mind. Unlike the leftism and extremism measures, there is no

appropriate family classification that could be used for this, as even parties defined by their

Euroscepticism are coded into other families. For example, the United Kingdom Independence

Party is coded as belonging to the conservative family.

The method this chapter uses to classify parties is to use the party position measures from

the CHES dataset. The CHES project surveys experts regarding the parties in their countries.

Each party is positioned on a variety of scales, such as left to right, authoritarian to libertarian

and pro- to anti-integration.4 The experts use their own professional judgement to assign these

positions rather than specified policy items but the mean scores assigned by each expert are a

reliable measure of a party’s position. This approach has several benefits. There are a number

of relevant scales, so party positions on a particular issue can be identified more precisely

than with the other approaches. Because these are scales, rather than nominal measures, this

method also makes it possible to compare party positions, unlike the party family approach.

Finally, the CHES surveys include the relevant questions in each of its waves, unlike the EES

data, which lacks a key question in the 2014 wave.5

It must be noted that the CHES data is not a perfect fit either. There have been five waves of

the survey so far, taking place in 1999, 2002, 2006, 2010 and 2014, which do not correspond

precisely with the three relevant EES survey waves, which took place in 2004, 2009 and 2014.

4Most of these items are measured on an eleven-point scale ranging from 0 to 10. The support for furtherintegration item is measured on a five-point scale, which has been recoded to range from−3 to+3 for this analysis.

5Although the party unification position questions were intended to be included in the 2014 wave, this hasnot been released owing to ‘an error in the questionnaire development’ (Popa et al. 2015, 5). The project intendsto address this omission with a follow-up survey but at the time of writing this data is not available.


Table 5.1: Variation in party position over time

Left–right position Integration positionFixed effect Coeff. SE p Coeff. SE p

Intercept 5.206 (0.149) < 0.001 0.996 (0.115) < 0.001Year 2006 0.028 (0.055) 0.612 −0.109 (0.051) 0.034Year 2010 0.030 (0.066) 0.649 −0.071 (0.061) 0.249Year 2014 0.094 (0.073) 0.200 −0.094 (0.067) 0.162

Party random effect Var. SD Var. SD

Intercept 4.580 (2.140) 2.643 (1.626)Year 2006 0.143 (0.378) 0.208 (0.457)Year 2010 0.410 (0.640) 0.421 (0.649)Year 2014 0.542 (0.736) 0.514 (0.717)

Multilevel model predicting expert assessment of party position for each year that the CHESsurvey was conducted. Sample size is 7661 responses for left–right position and 7878 re-sponses for European integration position for 226 parties in both cases. Pseudo R2 is 0.785for left–right position and 0.772 for integration position. Source: CHES

A model was constructed to estimate the amount of variation in two of the key measures among

the parties included in the CHES data. Table 5.1 summarises the results. The low magnitude of

the fixed effects indicates no evidence for a shift affecting all parties, which is to be expected.

The random effects are more interesting, as these show the degree to which individual parties

changed position over time. The variance of the intercepts associated with the year increases

over time, which is consistent with a random walk. The magnitude of these variances is also

quite low, suggesting that parties do tend to change position over time but somewhat slowly.

Since party positions appear to change slowly over time, the misalignment between the EES

and CHES survey waves should not pose exceptional difficulties. This misalignment has been

addressed by interpolating the party position at the required time.6 Luxembourg, Cyprus

and Malta were not included in the CHES dataset so these countries were excluded from this

chapter. The resulting data set includes 251 unique parties, or 464 total parties,7 within 22

countries.

Using another dataset to measure objective party position represents a change from the

practice used in Chapter 3, where each individual’s subjective measures were preferred. Apart

from the issues discussed above, there is a further reason for this difference. In that chapter, it

was argued that an individual’s support for a particular party ought to decrease as the distance6Not every party for which party support data exists in the EES surveys has corresponding data in the CHES

surveys. This affects approximately one quarter of the EES parties, but these are mostly parties of little practicalimportance, such as parties that only existed for a short time or which had a very low vote share. These partieshave been excluded from analysis in this chapter.

7As in previous chapters, parties in different years are treated as different entities because some countries havehighly fluid party systems.


Figure 5.1: Relationship between party support and position

European integration left–right

-3

-2

-1

0

1

2

3

-2 0 2 0 2 4 6 8 10party position

mea

n pa

rty p

refe

renc

e

Scatter plots showing the relationship between the mean relative support for a party and thatparty’s position on European integration, from −3 (anti-integration) to +3 (pro-integration),as well as its position on the left–right spectrum, from 0 (left) to 10 (right). The best linearfit is shown for the European integration plot. The left–right plot shows a relationship that isclearly non-linear, so it is shown with the best quadratic fit. Both curves are shown with their95% confidence intervals. Source: EES & CHES

increases between the party’s position and their own on the left–right scale, which was found

to be the case. In this spatial model of party support, it makes sense to use that individual’s

subjective measure of the party’s position, since the measure of their own party position is

necessarily subjective and this means the two positions can be meaningfully compared. Al-

though the spatial model remains the underlying model of party support used in this chapter,

and the subjective party distance variable is therefore still included, the focus here is on the

parties themselves. This is why, when classifying the parties, the individual’s perception of the

party is less important than its objective position.

5.3 How party position relates to voter support

Before building an individual-level model, it is necessary to examine the relationship between

aggregate party support and the new party position variables introduced in this chapter. The

5.3. HOW PARTY POSITION RELATES TO VOTER SUPPORT 125

mean relative party support8 was estimated for each party in the study, and this functions

as a measure of that party’s support relative to the other parties in the same country. For

this purpose, the data from all three survey years is combined but each party is treated as a

separate entity in all three years since the party position variables are time-dependent as well.

Figure 5.1 shows the relationship between mean party support and the party’s position on

European integration, in the first panel, and its left–right position, in the second panel. The

former ranges from −3 for strongly anti-integration parties, often referred to as ‘Eurosceptic’

parties, to +3 for strongly pro-integration, or ‘Europhile’, parties and the latter ranges from 0

for far-left parties to 10 for far-right parties.

The scatter plot in Figure 5.1 shows that pro-integration parties are more popular than

anti-integration parties and this relationship is approximately linear, as can be seen from the

linear fit line also shown. Similarly, centrist parties are more popular than far left and far

right parties. Curiously the parties that are in the very centre of the spectrum appear to be

rather less popular than those on the centre-left and the centre-right. This plot is shown with

a quadratic fit line, which is a good fit but does not account for the depressed support among

centrist parties. The quadratic fit is preferred to one attempting to account for this anomaly

for two reasons. First, there is a risk of overfitting the data. The quadratic fit is already a good

fit and the residuals from this model are not significantly different from a normal distribution.

Second, it has been shown in previous chapters that incumbent parties are more popular than

opposition parties and these parties tend to occupy the centre-left and centre-right. When

controlling for incumbency, the centrist parties are no longer outliers.

It is also clear from these plots that party position on either of these scales only explains a

small proportion of the variance in mean party support. The regression R2 is 0.18 for the party’s

integration position and 0.14 for its left–right position. Although these are small numbers, it

would be surprising if party position alone explained a large proportion of the variance in

party support. It is remarkable that even eighteen percent of this variance can be explained

by a party’s position on European integration alone, although it must be stressed that this

proportion is likely to be less when controlling for other variables. This is particularly likely

for the incumbency variable, as governing parties are generally more pro-integration than

opposition parties. The mean position on European integration among the cabinet parties in

the dataset is 1.85 [1.69, 2.02], and that for opposition parties is 0.46 [0.26, 0.67].8Party support is a measure of how likely an individual is to support a particular party on an eleven-point scale.

Relative party support is party support centred around each individual’s mean support level for the parties in theircountry. These measures are explained in detail in Chapter 2. Mean relative party support is thus an aggregatemeasure of a party’s support compared to other parties in the same country.


Figure 5.2: Relationship between integration position and salience

0.2

0.4

0.6

0.8

1.0

-3 -2 -1 0 1 2 3position

salie

nce

Scatter plot showing the relationship between a party’s position on European integration, from−3 (anti-integration) to +3 (pro-integration), and the salience of the integration issue in thatparty’s public image, from 0 (low) to 1 (high). This is shown along with the best quadratic fit,which is clearly a better fit than a linear one. The shaded area represents the 95% confidenceinterval around the curve. Source: CHES

The plots for the other party position variables are not shown here because they are similar

to these. The social and economic dimensions follow similar distributions to the left–right

spectrum. The integration salience measure is almost unrelated to party support, although

parties with high salience are slightly more popular, even when controlling for the party’s

position on that issue. It might be expected that salience would affect support differently

for parties with different positions on integration but surprisingly the interaction between

integration position and salience is not significant. There is no significant quadratic interaction

either. This is likely because these two variables are actually related. As Figure 5.2 shows,

salience is higher for parties that are more strongly pro- or anti-integration than it is for parties

with more moderate positions. This makes sense because it is difficult to see how a party could

make a moderate position on this issue an important part of its public image. Although the

correlation is not high (R2 = 0.23), this may explain why the expected interactions are not

observable. Because of this, the rest of the analysis will ignore the salience variable, focusing

on integration position only.

5.4. THE CHANGING FORTUNES OF PRO-EUROPEAN INTEGRATION PARTIES 127

Table 5.2: Regression model predicting general left–right position

Predictor Coeff. SE p

Intercept −0.197 (0.117) 0.094Economic position 0.683 (0.021) < 0.001Social position 0.412 (0.019) < 0.001

This model predicts a party’s general left–right position from its position on the economic andsocial dimensions. Sample size is 434 parties. Adjusted R2 is 0.847. Source: CHES

The relationship between the three left–right position variables was also explored. These

variables are the party’s positions on the general left–right spectrum as well as an economic

spectrum and a social spectrum. These latter variables allow parties to be positioned in a

two-dimensional political space, rather than along a single line, according to the theory that

positions on economic and social issues are often orthogonal (for example, see Lester 1994;

Swedlow 2008). The correlation between social and economic position is r = 0.38, which

suggests that these positions are not quite orthogonal but it is low enough that it may be

worth examining the two dimensions separately. It should be expected that a party’s general

left–right position can be predicted from its positions on the social and economic dimensions

and this is in fact the case. As Table 5.2 shows, position on the social and economic dimensions

explains 85 percent of the variation in general left–right position. It is interesting to note that

economic position contributes more strongly to a party’s general position than social position

does. This model shows that the economic and social position scores are a meaningful decom-

position of a party’s left–right position, so party support models will be constructed using both

the general left–right positions and the two-dimensional positions.

5.4 The changing fortunes of pro-European integration parties

The first hypothesis is that support for pro-integration parties fell after the Great Recession.

In order to test this, Model 5A was constructed. It is based on Model 3D, which measures

the economic vote using a three-way distinction between prime ministers’ parties, other cab-

inet parties and opposition parties. That model was extended by including as a predictor the

variable measuring each party’s position on European integration. As with the other terms

in the model, this predictor has been centred around the grand mean. In the original model,

economic voting is measured by the interaction between an individual’s prospective economic

assessment and a party’s incumbency status. This is because economic voting theory posits


that economic conditions affect government and opposition party support differently. A fur-

ther term measuring the three-way interaction between economic assessment, incumbency

and the survey year made it possible to discern how the strength of the economic voting effect

has changed over time.

The new model includes a similar interaction between economic assessment and party po-

sition on European integration, so that it can be tested whether support for pro-integration

parties is higher among optimistic or pessimistic voters. In other words, this inclusion makes

it possible to determine whether economic voting affected not just incumbent parties but also

pro-integration parties. Furthermore, a three-way integration term between economic assess-

ment, European integration position and survey year has been included so as to be able to

determine whether these effects changed over time.9 The incumbency terms mentioned are

still included so as to control for the standard economic voting effect, which has already been

shown to have existed.

In addition to the random effects terms included in the original model, corresponding

random effects terms were considered for each of the new fixed effects. Since the European

integration position variable was measured at the party level, it would only be meaningful to

include these random effects at the country level. It was previously shown that the country-

level variance in this data is almost zero, which is why the basic model does not include a

random intercept for the country level, so it is not surprising that the variance of all of these

random effects terms was also very small and in most cases their inclusion did not improve the

model fit significantly. A random effect for the integration position variable alone was included

because, although the measured variance was very low, it was large enough to improve the

model fit significantly and does not hinder convergence.

As in previous chapters, post-estimation simulation has been used to derive key predic-

tions along with the corresponding uncertainty estimates from the model coefficients and it

is these that will form the basis of the following discussion. The table of coefficients can be

found in Appendix B. Figure 5.3 shows how support changed over time for parties with differ-

ent positions towards European integration. The points plotted correspond to the predicted

support of an individual with a neutral prospective economic assessment towards each type

of party in each survey year. According to this plot, neutral voters were more likely to sup-

port pro-integration parties than anti-integration parties across all three years, although the

9Although not always mentioned explicitly, the models discussed in this chapter also include the additionalterms implied by any interactions mentioned. For example, where a three-way interaction is mentioned, all of thepairwise two-way interactions as well as the single factor terms are also included.

5.4. THE CHANGING FORTUNES OF PRO-EUROPEAN INTEGRATION PARTIES 129

Figure 5.3: Party support by position on European integration

3.4

3.6

3.8

4.0

4.2

4.4

2004 2009 2014year

pred

icte

d pr

efer

ence

party position

pro-integration

neutral

anti-integration

Predicted support in each survey year of a voter holding a neutral prospective economic as-sessment for a strongly pro-integration party, a neutral party and a strongly anti-integrationparty. These correspond to measured European integration positions of +3, 0 and −3 respect-ively. The modelled relationship between a party’s integration position and a voter’s supportfor that party in a particular year is linear, so the predictions for moderate parties lie betweenthose shown here. Source: EES, ParlGov & CHES

magnitude of this difference decreased dramatically over time. In 2004, a neutral voter has

a predicted support of 4.54 points (SE = 0.10, p < 0.001) for a pro-integration party, 3.99

points (SE = 0.07, p < 0.001) for a neutral party and 3.44 points (SE = 0.14, p < 0.001)

for an anti-integration party. This means that, compared to an anti-integration position, a

pro-integration position was associated with an extra 1.10 points (SE = 0.20, p < 0.001) of

support. In 2009, this difference had declined to 0.76 points (SE = 0.20, p < 0.001) and by

2014, the difference was no longer statistically significant (0.15 points, SE= 0.19, p = 0.41).

This suggests that voters were quite supportive of a pro-European integration position before

the beginning of the crisis, slightly less supportive after the crisis had begun and indifferent to

such a position after the end of the crisis.

So far, this analysis has only discussed voters with a neutral assessment of the course of

their country’s economy. Figure 5.4 compares the predicted difference in support for pro-

integration and anti-integration parties for optimistic, neutral and pessimistic voters. A pos-

itive value in a given year, indicates that such a voter is expected to prefer a pro-integration


Figure 5.4: Relative support for pro-integration parties

-0.5

0.0

0.5

1.0

2004 2009 2014year

pred

icte

d di

ffere

nce

group

optimistic

neutral

pessimistic

Difference in predicted support for a strongly pro-integration and anti-integration party ineach year among voters holding different prospective economic assessments. This differencerepresents the relative support for a pro-integration party (measured position of +3) over ananti-integration party (−3). The economic assessments shown are those for highly optimisticvoters (measured assessment of +2), neutral voters (0) and highly pessimistic voters (−2).Source: EES, ParlGov & CHES

party and a negative value indicates that the voter would prefer an anti-integration party. In

2004, pro-integration parties were preferred by all three voter groups. The pessimistic group

had a relative support towards these parties of +0.89 points (SE = 0.24, p < 0.001) and the

optimistic group had a relative support of +1.31 points (SE = 0.20, p < 0.001). The differ-

ence between the groups was not significant however (0.42 points, SE = 0.28, p = 0.14). By

2009, it appears that this relative support had fallen in each group by an approximately equal

amount but this change is not significant either (0.34 points for the neutral group, SE= 0.26,

p = 0.18). By 2014, the three groups had become less similar to each either. Optimistic voters

still had a strong relative support towards pro-integration parties, of +0.89 points (SE= 0.23,

p < 0.001). This is a mere 0.04 points (SE= 0.31, p = 0.87) less than in 2009. Neutral voters

had become almost indifferent to a party’s European integration position, with a relative sup-

port of +0.15 points (SE = 0.19, p = 0.41), a 0.60 point (SE = 0.23, p = 0.01) fall from

2009. Pessimistic voters in 2014 were the only group to prefer anti-integration parties, with a

5.5. LEFT–RIGHT POSITION: A SHIFT TO THE EXTREMES 131

relative support of −0.59 points (SE= 0.23, p < 0.001). This is a huge 1.16 point (SE= 0.30,

p < 0.001) fall from 2009.

In short, parties arguing for closer European integration were at an advantage in 2004,

before the beginning of the recession and this was probably independent of a voter’s eco-

nomic assessment. In 2009, during the recession, this was still the case, except that the size of

that advantage had declined slightly. By 2014, well after the recession, voters with different

economic assessments had started to behave differently in this respect. Optimistic voters con-

tinued to prefer the pro-integration parties but pessimistic voters now preferred those parties

arguing for weaker European integration, with voters of a neutral assessment having no clear

preference. This seems to be evidence in support of the hypothesis that the recession provoked

a shift away from pro-integration parties. It is worth noting, however, that the bulk of the shift

and the stratification according to economic assessment did not take place until well after the

recession.

5.5 Left–right position: a shift to the extremes

At the beginning of this chapter, it was hypothesised that voters shifted their support away

from leftist parties and towards those on the right during the recession. This is not necessarily

expected to be a long-term shift, as earlier work has found that such shifts tend to be short-

lived (Lindvall 2014). It was also hypothesised that it was extreme parties that benefited from

the recession. These hypotheses can be tested by modelling the impact of a party’s left–right

position on voter support for that party. Model 5B is based on, once again, Model 3D, which

has this time been extended to measure the impact of a party’s general left–right position on

party support. It was shown earlier in this chapter that the relationship between mean support

for a party and that party’s left–right position is approximately quadratic, with voters gener-

ally preferring centrist parties over extreme parties. The new model therefore represents this

relationship quadratically. Both the linear and quadratic terms are interacted with economic

assessment and time, allowing this relationship to be compared between the different groups

of voters in each year, as with previous models. No further random effects terms have been

included in this model because the variance of these terms is extremely low.

This model was estimated and used to predict the support in each year of optimistic, pess-

imistic and neutral voters for various party left–right positions. These predictions are sum-

marised visually by Figure 5.5. This figure shows that, in 2004, all groups of voters generally

preferred centrist parties over extremist parties, leading to an inverted U-shaped curve for


Figure 5.5: Party support by general left–right position

2004 2009 2014

2

3

4

5

2

3

4

5

2

3

4

5

optimistic

neutralpessim

istic

0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10party position

pred

icte

d pr

efer

ence

Predicted party support by party’s general left–right position in each year among voters holdingdifferent prospective economic assessments. These are shown along with their 95% predictiveintervals. Party position ranges from 0 (far left) to 10 (far right). The economic assessmentsshown are those for highly optimistic voters (measured assessment of +2), neutral voters (0)and highly pessimistic voters (−2). Source: EES, ParlGov & CHES

each. It can also be seen that optimistic voters leant slightly to the right and pessimistic voters

leant slightly to the left, with neutral voters leaning in neither direction. This is an interesting

pattern, because it suggests that before the recession, voters had a small tendency to support

the right when they believed that the economy was likely to do well and the left when they

believed otherwise. By 2009, this pattern had changed, with all three groups preferring the

centre. This suggests that, rather than enhancing the tendency for optimists to support the

right and pessimists the left, the recession in fact suppressed it. The year 2014 is particularly

interesting because it shows a rather different pattern altogether. Optimists in 2014 still pre-

ferred centrist parties overall but the flatter shape of the curve indicates an increased openness

to more extreme parties. The curve for neutral voters appears completely flat, suggesting that

this group of voters was indifferent between extreme and centrist parties. Finally, pessimistic

5.5. LEFT–RIGHT POSITION: A SHIFT TO THE EXTREMES 133

Table 5.3: Left–right and extreme tendency by year and voter group

Group Year Left–right tendency Extreme tendency

Optimistic2004 +1.45 [+0.08, +2.82] −0.102 [−0.149, −0.055]2009 −0.54 [−1.72, +0.65] −0.079 [−0.120, −0.037]2014 −0.08 [−1.29, +1.13] −0.036 [−0.078, +0.006]

Neutral2004 −0.23 [−1.29, +0.84] −0.076 [−0.113, −0.039]2009 −0.60 [−1.53, +0.34] −0.069 [−0.101, −0.036]2014 −0.30 [−1.24, +0.65] +0.003 [−0.030, +0.035]

Pessimistic2004 −1.90 [−3.22, −0.59] −0.050 [−0.095, −0.005]2009 −0.65 [−1.82, +0.51] −0.059 [−0.098, −0.019]2014 −0.52 [−1.70, +0.67] +0.041 [+0.001, +0.081]

Left–right tendency is the difference between a group’s cumulative predicted support for right-wing and left-wing positions, positive for a right-wing tendency and negative for a left-wingtendency. Extreme tendency is the curvature of the quadratic function describing the relation-ship between party position and predicted support for that party. A positive curvature indicatesa preference for more extreme parties and a negative curvature a preference against extremeparties. These figures are shown with their 95% predictive intervals. The three groups corres-pond to highly optimistic voters (measured assessment of +2), neutral voters (0) and highlypessimistic voters (−2). Source: EES, ParlGov & CHES

voters in 2014 appear to be more supportive of extreme parties than centrist parties, with a

U-shaped curve that is no longer inverted.

In order to determine whether these apparent differences are statistically significant, it is

helpful to find a relevant quantity that can be measured. In order to determine whether a

particular group had an overall left-wing or right-wing tendency, the total area under the left

half of the curve was compared to that under the right half.10 This difference is positive in

the case of a right-wing tendency and negative in the case of a left-wing tendency. Similarly,

support for extreme parties over centrist parties can be described by the curvature of the

function.11 A positive curvature indicates that support increases towards the extremes and

a negative curvature indicates that support decreases towards the extremes. A curvature of

zero implies that the curve is in fact a line. This curvature will be referred to as the extreme

tendency.

The left–right and extreme tendencies for optimistic, pessimistic and neutral voters in each

year are given in Table 5.3 along with their predictive intervals. These figures confirm the10A more precise description of this quantity follows: the party support curve for a particular year and group

can be described by a quadratic expression of the form y = ax2+ bx + c, where x represents the party’s left–right

position and y support for that party. Let the function C(a, b) =∫ b

aax2 + bx + c d x be the cumulative support

between a and b. Then ∆ = C(5,10)− C(0,5) = (875a/3+ 75b/2+ 5c)− (125a/3+ 25b/2+ 5c) = 250a+ 25bdescribes the difference in support between all right-wing and all left-wing parties. This quantity was computedalong with its predictive interval to produce the figures described here.

11The curvature of a function is described by its second derivative. In this case, d2 y/d x2 = 2a. Once again,this was computed with its predictive interval to produce the figures given here.


visual findings described above. In 2004, optimistic voters had a rightward tendency and

pessimistic voters a leftward tendency. Neutral voters in 2004 and all voters in the other years

had no significant left–right tendency. In 2004 and 2009, all groups of voters had support

functions with negative curvature, that is, they tended to prefer parties with more centrist

positions than parties at the extremes. By 2014, this was no longer the case. Pessimistic voters

had a positive support curvature, meaning that extreme parties were supported more strongly

than centrist parties. Neutral and optimistic voters had support curvatures not significantly

from zero and the point estimate for neutral voters was very close to zero. The point estimate

for optimistic voters was somewhat less than zero, hence the visibly curved shape in Figure 5.5,

but a flat line or a curve in the other direction are both plausible. On this evidence, it is likely

that both groups of voters were almost indifferent to a party’s left–right position in that year.

5.6 Beyond left and right: social and economic dimensions

Political position is sometimes modelled in two dimensions rather than the traditional con-

tinuum. These dimensions are usually a left–right economic dimension and a libertarian–

authoritarian social dimension. The argument for this more complex approach is that the two

dimensions are in practice orthogonal, so it is not unusual to find that parties that are, for

example, right-wing on economic issues still have a diversity of positions on social issues. This

being the case, the hypothesis that voters moved to leftist parties as a result of the recession

could be more accurately tested using a two-dimensional model, since movement along each

dimension can be analysed separately. In two-dimensional terms, this hypothesis posits that

voters would have moved left along the economic dimension, with no motion along the social

dimension. The alternative hypothesis that voters were inclined to move to the extremes also

applies particularly to the economic dimension. There is no expectation of movement along

the social dimension. This is because in a harsh economic environment voters may be more

open to more radical economic policies but there is no obvious reason why this should lead to

a change in their other views.

The one-dimensional model discussed in the previous section has been modified to produce

a two-dimensional model, designated Model 5C. As discussed earlier in the chapter, party

position has been measured along both the economic and social dimensions, as well as a

general left–right dimension. The general dimension used in the previous model has been

substituted with the economic and social dimensions, each of which is modelled quadratically

like the original. The two dimensions are not modelled as interacting because it as assumed

5.6. BEYOND LEFT AND RIGHT: SOCIAL AND ECONOMIC DIMENSIONS 135

Figure 5.6: Party support by economic left–right position

2004 2009 2014

2.53.03.54.04.55.05.5

2.53.03.54.04.55.05.5

2.53.03.54.04.55.05.5

optimistic

neutralpessim

istic

0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10party position

pred

icte

d pr

efer

ence

Predicted party support by party’s economic left–right position in each year among votersholding different prospective economic assessments. These are shown along with their 95%predictive intervals. Party position ranges from 0 (far left) to 10 (far right). The economicassessments shown are those for highly optimistic voters (measured assessment of+2), neutralvoters (0) and highly pessimistic voters (−2). Source: EES, ParlGov & CHES

that motion along one dimension is completely independent of motion along the other. As

before, no further random effects terms have been included owing to the very low variance of

those terms.

Figure 5.6 shows predicted support according to year and economic assessment towards

a party at different positions along the economic dimension. In making these predictions, the

party’s social position has been held to be in the centre. This only affects the vertical offset

of the curve, not its shape, as the model does not allow the two dimensions to interact. Once

again, it appears that optimistic voters in 2004 had a tendency to prefer right-wing parties

and pessimistic voters left-wing voters, although these tendencies appear stronger than in

the unidimensional model. Neutral voters still appear to prefer the centre. Similarly, the

magnitude of these tendencies has declined by 2009 but unlike the unidimensional model,

the optimistic and pessimistic voters still have a visibly right-wing and left-wing tendency


Table 5.4: Directional and extreme tendency along economic dimension

Group Year Directional tendency Extreme tendency

Optimistic2004 +3.09 [+1.50, +4.68] −0.042 [−0.093, +0.009]2009 +1.06 [−0.40, +2.52] −0.057 [−0.102, −0.013]2014 +0.72 [−0.64, +2.08] −0.015 [−0.058, +0.027]

Neutral2004 +0.07 [−1.19, +1.32] −0.048 [−0.089, −0.007]2009 −0.52 [−1.69, +0.64] −0.064 [−0.099, −0.028]2014 −0.54 [−1.59, +0.52] +0.005 [−0.029, +0.038]

Pessimistic2004 −2.96 [−4.47, −1.44] −0.054 [−0.103, −0.005]2009 −2.10 [−3.50, −0.70] −0.070 [−0.112, −0.027]2014 −1.80 [−3.10, −0.50] +0.025 [−0.015, +0.065]

Similar to Table 5.3, except that the dimension analysed is the economic left–right spectrumrather than the general left–right spectrum. A positive directional tendency here indicates apreference for the economic right and negative a preference for the economic left. Source:EES, ParlGov & CHES

respectively. Each group in both years prefers parties near the vertex over parties further away

but given the extreme location of the vertices for optimistic and pessimistic voters in 2004, this

may not be a particularly useful measure for those groups. In 2014, the directional tendency

of each groups appears to be maintained, with optimistic voters preferring right-wing position

and pessimistic voters left-wing positions, with neutral voters still close to indifferent. Each of

the 2014 curves has flattened somewhat compared to the previous years. This suggests that

voters in 2014 were more open to extreme positions and less open to centrist positions while

retaining the same directional preference as voters in previous years.

Numeric measures of each group’s directional and extreme tendencies are given in Table 5.4

along with their predictive intervals. These mostly confirm the predictions made above. Op-

timistic voters do have a rightward directional tendency in 2004, with rightward point estim-

ates in the other years as well, although the latter are not significant. Similarly, pessimistic

voters have a leftward directional tendency and this is significant in every year. Neutral voters

have no significant directional tendency in any year. The extreme tendency, described by the

quadratic curvature, of optimistic voters is negative but not quite significant in 2004. In the

context of the strong rightward tendency for this group, this figure is not especially meaning-

ful. In 2009, however, this figure is significantly negative suggesting that optimistic voters in

this year generally preferred centrist parties over more extreme parties. By 2014, this figure

was no longer significant with a point estimate close to zero, suggesting that these voters were

less averse to extreme economic positions than they had previously been. Pessimistic voters

had a similar trajectory, with their extreme tendency significantly negative in 2004 and 2009


but with a positive albeit no longer significant point estimate in 2014. The extreme tendency

for neutral voters was not significant in any year, although even in this group the point es-

timate in 2014 was significantly greater than in 2009 (difference is 0.068 points, SE= 0.025,

p < 0.01).

In summary, generally speaking, voters who believed the economy was improving gener-

ally supported economically right-wing parties and those who believed it was getting worse

generally supported economically left-wing parties, even when controlling for the spatial dis-

tance between the party and the voter. Not much changed between 2004 and 2009, other

than a slight weakening of this tendency. This suggests that the recession sparked no major

immediate shift in economic policy preferences. There was little if any change in this tend-

ency between 2009 and 2014 but there was an increased tendency to prefer more extreme

economic positions over centrists ones. Since this took place in the post-crisis period, it sug-

gests that the political response to the crisis did lead to a shift in policy preferences away from

the economic mainstream towards the extremes.

Turning to the social dimension, the relevant predictions are summarised in Figure 5.7.

For these predictions, the economic position is held in the centre while the social position is

varied. The expectation was that these relationships would not have changed over time but

this does not appear to have been the case. Optimistic voters do have similar looking curves

in each year, with a directional tendency towards libertarian positions in all years, though

somewhat less so in 2014. Pessimistic voters appear to have a strong centrist tendency in

2004, with an evidently negative curvature and a vertex close to the centre. By 2009, the

curvature has flattened considerably, though the vertex is still close to the centre. In 2014,

the curve is still quite flat but there is now a clear tendency to prefer authoritarian positions.

Neutral voters are quite similar, starting with a clear negative curvature and a preference

for libertarian positions. In 2009, the curve starts to flatten and by 2014 the curve remains

relatively flat but the vertex has moved to the centre. In other words, it appears that optimistic

voters in 2004 were libertarian-leaning and pessimistic voters centrist but by 2014, all groups

had shifted to the right, so that optimists were centrist and pessimists authoritarian-leaning.

Table 5.5 presents the directional tendency and curvature of the relationships between

party preference and social political position in the same manner as previously. Once again,

these confirm the patterns shown in the visual analysis. In 2004, optimistic voters had a clearly

libertarian tendency, neutral voters a not quite significant tendency in the same direction and

pessimistic voters an approximately centrist tendency. The point estimates of directional tend-


Figure 5.7: Party preference by social libertarian–authoritarian position

2004 2009 2014

3.0

3.5

4.0

4.5

5.0

3.0

3.5

4.0

4.5

5.0

3.0

3.5

4.0

4.5

5.0

optimistic

neutralpessim

istic

0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10party position

pred

icte

d pr

efer

ence

Predicted party preference by party’s social libertarian–authoritarian position in each yearamong voters holding different prospective economic assessments. These are shown alongwith their 95% predictive intervals. Party position ranges from 0 (libertarian/post-materialist)to 10 (traditional/authoritarian). The economic assessments shown are those for highlyoptimistic voters (measured assessment of +2), neutral voters (0) and highly pessimisticvoters (−2). Source: EES, ParlGov & CHES

ency have changed slightly between 2004 and 2009 but these changes are not significant. The

largest of these differences, that for optimistic voters, is 0.53 points (SE = 0.92, p = 0.57).

Between 2009 and 2014, on the other hand, the estimates for all three groups has shifted

in the authoritarian direction. In the case of optimistic voters, this change is not significant

(difference is 1.11 points, SE = 0.86, p = 0.20) but it is significant for neutral (1.38 points,

SE = 0.69, p = 0.04) and pessimistic voters (1.66 points, SE = 0.84, p = 0.05). Interest-

ingly, not only were pessimistic voters the most authoritarian group in 2004 but they were

also the group with the greatest shift in that direction between 2009 and 2014. As for the

extreme tendency or curvature, this was not significantly different from zero for optimistic

voters in any year. For neutral voters, this was significantly negative in 2004 but not in 2009

and 2014 and the increase between 2004 and 2009 was not significantly different in any case

(0.022 points, SE = 0.023, p = 0.34). The curvature for negative voters was significant and


Table 5.5: Directional and extreme tendency along social dimension

Group Year Directional tendency Extreme tendency

Optimistic2004 −1.90 [−3.21, −0.59] −0.014 [−0.057, +0.028]2009 −2.42 [−3.65, −1.20] −0.013 [−0.051, +0.025]2014 −1.32 [−2.50, −0.14] −0.020 [−0.060, +0.019]

Neutral2004 −1.08 [−2.12, −0.03] −0.036 [−0.070, −0.002]2009 −1.27 [−2.24, −0.30] −0.014 [−0.044, +0.015]2014 +0.11 [−0.82, +1.04] −0.014 [−0.045, +0.017]

Pessimistic2004 −0.25 [−1.50, +0.99] −0.058 [−0.099, −0.017]2009 −0.12 [−1.28, +1.04] −0.016 [−0.052, +0.020]2014 +1.54 [+0.40, +2.69] −0.007 [−0.046, +0.031]

Similar to Tables 5.3 and 5.4, except that the dimension analysed is the social libertarian–authoritarian spectrum rather than the general left–right spectrum. A positive directionaltendency here indicates a preference for the more traditional or authoritarian parties andnegative a preference for more libertarian or post-materialist parties. Source: EES, ParlGov &CHES

negative in 2004 but by 2014 this curvature was almost zero. This difference was not quite

significant (0.051 points, SE= 0.029, p = 0.08).

The changes in directional tendency over time are surprising. The fact that there is no

significant change between 2004 and 2009 suggests that there was no immediate reaction to

the crisis in terms of social preferences, which accords with the expectation that the social

dimension would be unaffected by the crisis. On the other hand, the change in directional

tendency between 2009 and 2014 indicates that in the years after the initial recession, there

was a general shift in preference towards traditional or authoritarian positions. Furthermore,

the more pessimistic the voter about economic conditions, the stronger the shift. This is partic-

ularly interesting, since pessimistic voters were also the least libertarian in the first place. This

suggests that there is a link between concern for the condition of the economy and embracing

traditional values. As for extreme tendency, given that the changes in curvature between the

years was not significant for any of the groups, if there was any increase in openness towards

extreme positions on the social spectrum, this would have been modest at best. In fact, in

most of the cases the curvature is not significantly different from zero, suggesting that a linear

relationship is possible. In these cases, there is also a clear preference for one or the other end

of the social dimension. This suggests that, unlike for the general and economic dimensions,

the social dimension is not one on which there is a strong tendency to prefer the centre over

the extremes.


5.7 Conclusion

This chapter has tested four hypotheses relating to the changing patterns of support for parties

occupying different political positions during the Great Recession and its aftermath. In partic-

ular, levels of support were examined for parties at different positions along several political

dimensions. The first hypothesis concerned party position along the European integration

spectrum. The hypothesis was that pro-integration parties lost support to anti-integration

parties as a result of the recession. It was found that there was such a shift in support but that

it mostly occurred not so much during the recession but in the following years. It was further

found that this change was mediated by an individual’s prospective economic assessment—the

more pessimistic the voter, the more likely that voter was to support anti-integration parties.

This was only true in 2014, voters in previous years having preferred pro-integration parties

irrespective of economic assessment. Taken together, these findings indicate that an increase

in anti-integration sentiment did not arise during the recession but in the post-recession period

and that this increase was economically motivated. This suggests that voters started to feel

that the institutions of the EU were at least partly responsible for either the continuing eco-

nomic problems or the unpopular austerity policies introduced in response. This would also

explain the division according to economic assessment, as those voters who were optimistic

about the economy at that point were presumably those who believed that austerity polices

would be effective.

The second and third hypotheses concerned support for parties at different positions along

the left–right political spectrum. The hypotheses were that the Great Recession led to a short-

term increase in support for right-wing positions and to an increase in support for extreme

positions respectively. The results supported the hypothesis of increased extreme support but

not the hypothesis of increased right-wing support. It was found rather that voters became

increasingly indifferent to whether a party was on the left or the right immediately following

the recession, whereas voters beforehand had tended to prefer right-wing parties when they

were optimistic about the economy and left-wing parties when pessimistic. Support for more

extreme parties increased both during the recession and during the years following it. These

findings challenge both the luxury goods model of Durr (1993), in which optimists are sup-

posed to prefer the left and pessimists the right, and the idea that the Great Recession pushed

voters to the right (for example, Lindvall 2013; Bartels 2014). This thesis differentiates itself

from those earlier studies by controlling for the economic voting effect, so it is likely that this

5.7. CONCLUSION 141

discrepancy can be explained by left-leaning and right-leaning voters having different attitudes

towards the economy.

The second and third hypotheses were also tested using a two-dimensional model of the

political space rather than the usual spectrum. The results for the economic dimension were

similar to those for the single left–right spectrum. Before the recession, optimism was a pre-

dictor of right support and pessimism of left support but this difference closed over time. Most

of the change occurred during the recession rather than in the period following it. Unlike the

single dimension, there was still a discernible difference between optimistic and pessimistic

voters by 2014. There was also an increase in support for more extreme economic positions

and this mostly took place in the aftermath period. It was expected that the social dimen-

sion would be unaffected by the recession but this proved not to be the case. There was an

increase in support for parties holding traditional or authoritarian values, particularly in the

period between 2009 and 2014, and this increase was strongest among pessimistic voters. Un-

like the economic dimension, voters seemed to be indifferent to the extremeness of a position

on the social spectrum, a linear relationship between position and support being plausible in

most cases.

The final hypothesis was that the other hypothesised effects were strongest among those

voters who believed their national economy was worsening and this was indeed found to be

the case. This is an important finding because this links the other findings with the economy

and is evidence that those results are in fact related to the recession and the timing is not

merely a coincidence.

Taken together, these findings suggest that the changes in party support during the re-

cession itself were quite distinct from those of the following period, during which austerity

measures were widely introduced. By far the most prominent change during the recession

itself was a shift towards the centre on economic issues, regardless of economic perception.

Then, in the period following the recession, support for extreme economic positions increased,

with no strong pattern of preference for whether those positions were on the left or the right.

At the same time, support shifted from parties holding libertarian or post-materialist social

views to those holding traditional or authoritarian ones. Similarly, there was a shift in sup-

port from pro-integration to anti-integration parties. As previously mentioned, all of these

effects were strongest among those voters who were pessimistic about the economy. This is

interesting because it suggests that the most striking political effects of the Great Recession


may actually have been induced more by the political response to the recession than by the

recession itself.

This and the two preceding chapters have examined patterns of party preference during

the Great Recession from several different perspectives. In order to gain a more complete

understanding of the political consequences of the recession, the following chapters explore

the same time period using different dependent variables. The next chapter looks at the change

in patterns of turnout during the crisis and whether that too is mediated by an individual’s

prospective economic assessment. Chapter 7 then examines voters’ attitudes towards the EU

and further European unification, so as to establish whether the voter response to austerity

politics was indeed stronger than the response to the recession itself.

Chapter 6

Economic abstention: turnout intention in the face

of economic pessimism

An understanding of what drives voter turnout is an essential part of electoral behaviour ana-

lysis. Studying vote choice alone risks missing an important part of the electoral decision pro-

cess and therefore introducing a bias into the analysis, since voters usually have the option of

not voting at all and the exercise of this option may depend upon some of the same factors that

also influence the party choice of those who do vote. Despite this possibility, voter turnout has

long been neglected in studies of economic voting (Weschle 2014). It is widely acknowledged

that turnout is in long-term decline in most of the mature democracies (Gray and Caul 2000;

Blais 2000, 34–36), so this omission is becoming increasingly problematic. Furthermore, low

turnout potentially has serious implications for democratic government, since certain social

groups may be less likely to vote than others, and if these groups would vote differently from

those who do vote, then this would introduce biases into the electoral process. In fact, there

is some evidence that these biases exist, particularly in the United States (Leighley and Na-

gler 2014). Recognising the importance of voter turnout in electoral behaviour studies, this

chapter builds upon the economic voting analysis of the previous chapters by testing for the

existence of a hypothesised economic turnout relationship.

This chapter begins by discussing the previous research on the relationship between eco-

nomic conditions and turnout, which has produced conflicting results. There were mixed res-

ults within the earlier literature, with some scholars theorising that harsh economic conditions

would motivate more people to vote, either in order to alleviate the situation (Lipset 1981,

192; Schlozman and Verba 1979, 235–239) or to punish the incumbent government (Kernell

1977), and others proposing the opposite, that such conditions would lead people to become

too preoccupied with day-to-day survival to become politically involved (Wolfinger and Ro-

senstone 1980, 20–22). There was also a third perspective arguing that in fact no relationship

143

144 CHAPTER 6. ECONOMIC ABSTENTION

between economic conditions and voter turnout ought to be expected. The empirical evidence

at that time largely supported either the no effect (Lane 1959, 329–331; Fiorina 1978) or the

withdrawal viewpoints (Brody and Sniderman 1977; Schlozman and Verba 1979, 254; Rosen-

stone 1982, 23–26). More recent studies have tried to explain these contradictory findings,

either by looking at individual rather than purely aggregate data (Arceneaux 2003; Stevens

2007; Passarelli and Tuorto 2014) or by examining multiple countries comparatively (Radcliff

1992; Martins and Veige 2013). Nonetheless, conclusions remain divided. This chapter con-

tributes to this research in two key ways. First, although there have been a number of recent

individual-level studies and cross-national studies of the economic influences on turnout, very

few of them are both individual-level and cross-national. This chapter examines individual-

level behaviour across nations, which sets it apart methodologically from most of the existing

literature. Second, this analysis focuses on the Great Recession, which as previous chapters

have discussed is an excellent opportunity to examine voter behaviour under exceptionally

difficult economic circumstances. If economic conditions do indeed affect voter turnout then

this time is an excellent opportunity to examine these effects, since it represents an extreme

case of adverse economic conditions.

6.1 Theory

The extensive economic voting literature has largely focused solely on vote choice and not on

voter turnout. This is an omission that has been noted as early as Radcliff (1992, 451), who

argued that the same economic conditions that were expected to promote an increased op-

position vote also discourage those same individuals from voting at all. This same observation

has since been picked up by others. Lacy and Burden (1999, 235), for example, make the

point that vote choice models that do not account for the option of not voting are potentially

biased, as they necessarily exclude from consideration all individuals in the random sample

who did not vote—even though the decision not to vote is likely to have been conditioned on

some of the independent variables in the model—and these individuals may have different

opinions from those who do vote. In other words this approach risks introducing a selection

bias. Their empirical evidence supports this criticism (252). Even in more recent years, this

criticism is still being made of the economic voting literature (Stevens 2007; Tillman 2008;

Weschle 2014). Stevens (2007, 167) points out that abstention is in most cases an equally

viable option to supporting one or other of the parties and should therefore be included in

voting behaviour models. He even goes so far as to claim that the ‘the reward–punishment

6.1. THEORY 145

model of economic voting, which informs almost all research in this field, needs to broaden

its focus [because it] misses a large and important part of the story’ (183).

Given the increasing acknowledgement of the importance of abstention as an alternative

to making a vote choice decision, this chapter examines the factors that explain the turnout

decision. Despite the tendency for turnout to be overlooked in economic voting models, there

is still a literature examining the influence of economic conditions on turnout alone. Unfortu-

nately, this literature has long been divided about the direction of this influence. In his study

of the economy’s influence on voter turnout, Rosenstone (1982) identified three competing

schools of thought about the effect of poor economic conditions on voter turnout. The first of

these is that citizens are more likely to vote, which he refers to as ‘mobilisation’ (25–26). The

argument behind this claim is based on the reward–punishment model of economic voting. If

economic adversity encourages a portion of the population to punish the government elect-

orally, then they would need to vote in order to do so, hence turnout ought to increase. The

second, ‘withdrawal’, school of thought is that rather than increasing turnout, poor economic

conditions should have the opposite effect (26). The argument here is that a person whose

livelihood is at risk is going to be heavily focused on meeting their material needs and will thus

have few resources remaining for less immediate concerns such as politics. Lastly, Rosenstone

(1982, 27–28) also identifies a third, ‘no effect’, school of thought, namely that neither the

mobilisation nor the withdrawal theory is accurate and that there is no consistent link between

economic conditions and turnout. The argument for this position is that voters might consider

any economic suffering they experience to be a personal rather than a systemic issue and not

blame the government or alternatively that the modern welfare state has reduced economic

suffering to the point that it is no longer a dominant issue for voters.

More recent studies have tried to explain these disparate sets of findings in different ways.

For example, Radcliff (1992) argues that both mobilisation and withdrawal effects can be ob-

served, the former in developing countries and the latter in developed countries. This claim

is based on an analysis of aggregate national election data from 29 countries and US time-

series election data. He further argues that this effect is mediated by the level of development

of the welfare state and that withdrawal is a characteristic of marginal welfare states, and

mobilisation of both non-welfare and institutional welfare states. Martins and Veige (2013),

on the other hand, argue mobilisation occurs under both extremely positive and extremely

negative conditions, when the economy is presumably a salient issue, and that withdrawal

occurs otherwise. This is based on an examination of turnout at local government elections


in Portugal and Flanders. One limitation of these two studies is that they were both based on

aggregate data and so cannot examine the effects of individual-level differences. Arceneaux

(2003) criticises some previous papers, including Radcliff (1994), for trying to draw infer-

ences about individual behaviour from aggregate data. He argues that this is the source of

evidence supporting the withdrawal hypothesis and, using data from the American National

Election Studies 1990–98, finds that economic adversity increases turnout, particularly for

those individuals who blame the government for the state of the economy. Another recent

individual-level study of turnout and vote choice at Italian elections during the Great Reces-

sion and its aftermath found that voters critical of the government’s handling of the economy

divided their response between non-voting, voting for the mainstream opposition and voting

for a new radical anti-establishment party (Passarelli and Tuorto 2014). In other words, this

paper also offers qualified support for the withdrawal hypothesis. The withdrawal hypothesis

is further supported by Stevens (2007), based on American National Election Studies data

from 1956 to 2000.

It is interesting to note that, of these studies, the individual-level single-country studies

have tended to find some variety of withdrawal effect, whereas the cross-country aggregate

studies have found the direction of the effect to be somewhat dependent upon context, al-

though there is no agreement on precisely how this is mediated by the context. This chapter

aims to unify these different approaches by conducting multilevel analysis of turnout across

a number of EU member states. By using multilevel analysis, it is possible to conduct an

individual-level analysis while still taking into consideration relevant contextual factors. This

chapter also examines, as previous chapters have done, three survey waves, which took place

before, during and after the Great Recession. Comparing these surveys allows for an analysis

of the effect of unusual severity of economic conditions, which Martins and Veige (2013) found

to be an important contextual variable.

Looking beyond economic conditions, the turnout literature more generally identifies sev-

eral factors that are linked to turnout. There are a number of studies looking into the polit-

ical and systemic factors that influence turnout and these are widely acknowledged to out-

weigh the importance of individual variation. For example, Powell (1980) studied turnout in

30 democracies between 1960 and 1978 and found compulsory voting, ease of registration,

proportional representation and strong party–group alignment to be important predictors of

high turnout. In later studies he confirms the importance of compulsory voting, party–group

linkage and ease of registration, although with the exception of the United States, automatic

6.1. THEORY 147

registration is the norm for countries where registration is not compulsory (Powell 1982, 111–

132; 1986). Crewe (1981, 240–250) also discusses the strong effect of compulsory voting on

turnout rates. He discusses other administrative incentives, such as postal voting and auto-

matic registration, but largely dismisses the importance of these apart from in the United

States. He also finds that countries where the party system closely corresponds to social cleav-

ages generally experience a higher turnout (251–253). Jackman (1987) similarly finds com-

pulsory voting to be strongly correlated with high turnout. He also finds nationally competit-

ive districts and unicameralism to have a positive turnout effect. Multipartyism and electoral

disproportionality, on the other hand, were found to have a negative effect. This last is consist-

ent with the earlier findings that proportional representation systems generally have higher

turnout than majoritarian systems. These results were largely confirmed by Jackman and

Miller (1995). In their aggregate study of elections in twenty Western industrialised countries

between 1847 and 1982, Blais and Carty (1990) identify compulsory voting and use of pro-

portional representation as the most important contributors to high turnout, with population

size and female suffrage also having a modest negative effect on turnout.

More recent studies have once again confirmed many of these same findings. Franklin

(1996) compares turnout in 37 democracies, looking at both national-level and individual-

level differences, and finding that the differences between countries eclipse the differences

between individuals. At the country level, compulsory voting was once again found to be an

important predictor of high turnout, along with postal voting and proportionality. Blais (2000,

25–31) finds, in addition to the well-established effects of compulsory voting, multipartyism

and proportional representation, that the closeness of the election, degree of literacy and level

of economic development (but not economic growth) contribute to higher turnout rates. He

also argued that turnout was higher in elections that voters perceived to be close contests

(74). In his review of aggregate-level turnout research, Geys (2006) discusses the importance

of a number of aggregate variables, including population size and concentration, perceived

closeness of the election, campaign expenditure, political fragmentation, previous levels of

turnout and the characteristics of the political system.

Although the largest variation in turnout exists between countries rather than individu-

als, there have also been a number of studies seeking to explain this individual variation.

Wolfinger and Rosenstone (1980, 102–103) found education to be by far the most important

predictor of an individual’s likelihood to vote, with age the second most important. Students

and married individuals were also more likely to vote. Blais (2000, 51–52) used the Compar-


ative Study of Electoral Systems post-election surveys from 1996–97 to analyse some of the

causes of individual variation in turnout. He argued that education and age were the most

important factors, followed by religiosity then wealth and marriage. Lewis-Beck et al. (2008,

82–107), in their individual-level study of US electoral behaviour discuss the importance of

vote preference, political involvement, sense of political efficacy, sense of citizen duty, close-

ness of the election, education and age, mobilisation and registration. Of the socio-economic

variables, age and education consistently feature as the two most important in turnout studies

(Blais 2007, 630–631). Franklin (2004) sought to explain the long-term decline in turnout

among established democracies. This is a particularly interesting study because it uses both

aggregate and individual data to test its hypotheses. Franklin found that the decline can be

explained by a combination of changing demographics and political changes. Besides some

idiosyncratic political changes, such as the abolition of compulsory voting in a couple of coun-

tries, he identifies the extension of the franchise to eighteen year olds in most countries as

one of the key causes (213). He specifically rules out changes in civic virtue or disaffection as

causes (215).

It is clear from this review that the importance of particular variables has been confirmed

again and again. Compulsory voting is undoubtedly an important predictor of high turnout,

as is the use of a proportional rather than a majoritarian electoral system. Small population

sizes and elections that are perceived to be close are also associated with higher levels of

turnout. Some of the earlier studies considered the extension of the franchise to women and

eighteen year olds to be of some significance but, owing to the universality of these practices

within the European Union, these are no longer relevant considerations. Similarly, as none

of the countries studied in this chapter have a policy of compulsory voting that is also not

relevant here.1 As for individual-level variables, the most important variables mentioned in

the literature are education and age. Religiosity, wealth and marital status are also noteworthy.

6.2 Hypotheses

This chapter uses survey data to test several hypotheses drawn from the theory outlined above.

The competing ideas about the nature of the relationship between economic conditions and

turnout lead to the following two hypotheses:

1Belgium, Cyprus and Luxembourg do use compulsory voting but they are not included in the analysis for thischapter.

6.3. MEASURING TURNOUT INTENTION 149

Hypothesis 6.1 Voters are more inclined to vote when they believe that economic conditions will

worsen, that is, there is a mobilisation effect.

and alternatively:

Hypothesis 6.2 Voters are less inclined to vote when they believe that economic conditions will

worsen, that is, there is a withdrawal effect.

These hypotheses take a prospective orientation, that is, it is assumed that voters act upon

their beliefs about the future course of the economy rather than their judgement about its past

performance. This is because, as has been discussed in previous chapters, there is considerably

more variance in reported prospective assessments immediately after the crisis than there is

in retrospective assessments, which were almost uniformly negative at that time.

The third hypothesis is based on the idea that an economic crisis intensifies economically

driven electoral behaviour beyond what would normally be expected, which is one of the

motivating ideas behind this thesis. The importance of unusual severity of economic conditions

has also been advanced as a key factor in predicting turnout by Martins and Veige (2013). This

leads to the hypothesis that:

Hypothesis 6.3 The effect of economic perceptions on turnout was of a greater magnitude or

a different sign during the Great Recession than during a period of relatively normal economic

conditions.

The null hypothesis then is that the sign and magnitude were the same at both times. It is

important to note that this would still allow for a stronger apparent turnout effect during the

Great Recession, simply because voters would be expected to have stronger views about the

condition of the economy at that time. For this hypothesis to be supported, there would have

to be evidence that any observable difference in turnout cannot be explained by a simple linear

model.

6.3 Measuring turnout intention

In order to test these hypotheses, the EES survey data is used once again. As in previous

chapters, the 2004, 2009 and 2014 waves are compared, so as to be able to contrast beha-

viour during the Great Recession and its aftermath with behaviour during relatively normal

economic conditions. Compulsory voting is presumed to dampen any abstention effect consid-

erably, so this analysis excludes countries where compulsory voting is enforced. This means


that Belgium, Cyprus and Luxembourg are excluded, while Greece is still included because

compulsory voting is not enforced there and Switzerland is also included because voting is

only compulsory in one canton (Birch 2008, 36). The dependent variable, turnout intention,

was measured with the question ‘And if there was a general election tomorrow, which party

would you vote for?’2 In the first two waves, this was asked as an open-ended question and

no list of options was read out. Those respondents who named a party are coded as having

turnout intention and those who said they would not vote, would spoil their vote or would vote

blank are coded as having no turnout intention. Unfortunately, in the most recent wave, a list

of parties was read out, and this appears to have affected the response. Whereas 89.6 percent

and 87.8 percent of voters reported intending to vote in 2004 and 2009 respectively, a full

95.8 percent of voters reported intending to vote in 2014, when refusals are excluded. On the

other hand, the proportion of refusals also increased considerably in 2014. Furthermore, it has

been known for some time that survey respondents tend to overreport turnout (for example,

see Traugott and Katosh 1979). It has also been shown that offering face-saving responses

tends to mitigate this tendency (Zeglovits and Kritzinger 2014). Based on these ideas, it was

decided to code refusals to answer this question as as non-intention to vote. Although not

a perfect solution, this produces a similar turnout intention rate in year, with 72.7 percent,

76.4 percent and 65.9 percent of respondents coded as having turnout intention in 2004, 2009

and 2014 respectively.

These estimated turnout rates are much closer to the actual recorded turnout rates than

the implausibly high rates obtained by ignoring refusals. Figure 6.1 shows the average turnout

at national elections in the EU since 1990, with the survey years indicated by dashed vertical

lines. As the figure shows, the average turnout at national elections in the EU since the year

2000 has been between the mid-sixties and the low-seventies. The estimated turnout rates

also lie in this range, although they are still slightly higher than the recorded turnouts in the

corresponding years. This remaining discrepancy might result partly from the fact that the vote

choice question is hypothetical and therefore not a perfect indicator of what voters would do

at an actual election. Moreover, the recorded average turnout for a particular year is of course

affected by which countries actually held general elections in that year, as typical turnout rates

vary from country to country. It is also likely that the sample is not perfectly random.3 The

effects of this are mitigated by including appropriate controls in the model but it means that

2This question was worded identically in all three survey waves, except that ‘was’ became ‘were’ in 2014. Asalways, the local translations may vary.

3Sampling, as well as related questions like interview mode and response rates, are discussed in Chapter 2.

6.3. MEASURING TURNOUT INTENTION 151

Figure 6.1: Average turnout at national elections in EU countries

70.0

72.5

75.0

77.5

1990 1995 2000 2005 2010 2015year

EU

ave

rage

turn

out

Average turnout at national elections in the European Union (EU-27) by year. The three dashedlines indicate the survey years of 2004, 2009 and 2014. Source: Eurostat

there may be problems in estimating the population turnout rate from this data, which is not

the purpose of this chapter anyway. It can also be seen from this figure that turnout has been

in decline over the medium term. This decline has been seen across the developed world and

has attracted much scholarly attention (see for example Lyons and Alexander 2000; Franklin

2004; Blais 2007, 2013). Its relevance to this chapter is simply that this medium-term decline

should be kept in mind when making inferences about any change in the absolute turnout rate

across the period 2004–2014.

As was discussed earlier, the variables that are expected to predict turnout are largely

the same as those predicting vote choice. The substantive independent variables are party

identification and economic assessment. Additionally, whether the voter reported voting at

the previous general election is a new independent variable for this chapter. This inclusion

is based on the evidence that voting tends to be habitual (Green and Shacar 2000; Plutzer

2002; Coppock and Green 2015), so those who have voted before are more likely to vote

again. This variable is measured in the same way as the dependent variable, except that it is

based on a question asking voters who they voted for at the preceding general election in their

country, rather than a hypothetical election held tomorrow. In addition to these, the same


Figure 6.2: Proportion intending to vote by year and economic assessment

2004 2009 2014

0.4

0.5

0.6

0.7

0.8

0.9

0.4

0.5

0.6

0.7

0.8

0.9



prop

ortio

n

Proportion of respondents intending to vote according to the respondent’s prospective (firstrow) and retrospective (second row) economic assessment in each survey year, shown with95% confidence intervals. Both forms of economic assessment are measured on a five-pointscale ranging from −2, indicating a very negative assessment, to +2, indicating a very positiveassessment, with 0 indicating a neutral assessment. Source: EES

control variables are used as in previous chapters. See Chapter 2 for a discussion of how these

variables were measured.

The mobilisation and withdrawal hypotheses (Hypotheses 6.1 and 6.2) predict that voters

are respectively more or less inclined to vote when they believe that economic conditions are

poor. The pattern of responses to the vote intention question suggests that voters were more

likely to abstain from voting during the recession than they had been beforehand and that they

became even more likely to abstain in the years after the recession. The mean proportion of

respondents indicating an intention to vote was 72.7 percent [72.1%,73.3%]4 in 2004, which

fell to 67.7 percent [66.7%, 68.0%] in 2009 and 65.9 percent [65.3%,66.5%] in 2014. These

changes are all statistically significant. This decline in turnout intention over the 2004–2014

4Square brackets indicate 95% confidence intervals.

6.4. AN ECONOMIC MODEL OF TURNOUT 153

decade is consistent with the decline in actual turnout at national elections shown over the

same period in Figure 6.1.

In order to shed light on the questions of mobilisation and withdrawal, it is necessary

to look at the relationship between economic assessment and turnout intention. Figure 6.2

shows the proportion of respondents indicating that they would vote at an election held the

following day disaggregated by their assessment of the economic conditions in their country.

The first row shows the breakdown by prospective economic assessment. An estimate of zero

indicates the belief that the economy will remain more or less the same over the following

year; negative numbers indicate the belief that it will get worse; and positive numbers indicate

the belief that it will improve. A similar pattern is observable in all three years. A positive

assessment of future economic conditions is associated with a greater likelihood of voting.

This pattern is preliminary evidence in support of the withdrawal hypothesis. The second row

shows the proportion intending to vote according to their retrospective economic assessment.

Here negative numbers indicate the belief that the economy has become worse over the past

year and positive numbers the belief that it has improved. In 2004 and 2014 the pattern of

turnout according to retrospective economic assessment closely resembles that according to

prospective assessment. In 2009, however, the proportion intending to vote does not increase

monotonically as the retrospective assessment improves and the confidence interval for the

most positive category is very wide because only a handful of voters claimed that the economy

in 2009 was much better than it had been previously. This mirrors the relationship between

mean party support and both kinds of economic assessment, which was discussed in Chapter 3.

Accordingly, the prospective measure will continue to be used, as it has throughout this thesis.

6.4 An economic model of turnout

A multilevel logistic regression model (Model 6A) has been constructed so as to test whether

there was a mobilisation or withdrawal effect and whether the magnitude of that effect was

altered during the crisis. This has two main advantages. First, it makes it possible to con-

trol for confounding variables that might otherwise explain the differences discussed above.

Second, because this is a comparative study of multiple countries, multilevel analysis helps to

to separate the variance due to individual variation from the variance due to the cross-national

design. This model predicts the probability of an individual intending to vote based primar-

ily on his or her prospective economic assessment, party identification and past behaviour

and past behaviour. Prospective economic assessment is on a five-point scale, with negative


scores indicating a pessimistic assessment, positive scores an optimistic assessment and zero

a neutral assessment. Since each country has its own set of political parties, the various party

identifications have been grouped into government identifiers, those who identify with one of

the parties included in the cabinet, opposition identifiers, who identify with any other party,

and non-identifiers, who did not feel close to any particular party. The reference group is non-

identifiers. Past behaviour is represented by a dummy variable indicating whether or not the

respondent reported having voted at the previous national election. Naturally, the survey year

is also included in the model so that changes over time can be detected. As is usual, 2004 is

the reference year and the other two years are represented by dummy variables. Interaction

terms are also included between the survey year dummies and each of these primary predict-

ors. In addition to these substantive variables, several demographic variables have also been

included in the model. As previously, these control for the sex, age, level of education and

labour force status of the respondent, as well as the population density of his or her area of

residence. Interaction terms have also been included between age and the party identifica-

tion variables, as in previous chapters. As a multilevel model, this model also controls for the

country where the interview occurred by means of a random intercept. Random slopes were

also included for the two party identification dummy variables, in order to account for the fact

that these do represent different parties in the different countries.

This model was used to predicted the probabilities of an individual reporting that they

would vote at an election held the following day under various circumstances. Figure 6.3

shows these predicted turnout probabilities according to the survey year and whether the

individual reported voting at the previous general election.5 These particular predictions are

for an individual holding a neutral prospective economic assessment. The predicted likelihood

of someone voting if they had voted at the previous election was 72 percent [68%, 75%] in

2004, 61 percent [57%,66%] in 2009 and 71 percent [68%, 75%] in 2014. For people who

had not voted at the previous election, these probabilities were much lower, at 29 percent

[25%,32%] in 2004, 20 percent [17%, 23%] in 2009 and 22 percent [20%, 26%] in 2014. All

of these differences are significant, except for the difference between 2004 and 2014 among

people who had previously voted (odds ratio 0.99, CI 0.91–1.07). It thus appears that people

were less inclined to vote during the recession than beforehand or afterwards, although this

inclination did not recover its previous level among people who had not voted previously. As

the figure shows, having voted at the previous election is a very strong predictor of future5Unless otherwise specified, all predictions are for a 40 year old male, who has completed high school but not

university, is employed and lives in a town.


Figure 6.3: Predicted probability of voting by past behaviour

0.2

0.3

0.4

0.5

0.6

0.7

2004 2009 2014year

pred

icte

d tu

rnou

t pro

babi

lity

at last electionvoted

did not vote

Predicted probability that an individual would vote in each survey year. The top line is for anindividual who reported voting at the previous national election and the bottom line is for anindividual who did not. These predictions are all for individuals with a neutral prospectiveeconomic assessment. Source: EES & ParlGov

voting intention. This is unsurprising in light of the evidence that voting is a habit (Green

and Shacar 2000; Plutzer 2002; Coppock and Green 2015). Voters in 2004 were 2.5 times as

likely (odds ratio 6.3, CI 5.8–6.8) to vote if they had voted at the previous election, compared

to 3.1 times as likely (OR 6.4, CI 5.9–6.9) in 2009 and 3.2 times as likely (OR 9.7, CI 8.0–

9.3) in 2014.6 The difference between 2004 and 2009 is not statistically significant (OR 1.01,

CI 0.91–1.13) but appears larger owing to the lower base rate in that year. The difference

between 2009 and 2014, on the other hand, is significant (OR 1.36, CI 1.22–1.51). This

suggests that there was increased divergence between habitual voters and non-habitual voters

in the years following the recession.

The other key independent variables were party identification and prospective economic

assessment. Figure 6.4 shows the relationship between these and the predicted likelihood of an

6In order to keep the interpretation of these results as intuitive as possible, the effect sizes reported here arerelative probabilities. This has the disadvantage that relative probabilities are sensitive to the initial conditions,that is, the same effect size can produce quite different relative probabilities if the base probability changes. Forthis reason, the corresponding odds ratios are also reported. The uncertainty of the estimates is indicated by givingthe 95% confidence interval around the odds ratio. The reason confidence intervals are used for this model, ratherthan the standard errors and p-values used elsewhere in the thesis, is that standard errors are only meaningful onthe log odds scale for logit models, not the odds ratio or relative probability scales.


Figure 6.4: Predicted probability of voting by economic assessment and party identification

2004 2009 2014

0.6

0.7

0.8

0.9

-2 -1 0 1 2 -2 -1 0 1 2 -2 -1 0 1 2prospective economic assessment

pred

icte

d tu

rnou

t pro

babi

lity

party ID government opposition none

Predicted probability that an individual would vote according to that individual’s prospectiveeconomic assessment as well as the survey year. Economic assessment ranges from −2 (verypessimistic) to +2 (very optimistic). The top line in each year represents an individual whoidentifies with one of the parties currently represented in the cabinet. The middle line rep-resents an individual identifying with one of the other parties and the bottom line representsan individual who lacks any party identification. These predictions are for an individual whoreported having voted at the previous national election. Source: EES & ParlGov

individual voting in each survey year. It can be seen that party identification plays a large role

in predicting the likelihood of someone voting. In all years, party identifiers are considerably

more likely to vote than non-party identifiers and those identifying with opposition parties are

slightly more likely to vote than those identifying with government parties. For example, in

2004, government identifiers were 27 percent more likely to vote (OR 4.3, CI 3.5–5.2) than

those identifying with no party and opposition identifiers were a further 3.1 percent as likely

to vote (OR 1.6, CI 1.3–1.9). The difference between non-party identifiers and government

identifiers gradually widened (OR 1.33, CI 1.14–1.55) between 2004 and 2014, while the

gap between government and opposition identifiers narrowed somewhat (OR 0.80, CI 0.66–

0.96) over the same period. The differences between 2009 and the other two years were

not significant, or in one case barely significant. As the figure shows, the likelihood of voting

dropped among all three groups in 2009 before recovering in 2014. It is interesting to note that


the likelihood of voting in 2014 actually exceeded that of voting in 2004 among government

identifiers, with this group 2.0 percent more likely (OR 1.31, CI 1.14–1.51) to vote than before.

The relationship of most substantive interest for this chapter is that between an indi-

vidual’s economic assessment and his or her likelihood of voting. It can be seen from Fig-

ure 6.4 that optimistic voters were more likely to vote than pessimistic voters in all three

years. This can be described as an economic abstention effect. Among non-party identifiers

who had voted at the previous election, highly optimistic voters were 13 percent more likely to

vote (OR 1.55, CI 1.33–1.81) than highly pessimistic voters in 2004, compared to 8.8 percent

more likely (OR 1.24, CI 1.08–1.43) in 2009 and 23 percent more likely (OR 2.06, CI 1.75–

2.42) in 2014. These findings support the withdrawal hypothesis, which states that pessimistic

voters are less inclined to vote, and contradict the mobilisation hypothesis, which asserts the

opposite. This economic abstention effect was weaker (OR 0.80, CI 0.65–2.05) in 2009 than

in 2004, which can be seen from the shallower slopes in the figure. The effect became stronger

again (OR 1.65, CI 1.34–2.05) in 2014, actually exceeding (OR 1.32, CI 1.05–1.66) its 2004

strength. This is reflected in the steeper slopes in the 2014 panel in the figure. These findings

contradict Hypothesis 6.3, which predicts that the Great Recession would either strengthen

the magnitude or reverse the sign of any economic abstention effect. In fact, these findings

suggest that the effect was partly suppressed during the recession. On the other hand, the

effect was strengthened beyond its initial level during the post-recession period. This fits the

pattern, found in previous chapters, of economic voting effects becoming weaker during the

crisis itself, with the most striking effects occurring in its aftermath.

In addition to these effects, there were also some small demographic differences in the

likelihood of voting. Women were 3.7 percent less likely (OR 0.88, CI 0.84–0.92) to vote than

men.7 The model allows age to interact with party identification, so the effect of age var-

ies accordingly. For non-party identifiers a 30 year old was 3.0 percent less likely (OR 1.04,

CI 1.01–1.07) than a 20 year old, compared to 0.4 percent less likely for either a govern-

ment (OR 0.94, CI 0.92–0.97) or opposition (OR 0.92, CI 0.89–0.94) identifier. It is surprising

that older people are apparently less likely to vote than younger people but this is only true

because everything else is held equal. In particular, many older poeple are no longer in the

workforce and older people are also less likely to hold a pessimistic economic assessment

than younger people. Education had a small effect, with those who had not finished high

school 2.7 percent more likely (OR 1.11, CI 1.04–1.18) to vote than those who had, when

7All of these comparisons are for a non-party identifier in 2004 holding a neutral economic assessment.


everything else is held equal. University graduates were not significantly different from high

school graduates (OR 1.03, CI 0.98–1.09). Urban density had no significant effect on turnout.

Employed people were not significantly different from the unemployed (OR 0.99, CI 0.91–

1.07) but those not in the workforce, which includes retired people, were 4.9 percent more

likely (OR 1.20, CI 1.14–1.26) to vote than the employed.

As this is a multilevel model, it is also possible to investigate the degree to which indi-

vidual countries deviate from what has already been discussed. The intercept varies consid-

erably between countries, which is unsurprising given that the actual level of turnout varies

from country to country. According to this model, among those holding no party identifica-

tion, respondents in the United Kingdom, Ireland and Denmark were the most likely to report

intending to vote, whereas voters in Portugal, Malta and Poland were the least likely to do

so. Among those identifying with government parties, however, respondents in Finland, Den-

mark and Slovakia were the most likely to indicate an intention to vote, whereas voters in

Ireland were the least likely, followed by Slovenia and Malta. In fact, the random intercept

is negatively correlated with the random slopes for both government (ρ = −0.70) and op-

position (ρ = −0.89) identification, which means that the countries with the highest turnout

among non-party identifiers also tend to be those with the lowest turnout among party iden-

tifiers. The two party identification random slopes are positively correlated (ρ = +0.55),

indicating that turnout among the government and opposition identifiers of a country tends

to be similar, given that the corresponding fixed effects are also close.

6.5 Conclusion

Previous research has produced inconsistent findings on the nature of the relationship between

economic conditions and voter turnout. The competing hypotheses are that voters are mobil-

ised to vote by poor economic conditions in order to hold the government responsible and

that voters withdraw under poor economic conditions because they are demotivated by the

situation (Rosenstone 1982). A multilevel analysis of survey data collected from 22 countries

in three different years was conducted in order to shed further light on this question. This

approach has the advantage of being less sensitive to country-specific circumstances as well as

the increased statistical power of the larger sample size. This analysis supports the withdrawal

hypothesis and is inconsistent with the mobilisation hypothesis. In all three years, it was found

that voters with a more optimistic prospective economic assessment were more likely to vote,

often considerably more likely.

6.5. CONCLUSION 159

It was also hypothesised that any economic abstention effect would be either stronger

or reversed in direction during a time of deep international crisis than under more typical

economic conditions. Although the other hypotheses have been examined by earlier studies,

there have been comparatively few that have studied voter behaviour under crisis conditions.

It has previously been argued that withdrawal occurs during ordinary conditions but that

mobilisation takes place under extremely positive or negative conditions (Martins and Veige

2013). By comparing the two surveys, one collected during the Great Recession and one well

before it began, this chapter has been able to examine this hypothesis. It was found that the

withdrawal effect still prevailed during the crisis, although it was weaker, as well as after

the crisis, when it was even stronger than it had been previously. This evidence does not

support the hypothesis. Rather, as in previous chapters, it suggests that the normal economic-

driven electoral tendencies of voters were muted during the crisis itself, with the strongest

effects taking place well after the initial crisis. This pattern suggests that voters may have

been responding more to the political reaction to the recession than to its mere occurrence.

Since this pattern has emerged in the findings made in this thesis, focus will now turn to

establishing whether there is other evidence that the political reaction to the Great Recession

has had a greater impact on voters than the crisis itself. A possibility that presents itself is

that voters were primarily reacting against the austerity policies implemented in the wake of

the crisis. If this is indeed the case, then it ought to affect attitudes towards the European

institutions, which in many cases promoted those policies. The next chapter will examine

European voters’ attitudes towards the European Union and the prospect of further integration.

By examining the relationship between these attitudes and voters’ economic assessments, as

well as the evolution in these patterns over time, further light will be shed on this question.

Chapter 7

Attitudes towards European integration and

institutions

On 23 June 2016, the citizens of the United Kingdom voted to leave the European Union.

Despite the government and all three major parties opposing the move, 51.9 percent of the

votes cast in the referendum were in favour of leaving, with a high turnout of 72.2 percent

(BBC News 2016a). This surprising result is just the most visible manifestation of increasing

anti-EU sentiment. It follows the 2014 European Parliament elections, which delivered a re-

cord number of seats to overtly Eurosceptic parties. The three strongly pro-EU groups lost

65 seats at the election, largely to radical right, Eurosceptic groups (Brack and Startin 2015,

242). The Greek results, which saw a surge of support to the far left Syriza and far right

Golden Dawn at the expense of the centrist parties, have been described as a ‘seismic shock’

(Verney 2015). These events followed not just the Great Recession but also a period in which

many European governments introduced harsh austerity policies in attempt to stave off fur-

ther economic troubles. Do these results represent an underlying shift in sentiment away from

the institutions of the EU, and if so, can this shift be linked to the politics of austerity? This

chapter explores these questions.

A pattern has emerged in the findings made in this thesis so far, which is that the most

striking shifts in voter behaviour took place later than expected. It was shown in Chapter 3

that the economic vote was strongest not in 2009, when the first wave of the Great Recession

was at its peak, but in 2014 when this had given way to a period of austerity politics. A

similar pattern was found for voter turnout in Chapter 6. It was also shown in Chapter 5 that

voters were more likely to show support for parties holding far left and far right positions

in this post-recession period than they were either before or during the recession. In all of

these cases, voters holding pessimistic views about the economy showed stronger changes

in behaviour than those holding optimistic views. These trends suggest that the instigator for

161

162 CHAPTER 7. ATTITUDES TOWARDS EUROPEAN INTEGRATION AND INSTITUTIONS

the change in voter behaviour may have been the unpopular austerity policies introduced after

the recession, rather than the mere fact of the recession itself. These policies were strongly

associated with the European Union, partly because the EU did encourage them and partly

because national governments found it convenient to remind voters of the European scope

of many of the economic problems facing their countries in order to diffuse responsibility

for politically troublesome measures (Hobolt and Tilley 2014, 100–119). If this theory is

true, then accordingly there ought to be evidence of a change in attitudes towards European

institutions between 2009 and 2014. This chapter analyses survey data in order to test whether

such a change took place.

The chapter begins by discussing the data that was used to explore these questions. The

European Election Studies surveys used throughout this thesis include a number of questions

addressing attitudes towards different aspects of EU politics and this section explains which

were used for this analysis and the reasons behind that selection. Following that, the chapter

introduces the four hypotheses that will be tested and discusses the theoretical motivation

behind them. Three statistical models have been constructed to test these hypotheses and

these are discussed in the following sections. The first of these concerns attitudes towards

European unification and whether voters believe that this should go further or that it has

already gone too far. The second looks at whether voters believe that membership of the EU

is a good thing or a bad thing for their country. The final model addresses the attribution

of responsibility for the economy and the degree to which voters hold the EU as opposed to

their national governments economically responsible. The evidence offered by these models

supports the theory that the post-recession austerity period produced a greater response in

voters than the recession itself did. The chapter concludes with a summary of these findings

and a discussion of their implications.

7.1 Austerity and the European Union

The European Union has had an active role in introducing austerity measures to four of its

member states: Ireland, Greece, Portugal and Cyprus. This took the form of ‘economic adjust-

ment programmes’—bailouts conditional on specified economic reforms—negotiated by the

European Commission in conjunction with the European Central Bank and the International

Monetary Fund. These three institutions are sometimes referred to collectively as the Troika.1

1The term ‘Troika’ is occasionally even used in official documents, e.g. European Commission (2011b, 4).

7.1. AUSTERITY AND THE EUROPEAN UNION 163

In May 2010, Greece was the first country to negotiate such an agreement, receiving 80 bil-

lion euros in loans from other Eurozone countries and a further 30 billion euros from the IMF.

In return, Greece was required to implement a number of structural reforms, including ex-

penditure cuts amounting to seven percent of its GDP, much of which came from pensions and

government wages, and new tax measures equivalent to four percent of GDP (European Com-

mission 2010). In December 2010, in the wake of its banking crisis, Ireland was the second

country to negotiate a bailout package with the Troika, receiving 85 billion euros of financial

assistance while being required to increase taxation and cut expenditure drastically, among

other things (European Commission 2011a). This pattern of loans granted by the Troika to

member states contingent upon the imposition severe austerity measures was continued in the

following years, with an agreement signed by Portugal in May 2011 (European Commission

2011b), a second agreement signed by Greece in March 2012 (European Commission 2012),

and finally an agreement with Cyprus signed in March 2013 (European Commission 2013).

Although European institutions have only been actively involved in the austerity policies of

these four countries, their influence is broader than that. The rules of the Economic and Mon-

etary Union, which are binding on Eurozone members, include a ‘Stability Growth Pact’ (SGP).

The SGP requires member states to keep government deficits to within three percent of GDP

and also requires that they aim for surpluses in the medium term, with strict sanctions specified

for breaches, although these are not automatic in operation and have proven easily avoidable

(Heipertz and Verdun 2010, 3–7). These rules have effectively institutionalised the monetarist

position, which is favoured by German politicians, but conflicts with competing economic the-

ories preferred by others, such as Keynesianism (Scharpf 2013, 109–114). It has been argued

that monetarist policy has not been effective for the Eurozone, owing to the large size of the

currency union (114–125). McBride (2015) argues that the SGP and other EU rules such as

the limit the capacity of member states to adopt any expansionary policies—in other words,

austerity is enforced from the centre. It has also been argued that not only do the economic

adjustment programmes reflect a German belief in austerity as the only option, but they also

put economic pressure on the remaining member states to adopt similar policies, if they wish

to avoid losing competitiveness (Flassbeck and Lapavitsas 2015).

The EU is thus directly responsible for the austerity programmes of four countries and

arguably indirectly responsible for austerity measures taken elsewhere. Even in countries

where austerity has not been adopted, taxpayers have funded a large proportion of the bailout

loans paid out under the four economic adjustment programmes. Moreover, there is evidence


that national politicians have been able to persuade many voters that the European Union,

rather than themselves, is responsible for disliked policies, economic or otherwise (Hobolt

and Tilley 2014). It will also be shown later in this chapter that voters typically regarded the

EU as having more responsibility for the economy in 2014 than they had done in 2009. This

suggests that attitudes towards the EU would be likely to reflect any anti-austerity sentiment

arising in the post-recession period.

7.2 Measuring attitudes towards the European Union

One important question that arises when examining attitudes towards the EU is whether these

attitudes are merely a reflection of the voter’s position on the left–right spectrum, or whether

they represent an additional, orthogonal dimension to Europeans’ political views. There are

competing views in the literature as to the relationship between these two dimensions. The

‘international relations model’ sees the European question as unrelated to domestic left–right

questions, pertaining instead to questions of national interest, so that the major parties in a

particular country might be expected to show some agreement on the question of integration

(Steenbergen and Marks 2004, 5–6). Hix and Lord (1997, 26), on the other hand, argue

that the integration question is a second, orthogonal dimension to the politics of EU mem-

ber states. This still assumes that the two dimensions are independent but it also permits

intranational contestation of the integration question. An analysis of European manifestos

has given this model some empirical support (Hix 1999). A third model is the ‘regulation

model’ (Steenbergen and Marks 2004, 7–8; Tsebelis and Garrett 2000), which argues that the

European dimension is not independent of the left–right dimension. In this model, those on

the left are supposed to be in favour of higher regulation and therefore greater European in-

tegration and those on the right of less regulation and so looser European integration. Finally,

Hooghe and Marks (1999) argue that the two dimensions cannot be collapsed together but

are not independent either. Political positions along these two dimensions are constrained by

‘the emergence of a cleavage ranging from center-left supranationalists who support regulated

capitalism to rightist nationalists who support neoliberalism’ (76).

Efforts to compare these models empirically have made similar findings, although their in-

terpretations differ in some respects. Gabel and Hix (2004, 111) found that the best perform-

ing model was the traditional unidimensional left–right model in their analysis of European

election manifestos. Gabel and Anderson (2004, 30) agree that European politics is effectively

7.2. MEASURING ATTITUDES TOWARDS THE EUROPEAN UNION 165

unidimensional in their study of citizen attitudes, although they appear to favour a Hooghe-

Marks interpretation of that dimension. According to Hooghe, Marks and Wilson (2004, 139),

European integration is structured by the traditional left–right dimension but the two dimen-

sions are not identical because parties at the extreme left and right are less supportive of the

EU than parties occupying the centre. Others argue that attitudes towards the EU are a reflec-

tion of not so much a person’s political views but their perceptions of their own government.

Harteveld, van der Meer and De Vries (2013, 561) observed that trust in national institutions

was an important predictor of trust in the EU in Eurobarometer survey data from 2009. This

was seen as confirming their hypothesised ‘logic of extrapolation’, which proposed that ‘the

legitimacy of the EU is actually derived indirectly, through the legitimacy of the individual

member states’ (546–547).

Even when a second, European dimension of political space has been recognised, its im-

portance has not been undisputed. Van der Eijk and Franklin (2004, 32) have shown that this

dimension has not had a large influence on European political behaviour, although this work

took place well before the Great Recession. They argue that this is not because this dimension

is unimportant to European voters but rather because the parties on offer at EU elections do

not cover the entire political space, which means that voters are not currently free to select

parties according to both their left–right preferences and their preferences towards the EU.

For this reason, they describe this European preference dimension as a ‘sleeping giant’ that

has the potential to awaken as a new motivator for vote choice if either new parties emerge to

fill those empty parts of the political space or unforeseen events lead voters to prioritise their

EU preferences over their left–right preferences (van der Eijk and Franklin 2004). One of the

goals of this chapter is to determine whether either the Great Recession or the austerity period

could be said to have constituted such an event.

For the purposes of this chapter, it is assumed that there is an EU dimension to Europeans’

political views and that this can be measured through survey questions. It is not however

assumed that this is completely independent of the left–right dimension and this is why the

models introduced later in the chapter control for the individual’s left–right position. As will

be seen, the results from those models suggest that, while there is a relationship between the

two dimensions, this relationship is not overwhelmingly powerful nor is left–right position the

dominant predictor of attitudes towards the EU.

Precisely how to measure these attitudes is also not uncontested. Boomgaarden et al.

(2011, 242) argue that discussion of attitudes towards the EU requires more precise terms


than ‘EU support’ or ‘Euroscepticism’. To this end, they conducted a survey in the Netherlands

in 2008 asking various different questions about attitudes towards the EU (246). A factor

analysis of their results found five factors, each consisting of five items. These factors were

identified as negative affection, identity, performance, utilitarianism and strengthening (247–

251). Unfortunately, the EES survey data does not include items that would allow each of

these dimensions to be measured in every year but it does include one question corresponding

to the utilitarianism dimension and one corresponding to the strengthening dimension. These

are the questions asking whether membership of the EU is a good thing and whether European

integration has gone too far respectively. Since these two questions appear to be measuring

somewhat different things, they are both analysed in this chapter.

There are several questions in the EES surveys asking about attitudes towards the European

Union. Often there are corresponding questions about the respondent’s national government,

which would allow attitudes towards the EU and towards national governments to be com-

pared directly. Unfortunately, the set of questions asked is different in each wave and some

otherwise promising questions were not asked in each of the years under study. For example,

questions about satisfaction with democracy in both the EU and the respondent’s country

were asked in both 2004 and 2009 but not in 2014. Questions about trust in the national and

European parliaments were missing from the 2009 survey and questions about whether the

national and European parliaments respond to the concerns of citizens were not asked in the

2004 survey.

Despite these difficulties, a set of survey questions was selected that does offer a useful

comparison of attitudes towards the EU across the years of interest. The first of these questions

asked voters to give an opinion as to whether further European unification is desirable. The

precise wording of the question is:

Some say European unification should be pushed further. Others say it already has

gone too far. What is your opinion? Please indicate your views using a scale from 0

to 10, where 0 means unification ‘has already gone too far’ and 10 means ‘it should

be pushed further’. What number on this scale best describes your position?

This wording is almost identical in all three years, except that in 2004 a ten-point scale was

used rather than an eleven-point scale. The use of ten-point scales was a consistent feature

of the 2004 survey wave. In order to make these responses comparable to the responses in

the later waves, the same correction was applied as was used for similar scales in previous

chapters. See Chapter 2 for full details.

7.2. MEASURING ATTITUDES TOWARDS THE EUROPEAN UNION 167

The second question asks voters their opinion of their country’s EU membership. The

wording is almost identical in all three years:

Generally speaking, do you think that [country’s] membership of the European

Union is a good thing, a bad thing, or neither good nor bad?

Although this question is superficially similar to the first question, they are measuring different

things. It is quite consistent to believe, for example, that European integration has gone too far

while still holding that membership of the EU is a good thing. This interpretation is supported

by the fact that the two variables are only weakly correlated (Spearman’s ρ = 0.38). If the

two variables were in fact measuring the same thing, this correlation ought to be considerably

stronger.

The final two questions ask voters to attribute a degree of responsibility for their country’s

economic conditions to their country’s government and to the EU respectively. The 2009 survey

worded the questions as following:

Now I would like to ask you some questions about how much responsibility the

[country’s] government and the European Union have for some of the things going

on in [country]. Of course you may think that neither is responsible.

First, thinking about the economy, how responsible is the [country’s] government

for economic conditions in [country]? Please indicate your views using any num-

ber on a scale from 0 to 10, where 0 means ‘no responsibility’ and 10 means ‘full

responsibility’.

And what about the European Union, how responsible is the EU for economic

conditions in [country]?2

Unfortunately, these questions were only introduced in 2009, so they are not included in the

2004 survey. These questions are, however, so directly relevant to this thesis that they are

worth using despite this omission. This means that it is not possible to gain as complete a pic-

ture of this variable as it is other variables, in that 2004 cannot serve as a baseline comparison

year in this case. On the other hand, this pair of variables does make it possible to examine the

relative responsibility assigned to governments compared to the EU in the key years of 2009

and 2014.2The wording used in the 2014 survey is: ‘Now I would like to ask you some questions about how much

responsibility the different institutions have in the current economic situation in [country]. Please use a scale from0 to 10, where 0 means that you think they have “no responsibility” and 10 means that they “full responsibility”.’The country’s government and the European Union are the first two institutions read out.


7.3 Hypotheses

The focus of this chapter is voter attitudes towards the European Union. The objective is to

determine whether there was an observable shift in attitudes in the post-recession period.

Such a shift could help to explain the results of the previous chapters, in which it was found

that the most notable changes took place in the period following the recession rather than

during the recession itself. Based on these findings, it was theorised that it was chiefly the

austerity policies characterising the post-recession period that voters were reacting against.

If this theory is accurate, then it would be expected that voters reacted against the EU in

particular during this period, as the EU was strongly associated with these austerity policies.

This chapter tests four specific hypotheses derived from these ideas.

The first hypotheses concern support for the institutions of the EU. This chapter has in-

troduced two relevant variables, namely the respondent’s opinion about the desirability of

continued European integration and whether membership of the EU is a good thing for the

respondent’s country. Although it has been argued that these variables are measuring differ-

ent things, it is hypothesised that they have been affected similarly by the events of the Great

Recession and its aftermath. One of the key ideas in previous chapters has been that the eco-

nomic voting concept is generalisable beyond vote choice to related variables, such as turnout.

Accordingly, it is expected that support for EU institutions is likewise influenced by a citizen’s

prospective economic assessment. This leads to the chapter’s first hypothesis:

Hypothesis 7.1 Support for both EU membership and further European integration is greater

among voters who have an optimistic economic assessment.

Previous work using Eurobarometer survey data has already shown that support for both of

these is linked to an individual’s retrospective economic assessment (Gabel and Whitten 1997).

It has also been shown, using the 2009 EES survey data, that a positive retrospective assess-

ment is associated with greater satisfaction with democracy at the EU level (Hobolt 2012, 99).

Although this thesis uses some of the same data, it is still useful to test this hypothesis for two

reasons. Firstly, those studies tested a retrospective version of the hypothesis, whereas this

thesis is focused on the effects of prospective economic assessment.3 The second reason is

that this study includes the data from not only 2009 but also 2004 and 2014, which provides

3The main reason for this is that the respondents in the 2009 EES survey were broadly in agreement that theeconomy had worsened over the past year, whereas there was considerably more variation in the responses to theprospective question. Chapter 1 discusses this issue in greater depth.

7.3. HYPOTHESES 169

an opportunity to examine how this relationship has evolved over the course of the recession

and its aftermath.

The main idea that this chapter seeks to test is that the post-crisis austerity policies are

responsible for a voter backlash against the European Union. If this is indeed the case, then

it would be expected that there would be an observable decline in support for the institutions

of the EU in the period following the crisis. The second hypothesis is thus:

Hypothesis 7.2 Support for both EU membership and further European integration fell between

2009 and 2014.

Some existing studies have looked at related questions. For example, Armingeon and Ceka

(2014, 83) observe that trust in the EU has fallen considerably over the course of the Great

Recession, particularly in Greece. They argue that this can largely be explained by falling sup-

port for national governments (103). In a multilevel analysis of European Social Survey data

from 2002, Kumlin (2009, 416) found a link between support for further European integration

and both national public service dissatisfaction and national social spending. In particular, he

found that both greater public service dissatisfaction and national social spending were associ-

ated with reduced integration support. He also found an interaction between these effects, so

that the effect of public service dissatisfaction was even stronger in countries with high social

spending. Mau (2005) looked at support for EU membership and for social policy-making at

the European level. Most relevantly for this study, he found that support for EU membership

was greater among those with higher socioeconomic status (79). Garry and Tilley (2014) ar-

gue that European citizens are more likely to favour European integration when they perceive

that EU policies are preferable to those of the national government. For example, left-leaning

citizens in countries where the prevailing consensus is right-leaning ought to be more support-

ive of the EU than right-leaning citizens. Their analysis of the 2009 EES survey data supports

this theory. While none of these studies address this hypothesis directly, they do add weight to

the idea that any decline in economic wellbeing ought to be linked to a reduction in support

for the EU.

Furthermore, it would be expected that this decline in support would be concentrated

among those voters holding a pessimistic assessment of the economy. The reason for this is

that optimistic voters presumably believe that the current economic policy is effective, whereas

pessimistic voters would be more critical. Since this relationship has already been hypothes-

ised to exist, it is expected that it would become stronger during the post-crisis period. The

third hypothesis is thus:


Hypothesis 7.3 The effect of economic assessment on support for both EU membership and fur-

ther European integration was strengthened between 2009 and 2014.

The final hypothesis concerns the way citizens allocate responsibility for the economy

between their national governments and the European Union. If voters did hold the EU to

some degree responsible for the austerity policies that were put in place following the reces-

sion, then it would be expected that more responsibility for the economy was attributed to

the EU during this period than beforehand. Another way of measuring the allocation of re-

sponsibility is to look at the difference between how responsible an individual holds the EU

and how responsible that same individual holds the national government for the condition of

the economy. This then gives an indication of where that individual believes the balance of

responsibility between the two institutions lies. For the same reasons as before, it is expected

that this balance would have shifted towards the EU during the post-recession period. The

final hypothesis is therefore:

Hypothesis 7.4 More economic responsibility was attributed to the EU, both in absolute terms

and relative to that attributed to national governments, in 2014 than in 2009.

Hobolt and Tilley (2014, 33–34) looked at the responsibility questions in the 2009 EES survey,

including those for non-economic areas of responsibility and they also conducted an expert

survey asking the same questions. They noted that citizens tended to attribute more respons-

ibility for the economy to the EU than experts did (35). While both citizens and experts show

some awareness of the various competencies of the EU and of national governments, the com-

plexity of EU structures seems to make it difficult for people to make a definitive attribution of

responsibility (44-45). This is concerning because it means that European voters’ perceptions

of responsibility do not accord with the reality of which institutions hold the relevant powers

and this discord is likely to damage long-term trust in those institutions (147). This hypothesis

offers an opportunity to further this analysis with respect to responsibility for the economy by

examining whether voters attributed responsibility differently following the implementation

of austerity policies in so many countries.

7.4 Attitudes towards further European integration

The first three hypotheses all concern attitudes towards continuing European integration as

well as EU membership. This section focuses on integration, with the following section dis-

cussing membership. In order to test these hypotheses, a model was constructed to explain

7.4. ATTITUDES TOWARDS FURTHER EUROPEAN INTEGRATION 171

attitude towards European unification in terms of a number of relevant predictors. As in pre-

vious chapters, a multilevel model has been used so as to take into account the hierarchical

structure of the data. In particular, it is believed that two individuals randomly selected from

within a single country are more likely to have similar attitudes towards the EU than two in-

dividuals randomly selected from the entire EU. In order to take this into account, a two-level

model has been built in which individuals are nested within countries.

Model 7A describes a voter’s attitude towards European integration chiefly in terms of time,

prospective economic assessment and left–right position. As has been the case throughout

this thesis, time is modelled as a categorical variable so as not to assume anything about the

pattern of change in the dependent variable across the years under study. This means that

two dummy variables are included in the model to represent the 2009 and 2014 waves of the

survey respectively, with the 2004 wave being the base case. Prospective economic assessment

is included in the model in order to test the first hypothesis, namely that voters with a more

positive assessment are likely to be more in favour of European unification. The voter’s left–

right position is included as a predictor because it is likely that a voter’s assessment of the

economy is influenced by their own political beliefs. In particular, austerity policies are likely

to be more palatable to voters on the right than those on the left. Controlling for left–right

position makes it possible to examine these effects separately and ensure that any relationship

found between support for integration and prospective economic assessment is not merely

an artefact of the voter’s ideology. Left–right position is modelled quadratically to account

for the possibility of a curvilinear relationship, like those found between voter support for

parties and the left–right position of those parties in Chapter 5. Indeed, it has been argued

that Eurosceptic positions should be more common among the far left and the far right, simply

because policies made at the European level will generally represent centrist preferences (Hix

2007, 136–137). The modelled quadratic relationship can account for this.

This model additionally includes variables controlling for a number of demographic vari-

ables. As in previous chapters, these variables are age, gender, level of education, urban

density and workforce status. This choice of controls has been informed by previous research

showing that many of these factors are related to attitudes towards the EU. It is well-known,

for example, that Euroscepticism is more prevalent among people with lower levels of edu-

cation than among the highly educated (Hakhverdian et al. 2013, 523). Hakhverdian et al.

(2013, 526–529) argue that this effect ought to have become stronger over time as a result of

continuing market integration, the development of organised Eurosceptic political parties and


weakened national sovereignty. Using Eurobarometer data from 1973 to 2010, they show that

there has indeed been an increasing educational gap in Euroscepticism since the signing of the

Maastricht Treaty (531–535). Other political beliefs are also believed to be related. For ex-

ample, C. J. Anderson (1998, 574–575) argues that people are not sufficiently well-informed

about international politics, including EU politics, for direct questions about attitudes towards

European integration to be independent of domestic attitudes. He proposes that models of

public opinion regarding the integration process control for other relevant political factors

(594). Similarly, Tillman (2013) argues that individuals holding authoritarian values are less

likely to support European integration, supporting this argument with data from the 2008

wave of the European Values Survey. Domestic political beliefs are accounted for by the inclu-

sion of the individual’s left–right position in the model.

The model also includes interactions between time and each of the other variables. Without

these interactions the model would assume a static relationship between the dependent and

independent variables, which is unrealistic. The interaction between time and prospective eco-

nomic assessment in particular is necessary in order to test the third hypothesis, which asserts

that this relationship has strengthened between 2009 and 2014. As different subpopulations

are likely to have been affected to different degrees by both the recession and the austerity

policies that followed it, it also makes sense to allow for the possibility that the level of sup-

port for European integration has evolved in different ways among those different groups.

Including interactions between time and each of the control variables achieves this goal.

Finally, as this is a multilevel model, it was necessary to determine which, if any, random

slopes should be included in addition to the random intercept for the country. Random slopes

were included for the two time dummy variables, as it expected that there are country-specific

circumstances that would cause the level of support for unification to evolve in different ways

in different countries. Random slopes were also included for the economic assessment term

and its interaction with the time dummy variables. Random slopes were not included for any

of the controls. The reason for this was that the estimated variance of these random slopes was

very small, which can lead to convergence problems and also means that they do not improve

the model particularly.

This model has been used to predict an individual’s attitude towards further European

integration based on his or her prospective economic assessment.4 These predictions are dis-

played in Figure 7.1, which shows that support for unification progressively decayed across4Unless otherwise specified, all predictions are for a 40-year-old employed male having completed high school

but not university and living in a mid-sized town.


Figure 7.1: Support for European integration by prospective economic assessment

3.5

4.0

4.5

5.0

5.5

6.0


supp

ort f

or in

tegr

atio

n

year

2004

2009

2014

Predicted support for European integration in each survey year according to the individual’sprospective economic assessment, ranging from −2 for a highly pessimistic assessment to +2for a highly optimistic assessment. These predictions are for an individual holding a left–rightposition in the centre of the spectrum. Source: EES

the survey years irrespective of economic assessment. For a voter with a neutral economic as-

sessment, this decay in support amounts to a 0.41 point (SE = 0.14, p < 0.01) drop between

2004 and 2009 with a further 0.70 point (SE = 0.14, p < 0.001) drop by 2014. From this

plot, it is also apparent that a more optimistic economic assessment is associated with a greater

support for further European unification, although the shallower slope of the 2009 line sug-

gests that this effect may have been weaker then than in the other two years. The size of

this effect in a given year can be summarised by the difference in the predicted support for

further integration of an individual holding a strongly optimistic and one holding a strongly

pessimistic economic assessment. This is the greatest change in support that can be attributed

to variation in economic assessment. This amounts to 1.49 points (SE = 1.49, p < 0.001) in

2004, 0.91 points (SE = 0.17, p < 0.001) in 2009 and 1.89 points (SE = 0.24, p < 0.001) in

2014. As the plot suggests, the effect size in 2009 is significantly smaller than in both 2004

(∆= 0.58 points, SE= 0.19, p < 0.01) and 2014 (∆= 0.99 points, SE= 0.32, p < 0.01). The

difference between 2004 and 2014 is not significant (∆= 0.41 points, SE= 0.29, p = 0.16).

Figure 7.2 shows the relationship between a voter’s left–right position and that voter’s pre-

dicted level of support for further European integration. Comparing to Figure 7.1 also shows


Figure 7.2: Support for European integration by left–right position

4.4

4.6

4.8

5.0

5.2

5.4

5.6

5.8

0 2 4 6 8 10left–right position

supp

ort f

or in

tegr

atio

n

year

2004

2009

2014

Predicted support for European integration in each survey year according to the individual’sassessment of their own left–right position, ranging from 0 (left) to 10 (right). These predic-tions are for an individual holding a neutral prospective economic assessment. Source: EES

that the effect of left–right position is very modest compared to that of prospective economic

assessment. Once again, it is clear that the level of support has generally declined across the

survey years. It also appears that support has decayed more among right-wing voters than left-

wing voters. In 2004 and 2009, it appears that left-wing and right-wing voters are both more

supportive of integration than those in the centre, with right-wing voters perhaps slightly more

supportive than left-wing voters. By 2014, this pattern has changed, with left-wing voters the

most supportive of further integration and right-wing voters the least supportive in an approx-

imately linear relationship.

In order to confirm the relationships suggested by this figure, it is helpful to derive numeric

estimates of the quantities of interest as well as estimates of their statistical significance. In this

case, there are two key quantities of interest. The first is the degree, if any, to which right-wing

voters are more likely to support integration than left-wing voters.5 This directional tendency

is positive if voters on the right are more supportive than voters on the left and negative if

the opposite is the case. In 2004, the directional tendency was +0.44 (SE = 0.22, p = 0.05),

5The measurement of both directional tendency and curvature is discussed in more detail in Chapter 5. Insummary, directional tendency is the difference between the definite integrals of the two halves of the curve andcurvature is the second derivative of the curve function, which in the case of a quadratic is a constant.


indicating that support for integration was very slightly higher among right-wing than left-

wing voters. The directional tendency in 2009 was not significant at +0.08 (SE = 0.20, p =

0.71). This means that it was unlikely that there was a difference in support between those

on the left and the right in that year. In fact, the difference between the directional tendency

of the two years is also not significant (∆ = 0.37, SE = 0.30, p = 0.22). In any case, the

difference between left- and right-wing voters is either small or absent in both years. The

directional tendency for 2014 on the other hand is −0.71 (SE = 0.21, p < 0.001), which is

significantly different from the other years (∆ = 0.78, SE = 0.29, p < 0.01, compared to

2009). This is still a modest tendency but it does suggest that by 2014 European unification

had become somewhat more popular on the left than on the right, albeit less popular among

both groups than in previous years.

The second quantity of interest is the curvature of the relationship in each year. This

represents the degree to which voters near the extremes differ from those in the centre.6

A positive curvature indicates that support is greater towards the extremes and a negative

curvature indicates that support is greatest at the centre. If the curvature is zero, then the

curve is in fact a line. The estimated curvature is +0.008 (SE = 0.005, p = 0.11) in 2004,

+0.016 (SE = 0.005, p < 0.001) in 2009 and −0.002 (SE = 0.005, p = 0.76). The only year

in which this is significantly different from zero is 2009, in which there is a slight tendency

for centrist voters to be less supportive of integration than those on the left and the right.

The estimated curvature for 2009 is not significantly different from that of 2004 (∆= 0.008,

SE = 0.007, p = 0.27) but it is significantly different from that of 2014 (∆ = 0.018, SE =

0.007, p < 0.01). Taken together, the directional tendency and curvature findings provide

evidence that there was a change in the relationship between left–right position and support

for European integration between 2009 and 2014. There is only weak evidence of a change

between 2004 and 2009. It must also be emphasised that this is a weak effect in any case.

Many of the control variables included in the model were also found to be related to

support for unification. As the effect sizes are small for these variables, the results will be

summarised briefly. Older people appear to be very slightly less supportive of European in-

tegration than younger people, which each additional decade of age associated with a 0.060

point (SE = 0.015, p < 0.001) decrease in support in 2004. The size of this effect was not

significantly different in the other years, being 0.012 points (SE = 0.021, p = 0.58) stronger6The ‘centre’ in this case is not necessarily the centre of the political spectrum but rather the vertex of the

curve. If, however, the directional tendency is close to zero, which has been shown to be the case, then the vertexwill be close to the point corresponding to the political centre. The location of the vertex can also be visuallyascertained from the plot.


in 2009 and 0.016 points (SE = 0.021, p = 0.44) stronger in 2014 per decade of age than

in 2004. Gender also played a very modest role, with women being 0.21 points (SE = 0.04,

p < 0.001) less supportive of integration than men in 2004, 0.13 points (SE= 0.04, p < 0.01)

less in 2009 and 0.14 points (SE = 0.04, p < 0.01) less in 2014, all other things being equal.

Once again, the differences between the years were not significant.

Education played a somewhat stronger role. There were three levels of education mod-

elled, indicating the highest level of education attained. This is a university degree in the case

of high education, a high school certificate in the case of medium education and neither of the

above in the case of low education. There was a significant difference in support for European

integration among these educational groups in each year, with higher education consistently

associated with greater support. In 2004, an individual with medium education was predicted

to be 0.36 points (SE = 0.06, p < 0.001) more supportive of unification than an individual

with low education. An individual with high education was predicted to be a further 0.37

points (SE = 0.05, p < 0.001) more supportive. These relationships were not significantly

different in the other years, except that the predictive difference between low and medium

levels of education was 0.20 points (SE = 0.09, p = 0.02) smaller in 2009 and 2004. This

particular result is likely to be statistical noise, given that the error is relatively high compared

to the difference and that the other differences are not significant.

Unlike education, the importance of urban density did appear to change over time. In

2004, voters living in urban areas were the most supportive of European integration. An urban

voter was predicted to be 0.21 points (SE = 0.06, p < 0.001) more supportive than a voter

living in a town and 0.31 points (SE= 0.05, p < 0.001) more supportive than a rural voter. By

2009, this gap had closed slightly, so that an urban voter was only 0.12 points SE= 0.05, p =

0.02 points more supportive than a rural voter, a 0.19 point (SE= 0.07, p = 0.01) difference.

By 2014, the direction had changed, so that rural voters were now more 0.16 points (SE =

0.06, p < 0.01) more supportive of integration than urban voters, a 0.28 point (SE = 0.08,

p < 0.001) change from 2004.

Lastly, workforce status is a factor that would very much be expected to have some influ-

ence. In particular, it is expected that unemployed respondents would be harsher critics of

European institutions than the employed or those not in the workforce. In fact, the estimated

effect size of workforce status, all other things being equal, is very low and for the most part

not significant. In 2004, unemployed voters are not significantly different from either em-

ployed voters (∆ = 0.087, SE = 0.095, p = 0.36) or those not in the workforce (∆ = 0.076,

7.5. POPULAR SUPPORT FOR EU MEMBERSHIP 177

SE= 0.102, p = 0.46). Employed people are predicted to be 0.16 points (SE= 0.05, p < 0.01)

more supportive of further integration than those not in the workforce. Although these point

estimates do differ slightly in the other years, none of the differences are significant.

As this is a multilevel model, there are also estimates of the variances and covariances of

the random effect terms. The random intercept accounts for 45 percent of the country-level

variance and the random slopes for the time dummy variables account for a further 42 percent.

That is, the bulk of the country-level variance was related to variation in the underlying level

of support for European integration in each country and how that evolved over time. Only

13 percent of that variance was related to variation in the effect of prospective economic

assessment in different countries. There is a moderate negative correlation (−0.50) between

the random intercept and the slope for the time dummy variable indicating the year 2009. This

indicates that countries that had high levels of support for unification in 2004 had typically

lost some support by 2009 and those with low levels had gained some. Similarly, there is

a weak negative correlation (−0.24) between the effective level of support in 2009 and the

difference between 2009 and 2014, indicating a similar pattern between those two years. This

is consistent with a regression to the mean over time. Combined, the country-level random

effects account for 16 percent of the total residual variance in the model. This means that

most of the variation is actually between individuals rather than between countries.

In summary, support for further European integration was strongest in 2004, decaying

between 2004 and 2009 and particularly between 2009 and 2014. Voters holding optimistic

economic assessments were typically more supportive of integration than those holding pess-

imistic assessments, although the size of this effect was depressed in 2009 before recovering

in 2014. A voter’s left–right position had only a small effect on his or her level of support but

there was a slight tendency for right-wing voters to be more pro-integration than left-wing

voters in 2004, a tendency which had reversed by 2014. These results are consistent with this

chapter’s first three hypotheses, which predicted a positive relationship between economic

assessment and support for integration, becoming stronger after 2009, as well as an overall

decline in support for integration after 2009.

7.5 Popular support for EU membership

Another potential indicator of satisfaction or dissatisfaction with the European Union is whether

or not voters believe that EU membership is a good thing for their country. The same set of


hypotheses that applied to attitudes towards European integration also apply to attitudes to-

wards EU membership. That is, it is expected that EU membership was seen less positively

after 2009 and that there is a positive relationship between prospective economic assessment

and attitude towards EU membership, which became stronger after 2009. Model 7B was con-

structed in order to test these hypotheses. This variable only has three levels of measurement,

indicating a positive view, a negative view and a neutral view of EU membership. As a result, it

is preferable to use a model that assumes an ordinal scale rather then the stronger assumption

of an interval scale. To this end, a multilevel ordinal logit model has been used to model the

likelihood of individuals believing that EU membership is a good thing or a bad thing for their

country.

The selection of predictors modelled is the same as for the integration model. This reflects

the fact that both models seek to test the same hypotheses. Once again, the key predictors

are time, prospective economic assessment and left–right position, which is modelled quad-

ratically. Age, gender, education, urban density and workforce status are included as controls.

Time is represented by two dummy variables, one for the year 2009 and one for 2014, with

2004 the base case. These dummy variables are interacted with each of the other variables so

as to allow for variation in the strength of those effects over time. As before, this is a multi-

level model, with individuals nested within countries. As well as the random intercept, this

model includes random slopes for economic assessment, the time dummy variables and their

interactions.

This model was estimated and post-estimation simulation was used to predict the prob-

abilities of an individual evaluating EU membership positively, negatively or neither in each

survey year. Figure 7.3 shows these predicted probabilities for an individual with a neutral

prospective economic assessment and who sits in the centre of the political spectrum. This

figure shows that the majority of voters evaluated the EU positively in all three years, ranging

from 55 percent [50%, 61%]7 in 2014 to 68 percent [61%,74%] in 2009. Negative evalu-

ations were comparatively rare, ranging from 7.6 percent [5.8%,9.9%] in 2009 to 13 percent

[10%,15%] in 2014. As the figure shows, EU membership was seen more positively in 2009,

during the initial crisis, than it was before or afterwards. Voters in 2009 were thirteen percent

more likely (odds ratio 1.39, CI 1.04–1.86) than in 2004 and eight percent less likely (OR 0.59,

CI 0.46–0.76) than in 2014 to see EU membership as good for their country. It appears from

7Square brackets indicate 95% confidence intervals.


Figure 7.3: Evaluation of EU membership over time

0.1

0.2

0.3

0.4

0.5

0.6

0.7

2004 2009 2014year

pred

icte

d pr

obab

ility

evaluation

good

neither

bad

Predicted probabilities of an individual holding a particular evaluation of EU membership ineach year. These possible evaluations are that EU membership is good for that individual’scountry, bad for the country, or neither good nor bad. These predictions are for an individualholding a neutral economic assessment and a left–right position in the centre of the spectrum.Source: EES

the figure as though assessments of EU membership were less positive in 2014 than in 2004

but this difference is not significant (OR 0.92, CI 0.79–1.04).8

Figure 7.4 shows the relationship between an individual’s prospective economic assess-

ment and that individual’s probability of evaluating EU membership positively, assuming a

neutral assessment of the economy. From this point onwards, focus will be limited to the

predicted probability of a ‘good’ response. This is because the relationships between the prob-

abilities of the various responses have been modelled as independent of the predictors,9 so

examining the other responses separately would not affect the inferences that could be drawn.

As the figure shows, there is a strong relationship between prospective economic assessment

and the likelihood of a positive evaluation of EU membership. An individual who was highly

optimistic about the economy was more than twice as likely (OR 5.82, CI 3.76–8.95) to evalu-

ate EU membership positively in 2004 than one who was highly pessimistic about the economy.8As in the previous chapter, the percentage differences reported in the text are relative probabilities, with odds

ratios and confidence intervals given alongside in brackets.9In particular, the likelihood of the ‘bad’ response is given by odds(bad) = k/odds(good) where k =

0.172 [0.168,0.177] was estimated from the data. As the three possible responses are mutually exclusive, thisgives enough information to infer the likelihood of the ‘neither’ response: P(neither) = 1− P(good)− P(bad).


Figure 7.4: Positive evaluation of EU membership by prospective economic assessment

0.3

0.4

0.5

0.6

0.7

0.8


pred

icte

d pr

obab

ility

year

2004

2009

2014

Predicted probability of an individual evaluating EU membership as good for his or her country.Probabilities are shown according to the survey year and the individual’s prospective economicassessment, ranging from −2 for a highly pessimistic assessment to +2 for a highly optimisticassessment. These predictions are for an individual holding a left–right position in the centreof the spectrum. Source: EES

This effect became significantly weaker (OR 0.60, CI 0.43–0.86) in 2009, when optimists were

only one and a half times as likely (OR 3.52, CI 2.67–4.61) as pessimists to see EU membership

positively. By 2014, this effect had strengthened once again (OR 3.09, CI 1.81–5.26), becom-

ing a three-fold difference (OR 10.9, CI 7.51–15.6) between optimists and pessimists. The

difference between the 2004 and 2014 effect sizes is not quite significant (OR 1.87, CI 0.95–

3.66). This figure also shows that most of the divergence over time occurred among pessimistic

voters, with optimistic voters remaining quite consistent.

Figure 7.5 shows how the predicted probability of a positive evaluation of EU membership

varies according to the individual’s reported left–right position. It is clear from this figure

that those on the right typically see EU membership more positively than those on the left.

Another thing that this figure shows is that the size of this effect is greatest in 2009. Com-

paring this figure with Figure 7.4 also shows that left–right position has a smaller impact on

this evaluation than economic assessment does. As this relationship is modelled quadratic-

ally, it will be summarised using the same measures of directional tendency and curvature as

previously, although the underlying units in this case are on the logit scale. The directional


Figure 7.5: Positive evaluation of EU membership by left–right position

0.50

0.55

0.60

0.65

0.70

0 2 4 6 8 10left–right position

pred

icte

d pr

obab

ility

year

2004

2009

2014

Predicted probability of an individual evaluating EU membership as good for his or her coun-try. Probabilities are shown according to the survey year and the individual’s assessment oftheir own left–right position, ranging from 0 (left) to 10 (right). These predictions are for anindividual holding a neutral prospective economic assessment. Source: EES

tendency was +0.72 (SE = 0.16, p < 0.001) in 2004, +1.26 (SE = 0.15, p < 0.001) in 2009

and +0.68 (SE = 0.15, p < 0.001) in 2014. This confirms that the probability of a positive

evaluation was greater among right-wing voters than left-wing voters in all years. The pattern

over time was that this imbalance increased (∆ = 0.54, SE = 0.22, p = 0.01) from 2004 to

2009 before decreasing again (∆ = 0.58, SE = 0.22, p < 0.01) by 2014 to very close to its

original level (∆ = 0.04, SE = 0.22, p = 0.86). As for curvature, there was a small negative

curvature (−0.010, SE = 0.004, p < 0.01) in 2009, indicating a slight tendency for those in

the centre to see EU membership more positively than those at the extremes. This was very

close to zero in both 2004 (+0.004, SE = 0.004, p = 0.25) and 2014 (−0.004, SE = 0.004,

p = 0.32), indicating that the relationship in those years was approximately linear.

Some of the control variables are much stronger predictors of an individual’s assessment

of EU membership than others. The effect of age, for example, on the probability of a positive

assessment is not significantly different from zero (OR 1.00, CI 0.98–1.03) in 2004, nor is the

effect significantly stronger in either of the other years. Men were approximately ten percent

more likely (OR 1.24, CI 1.17–1.32) than women to assess EU membership positively in 2004.


This effect was not significantly different in 2009 but by 2014 it had weakened (OR 0.87,

CI 0.80–0.95) to a four percent increase in probability (OR 1.08, CI 1.02–1.15). In general,

higher levels of education were associated with a more positive assessment of EU membership.

In 2004, voters with a university education were 35 percent more likely (OR 2.16, CI 1.98–

2.37) to see EU membership positively than voters who had not completed high school. This

difference grew (OR 1.27, CI 1.12–1.44) between 2004 and 2009, to a 37 percent increase

in probability (OR 2.75, CI 2.52–3.00).10 There was no significant difference between 2009

and 2014 (OR 0.97, CI 0.86–1.11). Similarly, voters interviewed in areas of higher population

density are more likely to assess EU membership positively. In 2004, voters in cities were

eleven percent more likely (OR 1.29, CI 1.20–1.40) to do so than voters living in rural areas.

This effect was about the same in 2009 but by 2014 had weakened (OR 0.78, CI 0.69–0.87)

such that there was no longer a statistically significant difference between rural areas and

cities (OR 1.01, CI 0.93–1.09). There was also a small workforce status effect. In 2004, voters

who were unemployed were seven percent less likely (OR 0.83, CI 0.73–0.95) and voters

who were neither employed nor unemployed—frequently retirees—were four percent more

likely (OR 1.11, CI 1.03–1.19) to evaluate EU membership positively than those who were

employed. These effects were remarkably consistent over time, with no significant differences

between any pair of years.

As this is a multilevel model, some things can also be said about the variances and co-

variances of the random effects terms. There is once again a moderate negative correla-

tion (−0.41) between the random intercept and the random slope for the 2009 time dummy

variable, which indicates that in those countries where assessments of EU memberships were

highly positive or highly negative in 2004 tended to be have more moderate assessments in

2009. Similarly, there is a strong negative correlation (−0.66) between the 2014 time dummy

variable and the effective 2009 level for each country. As this is consistent with a regression to

the mean effect, it suggests that the measured country-specific levels of support owe much to

chance. The remaining random slopes explain only sixteen percent of the total country-level

variance. Unfortunately, with an ordered logit model such as this, the concept of residual

variance is not well-defined, so it is difficult to talk about the total proportion of the variance

explained by the random effects.

In summary, EU membership is seen by most citizens as good, or at least neutral, for their

respective countries. Support for the idea that EU membership is a good thing peaked in 2009,10The reason that the considerably higher odds ratio only corresponds to a small increase in relative probability

is that the base probability is higher in 2009 than in 2004.

7.6. ATTRIBUTION OF RESPONSIBILITY FOR THE ECONOMY 183

during the initial phase of the Great Recession, before returning to roughly its 2004 levels by

2014. An optimistic economic assessment was a strong predictor of a positive evaluation of EU

membership in all years. Those holding a pessimistic economic assessment were most critical

of the EU in 2014 and least critical in 2009. Left–right position also had a modest influence,

with those on right generally being more positive in their assessment than those on the left,

especially in 2009. These results are consistent with the three hypotheses this model was

constructed to test, which were that support for EU membership fell between 2009 and 2014,

that support for EU membership was positively linked to economic assessment, and that the

strength of that relationship grew between 2009 and 2014.

7.6 Attribution of responsibility for the economy

The final hypothesis concerns the attribution of responsibility for the economy between the

European Union and the appropriate national government. This hypothesis states that the EU

was held more responsible in 2014 than it was in 2009, both in absolute terms and relative

to national governments. A model was constructed in order to test this hypothesis. The de-

pendent variable in Model 7C is the level of responsibility a particular individual attributes

to a particular institution on an eleven-point scale and a dummy predictor variable indicates

whether the institution is the EU or the national government. In other respects this model

is similar to the earlier model measuring support for European integration. Like that model,

this one includes as predictors the individual’s prospective economic assessment, left–right

position, age, gender and level of education. It also includes a time dummy variable, which is

interacted with each of those other variables. Unlike the other model, this one only includes a

single time variable because there is no data available from 2004. The base year in this model

is 2009 and the time dummy variable indicates the year 2014. Furthermore, each other vari-

able, including the interactions, is interacted with the institutional dummy variable. This was

done so as not to assume that any of those variables affect the amount of responsibility ac-

corded to the different institutions in precisely the same way. Finally, like the other models

in this chapter, this is a multilevel model with individuals nested within countries. Random

slopes have been included for, as previously, time and prospective economic assessment and,

additionally, their interactions with the institution dummy variable.

According to this model, a typical centrist voter with a neutral prospective economic as-

sessment would have held the government more responsible than the EU for the economy in


Figure 7.6: Attribution of economic responsibility by prospective economic assessment

2009 2014

5.5

6.0

6.5

7.0

7.5

-2 -1 0 1 2 -2 -1 0 1 2prospective economic assessment

pred

icte

d re

spon

sibi

lity

institution national government European Union

Predicted level of responsibility attributed to the national government and to the EuropeanUnion in 2009 and 2014 according to the individual’s prospective economic assessment, whichranges from −2 for a highly pessimistic assessment to +2 for a highly optimistic assessment.These predictions are for an individual holding a left–right position in the centre of the spec-trum. Source: EES

both years. The predicted responsibility scores that this voter would have assigned the gov-

ernment were 7.15 (SE= 0.13, p < 0.001) in 2009 and 7.41 (SE= 0.12, p < 0.001) in 2014.

The difference between the two years is not significant (∆ = 0.25, SE = 0.17, p = 0.13).

The predicted scores assigned to the EU were 5.58 (SE = 0.10, p < 0.001) in 2009 and

6.37 (SE = 0.13, p < 0.001) in 2014. This means that the responsibility attributed to the EU

increased by 0.79 points (SE = 0.10, p < 0.001) between the two years. In relative terms,

the responsibility attributed to the EU in 2009 was 1.58 points (SE = 0.16, p < 0.001) less

than that attributed to the national government, whereas by 2014 this difference was only

1.04 points (SE= 0.12, p < 0.001; ∆= 0.53, SE= 0.17, p < 0.01). In other words, this class

of voter generally considered national governments to be more responsible for the economy

than the EU but this gap has closed somewhat over time as voters have started to attribute a

greater amount of responsibility to the EU.

7.6. ATTRIBUTION OF RESPONSIBILITY FOR THE ECONOMY 185

Figure 7.7: Attribution of economic responsibility by left–right position

2009 2014

5.5

6.0

6.5

7.0

7.5

0 2 4 6 8 10 0 2 4 6 8 10left–right position

pred

icte

d re

spon

sibi

lity

institution national government European Union

Predicted level of responsibility attributed to the national government and to the EuropeanUnion in 2009 and 2014 according to the individual’s assessment of their own left–right po-sition, ranging from 0 (left) to 10 (right). These predictions are for an individual holding aneutral prospective economic assessment. Source: EES

Figure 7.6 shows how these predicted responsibility scores vary for voters with differing

economic assessments. The first thing that stands out in this figure is that voters in 2014 were

generally more likely to hold the EU responsible for economic conditions than voters in 2004.

It is also clear that in 2009, individuals who were highly optimistic about the economy typically

assigned national governments less responsibility for the economy than those who were highly

pessimistic. This effect accounts for 0.63 points (SE= 0.24, p < 0.01) of difference. It appears

from the plot that there was a similar, albeit weaker, effect for the responsibility attributed to

the EU but this effect was not statistically significant (0.12 points, SE = 0.15, p = 0.43), nor

was there a significant effect in 2014 (0.14 points, SE = 0.25, p = 0.59). This effect had

also weakened (∆ = 0.79, SE = 0.31, p < 0.01) for governments so that it was no longer

significant (−0.16 points, SE= 0.32, p = 0.61) by 2014.

Figure 7.7 shows the relationship between a voter’s left–right position and that individual’s

predicted responsibility scores under this model, assuming in this case a neutral economic


assessment. One thing that stands out in this figure, as in the last, is that all groups of voters

held the European Union more responsible for the economy than they did in 2009. It also

appears that right-wing voters in 2009 were more likely to hold governments accountable for

the condition of the economy and that otherwise voters on both the left and the right are

more inclined to assign higher responsibility scores than voters in the centre. The directional

tendency for government responsibility was +1.03 (SE = 0.17, p < 0.001) in 2009 but had

fallen (∆ = 0.90, SE = 0.25, p < 0.001) to almost zero (+0.13, SE = 0.18, p = 0.47)

in 2014. This confirms the impression that right-wing voters tended to hold governments

more responsible for the economy than left-wing voters but only in 2009. The directional

tendency for EU responsibility was positive but not quite significant at +0.32 (SE = 0.17,

p = 0.06) in 2009 and not significant in 2014 (+0.15, SE = 0.24, p < 0.01). The curvature

was positive in all cases, as the figure suggests. Specifically, for government responsibility, this

was +0.009 (SE = 0.004, p = 0.03) in 2009 and +0.010 (SE = 0.004, p = 0.01) in 2014

and for EU responsibility, +0.018 (SE = 0.004, p < 0.001) in 2009 and +0.020 (SE = 0.004,

p < 0.001) in 2014. These figures confirm that centrist voters were somewhat less inclined to

assign high responsibility scores than voters on the left and right. This may simply be a result

of higher political interest among those voters who identify as being further from the centre.

As the control variables are of less substantive interest, their analysis will be limited to their

effect on the relative responsibility attributed to national governments over the EU. In general,

older people are more likely to attribute economy responsibility to their national governments

than the EU, although the effect is very small, amounting to an extra 0.04 points (SE = 0.02,

p = 0.01) of difference in 2009 and a similar amount in 2014. Gender had a stronger effect,

with men’s relative responsibility scores 0.23 points (SE = 0.05, p < 0.001) greater than

women’s in 2009. By 2014 however, this effect had declined in magnitude (∆ = 0.15, SE =

0.07, p = 0.03) so that it was no longer significant. In 2009, higher levels of education were

associated with more responsibility assigned to the government over the EU, there being a

0.36 point (SE = 0.07, p < 0.001) difference between university educated voters and those

who did not complete high school. By 2014, there was no longer a significant difference

between these groups (∆ = −0.29, SE = 0.11, p < 0.01). Urban density did not have a

significant effect in 2009 but by 2014 a pattern had emerged in which those living in urban

areas attributed more responsibility to the EU relative to the national government than those

in rural areas. This amounted to a 0.18 point (SE = 0.07, p < 0.01) difference between

cities and rural areas. Lastly, there was also a mild workforce status effect in 2009, when

7.7. CONCLUSION 187

employed voters attributed more responsibility to the government relative to the EU than both

unemployed voters (∆= 0.28, SE= 0.11, p < 0.01) and those not in the workforce (∆= 0.14,

SE = 0.06, p = 0.02). These effects were not significant in 2014 but the differences between

the years were also not significant.

The random effects account for 26 percent of the total variance in this model, which is

considerably more than in the European integration model. This suggests that country-level

differences are more important influences on beliefs about who is responsible for the eco-

nomy than on attitudes towards continuing integration. There is a strong negative correla-

tion (−0.77) between the random intercept and the random slope for the institution dummy

variable. This indicates that, in 2009, countries where the government was seen as highly

responsible for the economy tended to see the EU as less responsible and vice versa. There

is also a strong negative correlation (−0.68) between the time random slope and the random

intercept and a similar correlation (−0.73) between the time random slope and the random

slope for the time–institution interaction. These are both consistent with regression to the

mean. The institution and time dummy variables along with their interactions account for 87

percent of the country-level variance.

These results support the fourth hypothesis, which states that the level of responsibility

for the economy attributed to the EU increased between 2009 and 2014 and that addition-

ally the balance between the responsibility assigned to the EU and to national governments

shifted towards the EU over the same time period. The post-estimation predictions from this

model show that these patterns are indeed observable. It was somewhat surprising to see that

prospective economic assessment played only a minor role in the attribution of responsibil-

ity, really only being relevant for the EU and in 2009 and even then on a relatively modest

scale. The reason this is surprising is that this variable has been shown to be an important

predictor in the other models not just in this chapter but throughout this thesis, predicting

party preference and turnout likelihood as well as attitudes towards European unification EU

membership. Based on this model it appears that voters’ attribution of responsibility between

the institutions of government and the EU is, unlike their voting behaviour, relatively robust

to their assessment of the economy.

7.7 Conclusion

A recurring pattern has emerged in the preceding chapters, which is that voters’ initial re-

sponse to the Great Recession was muted, with the strongest effects taking place well after


the initial shock. This pattern suggests that European voters may be responding more to the

political reaction to the crisis—the austerity programmes implemented in its wake—than to

the recession itself. If this were indeed the case, then there ought to be evidence of a change

in voters’ attitudes towards the EU in the years since the recession, since it played a direct role

in introducing austerity into four countries and in indirect role in the remaining countries,

owing to the Stability Growth Pact, as discussed earlier in the chapter. By testing four hypo-

theses, this chapter has shown that there was such a change in attitudes. The first of these

was that voters who have an optimistic view of the economy are more likely to see both EU

membership and further European unification positively. This was supported by the data, with

both the membership and integration models showing a positive relationship with prospect-

ive economic assessment in all years. This is confirmation of the idea that voters’ attitudes

towards the EU are linked to their evaluation of economic conditions. This resembles the eco-

nomic voting effect, except that it is not the only individual political parties that are being held

accountable but the European institutions themselves.

The second hypothesis was that this support for the EU fell between 2009 and 2014 and

the third hypothesis was that the effect of an optimistic economic assessment became stronger

over the same period. Both of these hypotheses were supported by both the membership

and integration models. Support for further integration fell between 2009 and 2014, having

previously fallen less severely between 2004 and 2009. Support for EU membership actually

rose by 2009 before falling again by 2014. The strength of the economic assessment effect in

both models fell from 2004 to 2009 before rising again by 2014. The changes between 2009

and 2014 are very different from those between 2004 and 2009. This implies that the changes

that took place in the later time period were not merely a continuation of those taking place in

the earlier period. This pattern supports the argument that the recession itself cannot entirely

explain recent European voter attitudes and behaviour. On the other hand, the economic link

in 2014 is even stronger than before and any explanation has to take account of this. These

trends are consistent with the idea that the austerity politics of the post-recession period has

had a strong impact on voters.

The final hypothesis was that the level of responsibility for the economy that voters at-

tributed to the EU was greater in 2014 than in 2009. This was measured in two ways—both

in absolute terms and relative to that responsibility attributed to the national government.

The data supported the hypothesis under both measures. Although governments were held

more responsible than the EU in both years, this gap closed between the two years, owing to

7.7. CONCLUSION 189

an increase in the level of responsibility attributed to the EU in 2014. Unfortunately no data

for the relevant questions was available from 2004 so it is not possible to compare with that

year as in the other cases. It is interesting to note that prospective economic assessment had

little if any impact on the attribution of responsibility in 2014. That is, by that year, optimistic

and pessimistic voters were nearly in agreement on the question. Once again, these findings

support the argument that there was a shift in attitudes about the economy that took place

too late to be attributed purely to the recession. This finding also supports Hobolt and Tilley’s

(2014) argument that voters increasingly tend to blame the EU for problems it may not be

directly responsible for.

This chapter’s findings offer support to the argument that the austerity policies implemen-

ted in the immediate aftermath of the Great Recession played a larger role in affecting voters’

political behaviour and attitudes than the mere fact of the recession itself, which is one of the

central arguments of the thesis. The following and final chapter summarises the findings of

the entire thesis and discusses their implications in general and for economic voting theory in

particular.

Chapter 8

Conclusion: revising theories of economic voting

This thesis has been motivated by two key questions. First, what were voters’ electoral inten-

tions in the wake of the Great Recession and how did these differ from their intentions at other

times? Secondly, how much of this response can be attributed to the austerity programmes

implemented in the wake of the recession, rather than the economic hardships themselves?

In summary, it was found that the economic voting response was depressed during the Great

Recession and still had not fully recovered by 2014. Voters’ economic dissatisfaction was also

expressed in other observable changes in attitudes and intentions but the most striking of these

did not take place until after 2009. By this point, the economies in most of the surveyed coun-

tries had recovered from the initial crisis but many of them had also implemented unpopular

austerity measures, suggesting that it was the political reaction to the crisis, rather than the

economic events themselves, that produced the strongest reaction from voters.

These questions were researched by using European Election Studies survey data to con-

struct several models assessing the influence of prospective economic assessment on a voter’s

intended electoral behaviour. By contrasting responses from the 2004, 2009 and 2014 survey

waves, the thesis was able to compare these relationships at a time before the crisis, a time

when the initial recession was at its peak and a time well after the initial recession but when

many countries were still struggling with the economic consequences of the situation. In do-

ing so, it was possible to make an argument about how these relationships evolved over the

course of the crisis. This thesis models the economic vote by using individuals’ prospective

sociotropic assessments of the economy to predict their support for the parties in their coun-

try. This means that a voter’s prospective economic perceptions are expected to influence his

or her level of support for each party—positively for for government parties and negatively

for opposition parties—and it is these levels of support which ultimately determine that indi-

vidual’s vote choice. Party support is thus used as the dependent variable for the empirical

economic voting models discussed in this thesis.

191

192 CHAPTER 8. CONCLUSION

A measurable economic vote was observed in all three years. This means that an optim-

istic economic assessment consistently improved a voter’s support for government parties and

reduced a voter’s support for opposition parties, while a pessimistic assessment had the op-

posite effect. The only clarity of responsibility effect involved the ideological cohesion of the

government, that is, the proportion of the government’s members of parliament who are on

the same side of the political spectrum as the prime minister. The economic vote tended to

be stronger in countries where the government was more ideologically cohesive. Other clarity

effects were tested for but not found, nor was there any evidence that the government’s time

in office affected the strength of the economic vote. The economy was also found to have an

influence on election turnout. This was a withdrawal effect, meaning that those who were

pessimistic about the economy were less likely to vote compared to those holding more op-

timistic views. Finally, support for both EU membership and further European integration was

higher among optimistic voters than pessimistic voters in all three years. In summary, people

who are optimistic about the economy are more likely to vote and more likely to support the

incumbent government as well as the European project, whereas pessimists are less likely to

vote, more likely to support opposition parties and more sceptical of the European project.

Several changes were observed between 2004 and 2009, during which period the Great

Recession began and, in most countries as well as the EU overall, peaked. During this period

of economic turmoil, the economic vote actually became weaker, not stronger. The mediating

effect of ideological cohesion was suppressed during this time, as was the withdrawal effect on

turnout. Support for EU membership and European integration also fell and the link between

prospective economic assessment and support for these also weakened. Voters started to prefer

parties further from the centre than they previously had. In economic terms, pessimistic voters

had tended to prefer left-leaning parties and optimistic voters right-leaning parties but this dif-

ference had become much smaller by 2009. In other words, every key link identified between

economic perceptions and voter attitudes or intentions became weaker during the Great Re-

cession. The only stronger effect that was identified was the increase in support for parties

further from the centre.

The most striking changes were found between 2009 and 2014. This period followed the

Great Recession, although some countries did experience subsequent periods of recession, and

is largely characterised by the European debt crisis. During this time, austerity programmes

were implemented in many EU member states. In some countries, these were imposed by the

193

European Troika1 in return for access to bailout funds. Over this period, the economic vote

started to recover, although not to its pre-crisis levels. The clarity effect of ideological cohesion

was also partly restored. As for turnout, the withdrawal effect became even stronger than it

had been in 2004. On the other hand, support for the European project became much weaker

and the link between economic perceptions and this support also became stronger than it had

been before the crisis. Support increased for parties even further away from the centre than

previously and, on the social dimension, there was also an increase in support for tradition-

alist/authoritarian parties, especially among pessimistic voters. Whereas in previous years all

voters had preferred pro-integration parties, by 2014 there was a shift towards anti-integration

parties, particularly among pessimistic voters once again. Finally, between 2009 and 2014,

voters had become much more likely to regard the EU as responsible for economic issues than

they had previously, although still not to the same degree as their national governments.

These findings suggest that it was the political response to the crisis, in the form of austerity

programmes, rather than the recession itself, that produced the greatest change in voters’

attitudes and intentions. Other explanations were also considered but these do not fit the

findings so well. For example, one possibility is that voters wanted to see how the crisis would

play out before reacting strongly. There are two problems with this explanation, however.

First, this would imply that the twelve months that the EU had been in recession by the time

of the 2009 survey wave were not sufficient to produce a full economic voting response, in

spite of the evidence that a twelve month window is a good choice for economic voting studies

(Hellwig and Marinova 2015). Second, this idea cannot explain the increase in Euroscepticism

or the increasing popularity of the view that the EU is responsible for the economy. Similarly, it

is unlikely that voters were simply responding more forcefully after 2009 because there were

additional waves of recession. This alone could not explain the change in attitudes towards

the EU. Yet another possibility is that the increase in Euroscepticism is a coincidence, that is it

has an unrelated cause. There must be some link between this Euroscepticism and economic

issues, however, since it was predominantly pessimistic voters who became more Eurosceptic

after 2009 and 2014 was the first year in which optimistic and pessimistic voters had different

attitudes towards Eurosceptic parties. It is difficult to give an account of events that would

explain all of these findings and yet allow no role for the politics of austerity.

It must be conceded that some key findings of this thesis were foreshadowed elsewhere.

In the introduction to their edited volume on politics during the Great Recession, Bermeo and1The European Commission, European Central Bank and International Monetary Fund. The role of these

institutions in introducing austerity measures was discussed in Section 7.1.


Bartels (2014) commented that:

Voters did punish incumbents as predicted; but contrary to expectations, the com-

bustible potential of the Great Recession was realized in only a few of the coun-

tries we studied. This observation constitutes the first of our volume’s two major

themes: in most countries, popular reactions to the Great Recession were surprisingly

muted and moderate. (3, emphasis in original)

They went on to state that:

Our analysis of these exceptional cases yields the second theme of our project:

dramatic political reactions to the Great Recession were associated less with the direct

economic repercussions of the crisis than with government initiatives to cope with

those repercussions. Radical reactions were less likely to be triggered directly by

declining growth or escalating unemployment than by the austerity and bailout

programs that policy makers adopted in response to crisis trends. (4, emphasis in

original)

In other words, the authors also found that the economic voting response to the Great Reces-

sion was surprisingly weak and that the strongest reactions appear to have been in response to

the austerity programmes implemented after the recession began. Despite not being the first

to make these observations, this thesis still makes an important contribution to this literature.

Bermeo and Bartels’s comments were based on very different evidence from that here. The

individual studies in their book were based on aggregate data and they did not use the EES

surveys but various other datasets. This thesis thus corroborates their evidence by showing

that similar conclusions can be reached using completely different data and methods.

This thesis further contributes to the economic voting literature in several other ways.

First, it provides a much-needed multinational account of individual-level behaviour during

the Great Recession. Although a number of studies of economic voting behaviour during this

period have emerged, they either study a single country or they use only aggregate data. Both

of these approaches have their place but they are also limited. Single country studies, while

useful for gaining a deep understanding of the situation in those countries, cannot necessarily

be generalised to other countries which might have different peculiar circumstances. For ex-

ample, studies in Sweden (Lindvall, Martinsson and Oscarsson 2013; Martinsson 2013) found

a strengthened economic vote during the Great Recession, while studies in Portugal (Freire

195

and Santana-Pereira 2012) and Turkey (Çarkoglu 2012) found a weakened economic vote.

By studying multiple countries together, this thesis has been able to show that it is the latter

result which predominates, at least in the EU. Similarly, aggregate studies are useful because

aggregate data is often readily available and they offer a means to compare countries but not

all changes in individual attitudes are necessarily manifested at the aggregate level. This thesis

fills this gap with an analysis that is both multinational and individual-level.

Second, this thesis also demonstrates that the novel method for measuring the economic

vote proposed by van der Brug, van der Eijk and Franklin (2007) can be replicated using newer

data. Whereas most economic voting studies use logistic regression models to predict the ef-

fect of changing economic conditions on party choice, they argued that that economic voting

could be more accurately and more reliably measured using party support as a dependent vari-

able. According to them, this has several advantages, such as being more sensitive to shifts in

support that might not necessarily be expressed in a change in vote likelihood, a characteristic

which makes this method particularly appropriate for multiparty systems such as those pre-

valent throughout Europe. This thesis has adopted this method, with some key differences,

notably the use of multilevel modelling rather than regression with robust standard errors. By

showing that this method can be applied to different data yielding similar results, this thesis

has contributed further evidence in support of their argument that party support models can

correct the instability problems that have plagued party choice models.

Third, this thesis has implications for the clarity of responsibility literature. Powell and

Whitten (1993) were trying to solve the same problem with their clarity of responsibility the-

ory as van der Brug, van der Eijk and Franklin (2007) were with their party support model

of economic voting, namely the instability problem of economic voting results. Of the many

clarity components tested in this thesis, only the ideological cohesion of the government was

shown to have any effect on the strength of the economic vote. Since these models are based

on the party support theory of economic voting, this supports van der Brug, van der Eijk and

Franklin’s argument that model misspecification may account for much of the instability of

economic voting results. It also suggests that clarity of responsibility effect may be at least

in part an artefact of this misspecification. Given the large extant literature on clarity of re-

sponsibility, these results are hardly definitive but it suggests an interesting area for further

research. A study could be designed to test whether the existence of an apparent clarity effect

depends on whether the economic vote is measured in terms of vote choice or party support.


The current study has also contributed to the turnout literature, particularly that part of

the literature examining the idea that, much like the economic voting effect, an individual’s

inclination to vote is affected in some way by his or her assessment of the economy. One of

the two competing theories is that, in the face of economic misfortune, people are mobilised

to vote so that they can show their dissatisfaction. An alternative theory is that people are

not mobilised but in fact too concerned with the increasingly difficult task of meeting their

daily needs to focus on more abstract concerns like politics and will as a result withdraw

from political involvement (Rosenstone 1982, 25–26). One argument for this hypothesis is

that mobilisation occurs when the economy is performing especially well or especially poorly,

owing to the increased salience of economic issues, and withdrawal occurs otherwise (Martins

and Veige 2013). The findings of the current research support the withdrawal argument,

although as the withdrawal effect was weakened during the recession, it also offers partial

support to the theory that withdrawal is the norm but that mobilisation also occurs during

extreme economic conditions.

These results also have broader implications for European politics. The finding that voters

were relatively cautious about condemning their governments for an international recession

originating beyond their borders is presumably good news for politicians, as it suggests that at

least some voters will allow them scope to react to events outside of their immediate control.

On the other hand, the strong reaction against the austerity programmes introduced after the

crisis also suggests that there are limits to the policy measures that their citizens are willing to

tolerate, even in response to such a crisis. That voters would object to pension and pay cuts,

tax hikes and the loss of government services is not on the face of it surprising but this idea

also appears to stand in contrast to traditional accounts of the economic vote, which argue

that voters are not concerned with policy instruments, only with outcomes (Fiorina 1981, 8–

9). This is only the case, however, if voters see austerity purely as a policy instrument. It is

very likely that an individual facing a cut to his or her own pension sees this as an undesirable

policy outcome, rather than a mere instrument. This would explain why economic voting and

anti-austerity sentiment can exist side by side.

The change in attitudes towards the EU and European integration should be of concern to

European politicians, as well as the increased support for Eurosceptic parties. These results

show that the consensus in favour of the existing European institutions is under threat. The

recent decision by British voters to leave the EU is a visible indicator of this but these results do

not suggest that this shift in attitudes is confined to the UK. Furthermore, the large increase in

197

the degree to which voters consider the EU to be responsible for the condition of the economy

does not correspond to a new centralisation of economic power at the European level. As

Hobolt and Tilley (2014) have observed, the disconnect between the powers of the European

institutions and what voters hold them responsible for is creating an accountability deficit.

Unless this is addressed, there is no reason to expect support for the status quo to return to its

previous levels.

There are several limitations in the thesis. One obvious limitation is that the findings are

not necessarily generalisable beyond the countries of Europe. These countries are predomin-

antly industrialised parliamentary democracies, so many of the findings are likely to be valid

for similar countries like Canada or New Zealand; the situation in countries using a presidential

system, like the United States and most of South America, or in less industrialised countries,

may be different. It would be interesting to investigate the situation in these other countries

but this is made difficult by the lack of comparable survey data collected at the same time.

It was the convenient timing of the EES survey data that led to the choice to study Europe

in this thesis. Nonetheless, it may be possible to use surveys such as the Comparative Study

of Electoral Systems to perform similar comparative research across a different selection of

countries.

Two limitations arise from the constraints of the surveys used. First, as in any survey-

based research, the sample is not perfectly random, the response rates are not as high as

would be preferred and there are missing responses. Second, the questions are not always

worded identically in each year and sometimes a particular question is not included at all in

a particular year’s survey. Even with a single year, there are sometimes variations among the

specific national surveys. This is another reason why it would be useful to replicate this study

using a different comparative survey project. Although these constraints are to some degree

unavoidable, it would increase the confidence in these findings if they were repeated using

different data.

Finally, an important priority for further research would be to look at voters’ attitudes

towards austerity. As has already been discussed, the results of this thesis are highly suggestive

that it was the austerity measures rather than the Great Recession itself that produced the

greatest change in voters’ attitudes and intentions. However, this was not an a priori hypothesis

but rather one that has emerged from the data, being suggested by the pattern in the findings of

Chapters 3–6. Indirect evidence was used to test this hypothesis in Chapter 7 and the evidence

found supports it but the surveys used in this thesis do not include the questions that would be


needed to test it directly. As a result, it would be valuable to study attitudes towards austerity

directly, including the different individual policy measures that are collectively described as

austerity measures.

Appendix A

Countries and parties

This appendix lists all of the countries and parties involved in the analysis presented in this

thesis. The countries are listed in alphabetical order and under each country heading, the

parties for each year are listed in the order that survey respondents were asked about them in

the relevant European Election Study (EES) survey. For each party, it is noted whether it has

been treated as the prime minister’s party or as another cabinet party at that time. All other

parties have been treated as opposition parties.

199

200 APPENDIX A. COUNTRIES AND PARTIES

Austria

2004

Status Party

Social Democratic Party of AustriaPM Austrian People’s Partycabinet Freedom Party of Austria

The Greens—The Green Alternat-ive

Communist Party of Austria

2009

Status Party

PM Social Democratic Party of Austriacabinet Austrian People’s Party

Freedom Party of AustriaAlliance for the Future of AustriaThe Greens—The Green Alternat-

iveHans-Peter Martin’s ListJunge LiberaleCommunist Party of Austria

2014

Status Party

cabinet Austrian People’s PartyPM Social Democratic Party of Austria

NEOS—The New Austria andLiberal Forum

The Greens—The Green Alternat-ive

Freedom Party of AustriaAlliance for the Future of Austria

Belgium

2004

No party support questions were included in

this year’s survey.

2009

Status Party

PM Christian Democratic and Flemishcabinet Open Flemish Liberals and Demo-

cratsSocialist Party DifferentFlemish InterestGreenNew Flemish AllianceList DedeckerSocial Liberal PartyWorkers’ Party of Belgium

cabinet Humanist Democratic Centrecabinet Reformist Movementcabinet Socialist Party

National FrontEnvironmentalists

2014

Status Party

Workers’ Party of Belgiumcabinet Christian Democratic and Flemishcabinet Socialist Party Differentcabinet Open Flemish Liberals and Demo-

cratsNew Flemish AllianceGreenFlemish Interest

cabinet Humanist Democratic CentrePM Socialist Partycabinet Reformist Movement

EnvironmentalistsPeople’s PartyWorkers’ Party of Belgium (Wallo-

nia)

201

Cyprus

2004

Status Party

cabinet Progressive Party of WorkingPeople

Democratic RallyPM Democratic Partycabinet Movement for Social Democracy

2009

Status Party

PM Progressive Party of WorkingPeople

Democratic Rallycabinet Democratic Partycabinet Movement for Social Democracy

European PartyEcological and Environmental

Movement

2014

Status Party

PM Democratic RallyDemocratic PartyMovement for Social DemocracyProgressive Party of Working PeopleEcological and Environmental

MovementCitizens’ Alliance

Czech Republic

2004

Status Party

PM Czech Social Democratic Partycabinet Christian and Democratic Union—

Czechoslovak People’s PartyCommunist Party of Bohemia and

MoraviaCivic Democratic PartyGreen Party

2009

Status Party

Czech Social Democratic Partycabinet Christian and Democratic Union—

Czechoslovak People’s PartyCommunist Party of Bohemia and

MoraviaPM Civic Democratic Partycabinet Green Party

2014

Status Party

cabinet Christian and Democratic Union—Czechoslovak People’s Party

Tradition Responsibility Prosperity(TOP 09)

PM Czech Social Democratic PartyCivic Democratic PartyCommunist Party of Bohemia and

MoraviaNational Socialists—21st Century

Leftcabinet ANO 2011

Party of Free Citizens


Denmark

2004

Status Party

Social DemocratsDanish Social Liberal Party

cabinet Conservative People’s PartySocialist People’s PartyDanish People’s Party

PM Venstre

2009

Status Party

Social DemocratsDanish Social Liberal Party

cabinet Conservative People’s PartySocialist People’s PartyDanish People’s Party

PM VenstreLiberal AllianceJune MovementPeople’s Movement against the EU

2014

Status Party

PM Social DemocratsVenstreSocialist People’s PartyDanish People’s Party

cabinet Danish Social Liberal PartyLiberal AllianceConservative People’s PartyPeople’s Movement against the EU

Estonia

2004

Status Party

Estonian Centre PartyPM Res Publica Partycabinet Estonian Reform Partycabinet People’s Union of Estonia

Pro Patria UnionSocial Democratic PartyEstonian United People’s PartyEstonian Christian UnionEstonian Social Democratic Labour

Party

2009

Status Party

PM Estonian Reform PartyEstonian Centre Party

cabinet Pro Patria and Res Publica UnionSocial Democratic PartyEstonian GreensPeople’s Union of Estonia

2014

Status Party

Pro Patria and Res Publica Unioncabinet Social Democratic PartyPM Estonian Reform Party

Estonian Centre PartyEstonian Greens

203

Finland

2004

Status Party

cabinet Social Democratic Party of FinlandPM Centre Party

National Coalition PartyLeft AllianceGreen League

cabinet Swedish People’s Party of FinlandChristian DemocratsTrue Finns

2009

Status Party

Social Democratic Party of FinlandPM Centre Partycabinet National Coalition Party

Left Alliancecabinet Green Leaguecabinet Swedish People’s Party of Finland

Christian DemocratsTrue Finns

2014

Status Party

PM National Coalition Partycabinet Christian Democratscabinet Social Democratic Party of Finland

Centre Partycabinet Swedish People’s Party of Finlandcabinet Green League

Left AllianceTrue Finns

France

2004

Status Party

Far left (Workers’ Struggle / NewAnticapitalist Party)

French Communist PartySocialist PartyThe Greens

cabinet Union for French Democracy /other right parties

PM Union for a Popular MovementNational Front / Gathering for

France

2009

Status Party

Far left (Workers’ Struggle / NewAnticapitalist Party)

French Communist PartySocialist PartyThe GreensDemocratic Movement

PM Union for a Popular MovementNational FrontLeft Party

2014

Status Party

Union for a Popular MovementPM Socialist Party

National FrontThe GreensLeft FrontUnion of Democrats and Independ-

ents / Democratic MovementFrance AriseNew Anticapitalist Party


Germany

2004

Status Party

Christian Democratic Union /Christian Social Union

PM Social Democratic Partycabinet Alliance 90 / The Greens

Party of Democratic SocialismFree Democratic PartyThe Republicans

2009

Status Party

PM Christian Democratic Union /Christian Social Union

cabinet Social Democratic PartyAlliance 90 / The GreensThe LeftFree Democratic Party

2014

Status Party

PM Christian Democratic Union /Christian Social Union

cabinet Social Democratic PartyFree Democratic PartyAlliance 90 / The GreensThe LeftAlternative for GermanyPirate Party

Greece

2004

Status Party

PM New DemocracyPanhellenic Socialist MovementCommunist Party of GreeceCoalition of LeftPopular Orthodox RallyDemocratic Social Movement

2009

Status Party

PM New DemocracyPanhellenic Socialist MovementCommunist Party of GreeceCoalition of the Radical LeftPopular Orthodox RallyEcologist Greens

2014

Status Party

PM New DemocracyCoalition of the Radical Left

cabinet Panhellenic Socialist MovementIndependent GreeksGolden DawnDemocratic LeftCommunist Party of GreeceThe River

205

Hungary

2004

Status Party

Fidesz—Hungarian Civic AllianceHungarian Democratic ForumHungarian Justice and Life Party

PM Hungarian Socialist PartyHungarian Workers’ Party

cabinet Alliance of Free Democrats

2009

Status Party

Fidesz—Hungarian Civic AllianceJobbikHungarian Communist Workers’

PartyHungarian Democratic Forum

PM Hungarian Socialist PartyAlliance of Free DemocratsChristian Democratic People’s Party

2014

Status Party

JobbikPolitics Can Be Different

PM Fidesz / Christian DemocraticPeople’s Party

Hungarian Socialist PartyTogether 2014 / Dialogue for Hun-

garyDemocratic Coalition

Ireland

2004

Status Party

PM Fianna FailFine GaelGreen PartyLabour Party

cabinet Progressive DemocratsSinn Fein

2009

Status Party

PM Fianna FailFine Gael

cabinet Green PartyLabour PartySinn FeinLibertas

2014

Status Party

PM Fine Gaelcabinet Labour Party

Fianna FailGreen PartySinn FeinSocialist Party


Italy

2004

Status Party

Communist Refoundation PartyDemocrats of the LeftDemocracy is Freedom—The DaisyParty of Italian CommunistsFederation of the GreensItalian Democratic SocialistsUnion of Democrats for EuropeItaly of Values

PM Forza Italiacabinet National Alliancecabinet Union of the Centrecabinet North League

New Italian Socialist PartyItalian Radicals / Pannella-Bonino

List

2009

Status Party

PM The People of Freedomcabinet North League

Democratic PartyItaly of ValuesUnion of the CentreCommunist Refoundation Party /

Party of Italian CommunistsLeft and FreedomThe Right

2014

Status Party

PM Democratic PartyForza ItaliaNorth LeagueFive Star Movement

cabinet Union of the CentreThe Other Europe

cabinet New Centre-RightBrothers of Italy

Latvia

2004

Status Party

New Era PartyFor Human Rights in a United

Latviacabinet People’s PartyPM Union of Greens and Farmerscabinet Latvia’s First party

For Fatherland and FreedomLatvian Way

2009

Status Party

cabinet People’s Partycabinet Union of Greens and FarmersPM New Era Party

Harmony CentreLatvia’s First Party / Latvian Way

cabinet For Fatherland and FreedomFor Human Rights in a United

Latviacabinet Civic Union

Society for Other Politics

2014

Status Party

PM UnityHarmony

cabinet National Alliancecabinet Union of Greens and Farmerscabinet Reform Party

Latvian Russian Union

207

Lithuania

2004



2009

Status Party

PM Homeland Union—LithuanianChristian Democrats

Social Democratic Party ofLithuania

cabinet National Resurrection PartyOrder and Justice

cabinet Liberal MovementLabour Party

cabinet Liberal and Centre UnionElection Action of Poles in

LithuaniaLithuanian Popular Peasants’

UnionNew Union (Social Liberals)

2014

Status Party

Homeland Union—LithuanianChristian Democrats

PM Social Democratic Party ofLithuania

Liberal Movementcabinet Labour Partycabinet Order and Justicecabinet Election Action of Poles in

LithuaniaLithuanian Popular Peasants’

Union

Luxembourg

2004



2009

Status Party

The Greenscabinet Luxembourg Socialist Workers’

PartyDemocratic Party

PM Christian Social People’s PartyAlternative Democratic Reform

PartyThe LeftCommunist Party of LuxembourgCitizens’ List

2014

Status Party

Christian Social People’s Partycabinet Luxembourg Socialist Workers’

PartyPM Democratic Partycabinet The Greens

The LeftAlternative Democratic Reform

Party


Malta

2004



2009

Status Party

PM Nationalist PartyLabour PartyDemocratic AlternativeNational Action

2014

Status Party

PM Labour PartyNationalist PartyDemocratic Alternative

The Netherlands

2004

Status Party

Labour PartyPM Christian Democratic Appealcabinet People’s Party for Freedom and

Democracycabinet Democrats 66

GreenLeftPim Fortuyn ListChristianUnionReformed Political PartySocialist Party

2009

Status Party

cabinet Labour PartyPM Christian Democratic Appeal

People’s Party for Freedom andDemocracy

Democrats 66GreenLeftParty for the Animals

cabinet ChristianUnionReformed Political PartySocialist PartyParty for FreedomProud of the Netherlands

2014

Status Party

PM People’s Party for Freedom andDemocracy

cabinet Labour PartyParty for FreedomSocialist PartyChristian Democratic AppealDemocrats 66ChristianUnionGreenLeft

209

Poland

2004

Status Party

League of Polish Familiescabinet Polish People’s Party

Law and JusticeCivic PlatformSelf-Defence of the Republic of

Polandcabinet Social Democracy of PolandPM Democratic Left Alliancecabinet Labour United

Freedom Union

2009

Status Party

cabinet Polish People’s PartyLibertas PolandCoalition Agreement for the FutureDemocratic Left Alliance

PM Civic PlatformLaw and Justice

2014

Status Party

PM Civic Platformcabinet Polish People’s Party

Democratic Left AllianceLaw and JusticeYour MovementCongress of the New RightUnited Poland

Portugal

2004

Status Party

Left Bloccabinet Democratic and Social Centre—

People’s PartyDemocratic Unitarian CoalitionNew Democracy PartySocialist Party

PM Social Democratic Party

2009

Status Party

Left BlocDemocratic and Social Centre—

People’s PartyDemocratic Unitarian Coalition

PM Socialist PartySocial Democratic Party

2014

Status Party

PM Social Democratic Partycabinet Democratic and Social Centre—

People’s PartySocialist PartyDemocratic Unitarian CoalitionLeft BlocEarth Party


Slovakia

2004

Status Party

People’s Party—Movement for aDemocratic Slovakia

Direction—Social DemocracyCommunist Party of Slovakia

PM Slovak Democratic and ChristianUnion

cabinet Party of the Hungarian Coalitioncabinet Christian Democratic Movementcabinet Alliance of the New Citizen

Slovak National Party

2009

Status Party

cabinet People’s Party—Movement for aDemocratic Slovakia

PM Direction—Social DemocracySlovak Democratic and Christian

Union—Democratic PartyParty of the Hungarian CoalitionChristian Democratic Movement

cabinet Slovak National PartyCommunist Party of SlovakiaFree Forum

2014

Status Party

Christian Democratic MovementSlovak Democratic and Christian

Union—Democratic PartyParty of the Hungarian Coalition

PM Direction—Social DemocracyNew MajorityFreedom and SolidarityOrdinary PeopleMost-Híd

Slovenia

2004

Status Party

cabinet Democratic Party of Pensioners ofSlovenia

PM Liberal Democracy of SloveniaNew Slovenia—Christian People’s

PartySlovenian People’s PartyYouth Party of SloveniaSlovenian National PartySlovenian Democratic Party

cabinet United List of Social DemocratsSlovenia is Ours

2009

Status Party

cabinet Democratic Party of Pensioners ofSlovenia

cabinet Liberal Democracy of SloveniaSlovenian People’s PartySlovenian National PartySlovenian Democratic Party

PM Social Democratscabinet Zares—New Politics

New Slovenia—Christian People’sParty

Youth Party of Slovenia

2014

Status Party

PM Positive SloveniaSlovenian Democratic Party

cabinet Social Democratscabinet Civic Listcabinet Democratic Party of Pensioners of

SloveniaNew Slovenia—Christian People’s

PartyLiberal Democracy of SloveniaSlovenian People’s Party

211

Spain

2004

Status Party

People’s Party / Navarrese People’sUnion

PM Spanish Socialist Workers’ PartyUnited Left / Initiative for Catalonia

Greens

2009

Status Party

People’s PartyPM Spanish Socialist Workers’ Party

United Left / Initiative for CataloniaGreens

UnionConvergence and UnionRepublican Left of CataloniaBasque National PartyGalician Nationalist BlocCanarian Coalition / Canarian

Nationalist PartyYes to NavarreBasque SolidarityNavarrese People’s Union

2014

Status Party

PM People’s PartySpanish Socialist Workers’ PartyUnited Left / Initiative for Catalonia

GreensUnionCoalition for EuropeRepublican Left of CataloniaCitizens—Party of the CitizenryPodemos

Sweden

2004

Status Party

Left PartyPM Swedish Social Democratic Party

Centre PartyLiberal People’s PartyModerate PartyChristian DemocratsGreen PartyJune List

2009

Status Party

Left PartySwedish Social Democratic Party

cabinet Centre Partycabinet Liberal People’s PartyPM Moderate Partycabinet Christian Democrats

Green PartySwedish DemocratsPirate Party

2014

Status Party

Swedish Social Democratic PartyPM Moderate Party

Green Partycabinet Liberal People’s Partycabinet Centre Party

Swedish Democratscabinet Christian Democrats

Left Party


United Kingdom

2004

Status Party

PM LabourConservativesLiberal DemocratsUK Independence PartyScottish National PartyPlaid Cymru

2009

Status Party

PM LabourConservativesLiberal DemocratsScottish National PartyPlaid CymruUK Independence PartyBritish National PartyGreen Party

2014

Status Party

PM ConservativesLabour

cabinet Liberal DemocratsGreen PartyUK Independence PartyScottish National PartyPlaid Cymru

Appendix B

Coefficient tables

Model 3A. Economic voting for PMs’ parties


Intercept 0.067 (0.095) 0.492Year 2009 0.490 (0.183) 0.014Year 2014 0.159 (0.130) 0.235Left–right distance −0.428 (0.019) < 0.001Left–right distance2 0.023 (0.001) < 0.001Party ID 4.706 (0.129) < 0.001Prospective assessment 0.323 (0.053) < 0.001Female 0.054 (0.021) 0.011Age (decades) −0.007 (0.008) 0.353High education 0.084 (0.025) 0.001Low education 0.028 (0.031) 0.357Urban area −0.018 (0.026) 0.495Rural area 0.065 (0.027) 0.016Unemployed −0.083 (0.043) 0.055Not in workforce 0.052 (0.025) 0.038Distance × party ID 0.280 (0.029) < 0.001Age × party ID 0.179 (0.015) < 0.001Prosp. assess. × year 2009 −0.051 (0.048) 0.310Prosp. assess. × year 2014 −0.038 (0.065) 0.565

Country random effect Var. SD

Intercept 0.184 (0.429)Year 2009 0.773 (0.879)Year 2014 0.359 (0.599)Left–right distance 0.008 (0.091)Party ID 0.375 (0.612)Prospective assessment 0.055 (0.235)Distance × party ID 0.015 (0.123)Prosp. assess. × year 2009 0.036 (0.189)Prosp. assess. × year 2014 0.076 (0.275)

Dependent variable is support for the prime minister’s party. Sample size is 51962 individuals

within 25 countries. Pseudo R2 is 0.541.

213

214 APPENDIX B. COEFFICIENT TABLES

Model 3B. Economic voting for all parties, no incumbency status


Intercept −0.195 (0.052) < 0.001Year 2009 0.137 (0.063) 0.030Year 2014 0.160 (0.064) 0.013Left–right distance −0.345 (0.006) < 0.001Left–right distance2 0.021 (0.000) < 0.001Party ID 5.192 (0.051) < 0.001Prospective assessment −0.034 (0.023) 0.135Female 0.021 (0.011) 0.059Age (decades) −0.003 (0.005) 0.586High education 0.009 (0.015) 0.555Low education 0.013 (0.016) 0.413Urban area 0.017 (0.014) 0.207Rural area 0.020 (0.014) 0.146Unemployed 0.002 (0.018) 0.890Not in workforce 0.000 (0.011) 0.989Distance × party ID 0.212 (0.013) < 0.001Age × party ID 0.131 (0.008) < 0.001Prosp. assess. × year 2009 0.015 (0.030) 0.615Prosp. assess. × year 2014 0.002 (0.031) 0.953

Party random effect Var. SD

Intercept 0.555 (0.745)Left–right distance 0.018 (0.135)Party ID 0.936 (0.968)Prospective assessment 0.065 (0.255)Female 0.034 (0.185)Age (decades) 0.010 (0.100)High education 0.065 (0.255)Low education 0.052 (0.228)Urban area 0.043 (0.208)Rural area 0.044 (0.210)Unemployed 0.035 (0.187)Not in workforce 0.020 (0.142)Distance × party ID 0.037 (0.192)

Dependent variable is individual’s support for party. Sample size is 352050 measurements

within 497 parties. Pseudo R2 is 0.508.

215

Model 3C. Economic voting for all parties, 2-way incumbency status


Intercept −0.209 (0.065) 0.002Year 2009 0.129 (0.062) 0.038Year 2014 0.153 (0.063) 0.015Cabinet party 0.043 (0.094) 0.646Left–right distance −0.345 (0.006) < 0.001Left–right distance2 0.021 (0.000) < 0.001Party ID 5.186 (0.051) < 0.001Prospective assessment −0.036 (0.017) 0.038Female 0.021 (0.011) 0.061Age (decades) −0.003 (0.005) 0.591High education 0.009 (0.015) 0.538Low education 0.014 (0.016) 0.372Urban area 0.017 (0.014) 0.209Rural area 0.020 (0.014) 0.144Unemployed 0.002 (0.018) 0.909Not in workforce 0.000 (0.011) 0.984Distance × party ID 0.213 (0.013) < 0.001Age × party ID 0.130 (0.008) < 0.001Prosp. assess. × year 2009 0.031 (0.023) 0.173Prosp. assess. × year 2014 0.007 (0.024) 0.775Cabinet × year 2009 0.167 (0.125) 0.185Cabinet × year 2014 0.052 (0.127) 0.685Prosp. assess. × cabinet 0.431 (0.050) < 0.001Prosp. assess. × cabinet × year 2009 −0.136 (0.049) 0.006Prosp. assess. × cabinet × year 2014 −0.125 (0.051) 0.014




Intercept 0.037 (0.193)Prosp. assess. × cabinet 0.028 (0.167)

Dependent variable is individual’s support for party. Sample size is 352050 measurements

within 497 parties within 25 countries. Pseudo R2 is 0.508.


Model 3D. Economic voting for all parties, 3-way incumbency status


Intercept −0.207 (0.066) 0.003Year 2009 0.125 (0.062) 0.044Year 2014 0.151 (0.063) 0.016Cabinet party 0.116 (0.116) 0.320Prime minister’s party −0.164 (0.148) 0.269Left–right distance −0.345 (0.006) < 0.001Left–right distance2 0.021 (0.000) < 0.001Party ID 5.187 (0.051) < 0.001Prospective assessment −0.034 (0.017) 0.044Female 0.021 (0.011) 0.062Age (decades) −0.003 (0.005) 0.591High education 0.009 (0.015) 0.535Low education 0.014 (0.016) 0.372Urban area 0.017 (0.014) 0.207Rural area 0.020 (0.014) 0.144Unemployed 0.002 (0.018) 0.911Not in workforce 0.000 (0.011) 0.982Distance × party ID 0.214 (0.013) < 0.001Age × party ID 0.130 (0.008) < 0.001Prosp. assess. × year 2009 0.027 (0.022) 0.223Prosp. assess. × year 2014 0.004 (0.023) 0.872Cabinet × year 2009 −0.036 (0.156) 0.816Cabinet × year 2014 −0.042 (0.158) 0.789PM × year 2009 0.433 (0.200) 0.031PM × year 2014 0.201 (0.202) 0.321Prosp. assess. × cabinet 0.384 (0.054) < 0.001Prosp. assess. × PM 0.108 (0.058) 0.061Prosp. assess. × cabinet × year 2009 −0.161 (0.059) 0.007Prosp. assess. × cabinet × year 2014 −0.157 (0.061) 0.010Prosp. assess. × PM × year 2009 0.044 (0.078) 0.569Prosp. assess. × PM × year 2014 0.065 (0.080) 0.416

217




Intercept 0.038 (0.195)Prosp. assess. × cabinet 0.022 (0.150)Prosp. assess. × PM 0.000 (0.009)

Dependent variable is individual’s support for party. Note that as prime ministers’ parties are

also cabinet parties, the prime minister’s party term should be interpreted relative to other

cabinet parties rather than opposition parties. Sample size is 352050 measurements within

497 parties within 25 countries. Pseudo R2 is 0.508.


Model 3E. Mean party preference


Intercept 3.594 (0.110) < 0.001Year 2009 −0.346 (0.146) 0.029Year 2014 −0.434 (0.121) 0.003Prospective assessment 0.129 (0.025) < 0.001Female 0.096 (0.013) < 0.001Age (decades) −0.089 (0.004) < 0.001High education 0.035 (0.015) 0.021Low education 0.003 (0.018) 0.875Urban area −0.016 (0.016) 0.323Rural area 0.006 (0.016) 0.694Unemployed −0.042 (0.025) 0.095Not in workforce 0.028 (0.015) 0.063Prosp. assess. × year 2009 −0.052 (0.022) 0.027Prosp. assess. × year 2014 0.070 (0.034) 0.050


Intercept 0.262 (0.511)Year 2009 0.482 (0.694)Year 2014 0.316 (0.562)Prospective assessment 0.011 (0.103)Prosp. assess. × year 2009 0.004 (0.066)Prosp. assess. × year 2014 0.020 (0.140)

Dependent variable is the individual’s mean party preference. Sample size is 63286 individuals


219

Model 4A. Clarity of responsibility (PMs’ parties)


Intercept 0.543 (0.244) 0.033Year 2009 0.329 (0.173) 0.071Year 2014 0.039 (0.133) 0.773Left–right distance −0.438 (0.005) < 0.001Left–right distance2 0.023 (0.001) < 0.001Party ID 4.680 (0.118) < 0.001Prospective assessment 0.131 (0.091) 0.161Time in office (PM) 0.024 (0.016) 0.130Government clarity −0.893 (0.290) 0.004Institutional clarity 0.054 (0.261) 0.837Female 0.047 (0.021) 0.027Age (decades) −0.005 (0.008) 0.525High education 0.087 (0.025) 0.001Low education 0.035 (0.031) 0.260Urban area −0.018 (0.026) 0.503Rural area 0.068 (0.027) 0.012Unemployed −0.077 (0.044) 0.079Not in workforce 0.056 (0.025) 0.028Distance × party ID 0.266 (0.015) < 0.001Age × party ID 0.178 (0.016) < 0.001Prosp. assess. × year 2009 −0.083 (0.029) 0.004Prosp. assess. × year 2014 −0.066 (0.031) 0.033Prosp. assess. × time in office 0.005 (0.004) 0.168Prosp. assess. × govt clarity 0.186 (0.089) 0.037Prosp. assess. × inst. clarity 0.144 (0.121) 0.245


Intercept 0.195 (0.442)Year 2009 0.631 (0.795)Year 2014 0.353 (0.594)Party ID 0.313 (0.559)Prospective assessment 0.025 (0.159)




Model 4B. Components of government clarity (PMs’ parties)


Intercept 1.917 (0.414) < 0.001Year 2009 0.311 (0.186) 0.110Year 2014 −0.056 (0.140) 0.696Left–right distance −0.437 (0.005) < 0.001Left–right distance2 0.023 (0.001) < 0.001Party ID 4.674 (0.118) < 0.001Prospective assessment −0.131 (0.137) 0.343Time in office (PM) 0.006 (0.015) 0.690Single-party government 0.213 (0.193) 0.283Absence of cohabitation −0.267 (0.213) 0.237Ideological cohesion −1.580 (0.323) < 0.001Dominance of main party −0.142 (0.336) 0.675Institutional clarity −0.146 (0.257) 0.576Female 0.047 (0.021) 0.028Age (decades) −0.005 (0.008) 0.502High education 0.086 (0.025) 0.001Low education 0.034 (0.031) 0.267Urban area −0.018 (0.026) 0.501Rural area 0.067 (0.027) 0.013Unemployed −0.078 (0.044) 0.073Not in workforce 0.056 (0.025) 0.027Distance × party ID 0.266 (0.015) < 0.001Age × party ID 0.178 (0.016) < 0.001Prosp. assess. × year 2009 −0.071 (0.030) 0.016Prosp. assess. × year 2014 −0.037 (0.032) 0.250Prosp. assess. × time in office 0.007 (0.004) 0.069Prosp. assess. × single party −0.053 (0.061) 0.381Prosp. assess. × no cohabitation −0.026 (0.068) 0.696Prosp. assess. × cohesion 0.388 (0.086) < 0.001Prosp. assess. × dominance 0.035 (0.113) 0.757Prosp. assess. × inst. clarity 0.190 (0.117) 0.120





221

Model 4C. Ideological cohesion (PMs’ parties)


Intercept 1.395 (0.315) < 0.001Year 2009 0.339 (0.175) 0.065Year 2014 −0.061 (0.135) 0.654Left–right distance −0.437 (0.005) < 0.001Left–right distance2 0.023 (0.001) < 0.001Party ID 4.676 (0.118) < 0.001Prospective assessment −0.109 (0.103) 0.295Time in office (PM) 0.008 (0.014) 0.559Ideological cohesion −1.391 (0.297) < 0.001Institutional clarity −0.080 (0.235) 0.738Female 0.047 (0.021) 0.028Age (decades) −0.005 (0.008) 0.498High education 0.087 (0.025) 0.001Low education 0.036 (0.031) 0.245Urban area −0.018 (0.026) 0.504Rural area 0.067 (0.027) 0.013Unemployed −0.077 (0.044) 0.076Not in workforce 0.056 (0.025) 0.027Distance × party ID 0.266 (0.015) < 0.001Age × party ID 0.178 (0.016) < 0.001Prosp. assess. × year 2009 −0.071 (0.028) 0.012Prosp. assess. × year 2014 −0.034 (0.032) 0.284Prosp. assess. × time in office 0.007 (0.004) 0.069Prosp. assess. × cohesion 0.375 (0.082) < 0.001Prosp. assess. × inst. clarity 0.152 (0.114) 0.195






Model 4D. Ideological cohesion over time (PMs’ parties)


Intercept 1.702 (0.637) 0.016Year 2009 0.367 (0.871) 0.676Year 2014 −0.777 (0.764) 0.318Left–right distance −0.437 (0.005) < 0.001Left–right distance2 0.023 (0.001) < 0.001Party ID 4.674 (0.119) < 0.001Prospective assessment −0.447 (0.220) 0.043Time in office (PM) −0.034 (0.031) 0.293Ideological cohesion −1.288 (0.613) 0.053Institutional clarity −0.531 (0.401) 0.201Female 0.046 (0.021) 0.031Age (decades) −0.005 (0.008) 0.485High education 0.087 (0.025) 0.001Low education 0.036 (0.031) 0.244Urban area −0.018 (0.026) 0.488Rural area 0.066 (0.027) 0.015Unemployed −0.079 (0.044) 0.071Not in workforce 0.055 (0.025) 0.030Distance × party ID 0.266 (0.015) < 0.001Age × party ID 0.178 (0.016) < 0.001Prosp. assess. × year 2009 0.502 (0.225) 0.026Prosp. assess. × year 2014 −0.136 (0.228) 0.549Time in office × year 2009 0.041 (0.040) 0.309Time in office × year 2014 0.078 (0.048) 0.112Cohesion × year 2009 −0.685 (0.875) 0.440Cohesion × year 2014 −0.033 (0.716) 0.964Inst. clarity × year 2009 0.704 (0.825) 0.402Inst. clarity × year 2014 0.825 (0.557) 0.154Prosp. assess. × time in office 0.019 (0.010) 0.059Prosp. assess. × cohesion 0.824 (0.205) < 0.001Prosp. assess. × inst. clarity −0.061 (0.154) 0.695Prosp. assess. × time in office × year 2009 −0.016 (0.011) 0.132Prosp. assess. × time in office × year 2014 0.030 (0.015) 0.036Prosp. assess. × cohesion × year 2009 −0.681 (0.238) 0.004Prosp. assess. × cohesion × year 2014 −0.340 (0.220) 0.123Prosp. assess. × inst. clarity × year 2009 0.189 (0.142) 0.183Prosp. assess. × inst. clarity × year 2014 0.568 (0.136) < 0.001





223

Model 4E. Ideological cohesion over time (all parties)


Intercept 1.062 (0.436) 0.016Year 2009 −1.336 (0.474) 0.005Year 2014 −0.751 (0.463) 0.106Cabinet party −0.178 (0.771) 0.817Prime minister’s party 0.098 (0.083) 0.242Left–right distance −0.345 (0.006) < 0.001Left–right distance2 0.021 (0.000) < 0.001Party ID 5.188 (0.051) < 0.001Prospective assessment −0.083 (0.145) 0.569Time in office (cabinet) −0.018 (0.017) 0.293Ideological cohesion −1.077 (0.424) 0.011Institutional clarity −0.398 (0.265) 0.140Female 0.021 (0.011) 0.062Age (decades) −0.003 (0.005) 0.588High education 0.009 (0.015) 0.540Low education 0.014 (0.016) 0.375Urban area 0.017 (0.013) 0.204Rural area 0.021 (0.014) 0.139Unemployed 0.001 (0.018) 0.940Not in workforce 0.000 (0.011) 0.975Distance × party ID 0.214 (0.013) < 0.001Age × party ID 0.130 (0.008) < 0.001Prosp. assess. × year 2009 0.071 (0.162) 0.662Prosp. assess. × year 2014 −0.001 (0.164) 0.996Cabinet × year 2009 0.963 (0.863) 0.265Cabinet × year 2014 1.022 (0.860) 0.235Time in office × year 2009 0.012 (0.020) 0.534Time in office × year 2014 −0.001 (0.020) 0.947Cohesion × year 2009 1.294 (0.500) 0.010Cohesion × year 2014 0.844 (0.459) 0.067Inst. clarity × year 2009 0.430 (0.286) 0.133Inst. clarity × year 2014 0.119 (0.281) 0.672Cohesion × cabinet 0.207 (0.752) 0.783Inst. clarity × cabinet 0.061 (0.419) 0.884Prosp. assess. × cabinet −0.369 (0.302) 0.223Prosp. assess. × PM 0.154 (0.033) < 0.001Prosp. assess. × time in office −0.003 (0.007) 0.623Prosp. assess. × cohesion 0.021 (0.143) 0.886Prosp. assess. × inst. clarity 0.058 (0.078) 0.453Prosp. assess. × cabinet × year 2009 0.430 (0.339) 0.205Prosp. assess. × cabinet × year 2014 0.096 (0.340) 0.778Cohesion × cabinet × year 2009 −1.176 (0.889) 0.186Cohesion × cabinet × year 2014 −0.890 (0.852) 0.297Inst. clarity × cabinet × year 2009 0.357 (0.563) 0.526Inst. clarity × cabinet × year 2014 −0.411 (0.555) 0.459Prosp. assess. × cohesion × cabinet 1.031 (0.296) 0.001Prosp. assess. × inst. clarity × cabinet −0.358 (0.161) 0.027Prosp. assess. × time in office × year 2009 0.000 (0.008) 0.950Prosp. assess. × time in office × year 2014 0.012 (0.008) 0.132Prosp. assess. × cohesion × year 2009 −0.014 (0.167) 0.934Prosp. assess. × cohesion × year 2014 0.036 (0.163) 0.824Prosp. assess. × inst. clarity × year 2009 −0.059 (0.103) 0.568Prosp. assess. × inst. clarity × year 2014 −0.068 (0.107) 0.523Prosp. assess. × cohesion × cabinet × year 2009 −0.924 (0.351) 0.009Prosp. assess. × cohesion × cabinet × year 2014 −0.555 (0.338) 0.102Prosp. assess. × inst. clarity × cabinet × year 2009 0.488 (0.219) 0.026Prosp. assess. × inst. clarity × cabinet × year 2014 0.544 (0.220) 0.014





Intercept 0.034 (0.185)

Dependent variable is individual’s support for party. Sample size is 352050 individuals within


225

Model 5A. Party’s European integration position


Intercept −0.100 (0.050) 0.046Year 2009 0.107 (0.064) 0.094Year 2014 0.101 (0.064) 0.117Cabinet party −0.172 (0.112) 0.125Prime minister’s party 0.040 (0.090) 0.658Party European integration position 0.183 (0.034) < 0.001Left–right distance −0.354 (0.007) < 0.001Left–right distance2 0.022 (0.001) < 0.001Party ID 5.222 (0.056) < 0.001Prospective assessment −0.026 (0.018) 0.155Female 0.026 (0.012) 0.036Age (decades) −0.003 (0.006) 0.614High education 0.021 (0.016) 0.184Low education 0.009 (0.017) 0.581Urban area 0.010 (0.014) 0.475Rural area 0.013 (0.015) 0.376Unemployed −0.004 (0.019) 0.832Not in workforce 0.001 (0.012) 0.910Distance × party ID 0.287 (0.008) < 0.001Age × party ID 0.132 (0.008) < 0.001Props. assess. × year 2009 0.009 (0.025) 0.708Props. assess. × year 2014 −0.014 (0.026) 0.576Cabinet × year 2009 0.244 (0.145) 0.093Cabinet × year 2014 0.266 (0.147) 0.072Integration pos. × year 2009 −0.058 (0.043) 0.174Integration pos. × year 2014 −0.161 (0.041) < 0.001Prosp. assess. × cabinet 0.381 (0.044) < 0.001Prosp. assess. × PM 0.136 (0.036) < 0.001Prosp. assess. × integration pos. 0.017 (0.012) 0.140Prosp. assess. × cabinet × year 2009 −0.191 (0.057) 0.001Prosp. assess. × cabinet × year 2014 −0.217 (0.058) < 0.001Prosp. assess. × integration pos. × year 2009 −0.002 (0.016) 0.918Prosp. assess. × integration pos. × year 2014 0.045 (0.016) 0.006


Intercept 0.391 (0.626)Left–right distance 0.020 (0.141)Party ID 1.066 (1.033)Prospective assessment 0.036 (0.189)Female 0.037 (0.193)Age (decades) 0.011 (0.103)High education 0.065 (0.255)Low education 0.054 (0.233)Urban area 0.043 (0.208)Rural area 0.047 (0.217)Unemployed 0.035 (0.187)Not in workforce 0.020 (0.143)


Party European integration position 0.005 (0.067)




Model 5B. Party’s left–right position


Intercept 0.094 (0.070) 0.180Year 2009 0.106 (0.093) 0.256Year 2014 −0.091 (0.095) 0.337Cabinet party −0.056 (0.114) 0.627Prime minister’s party 0.026 (0.092) 0.774Party left–right position −0.021 (0.022) 0.335Party left–right position2 −0.037 (0.009) < 0.001Left–right distance −0.354 (0.007) < 0.001Left–right distance2 0.022 (0.001) < 0.001Party ID 5.228 (0.056) < 0.001Prospective assessment 0.005 (0.026) 0.847Female 0.026 (0.012) 0.035Age (decades) −0.003 (0.006) 0.608High education 0.021 (0.016) 0.187Low education 0.009 (0.017) 0.590Urban area 0.010 (0.014) 0.471Rural area 0.014 (0.015) 0.371Unemployed −0.004 (0.019) 0.825Not in workforce 0.002 (0.012) 0.899Distance × party ID 0.287 (0.008) < 0.001Age × party ID 0.131 (0.008) < 0.001Props. assess. × year 2009 −0.007 (0.036) 0.837Props. assess. × year 2014 0.008 (0.037) 0.839Cabinet × year 2009 0.192 (0.147) 0.192Cabinet × year 2014 0.200 (0.147) 0.175Left–right pos. × year 2009 −0.012 (0.029) 0.683Left–right pos. × year 2014 0.010 (0.029) 0.739Left–right pos.2 × year 2009 0.003 (0.012) 0.793Left–right pos.2 × year 2014 0.039 (0.012) 0.002Prosp. assess. × cabinet 0.346 (0.044) < 0.001Prosp. assess. × PM 0.142 (0.036) < 0.001Prosp. assess. × left–right position 0.032 (0.008) < 0.001Prosp. assess. × left–right position2 −0.006 (0.004) 0.069Prosp. assess. × cabinet × year 2009 −0.155 (0.057) 0.007Prosp. assess. × cabinet × year 2014 −0.129 (0.058) 0.026Prosp. assess. × left–right pos. × year 2009 −0.031 (0.011) 0.005Prosp. assess. × left–right pos. × year 2014 −0.030 (0.011) 0.007Prosp. assess. × left–right pos.2 × year 2009 0.004 (0.005) 0.400Prosp. assess. × left–right pos.2 × year 2014 −0.003 (0.005) 0.509

227




428 parties. Pseudo R2 is 0.504.


Model 5C. Party’s social and economic positions


Intercept 0.139 (0.090) 0.125Year 2009 0.073 (0.120) 0.543Year 2014 −0.093 (0.124) 0.454Cabinet party −0.031 (0.119) 0.796Prime minister’s party 0.027 (0.094) 0.774Party economic position 0.008 (0.025) 0.744Party economic position2 −0.024 (0.010) 0.020Party social position −0.042 (0.021) 0.050Party social position2 −0.019 (0.009) 0.031Left–right distance −0.354 (0.007) < 0.001Left–right distance2 0.022 (0.001) < 0.001Party ID 5.227 (0.056) < 0.001Prospective assessment −0.067 (0.032) 0.038Female 0.026 (0.012) 0.035Age (decades) −0.003 (0.006) 0.603High education 0.021 (0.016) 0.184Low education 0.009 (0.017) 0.590Urban area 0.010 (0.014) 0.470Rural area 0.014 (0.015) 0.371Unemployed −0.004 (0.019) 0.833Not in workforce 0.001 (0.012) 0.909Distance × party ID 0.287 (0.008) < 0.001Age × party ID 0.132 (0.008) < 0.001Props. assess. × year 2009 0.042 (0.043) 0.331Props. assess. × year 2014 0.055 (0.046) 0.235Cabinet × year 2009 0.193 (0.153) 0.207Cabinet × year 2014 0.176 (0.151) 0.244Economic pos. × year 2009 −0.018 (0.034) 0.597Economic pos. × year 2014 −0.033 (0.033) 0.324Economic pos.2 × year 2009 −0.008 (0.014) 0.573Economic pos.2 × year 2014 0.027 (0.013) 0.046Social pos. × year 2009 −0.007 (0.029) 0.797Social pos. × year 2014 0.048 (0.029) 0.093Social pos.2 × year 2009 0.011 (0.011) 0.325Social pos.2 × year 2014 0.012 (0.012) 0.316Prosp. assess. × cabinet 0.312 (0.043) < 0.001Prosp. assess. × PM 0.161 (0.035) < 0.001Prosp. assess. × economic position 0.060 (0.009) < 0.001Prosp. assess. × economic position2 0.002 (0.004) 0.681Prosp. assess. × social position −0.016 (0.008) 0.032Prosp. assess. × social position2 0.005 (0.003) 0.073Prosp. assess. × cabinet × year 2009 −0.133 (0.055) 0.016Prosp. assess. × cabinet × year 2014 −0.100 (0.056) 0.074Prosp. assess. × economic pos. × year 2009 −0.029 (0.012) 0.019Prosp. assess. × economic pos. × year 2014 −0.033 (0.012) 0.008Prosp. assess. × economic pos.2 × year 2009 0.000 (0.005) 0.998Prosp. assess. × economic pos.2 × year 2014 −0.007 (0.005) 0.185Prosp. assess. × social pos. × year 2009 −0.007 (0.010) 0.533Prosp. assess. × social pos. × year 2014 −0.012 (0.010) 0.244Prosp. assess. × social pos.2 × year 2009 −0.005 (0.004) 0.210Prosp. assess. × social pos.2 × year 2014 −0.007 (0.004) 0.099

229




428 parties. Pseudo R2 is 0.504.


Model 6A. Turnout intention


Intercept 0.314 (0.017) < 0.001Year 2009 −0.094 (0.008) < 0.001Year 2014 −0.063 (0.007) < 0.001Government identifier 0.245 (0.017) < 0.001Opposition identifier 0.287 (0.016) < 0.001Prospective assessment 0.015 (0.003) < 0.001Voted last time 0.358 (0.006) < 0.001Female −0.017 (0.003) < 0.001Age (decades) −0.020 (0.001) < 0.001High education 0.003 (0.003) 0.451Low education 0.014 (0.004) 0.001Urban area 0.003 (0.004) 0.362Rural area −0.002 (0.004) 0.581Unemployed 0.003 (0.006) 0.659Not in workforce 0.028 (0.004) < 0.001Government ID × year 2009 0.071 (0.010) < 0.001Government ID × year 2014 0.046 (0.010) < 0.001Opposition ID × year 2009 0.068 (0.009) < 0.001Opposition ID × year 2014 0.030 (0.009) 0.001Age × government ID 0.010 (0.002) < 0.001Age × opposition ID 0.006 (0.002) 0.002Prosp. assess. × year 2009 −0.007 (0.004) 0.053Prosp. assess. × year 2014 0.010 (0.004) 0.012Voted last time × year 2009 −0.007 (0.009) 0.435Voted last time × year 2014 0.041 (0.009) < 0.001


Intercept 0.006 (0.076)Government identifier 0.006 (0.074)Opposition identifier 0.005 (0.070)

This is a multilevel logistic regression model. The dependent variable is one for individuals

who indicated an intention to vote and zero for individuals who indicated they would not vote

as well as individuals who refused to answer the question. Sample size is 63483 individuals


231

Model 7A. Desirability of continued European unification


Intercept 5.506 (0.173) < 0.001Year 2009 −0.381 (0.126) 0.006Year 2014 −1.082 (0.131) < 0.001Prospective assessment 0.372 (0.056) < 0.001Female −0.206 (0.044) < 0.001Age (decades) −0.060 (0.016) < 0.001High education 0.369 (0.052) < 0.001Low education −0.355 (0.062) < 0.001Urban area 0.206 (0.054) < 0.001Rural area −0.106 (0.053) 0.046Unemployed 0.088 (0.096) 0.362Not in workforce 0.162 (0.053) 0.002Left–right position 0.020 (0.009) 0.025Left–right position2 0.004 (0.003) 0.107Prosp. assess. × year 2009 −0.145 (0.047) 0.007Prosp. assess. × year 2014 0.101 (0.072) 0.179Left–right pos. × year 2009 −0.013 (0.012) 0.281Left–right pos. × year 2014 −0.048 (0.012) < 0.001Left–right pos.2 × year 2009 0.004 (0.004) 0.272Left–right pos.2 × year 2014 −0.005 (0.004) 0.169Age × year 2009 −0.011 (0.021) 0.586Age × year 2014 −0.016 (0.021) 0.437Female × year 2009 0.079 (0.061) 0.191Female × year 2014 0.069 (0.061) 0.256Low education × year 2009 0.199 (0.087) 0.022Low education × year 2014 0.145 (0.090) 0.107High education × year 2009 0.071 (0.073) 0.325High education × year 2014 0.053 (0.071) 0.460Rural area × year 2009 0.002 (0.076) 0.980Rural area × year 2014 0.305 (0.077) < 0.001Urban area × year 2009 −0.188 (0.075) 0.013Urban area × year 2014 −0.168 (0.075) 0.025Unemployed × year 2009 −0.170 (0.131) 0.194Unemployed × year 2014 −0.182 (0.124) 0.144Not in workforce × year 2009 −0.022 (0.072) 0.759Not in workforce × year 2014 −0.137 (0.073) 0.061



Dependent variable is individual’s support for further European integration. Sample size is

54806 individuals within 25 countries. Pseudo R2 is 0.107.


Model 7B. Evaluation of EU membership


Intercept (‘bad’ response) −2.152 (0.150) < 0.001Intercept (‘bad’ or ‘neutral’ response) −0.392 (0.149) 0.009Year 2009 0.440 (0.137) 0.001Year 2014 −0.119 (0.149) 0.423Prospective assessment 0.440 (0.056) < 0.001Female −0.219 (0.031) < 0.001Age (decades) 0.005 (0.011) 0.662High education 0.440 (0.038) < 0.001Low education −0.331 (0.043) < 0.001Urban area 0.150 (0.039) < 0.001Rural area −0.108 (0.037) 0.004Unemployed −0.182 (0.066) 0.006Not in workforce 0.101 (0.038) 0.008Left–right position 0.030 (0.006) < 0.001Left–right position2 0.002 (0.002) 0.251Prosp. assess. × year 2009 −0.126 (0.044) 0.005Prosp. assess. × year 2014 0.157 (0.086) 0.070Left–right pos. × year 2009 0.018 (0.009) 0.045Left–right pos. × year 2014 −0.004 (0.009) 0.671Left–right pos.2 × year 2009 −0.007 (0.003) 0.005Left–right pos.2 × year 2014 −0.004 (0.003) 0.129Age × year 2009 0.022 (0.015) 0.153Age × year 2014 0.005 (0.015) 0.757Female × year 2009 0.018 (0.045) 0.693Female × year 2014 0.139 (0.043) 0.001Low education × year 2009 −0.093 (0.062) 0.133Low education × year 2014 −0.080 (0.062) 0.192High education × year 2009 0.148 (0.055) 0.007High education × year 2014 0.136 (0.052) 0.009Rural area × year 2009 0.108 (0.055) 0.051Rural area × year 2014 0.124 (0.054) 0.023Urban area × year 2009 0.047 (0.056) 0.403Urban area × year 2014 −0.130 (0.054) 0.016Unemployed × year 2009 −0.057 (0.092) 0.531Unemployed × year 2014 −0.084 (0.085) 0.322Not in workforce × year 2009 −0.018 (0.053) 0.741Not in workforce × year 2014 −0.014 (0.052) 0.787

233



This is a multilevel ordered logit model. The dependent variable is the individual’s evalu-

ation of EU membership as good, bad or neutral. Sample size is 56803 individuals within

25 countries. No Pseudo R2 has been computed as the residual variance is unknown. The

p-values given here are based on Wald tests, as the software used to estimate this model does

not produce Satterthwaite estimates of the degrees of freedom as for the other models.


Model 7C. Allocation of responsibility for the economy


Intercept 0.319 (0.118) 0.013Institution is EU −1.478 (0.147) < 0.001Year 2014 0.430 (0.158) 0.012Prospective assessment −0.159 (0.059) 0.014Female 0.185 (0.036) < 0.001Age (decades) −0.050 (0.012) < 0.001High education −0.010 (0.043) 0.811Low education −0.095 (0.051) 0.062Urban area 0.038 (0.045) 0.394Rural area 0.099 (0.047) 0.034Unemployed −0.058 (0.075) 0.443Not in workforce −0.186 (0.042) < 0.001Left–right position 0.043 (0.007) < 0.001Left–right position2 0.004 (0.002) 0.029Prosp. assess. × year 2014 0.200 (0.077) 0.016Left–right pos. × year 2014 −0.036 (0.010) < 0.001Left–right pos.2 × year 2014 0.001 (0.003) 0.772Age × year 2014 0.051 (0.017) 0.003Female × year 2014 −0.112 (0.050) 0.027Low education × year 2014 0.139 (0.075) 0.065High education × year 2014 0.289 (0.060) < 0.001Rural area × year 2014 0.021 (0.066) 0.748Urban area × year 2014 −0.037 (0.063) 0.559Unemployed × year 2014 0.081 (0.100) 0.417Not in workforce × year 2014 0.170 (0.060) 0.004EU × year 2014 0.419 (0.143) 0.007Prosp. assess. × EU 0.129 (0.041) 0.004Left–right pos. × EU −0.026 (0.010) 0.007Left–right pos.2 × EU 0.004 (0.003) 0.124Age × EU −0.042 (0.017) 0.016Female × EU 0.229 (0.051) < 0.001Low education × EU 0.169 (0.073) 0.020High education × EU −0.188 (0.061) 0.002Rural area × EU −0.021 (0.066) 0.750Urban area × EU −0.069 (0.064) 0.281Unemployed × EU 0.282 (0.107) 0.008Not in workforce × EU 0.141 (0.059) 0.017Prosp. assess. × EU × year 2014 −0.204 (0.058) 0.002Left–right pos. × EU × year 2014 0.030 (0.014) 0.034Left–right pos.2 × EU × year 2014 0.000 (0.004) 0.908Age × EU × year 2014 0.003 (0.024) 0.913Female × EU × year 2014 −0.155 (0.072) 0.031Low education × EU × year 2014 −0.250 (0.106) 0.019High education × EU × year 2014 0.033 (0.084) 0.693Rural area × EU × year 2014 −0.023 (0.094) 0.805Urban area × EU × year 2014 0.208 (0.089) 0.019Unemployed × EU × year 2014 −0.204 (0.142) 0.151Not in workforce × EU × year 2014 −0.129 (0.085) 0.128

235


Intercept 0.321 (0.567)Institution is EU 0.490 (0.700)Year 2014 0.568 (0.754)Prospective assessment 0.077 (0.278)Prosp. assess. × year 2014 0.125 (0.353)EU × year 2014 0.436 (0.660)Prosp. assess. × EU 0.027 (0.163)Prosp. assess. × EU × year 2014 0.046 (0.214)

Dependent variable is the degree to which the individual holds the institution responsble for

the condition of the economy. Sample size is 76658 individuals within 24 countries. Pseudo

R2 is 0.135.

References

Aidt, Toke S. 2000. ‘Economic voting and information’. Electoral Studies 19 (2–3): 349–362.

Alesina, Alberto, and Howard Rosenthal. 1995. Partisan Politics, Divided Government, and the

Economy. Cambridge: Cambridge University Press.

Allen, Trevor J. 2015. ‘Extreme right support and mainstream party failure after the Eurozone

Crisis’. Conference paper, APSA Annual Meeting, 5 September. Accessed 23 September

2015. http://convention2.allacademic.com/one/apsa/apsa15/

online_program.php.

Alwin, Duane F. 1997. ‘Feeling Thermometers Versus 7-Point Scales: Which are Better?’ Soci-

ological Methods and Research 25 (3): 318–340.

Alwin, Duane F., and Jon A. Krosnick. 1991. ‘The Reliability of Survey Attitude Measurement:

The Influence of Question and Respondent Attitudes’. Sociological Methods and Research

20 (1): 139–181.

Anderson, Cameron D. 2006. ‘Economic Voting and Multilevel Governance: A Comparative

Individiual-Level Analysis’. American Journal of Political Science 50 (2): 449–463.

Anderson, Christopher J. 1998. ‘When in Doubt, Use Proxies: Attitudes Toward Domestic Polit-

ics and Support for European Integration’. Comparative Poliitcal Studies 31 (5): 569–601.

. 2000. ‘Economic voting and political context: A comparative perspective’. Electoral

Studies 19 (2): 151–170.

Anderson, Christopher J., and Jason D. Hecht. 2012. ‘Voting when the economy goes bad,

everyone is in charge, and no one is to blame: The case of the 2009 German election’.

Electoral Studies 31 (1): 5–19.

237

238 REFERENCES

. 2014. ‘Crisis of Confidence? The Dynamics of Economic Opinions During the Great

Recession’. In Mass Politics in Tough Times: Opinions, Votes, and Protest in the Great Reces-

sion, edited by Nancy Bermeo and Larry M. Bartels, 40–71. New York: Oxford University

Press.

Arcelus, Francisco, and Allan H. Meltzer. 1975. ‘The Effect of Aggregate Economic Variables

on Congressional Elections’. American Political Science Review 69 (4): 1232–1239.

Arceneaux, Kevin. 2003. ‘The Conditional Impact of Blame Attribution on the Relationship

Between Economic Adversity and Turnout’. Political Research Quarterly 56 (1): 67–75.

. 2006. ‘The Federal Face of Voting: Are Elected Officials Held Accountable for the Func-

tions Relevant to Their Office?’ Political Psychology 27 (5): 731–754.

Arceneaux, Kevin, and David W. Nickerson. 2009. ‘Modeling Certainty with Clustered Data: A

Comparison of Methods’. Political Analysis 17 (2): 177–190.

Armingeon, Klaus, and Besir Ceka. 2014. ‘The loss of trust in the European Union during

the great recession since 2007: The role of heuristics from the national political system’.

European Union Politics 15 (1): 82–107.

Ashworth, Scott. 2005. ‘Reputational Dynamics and Political Careers’. Journal of Law, Econom-

ics & Organization 21 (2): 441–466.

Ashworth, Scott, and Ethan Bueno de Mesquita. 2008. ‘Electoral Selection, Strategic Chal-

lenger Entry, and the Incumbency Advantage’. Journal of Politics 70 (4): 1006–1025.

Aslanidis, Paris, and Cristóbal Rovira Kaltwasser. 2016. ‘Dealing with populists in government:

The SYRIZA–ANEL coalition in Greece’. Democratization 23 (6): 1077–1091.

Barro, Robert J. 1973. ‘The Control of Politicians: An Economic Model’. Public Choice 14 (1):

19–42.

Bartels, Larry M. 2012. ‘Elections in hard times’. Public Policy Research 19 (1): 44–50.

. 2013. ‘Political Effects of the Great Recession’. Annals of the American Academy of

Political and Social Science 650 (1): 47–75.

. 2014. ‘Ideology and Retrospection in the Electoral Responses to the Great Recession’.

In Mass Politics in Tough Times: Opinions, Votes, and Protest in the Great Recession, edited

by Nancy Bermeo and Larry M. Bartels, 185–223. New York: Oxford University Press.

REFERENCES 239

Bates, Douglas, Martin Mächler, Ben Bolker and Steve Walker. 2015. ‘Fitting Linear Mixed-

Effects Models Using lme4’. Journal of Statistical Software 67 (1): 1–48.

BBC News. 2009. ‘Huge protest over Irish economy’. Accessed 4 August 2016. http://

news.bbc.co.uk/2/hi/europe/7903518.stm.

. 2015. ‘Election 2015: Results’. Accessed 5 August 2016. http://www.bbc.com/

news/election/2015/results.

. 2016a. ‘EU Referendum Results’. Accessed 30 July 2016. http://www.bbc.

com/news/politics/eu_referendum/results.

. 2016b. ‘Scottish Parliament election 2016 results’. Accessed 15 July 2016. http:

//www.bbc.com/news/election/2016/scotland/results.

Bell, Robert M., and Michael L. Cohen. 2007. ‘Comment: Struggles with Survey Weighting and

Regression Modeling’. Statistical Science 22 (2): 165–167.

Bellucci, Paolo. 2012. ‘Government accountability and voting choice in Italy, 1990–2008’. Elect-

oral Studies 31 (3): 491–497.

Berglund, Frode, Sören Holmberg, Hermann Schmitt and Jacques Thomassen. 2005. ‘Party

Identification and Party Choice’. In The European Voter: A Comparative Study of Modern

Democracies, edited by Jacques Thomassen, 106–124. Oxford: Oxford University Press.

Bermeo, Nancy, and Larry M. Bartels. 2014. ‘Mass Politics in Tough Times’. In Mass Politics in

Tough Times: Opinions, Votes, and Protest in the Great Recession, edited by Nancy Bermeo

and Larry M. Bartels, 1–39. New York: Oxford University Press.

Birch, Sarah. 2008. Full participation: A comparative study of compulsory voting. Manchester:

Manchester University Press.

Blais, André. 2000. To Vote or Not to Vote?: The Merits and Limits of Rational Choice Theory.

Pittsburgh: University of Pittsburgh Press.

. 2007. ‘Turnout in Elections’. In The Oxford Handbook of Poltical Behavior, edited by

Russell J. Dalton and Hans-Dieter Klingemann, 621–635. Oxford: Oxford University Press.

. 2013. ‘The Source of Turnout Decline: New Values or New Contexts?’ Comparative

Political Studies 46 (1): 95–117.

240 REFERENCES

Blais, André, and R. K. Carty. 1990. ‘Does proportional representation foster voter turnout?’

European Journal of Political Research 18 (2): 167–181.

Bloom, Howard S., and H. Douglas Price. 1975. ‘Voter Response to Short-Run Economic Con-

ditions: The Asymmetric Effect of Prosperity and Recession’. American Political Science

Review 69 (4): 1240–1254.

Boomgaarden, Hajo G., Andreas R. T. Schuck, Matthijs Elenbaas and Claes H. de Vreese. 2011.

‘Mapping EU attitudes: Conceptual and empirical dimensions of Euroscepticism and EU

support’. European Union Politics 12 (2): 241–266.

Borges, Walt, Harold D. Clarke, Marianne C. Stewart, David Sanders and Paul Whiteley. 2013.

‘The emerging political economy of austerity in Britain’. Electoral Studies 32 (3): 396–403.

Borre, Ole. 1997. ‘Economic Voting in Danish Electoral Surveys 1987–94’. Scandinavian Polit-

ical Studies 20 (4): 347–365.

Bouvet, Florence, and Sharmila King. 2016. ‘Income inequality and election outcomes in OECD

countries: New evidence following the Great Recession of 2008–2009’. Electoral Studies

41 (1): 70–79.

Brack, Nathalie, and Nicholas Startin. 2015. ‘Introduction: Euroscepticism, from the margins

to the mainstream’. International Political Science Review 36 (3): 239–249.

Brody, Richard A., and Paul M. Sniderman. 1977. ‘From Life Space to Polling Place: The Re-

leveance of Personal Concerns for Voting Behavior’. British Journal of Political Science 7

(3): 337–360.

Busch, Kathrin Barbara. 2016. ‘Estimating parties’ left–right positions: Determinants of voters’

perceptions’ proximity to party ideology’. Electoral Studies 41 (1): 159–178.

Campbell, Angus, Philip E. Converse, Warren E. Miller and Donald E. Stokes. 1960. The Amer-

ican Voter. New York: John Wiley & Sons.

Çarkoglu, Ali. 2012. ‘Economic valuations vs. ideology: Diagnosing the sources of electoral

change in Turkey, 2002–2011’. Electoral Studies 31 (3): 313–521.

Carlsen, Fredrik. 2000. ‘Unemployment, inflation and government popularity—are there par-

tisan effects?’ Electoral Studies 19 (2–3): 141–150.

REFERENCES 241

Chappell Jr., Henry W., and Linda Gonçalves Veiga. 2000. ‘Economics and elections in Western

Europe: 1960–1997’. Electoral Studies 19 (2): 183–197.

Christensen, Rune Haubo Bojesen. 2015. ordinal: Regression Models for Ordinal Data. R pack-

age version 2015.6-28. http://www.cran.r-project.org/package=ordinal/.

Clarke, Harold D., and Guy D. Whitten. 2013. ‘Hard choices in hard times: Valence voting in

Germany’. Electoral Studies 32 (3): 445–451.

Connor, Gregery, Thomas Flavin and Brian O’Kelly. 2012. ‘The U.S. and Irish credit crises:

Their distinctive differences and common features’. Journal of International Money and

Finance 31 (1): 60–79.

Conover, Pamela Johnston, Stanley Feldman and Kathleen Knight. 1987. ‘The Personal and

Political Underpinnings of Economic Forecasts’. American Journal of Political Science 31

(3): 559–583.

Converse, Philip E. 1976. The Dynamics of Party Support: Cohort-Analyzing Party Identification.

Beverly Hills, California: Sage Publications.

Coppock, Alexander, and Donald P. Green. 2015. ‘Is Voting Habit Forming? New Evidence

from Experiments and Regression Discontinuities’. American Journal of Political Science.

doi:10.1111/ajps.12210.

Crewe, Ivor. 1981. ‘Electoral Participation’. In Democracy at the Polls: A Comparative Study of

Competitive National Elections, edited by David Butler, Howard R. Penniman and Austin

Ranney, 216–263. Washington, D.C.: American Enterprise Institute.

Cutler, Fred. 2004. ‘Government Responsibility and Electoral Accountability in Federations’.

Journal of Federalism 34 (2): 19–38.

. 2008. ‘Whodunnit? Voters and Responsibility in Canadian Federalism’. Canadian Journal

of Political Science 41 (3): 627–654.

Dalton, Russell J. 2016. ‘Stability and chance in party issue positions: The 2009 and 2014

European elections’. Electoral Studies. doi:10.1016/j.electstud.2016.06.

008.

Dalton, Russell J., David M. Farrell and Ian McAllister. 2011. Political Parties and Democratic

Linkage: How Parties Organize Democracy. Oxford: Oxford University Press.

242 REFERENCES

Dalton, Russell J., and Ian McAllister. 2015. ‘Random Walk or Planned Excursion? Continuity

and Change in the Left–Right Positions of Political Parties’. Comparative Political Studies

48 (6): 759–787.

Davis, Otto A., Melvin J. Hinich and Peter C. Ordeshook. 1970. ‘An Expository Development of

a Mathematical Model of the Electoral Process’. American Political Science Review 54 (2):

426–448.

Debus, Marc, Mary Stegmaier and Jale Tosun. 2014. ‘Economic Voting under Coalition Gov-

ernments: Evidence from Germany’. Political Science Research and Methods 2 (1): 49–67.

Döring, Holger, and Philip Manow. 2015. Parliament and government composition database

(ParlGov): An infrastructure for empirical information on parties, elections and governments

in modern democracies. Version 14/10. Accessed 27 June. www.parlgov.org.

Dorussen, Han. 2002. ‘Conclusions’. In Economic Voting, edited by Han Dorussen and Michaell

Taylor, 303–310. ECPR Studies in European Political Science 27. London: Routledge.

Dorussen, Han, and Harvey D. Palmer. 2002. ‘The context of economic voting: An introduc-

tion’. In Economic Voting, edited by Han Dorussen and Michaell Taylor, 1–14. ECPR Studies

in European Political Science 27. London: Routledge.

Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper & Row.

Duch, Raymond M. 2001. ‘A Developmental Model of Heterogeneous Economic Voting in New

Democracies’. American Political Science Review 95 (4): 895–910.

. 2007. ‘Comparative Studies of the Economy and the Vote’. In The Oxford Handbook

of Comparative Politics, edited by Charles Boix and Susan C. Stokes, 805–844. Oxford:

Oxford University Press.

Duch, Raymond M., and Iñaki Sagarzazu. 2014. ‘Crisis Perceptions and Economic Voting Among

the Rich and the Poor: The United Kingdom and Germany’. In Mass Politics in Tough Times:

Opinions, Votes, and Protest in the Great Recession, edited by Nancy Bermeo and Larry M.

Bartels, 224–265. New York: Oxford University Press.

Duch, Raymond M., and Randolph T. Stevenson. 2008. The Economic Vote: How Political and

Economic Institutions Condition Election Results. New York: Cambridge University Press.

Dumont, Patrick, Raphaël Kies and Philippe Poirier. 2010. ‘Luxembourg’. European Journal of

Political Research 49 (7–8): 1076–1088.

REFERENCES 243

Dumont, Patrick, and Philippe Poirier. 2005. ‘Luxembourg’. European Journal of Political Re-

search 44 (7–8): 1102–1118.

DuMouchel, William H., and Greg J. Duncan. 1983. ‘Using Sample Survey Weights in Multiple

Regression Analyses of Stratified Samples’. Journal of the American Statistical Association

78 (383): 535–543.

Durant, Henry. 1965. ‘Indirect influences on voting behaviour’. Polls 1:3–11.

Durr, Robert H. 1993. ‘What Moves Policy Sentiment?’ American Political Science Review 87

(1): 158–170.

Duverger, Maurice. 1959. Political Parties: Their Organization and Activity in the Modern State.

2nd ed. Translated by Barbara North and Robert North. University Paperbacks. London:

Methuen & Co.

EES. 2004. European Election Study 2004. Voter study. Version 2.http://eeshomepage.

net/ees-2004-study/voter-study.

. 2009. European Parliament Election Study 2009. Voter study. Advance release, 7 April

2010. http://www.piredeu.eu.

. 2014. European Election Study 2014. Voter study. Advance release, 1 January 2015.

http://eeshomepage.net/voter-study-2014.

EJPR. 2003–2015. ‘Political Data Yearbook’. Country entries not cited individually, European

Journal of Political Research 42–54.

. 2016. Political Data Yearbook: Interactive. European Journal of Political Research. Ac-

cessed 22 January 2016. http://www.politicaldatayearbook.com/.

Elgie, Robert. 1999. ‘The Politics of Semi-Presidentialism’. In Semi-Presidentialism in Europe,

edited by Robert Elgie, 1–21. Oxford: Oxford University Press.

. 2012. ‘The President of Ireland in Comparative Perspective’. Irish Political Studies 27

(4): 502–521.

Erikson, Robert S., Michael B. MacKuen and James A. Stimson. 2000. ‘Bankers or peasants

revisited’. Electoral Studies 19 (2–3): 295–312.

. 2002. The Macro Polity. Cambridge: Cambridge University Press.

244 REFERENCES

European Commission. 2009. Economic Crisis in Europe: Causes, Consequences and Responses.

European Economy 7. Directorate-General for Economic and Financial Affairs. Accessed

4 August 2016.http://ec.europa.eu/economy_finance/publications/

publication15887_en.pdf.

. 2010. The Economic Adjustment Programme for Greece. Occasional Papers 61. Directorate-

General for Economic and Financial Affairs. Accessed 29 July 2016. http://ec.

europa.eu/economy_finance/publications/occasional_paper/

2010/pdf/ocp61_en.pdf.

. 2011a. The Economic Adjustment Programme for Ireland. Occasional Papers 76. Directorate-




. 2011b. The Economic Adjustment Programme for Portugal. Occasional Papers 79. Directorate-




. 2012. The Second Economic Adjustment Programme for Greece. Occasional Papers 94.

Directorate-General for Economic and Financial Affairs. Accessed 29 July 2016. http:

//ec.europa.eu/economy_finance/publications/occasional_

paper/2012/pdf/ocp94_en.pdf.

. 2013. The Economic Adjustment Programme for Cyprus. Occasional Papers 149. Directorate-




Evans, Geoffrey, and Robert Andersen. 2006. ‘The Political Conditioning of Economic Percep-

tions’. Journal of Politics 68 (1): 194–207.

Evans, Geoffrey, and Mark Pickup. 2010. ‘Reversing the Causal Arrow: The Political Condition-

ing of Economic Perceptions in the 2000–2004 U.S. Presidential Election Cycle’. Southern

Political Science Association 72 (4): 1236–1251.

Fair, Ray C. 1978. ‘The Effect of Economic Events on Votes for President’. Review of Economics

and Statistics 60 (2): 159–173.

REFERENCES 245

Fearon, James D. 1999. ‘Electoral Accountability and the Control of Politicians: Selecting Good

Types versus Sanctioning Poor Performance’. In Democracy, Accountability, and Representa-

tion, edited by Adam Przeworski, Susan C. Stokes and Bernard Manin, 55–97. Cambridge:

Cambridge University Press.

Feld, Lars P., and Gebhard Kirchgässner. 2000. ‘Official and hidden unemployment and the

popularity of the government: an econometric analysis for the Kohl government’. Electoral

Studies 19 (2–3): 333–347.

Ferejohn, John. 1986. ‘Incumbent performance and electoral control’. Public Choice 50 (1–3):

5–25.

Fiorina, Morris P. 1978. ‘Economic Retrospective Voting in American National Elections: A

Micro-Analysis’. Midwest Political Science Association 22 (2): 426–443.

. 1981. Retrospective Voting in American National Elections. New Haven: Yale University

Press.

Fisher, Stephen D., and Sara B. Hobolt. 2010. ‘Coalition government and electoral accountab-

ility’. Electoral Studies 29 (3): 358–369.

Flassbeck, Heiner, and Costas Lapavitsas. 2015. ‘Confronting the failure of the European Mon-

etary Union’. In Asymmetric Crisis in Europe and Possible Futures: Critical political economy

and post-Keynesian perspectives, edited by Johannes Jäger and Elisabeth Springler, 131–

148. New York: Routledge.

Ford, Robert, and Matthew J. Goodwin. 2010. ‘Angry White Men: Individual and Contextual

Predictors of Support for the British National Party’. Political Studies 58 (1): 1–25.

Fraile, Marta, and Michael S. Lewis-Beck. 2012. ‘Economic and elections in Spain (1982–

2008): Cross-measures, cross-time’. Electoral Studies 31 (3).

. 2013. ‘Multi-dimensional economic voting in Spain: The 2008 election’. Electoral Stud-

ies 32 (3): 465–469.

Franklin, Mark N. 1996. ‘Electoral Participation’. In Comparing Democracies: Elections and Vot-

ing in Global Perspective, edited by Lawrence LeDuc, Richard G. Nieme and Pippa Norris,

216–235. Thousand Oaks, California: SAGE.

. 2004. Voter Turnout and the Dynamics of Electoral Competition in Established Democra-

cies since 1945. Cambridge: Cambridge University Press.

246 REFERENCES

Freire, André, and José Santana-Pereira. 2012. ‘Economic voting in Portugal, 2002–2009’.


Gabel, Matthew J., and Christopher J. Anderson. 2004. ‘The structure of citizen attitudes and

the European political space’. In European Integration and Political Conflict, edited by Gary

Marks and Marco R. Steenbergen, 13–31. New York: Cambridge University Press.

Gabel, Matthew J., and Simon Hix. 2004. ‘Defining the EU political space: an empirical study

of the European election manifestos, 1979–1999’. In European Integration and Political

Conflict, edited by Gary Marks and Marco R. Steenbergen, 93–119. New York: Cambridge

University Press.

Gabel, Matthew, and Guy D. Whitten. 1997. ‘Economic conditions, economic perceptions, and

public support for European integration’. Political Behavior 19 (1): 81–96.

Garry, John. 2014. ‘Holding parties responsible at election time: Multi-level, multi-party gov-

ernment and electoral accountability’. Electoral Studies 34 (1): 78–88.

Garry, John, and James Tilley. 2014. ‘Inequality, state ownership and the European Union:

How economic context and economic ideology shape support for the Europeana Union’.


Gelman, Andrew. 2007a. ‘Rejoinder: Struggles with Survey Weighting and Regression Model-

ing’. Statistical Science 22 (2): 184–188.

. 2007b. ‘Struggles with Survey Weighting and Regression Modeling’. Statistical Science

22 (2): 153–164.

Gelman, Andrew, and Jennifer Hill. 2007. Data Analysis Using Regression and Multilevel/Hier-

archical Models. New York: Cambridge University Press.

Gelman, Andrew, and Yu-Sung Su. 2015. arm: Data Analysis Using Regression and Multilevel/

Hierarchical Models. R package version 1.8-6. https://CRAN.R-project.org/

package=arm.

Geys, Benny. 2006. ‘Explaining voter turnout: A review of aggregate-level research’. Electoral

Studies 25 (4): 637–663.

Gosnell, Harold F., and William G. Colman. 1940. ‘Political Trends in Industrial America:

Pennsylvania an Example’. Public Opinion Quarterly 4 (3): 473–486.

REFERENCES 247

Gowrisankaran, Gautam, Matthew F. Mitchell and Andrea Moro. 2008. ‘Electoral design and

voter welfare from the US Senate: Evidence from a dynamic selection model’. Review of

Economic Dynamics 11 (1): 1–17.

Gray, Mark, and Miki Caul. 2000. ‘Declining Voter Turnout in Advanced Industrial Democra-

cies, 1950 to 1997: The Effects of Declining Group Mobilization’. Comparative Political

Studies 33 (9): 1091–1122.

Green, Donald P., and Ron Shacar. 2000. ‘Habit Formation and Political Behaviour: Evidence

of Consuetude in Voter Turnout’. British Journal of Political Science 30 (4): 561–573.

Green, Donald, Bradley Palmquist and Eric Schickler. 2002. Partisan Hearts and Minds. New

Haven, Connecticut: Yale University Press.

Grittersova, Jana, Indridi H. Indridason, Christina C. Gregory and Ricardo Crespo. 2015. ‘Aus-

terity and Political Extremism: The Electoral Consequences of Fiscal Reforms’. Confer-

ence paper, APSA Annual Meeting, 5 September. Accessed 23 September 2015. http:

//convention2.allacademic.com/one/apsa/apsa15/online_

program.php.

Hakhverdian, Armen, Erika van Elsas, Wouter van der Brug and Theresa Kuhn. 2013. ‘Euro-

scepticism and education: A longitudinal study of 12 EU member states, 1973–2010’.


Hansford, Thomas G., and Brad T. Gomez. 2015. ‘Reevaluating the sociotropic economic voting

hypothesis’. Electoral Studies 39 (1): 15–25.

Harteveld, Eelco, Tom van der Meer and Catherine E. De Vries. 2013. ‘In Europe we trust?

Exploring three logics of trust in the European Union’. European Union Politics 14 (4):

542–565.

Heipertz, Martin, and Amy Verdun. 2010. Ruling Europe: The Politics of the Stability and Growth

Pact. New York: Cambridge University Press.

Hellwig, Timothy, and Eva Coffey. 2011. ‘Public opinion, party messages, and responsibility

for the financial crisis in Britain’. Electoral Studies 30 (3): 417–426.

Hellwig, Timothy, and Dani M. Marinova. 2015. ‘More Misinformed than Myopic: Economic

Retrospections and the Voter’s Time Horizon’. Political Behaviour 37 (4): 865–887.

248 REFERENCES

Hellwig, Timothy, and David Samuels. 2008. ‘Electoral Accountability and the Variety of Demo-

cractic Regimes’. British Journal of Political Science 38 (1): 65–90.

Hernández, Enrique, and Hanspeter Kriesi. 2015. ‘The electoral consequences of the financial

and economic crisis in Europe’. European Journal of Political Research. doi:10.1111/

1475-6765.12122.

Hibbs, Douglas. 1993. Solidarity or Egoism? The Economics of Sociotropic and Egocentric Influ-

ences on Political Behavior: Denmark in International and Theoretical Perspective. Aarhus,

Denmark: Aarhus University Press.

Hix, Simon. 1999. ‘Dimensions and alignments in European Union politics: Cognitive con-

straints and partisan responses’. European Journal of Political Research 35 (1): 69–106.

. 2007. ‘Euroscepticism as Anti-Centralization: A Rational Choice Institutionalist Per-

spective’. European Union Politics 8 (1): 131–150.

Hix, Simon, and Christopher Lord. 1997. Political Parties in the European Union. Basingstoke,

Hampshire: Macmillan Press.

Hobolt, Sara B. 2012. ‘Citizen Satisfaction with Democracy in the European Union’. Journal of

Common Market Studies 50 (Supplement 1): 88–105.

Hobolt, Sara B., and Patrick Leblond. 2014. ‘Economic Insecurity and Public Support for the

Euro: Before and During the Financial Crisis’. In Mass Politics in Tough Times: Opinions,

Votes, and Protest in the Great Recession, edited by Nancy Bermeo and Larry M. Bartels,

128–147. New York: Oxford University Press.

Hobolt, Sara B., and James Tilley. 2014. Blaming Europe? Responsibility Without Accountability

in the European Union. Oxford: Oxford University Press.

Hobolt, Sara, James Tilley and Susan Banducci. 2013. ‘Clarity of responsibility: How govern-

ment cohesion conditions preformance voting’. European Journal of Political Research 52

(2): 164–187.

Hooghe, Liesbet, and Gary Marks. 1999. ‘The Making of a Polity: The Struggle over European

Integration’. In Continuity and Change in Contemporary Capitalism, edited by Herbert

Kitschelt, Peter Lange, Gary Marks and John D. Stephens, 70–97. Cambridge: Cambridge

University Press.

REFERENCES 249

Hooghe, Liesbet, Gary Marks and Carole J. Wilson. 2004. ‘Does left/right structure party po-

sitions on European integration?’ In European Integration and Political Conflict, edited by

Gary Marks and Marco R. Steenbergen, 120–140. New York: Cambridge University Press.

Hotelling, Harold. 1929. ‘Stability in Competition’. The Economic Journal 39 (153): 41–57.

Huber, Gregory A., Seth J. Hill and Gabriel S. Lenz. 2012. ‘Sources of Bias in Retrospective De-

cision Making: Experimental Evidence on Voters’ Limitations in Controlling Incumbents’.

American Political Science Review 106 (4): 720–741.

Irish Times. 2016. ‘Election 2016: Results Hub’. Accessed 4 August 2016. http://www.

irishtimes.com/election-2016/results-hub.

Jackman, Robert W. 1987. ‘Political Institutions and Voter Turnout in the Industrial Democra-

cies’. American Political Science Review 81 (2): 405–424.

Jackman, Robert W., and Ross A. Miller. 1995. ‘Voter Turnout in the Industrial Democracies

during the 1980s’. Comparative Political Studies 27 (4): 467–492.

Johns, Robert, James Mitchell and Christopher J. Carman. 2013. ‘Constitution or Competence?

the SNP’s Re-Election in 2011’. Political Studies 61 (1): 158–178.

Katz, Richard S., and Peter Mair. 1995. ‘Changing Models of Party Organization and Party

Democracy: The Emergence of the Cartel Party’. Party Politics 1 (1): 5–28.

. 2009. ‘The Cartel Party Thesis: A Restatement’. Perspectives on Politics 7 (4): 753–766.

Kenworthy, Lane, and Lindsay A. Owens. 2011. ‘The Surprisingly Weak Effect of Recessions

on Public Opinion’. In The Great Recession, edited by David B. Grusky, Bruce Western and

Christopher Wimer, 196–219. New York: Russell Sage Foundation.

Kernell, Samuel. 1977. ‘Presidential Popularity and Negative Voting: An Alternative Explan-

ation of the Midterm Congressional Decline of the President’s Party’. American Political

Science Review 71 (1): 44–66.

Kerr, W. A. 1944. ‘A Quantitative Study of Political Behavior, 1840–1940’. Journal of Social

Psychology 19 (2): 273–281.

Key Jr, V. O. 1961. Public Opinion and American Democracy. New York: Alfred A. Knopf.

. 1964. Politics, Parties, & Pressure Groups. 5th ed. New York: Thomas Y. Crowell Com-

pany.

250 REFERENCES

. 1966. The Responsible Electorate: Rationality in Presidential Voting 1936–1960. Cam-

bridge, Massachussets: Belknap Press of Harvard University Press.

Kinder, Donald R., and D. Roderick Kiewiet. 1979. ‘Economic Discontent and Political Beha-

vior: The Role of Personal Grievances and Collective Economic Judgments in Congres-

sional Voting’. American Journal of Political Science 23 (3): 495–527.

. 1981. ‘Sociotropic Politics: The American Case’. British Journal of Political Science 11

(2): 129–161.

King, Gary, James Honaker, Anne Joseph and Kenneth Scheve. 2001. ‘Analyzing Incomplete

Political Science Data: An Alternative Algorithm for Multiple Imputation’. American Polit-

ical Science Review 95 (1): 49–69.

Kirchheimer, Otto. 1966. ‘The Transformation of the Western European Party Systems’. In Polit-

ical Parties and Political Development, edited by Joseph LaPalombara and Myron Weiner,

177–200. Princeton, New Jersey: Princeton University Press.

Koole, Ruud. 1996. ‘Cadre, Catch-all or Cartel? A Comment on the Notion of the Cartel Party’.

Party Politics 2 (4): 507–523.

Kramer, Gerald H. 1971. ‘Short-Term Fluctuations in U.S. Voting Behavior, 1896–1964’. Amer-

ican Political Science Review 65 (1): 131–143.

. 1983. ‘The Ecological Fallacy Revisited: Aggregate- versus Individual-level Findings

on Economics and Elections, and Sociotropic Voting’. American Political Science Review 77

(1): 92–111.

Kriesi, Hanspeter. 2012. ‘The Political Consequences of the Financial and Economic Crisis in

Europe: Electoral Punishment and Popular Protest’. Swiss Poliitcal Science Review 18 (4):

518–522.

. 2014. ‘The Political Consequences of the Economic Crisis in Europe: Electoral Punish-

ment and Popular Protest’. In Mass Politics in Tough Times: Opinions, Votes, and Protest in

the Great Recession, edited by Nancy Bermeo and Larry M. Bartels, 297–333. New York:


Kroh, Martin. 2009. ‘The Ease of Ideological Voting: Voter Sophistication and Party System

Complexity’. In The Comparative Study of Electoral Systems, edited by Hans-Dieter Klinge-

mann, 220–236. Oxford: Oxford University Press.

REFERENCES 251

Krouwel, André. 2012. Party Transformations in European Democracies. Albany, New York:

State University of New York Press.

Kuklinski, James H., and Darrell M. West. 1981. ‘Economic Expectations and Voting Behavior

in United States House and Senate Elections’. American Political Science Review 75 (2):

436–447.

Kumlin, Staffan. 2009. ‘Blaming Europe? Exploring the variable impact of national public ser-

vice dissatisfaction on EU trust’. Journal of European Social Policy 19 (5): 408–420.

Kuznetsova, Alexandra, Per Bruun Brockhoff and Rune Haubo Bojesen Christensen. 2016.

lmerTest: Tests in Linear Mixed Effects Models. R package version 2.0-32. https://

CRAN.R-project.org/package=lmerTest.

Lacy, Dean, and Barry C. Burden. 1999. ‘The Vote-Stealing and Turnout Effects of Ross Perot in

the 1992 U.S. Presidential Election’. American Journal of Political Science 43 (1): 233–255.

Lane, Robert E. 1959. Political Life: Why People Get Involved in Politics. New York: Free Press.

LeDuc, Lawrence, and Jon H. Pammett. 2013. ‘The fate of governing parties in times of eco-

nomic crisis’. Electoral Studies 32 (3): 494–499.

Leighley, Jan E., and Jonathon Nagler. 2014. Who Votes Now? Demographics, Issues, Inequality,

and Turnout in the United States. Princeton: Princeton University Press.

Lester, J. C. 1994. ‘The Evolution of the Political Compass (And Why Libertarianism Is Not

Right-Wing)’. Journal of Social and Evolutionary Systems 17 (3): 231–241.

Lewis-Beck, Michael S. 1986. ‘Comparative Economic Voting: Britain, France, Germany, Italy’.

American Journal of Political Science 30 (2): 315–346.

Lewis-Beck, Michael S., William G. Jacoby, Helmut Norpoth and Herbert F. Weisberg. 2008.

The American Voter Revisited. Ann Arbor: University of Michigan Press.

Lewis-Beck, Michael S., Richard Nadeau and Angelo Elias. 2008. ‘Economics, Party, and the

Vote: Causality Issues and Panel Data’. American Journal of Political Science 52 (1): 84–95.

Lewis-Beck, Michael S., and Martin Paldam. 2000. ‘Economic voting: an introduction’. Electoral

Studies 19 (2–3): 113–121.

Lewis-Beck, Michael S., Rune Stubager and Richard Nadeau. 2013. ‘The Kramer problem:

Micro-macro resolution with a Danish pool’. Electoral Studies 32 (3): 500–505.

252 REFERENCES

Lindvall, Johannes. 2012. ‘The Political Consequences of the Great Depression and the Great

Recession: Remarkably Similar’. Swiss Political Science Review 18 (4): 514–517.

. 2013. ‘Economic crises as political opportunities’. In Before and Beyond the Global Eco-

nomic Crisis: Economics, Politics and Settlement, edited by Mats Benner, 132–150. Chel-

tenham: Edward Elgar.

. 2014. ‘The electoral consequences of two great crises’. European Journal of Political

Research 53 (4): 747–765.

Lindvall, Johannes, Johan Martinsson and Henrik Oscarsson. 2013. ‘Party choice in hard times:

Group-specific responses to economic downturns in Sweden’. Electoral Studies 32 (3).

Linek, Lukáš, and Tomáš Lacina. 2010. ‘Czech Republic’. European Journal of Political Research

49 (7–8): 939–946.

Lipset, Seymour Martin. 1981. Political Man: The Social Bases of Politics. Expanded Edition.

Baltimore: John Hopkins University Press.

Lyons, William, and Robert Alexander. 2000. ‘A Tale of Two Electorates: Generational Replace-

ment and the Decline of Voting in Presidential Elections’. Journal of Politics 62 (4): 1014–

1034.

MacKuen, Michael B., Robert S. Erikson and James A. Stimson. 1992. ‘Peasants or Bankers?

The American Electorate and the U.S. Economy’. American Political Science Review 86 (3):

597–611.

Mair, Peter. 2007. ‘Left–Right Orientations’. In The Oxford Handbook of Poltical Behavior, edited

by Russell J. Dalton and Hans-Dieter Klingemann, 206–222. Oxford: Oxford University

Press.

Marsh, Michael, and Slava Mikhaylov. 2012. ‘Economic voting in a crisis: The Irish election of

2011’. Electoral Studies 31 (3): 478–484.

Martins, Rodrigo, and Francisco José Veige. 2013. ‘Economic performance and turnout at na-

tional and local elections’. Public Choice 157: 429–448.

Martinsson, Johan. 2013. ‘Economic voting in Sweden 1985–2010’. Electoral Studies 32 (3):

470–475.

REFERENCES 253

Matakos, Konstantinos, and Dimitrios Xefteris. 2016. ‘Election note: The September 20, 2015

parliamentary election in Greece’. Electoral Studies. doi:10.1016/j.electstud.

2016.04.004.

Mau, Steffen. 2005. ‘Democratic demand for a social Europe? Preferences of the European

citizenry’. International Journal of Social Welfare 14 (2): 76–85.

Mavrogordatos, George Th., and Nikos Marantzidis. 2010. ‘Greece’. European Journal of Polit-

ical Research 49 (7–8): 991–1000.

Mayer, Nonna. 2014. ‘The Electoral Impact of the Crisis on the French Working Class: More

to the right?’ In Mass Politics in Tough Times: Opinions, Votes, and Protest in the Great

Recession, edited by Nancy Bermeo and Larry M. Bartels, 266–296. New York: Oxford

University Press.

McBride, Stephen. 2015. ‘The economics of austerity’. In Social Policy in Times of Austerity:

Global economic crisis and the new politics of welfare, edited by Kevin Farnsworth and Zoë

Irving, 67–86. Bristol: Policy Press.

McDonnell, Duncan, and James L. Newell. 2011. ‘Outsider parties in government in Western

Europe’. Party Politics 17 (4): 443–452.

Meehl, Paul E. 1977. ‘The Selfish Voter Paradox and the Thrown-Away Vote Argument’. Amer-

ican Political Science Review 71 (1): 11–30.

Meyer, Thomas M., and Daniel Strobl. 2016. ‘Voter perceptions of coalition policy positions in

multiparty systems’. Electoral Studies 41 (1): 80–91.

Michelitch, Kristin, Marco Morales, Andrew Owen and Joshua A. Tucker. 2012. ‘Looking to

the future: Prospective economic voting in 2008 Presidential Elections’. Electoral Studies

31 (4): 838–851.

Mylonas, Harris. 2013. ‘Greece’. European Journal of Political Research: Political Data Yearbook

52 (1): 87–95.

Nadeau, Richard, and Michael S. Lewis-Beck. 2001. ‘National Economic Voting in U.S. Presid-

ential Elections’. Journal of Politics 63 (1): 159–181.

Nadeau, Richard, Richard G. Niemi and Antoine Yoshinaka. 2002. ‘A cross-national analysis of

economic voting: taking account of the political context across time and nations’. Electoral

Studies 21 (3): 403–423.

254 REFERENCES

Nannestad, Peter, and Martin Paldam. 1994. ‘The VP-function: A survey of the literature on

vote and popularity functions after 25 years’. Public Choice 79 (3): 213–245.

. 1995. ‘It’s the government’s fault! A cross-section study of economic voting in Den-

mark, 1990/93’. European Journal of Political Research 28 (1): 33–62.

. 1997a. ‘From the Pocketbook of the Welfare Man: A Pooled Cross-Section Study of

Economic Voting in Denmark, 1986–92’. British Journal of Political Science 27 (1): 119–

136.

. 1997b. ‘The grievance asymmetry revisited: A micro study of economic voting in Den-

mark, 1986–92’. European Journal of Political Economy 13 (1): 81–99.

Nezi, Roula. 2012. ‘Economic voting under the economic crisis: Evidence from Greece’. Elect-

oral Studies 31 (3): 498–505.

Norpoth, Helmut. 2001. ‘Divided Government and Economic Voting’. Journal of Politics 63 (2):

414–435.

Nousiainen, Jaakko. 2001. ‘From Semi-presidentialism to Parliamentary Government: Political

and Constitutional Developments in Finland’. Scandinavian Political Studies 24 (2): 95–

109.

Ogburn, William F., and Lolagene C. Coombs. 1940. ‘American Government and Politics: The

Economic Factor in the Roosevelt Elections’. American Political Science Review 34 (4): 719–

727.

O’Malley, Eoin. 2012. ‘Ireland’. European Journal of Political Research: Political Data Yearbook

51 (1): 141–152.

Orriols, Lluis, and Álvaro Martínez. 2014. ‘The role of the political context in voting indecision’.


Paldam, Martin. 1991. ‘How Robust is the Vote Function? A Study of Seventeen Nations over

Four Decades: The Calculus of Support’. In Economics and Politics, edited by Helmut Nor-

poth, Michael S. Lewis-Beck and Jean-Dominique Lafay, 9–32. Manchester: Manchester

University Press.

Paldam, Martin, and Peter Nannestad. 2000. ‘What do voters know about the economy? A

study of Danish data, 1990–1993’. Electoral Studies 19 (2–3): 363–391.

REFERENCES 255

Palmer, Harvey D., and Guy D. Whitten. 2011. ‘Through thick and thin? The dynamics of

government support across income groups during economic crises’. Electoral Studies 30

(3): 427–437.

Palmer, Harvey D., Guy D. Whitten and Laron K. Williams. 2013. ‘Who sould be chef? The

dynamics of valence evalations across income groups during economic crises’. Electoral

Studies 32 (3): 425–431.

Panebianco, Angelo. 1988. Political parties: organization and power. Translated by Marc Silver.

Cambridge: Cambridge University Press.

Passarelli, Gianluca, and Dario Tuorto. 2014. ‘Not with my vote: Turnout and the economic

crisis in Italy’. Contemporary Italian Politics 6 (2): 147–158.

Pearson, F. A., and W. I. Myers. 1948. ‘Prices and Presidents’. Farm Economics (Ithaca, New

York), no. 163 (September): 4210–4218.

Plutzer, Eric. 2002. ‘Becoming a Habitual Voter: Inertia, Resources, and Growth in Young

Adulthood’. American Political Science Review 96 (1): 41–56.

Popa, Sebastian Adrian, Hermann Schmitt, Sara B. Hobolt and Eftichia Teperoglou. 2015. EES

2014 Voter Study Advance Release Notes. Mannheim: MZES, University of Mannheim.

Powell Jr., G. Bingham. 1980. ‘Voting Turnout in Thirty Democracies: Partisan, Legal, and

Socio-Economic Influences’. In Electoral Participation: A Comparative Analysis, edited by

Richard Rose, 5–34. London: SAGE.

. 1982. Contemporary Democracies: Participation, Stability, and Violence. Cambridge,

Massachussets: Harvard University Press.

. 1986. ‘American Voter Turnout in Comparative Perspective’. American Political Science

Review 80 (1): 17–43.

Powell Jr., G. Bingham, and Guy D. Whitten. 1993. ‘A Cross-National Analysis of Economic

Voting: Taking Account of the Political Context’. American Journal of Political Science 37

(2): 391–414.

R Core Team. 2016. R: A Language and Environment for Statistical Computing. Vienna, Austria:

R Foundation for Statistical Computing. https://www.R-project.org/.

256 REFERENCES

Radcliff, Benjamin. 1992. ‘The Welfare State, Turnout, and the Economy: A Comparative Ana-

lysis’. American Political Science Review 86 (2): 444–454.

. 1994. ‘Reward Without Punishment: Economic Conditions and the Vote’. Political Re-

search Quarterly 47 (3): 721–731.

Rees, Albert, Herbert Kaufman, Samuel J. Eldersveld and Frank Freidel. 1962. ‘The Effect of

Economic Conditions on Congressional Elections 1946–1958’. Review of Economics and

Statistics 44 (4): 458–465.

Reinhart, Carmen M., and Kenneth S. Rogoff. 2009. This Time is Different: Eight Centuries of

Financial Folly. Princeton, New Jersey: Princeton University Press.

Rihoux, Benoît, Audrey Vandeleene, Lieven de Winter, Pierre Baudewyns and Serge Deruette.

2015. ‘Belgium’. European Journal of Political Research: Political Data Yearbook 54 (1):

33–43.

Rogers, Jonathan. 2014. ‘A communotropic theory of economic voting’. Electoral Studies 36

(1): 107–116.

Rori, Lamprini. 2016. ‘The 2015 Greek parliamentary elections: from great expectations to no

expectations’. West European Politics. doi:10.1080/01402382.2016.1171577.

Rosenstone, Steven J. 1982. ‘Economic Adversity and Voter Turnout’. American Journal of Polit-

ical Science 26 (1): 25–46.

Royed, Terry J., Kevin M. Leyden and Stephen A. Borrelli. 2000. ‘Is “Clarity of Responsibil-

ity” Important for Economic Voting? Revisiting Powell and Whitten’s Hypothesis’. British

Journal of Political Science 30 (4): 669–698.

Rubin, Donald B. 1978. ‘Multiple Imputations in Sample Surveys: A Phenomenonological

Bayesian Approach to Nonresponse’. Proceedings of the Survey Research Methods Section of

the American Statistical Association: 20–34.

Rüdig, Wolfgang, and Georgios Karyotis. 2014. ‘Who Protests in Greece? Mass Opposition to

Austerity’. British Journal of Political Sicence 44 (4): 487–513.

Samuels, David. 2004. ‘Presidentialism and Accountability for the Economy in Comparative

Perspective’. American Political Science Review 98 (3): 425–436.

REFERENCES 257

Sanders, David. 2000. ‘The real economy and the perceived economy in popularity functions:

how much do voters need to know? A study of British data, 1974–97’. Electoral Studies

19 (2–3): 275–294.

Scharpf, Fritz W. 2013. ‘Monetary Union, Fiscal Crisis and the Disabling of Democratic Ac-

countability’. In Politics in the Age of Austerity, edited by Armin Schäfer and Wolfgang

Streeck, 108–142. Cambridge: Polity Press.

Schlozman, Kay Lehman, and Sidney Verba. 1979. Injury to Insult: Unemployment, Class, and

Political Response. Cambridge, Massachusetts: Harvard University Press.

Schmitt, Hermann. 2009. ‘Multiple Party Identifications’. In The Comparative Study of Electoral

Systems, edited by Hans-Dieter Klingemann, 137–157. Oxford: Oxford University Press.

Schmitt, Hermann, Matthew Loveless, Sascha Adam and Daniela Braun. 2009. European Elec-

tion Study 2004: Design, Data Description and Documentation. 2nd ed. Mannheim: MZES,

University of Mannheim.

Singer, Matthew M. 2011a. ‘When do voters actually think “It’s the Economy”? Evidence from

the 2008 presidential campaign’. Electoral Studies 30 (4): 621–632.

. 2011b. ‘Who Says “It’s the Economy”? Cross-National and Cross-Individual Variance

in the Salience of Economic Performance’. Comparative Political Studies 44 (3): 284–312.

. 2013. ‘The global economic crisis and domestic political agendas’. Electoral Studies 32

(3): 404–410.

Soroka, Stuart N. 2006. ‘Good News and Bad News: Asymmetric Responses to Economic In-

formation’. Journal of Politics 68 (2): 372–385.

Soroka, Stuart, and Christopher Wlezien. 2014. ‘Economic Crisis and Support for Redistribu-

tion in the United Kingdom’. In Mass Politics in Tough Times: Opinions, Votes, and Protest

in the Great Recession, edited by Nancy Bermeo and Larry M. Bartels, 105–127. New York:


Stanig, Piero. 2013. ‘Political polarization in retrospective economic evalations during reces-

sions and recoveries’. Electoral Studies 32 (4): 729–745.

Stavrakakis, Yannis, and Giorgos Katsambekis. 2014. ‘Left-wing populism in the European

periphery: the case of SYRIZA’. Journal of Political Ideologies 19 (2): 119–142.

258 REFERENCES

Steenbergen, Marco R., and Gary Marks. 2004. ‘Introduction: Models of political conflict in

the European Union’. In European Integration and Political Conflict, edited by Gary Marks

and Marco R. Steenbergen, 1–10. New York: Cambridge University Press.

Stegmaier, Mary, and Michael S. Lewis-Beck. 2011. ‘Shocks and oscillations: The political

economy of Hungary’. Electoral Studies 30 (3): 462–467.

Stevens, Daniel. 2007. ‘Mobilization, Demobilization and the Economy in American Elections’.

British Journal of Political Science 37 (1): 165–186.

Stevenson, Randolph T. 2001. ‘The Economy and Policy Mood: A Fundamental Dynamic of

Democratic Politics?’ American Journal of Political Science 45 (3): 620–633.

. 2002. ‘The economy as context: indirect links between the economy and voters’. In

Economic Voting, edited by Han Dorussen and Michaell Taylor, 45–65. ECPR Studies in

European Political Science 27. London: Routledge.

Stevenson, Randolph T., and Raymond Duch. 2013. ‘The meaning and use of subjective per-

ceptions in studies of economic voting’. Electoral Studies 32 (2): 305–320.

Stigler, George J. 1973. ‘General Economic Conditions and National Elections’. American Eco-

nomic Review 63 (2): 160–167.

Stubager, Rune, Niels W. Botterill, Michael S. Lewis-Beck and Richard Nadeau. 2014. ‘Scope

conditions of economic voting: The Danish exception?’ Electoral Studies 34 (1): 16–26.

Swedlow, Brendon. 2008. ‘Beyond liberal and conservative: Two-dimensional conceptions of

ideology and the structure of political attitudes and values’. Journal of Political Ideologies

13 (2): 157–180.

Taniguchi, Masaki. 2016. ‘The multi-store model for economic voting: Rome wasn’t built in a

day’. Electoral Studies 41 (1): 179–189.

The Economist. 2009. ‘The European elections: Swing low, swing right’. 11 June. Accessed

21 September 2015. http://www.economist.com/node/13832286.

Tibbitts, Clark. 1931. ‘Majority Votes and the Business Cycle’. American Journal of Sociology

36 (4): 596–606.

Tillman, Erik R. 2008. ‘Economic Judgments, Party Choice, and Voter Abstention in Cross-

National Perspective’. Comparative Political Studies 41 (9): 1290–1309.

REFERENCES 259

. 2013. ‘Authoritarianism and citizen attitudes towards European integration’. European

Union Politics 14 (4): 566–589.

To Vima: English. 2014. ‘Alexis Tsipras: “We do not support any sort of Euroscepticism” ’. 7 May.

Accessed 1 October 2015. http://www.tovima.gr/en/article/?aid=

593303.

Traugott, Michael W., and John P. Katosh. 1979. ‘Response Validity in Surveys of Voting Beha-

vior’. Public Opinion Quarterly 43 (3): 359–377.

Tsebelis, George, and Geoffrey Garrett. 2000. ‘Legislative Politics in the European Union’.


Tufte, Edward R. 1975. ‘Determinants of the Outcomes of Midterm Congressional Elections’.

American Political Science Review 69 (3): 812–826.

. 1978. Political Control of the Economy. Princeton, New Jersey: Princeton University

Press.

Van der Brug, Wouter, Cees van der Eijk and Mark Franklin. 2007. The Economy and the Vote:

Economic Conditions and Elections in Fifteen Countries. Cambridge: Cambridge University

Press.

Van der Eijk, Cees, and Mark N. Franklin. 2004. ‘Potential for contestation on European matters

at natioanl elections in Europe’. In European Integration and Political Conflict, edited by

Gary Marks and Marco R. Steenbergen, 32–50. New York: Cambridge University Press.

Van der Eijk, Cees, Hermann Schmitt and Tanja Binder. 2005. ‘Left–Right Orientations and

Party Choice’. In The European Voter: A Comparative Study of Modern Democracies, edited

by Jacques Thomassen, 167–191. Oxford: Oxford University Press.

Van Egmond, Marcel H., Eliyahu V. Sapir, Wouter van der Brug, Sara B. Hobolt and Mark N.

Franklin. 2010. EES 2009 Voter Study Advance Release Notes. Amsterdam: University of

Amsterdam.

Várnagy, Réka. 2010. ‘Hungary’. European Journal of Political Research 49 (7–8): 1001–1008.

Verney, Susannah. 2015. ‘Waking the ‘sleeping giant’ or expressing domestic dissent? Main-

streaming Euroscepticism in crisis-stricken Greece’. International Political Science Review

36 (3): 279–295.

260 REFERENCES

Weschle, Simon. 2014. ‘Two types of economic voting: How economic conditions jointly affect

vote choice and turnout’. Electoral Studies 34 (1): 39–53.

Whitaker, Richard. 2011. ‘United Kingdom’. European Journal of Political Research 50 (7–8):

1164–1174.

Whiteley, Paul. 2016. ‘What explains electoral responses to the ‘Great Recession’ in Europe?’

Asian Journal of Comparative Politics 1 (1): 62–76.

Whitten, Guy D., and Harvey D. Palmer. 1999. ‘Cross-national analyses of economic voting’.


Wickham, Hadley. 2009. ggplot2: Elegant Graphics for Data Analysis. New York: Springer-

Verlag. http://ggplot2.org.

Wickham, Hadley, and Romain Francois. 2015. dplyr: A Grammar of Data Manipulation. R

package version 0.4.3. https://CRAN.R-project.org/package=dplyr.

Wickham, Hadley, David A. James and Seth Falcon. 2014. RSQLite: SQLite Interface for R. R

package version 1.0.0.https://CRAN.R-project.org/package=RSQLite.

Wilcox, Clyde, Lee Sigelman and Elizabeth Cook. 1989. ‘Some Like It Hot: Individual Differ-

ences in Responses to Group Feeling Thermometers’. Public Opinion Quarterly 53 (2):

246–257.

Wilkinson, Thomas, and Hornell Hart. 1950. ‘Prosperity and Political Victory’. Public Opinion

Quarterly 14 (2): 331–335.

Wlezien, Christopher. 2015. ‘The myopic voter? The economy and US presidental elections’.


Wlezien, Christopher, Mark Franklin and Daniel Twiggs. 1997. ‘Economic Perceptions and Vote

Choice: Disentangling the Endogeneity’. Political Behavior 19 (1): 7–17.

Wolfinger, Raymond E., and Steven J. Rosenstone. 1980. Who Votes? New Haven: Yale Uni-

versity Press.

Woon, Jonathon. 2012. ‘Democratic Accountability and Retrospective Voting: A Laboratory

Experiment’. American Journal of Political Science 56 (4): 913–930.

Xu, Ronghui. 2003. ‘Measuring explained variation in linear mixed effects models’. Statistics

in Medicine 22 (22): 3527–3541.

REFERENCES 261

Zeglovits, Eva, and Sylvia Kritzinger. 2014. ‘New Attempts to Reduce Overreporting of Voter

Turnouts and Their Effects’. International Journal of Public Opinion Research 26 (2): 224–

234.

Economic voting and the Great Recession in Europe · 2020. 2. 5. · Abstract The Great Recession of 2007–09 was the worst global economic crisis since the Great Depres-sion of

Documents