Download The

THE UNEQUAL DESCRIPTIVE AND SUBSTANTIVE REPRESENTATION OF CLASS

Alexander Vincent Hemingway

B.A., Simon Fraser University, 2008

MSc, London School of Economics, 2009

MSc, London School of Economics, 2010

A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF

THE REQUIREMENTS FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY

THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES

(Political Science)

THE UNIVERSITY OF BRITISH COLUMBIA

(Vancouver)

December 2020

The unequal descriptive and substantive representation of class

submitted by Alexander Vincent Hemingway in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Political Science

Examining Committee:

Alan Jacobs, Political Science, UBC Supervisor

Richard Johnston, Political Science, UBC Supervisory Committee Member

Cesi Cruz, Political Science and Vancouver School of Economics, UBC Supervisory Committee Member

Fred Cutler, Political Science, UBC University Examiner

Sylvia Fuller, Sociology, UBC University Examiner

Abstract

This dissertation seeks to contribute to our understanding of unequal class representation among legislators in advanced democracies. In particular, the consequences of unequal class representation are examined quantitatively at the levels of both individual legislators and policy outcomes. This builds on a small number of recent studies showing that class representation does indeed matter, contrary to an earlier wave of literature and assumptions on this question. In addition, the possible causes of and potential solutions to the underrepresentation of working- class people are also explored.

Paper 1 studies the relationship between legislators’ class and their attitudes and self-reported behaviour, drawing on existing survey data from 15 countries including 73 national and subnational parliaments in Europe and Israel. The results show that legislators from business backgrounds are more likely to support income inequality and small government, as well as less likely to consult with labour groups, than those from working-class and other backgrounds. An exploratory analysis also suggests that these class-based differences between legislators may vary across different institutional contexts.

Paper 2 examines the relationship between the share of working-class representatives on Finnish municipal councils and the levels of social spending in those municipalities. Using an instrumental variables approach to exploit as-if-random variation in close elections, the analysis shows that a higher share of workers on these municipal councils is associated with higher levels of social spending. This represents one of the only studies showing that the effect of class carries through to policy outcomes.

iii

Paper 3 looks further back in the causal chain to explore the possible barriers to working-class people taking office, reviewing and analyzing the sparse literature on the topic and employing an exploratory analysis of the data sets used in the first paper to probe for further evidence. The paper also examines possible solutions and interventions that could help increase working-class representation. While recent research has examined these questions in some depth in the US case, this paper considers how we would expect barriers and solutions to vary across social and institutional contexts.

Lay Summary

This dissertation is composed of three papers that seek to contribute to our understanding of both the consequences and causes of unequal class representation among legislators in advanced democracies. Paper 1 studies the relationship between legislators’ class and their attitudes and behaviour, drawing on existing survey data from 15 mainly European countries. The results show that legislators from business backgrounds are more likely to support income inequality and small government, as well as less likely to consult with labour groups, than those from working-class and other backgrounds. Paper 2 uses more advanced quantitative methods to show that higher shares of working-class representatives on Finnish municipal councils are linked to higher levels of social spending. Paper 3 explores the barriers to working-class people taking office in the first place, as well as solutions and interventions that could help increase their representation, including how these barriers and solutions may vary across countries.

Preface

This dissertation is my own original, independent work. The quantitative analyses make use of a range of existing and newly compiled data sets, as cited and indicated in each paper.

A version of paper 1 has been accepted for publication in the peer-reviewed academic journal,

Government and Opposition. The initial online citation is:

Hemingway, A. (2020). Does Class Shape Legislators’ Approach to Inequality and

Economic Policy? A Comparative View. Government and Opposition, 1-24.

Table of Contents

Abstract ...... iii

Lay Summary ...... v

Preface ...... vi

Table of Contents ...... vii

List of Tables ...... x

List of Figures ...... xiv

Acknowledgements ...... xv

Dedication ...... xvi

Introduction ...... 1

Paper 1 – Does class shape legislators’ approach to inequality and economic policy? A comparative view ...... 5

1 Introduction ...... 5

2 Descriptive and substantive representation of class ...... 8

3 Research design ...... 12

3.1 Dependent variables ...... 13

3.2 Explanatory variables...... 15

3.3 Model specification ...... 17

4 Empirical analysis ...... 18

4.1 Legislators’ attitudes ...... 19

4.2 Legislators’ behaviour ...... 24

5 Electoral institutions as moderators of the influence of legislators’ class ...... 27

6 Alternative explanations and additional analysis ...... 33 vii

6.1 Political culture of a country or region ...... 34

6.2 Constituents’ preferences at the district level ...... 35

6.3 Alternative outcome variables ...... 37

6.4 Subsample of experienced and influential MPs ...... 38

6.5 Testing using a second data set: Comparative Candidate Survey...... 39

6.6 Additional robustness checks ...... 44

7 Conclusion ...... 46

Paper 2 – The effect of class representation on social spending in Finnish municipalities ...50

1 Introduction ...... 50

2 Linking the descriptive representation of class to substantive policy outcomes ...... 53

3 Empirical strategy ...... 56

3.1 Data and variables ...... 56

3.2 Identification strategy and model...... 58

4 Validity tests ...... 63

5 Results ...... 65

6 Robustness ...... 69

7 Channels of influence ...... 73

8 Conclusion ...... 76

Paper 3 – A comparative analysis and exploration of barriers to working-class representation ...... 78

1 Introduction ...... 78

2 Baseline levels of working-class representation ...... 79

3 Stages of selection and election ...... 82 viii

4 Barriers to working-class representation in comparative perspective ...... 85

4.1 Election stage and the role of voter attitudes ...... 88

4.1.1 Moderators ...... 93

4.2 Exploratory data analysis: election stage ...... 96

4.2.1 Moderators ...... 97

4.3 Self-selection and party recruitment and nomination factors ...... 101

4.3.1 Moderators ...... 105

4.4 Exploratory data analysis: self-selection and recruitment ...... 108

4.4.1 Moderators ...... 109

4.5 Disadvantages in material resources and time ...... 112

4.5.1 Moderators ...... 116

4.6 Exploratory data analysis: resource and time disadvantages ...... 118

4.6.1 Moderators ...... 122

5 Policy solutions and interventions ...... 125

5.1 Addressing resource disadvantages ...... 126

5.2 Recruiting workers to candidacy and office ...... 132

6 Conclusion ...... 137

Conclusion ...... 139

References ...... 143

Appendix A: Supporting materials to Paper 1 ...... 154

Appendix B: Supporting materials to Paper 2 ...... 187

Appendix C: Supporting materials to Paper 3 ...... 191

List of Tables

Table 1.1. Regression models on legislators’ attitudes ...... 20

Table 1.2. Regression models on legislators’ contact with workers’ organizations ...... 26

Table 2.1. Covariate balance tests...... 65

Table 2.2. Results for social expenditures: OLS and IV analysis with ε = 0 ...... 66

Table 2.3. Results for social expenditures: IV analysis with ε = 0.4 ...... 68

Table 2.4. Effect on contemporary vs. lagged social spending: IV analysis with ε = 0.4 ...... 72

Table 2.5. Results for social expenditures by council size ...... 76

Table 3.1. Summary of barriers to working-class descriptive representation by stage ...... 85

Table 3.2. Election-stage barriers to working-class representation: summary ...... 101

Table 3.3. Self-selection, recruitment barriers to working-class representation: summary ...... 112

Table 3.4. Resource barriers to working-class representation: summary ...... 125

Table A.1. Occupation categories by country ...... 155

Table A.2. Occupation categories by party type ...... 156

Table A.3. Occupational categories (reproduced and adapted from Carnes and Lupu 2014) .... 157

Table A.4. Regression models relating legislators’ attitudes (full results including controls) ... 159

Table A.5. Regression models on legislators' contact with workers' organizations (full results)161

Table A.6. Regression models on legislators' contact with workers’ organizations (workers as omitted reference category)...... 163

Table A.7. Regression models on legislators’ attitudes and behaviour, with parliamentary political culture controls ...... 165

Table A.8. Regression models on legislators’ attitudes, with average electorate attitude controls

...... 167 x

Table A.9. Regression models on legislators’ attitudes, with low district magnitudes excluded 169

Table A.10. Regression models on placebo attitude questions and alternative DV: full results 171

Table A.11. Regression models on legislators’ attitudes and behaviour, subsample of influential

MPs ...... 173

Table A.12. Regression models on candidates’ attitudes (CCS data) ...... 175

Table A.13. Regression models on legislators’ attitudes and behaviour, using ordered logit .... 177

Table A.14. Regression models on legislators' contact with workers’ organizations, union staff separated ...... 179

Table A.15. The relationship between class and inequality attitudes, conditioned by institution

...... 181

Table A.16. The relationship between class and contact with workers’ organizations, conditioned by institution ...... 184

Table B.1. Statistics Finland municipal expenditure categories, 2016 ...... 188

Table B.2. Robustness models: year as unit of analysis, non-social spending, and placebo thresholds ...... 190

Table B.3. Results for social expenditures: next largest party and whole council ...... 190

Table C.1. T-test of working-class share by elected status (CCS) ...... 191

Table C.2. T-test of working-class share by elected status (CCS; right parties only) ...... 191

Table C.3. T-test of working-class share by elected status (CCS; left parties only) ...... 191

Table C.4. T-test of working-class share by electoral institution (PARTIREP) ...... 191

Table C.5. T-test of working-class share by electoral institution (PARTIREP) ...... 191

Table C.6. T-test of working-class share by country polarization level (PARTIREP) ...... 191

Table C.7. T-test of worker share by selectorate scope (CCS) ...... 191 xi

Table C.8. T-test of party connections (past employment) by class (CCS) ...... 192

Table C.9. T-test of level of contestation of nomination by class (CCS) ...... 192

Table C.10. T-test of self-assessed chances of winning by class (CCS) ...... 192

Table C.11. T-test of working-class share by party type (PARTIREP) ...... 192

Table C.12. T-test of worker share by party type (CCS; elected only) ...... 192

Table C.13. T-test of worker share by party type (CCS) ...... 192

Table C.14. T-test of working-class share by country union density level (PARTIREP) ...... 192

Table C.15. T-test of worker share by country union density (CCS) ...... 192

Table C.16. T-test of union membership by class (CCS) ...... 193

Table C.17. T-test of elected status by union membership (CCS)...... 193

Table C.18. T-test of elected status by union membership (CCS; workers only) ...... 193

Table C.19. T-test of elected status by union membership (CCS; non-workers) ...... 193

Table C.20. T-test of campaign starting time by class (CCS) ...... 193

Table C.21. T-test of campaign starting time by class (CCS; elected only) ...... 193

Table C.22. T-test of full-time campaign starting time by class (CCS) ...... 193

Table C.23. T-test of full-time campaign starting time by class (CCS; elected only) ...... 194

Table C.24. T-test of campaign budget by class (CCS) ...... 194

Table C.25. T-test of campaign budget by class (CCS; elected only) ...... 194

Table C.26. Regression model relating class to campaign budgets (CCS) ...... 195

Table C.27. T-test of campaign staff size by class (CCS) ...... 196

Table C.28. T-test of campaign staff size by class (CCS; elected only) ...... 196

Table C.29. T-test of party-provided campaign staff size by class (CCS) ...... 196

Table C.30. T-test of party-provided campaign staff size by class (CCS; elected only) ...... 196 xii

Table C.31. T-test of worker share by country campaign expense (CCS) ...... 196

Table C.32. T-test of working-class share by elected status (PARTIREP) ...... 196

Table C.33. T-test of working-class share by country P90/P10 inequality level (PARTIREP) . 197

Table C.34. T-test of worker share by country P90/P10 inequality (CCS; elected only) ...... 197

Table C.35. T-test of worker share by country P90/P10 inequality (CCS) ...... 197

xiii

List of Figures

Figure 1.1. Estimated class-based differences in legislators’ attitudes ...... 21

Figure 1.2. Class-based differences in inequality attitudes, by institution ...... 31

Figure 1.3. Class-based differences in contact with workers’ groups, by institution ...... 33

Figure 1.4. Estimated class-based differences in candidate attitudes (CCS data) ...... 41

Figure 1.5. Estimated class-based differences in candidate attitudes (CCS data) ...... 43

Figure 2.1. Effect of worker representation on social spending ...... 70

Figure 3.1. Worker share in right parties ...... 98

Figure 3.2. Estimated class-based differences in campaign budgets (Euros) ...... 120

Figure 3.3. Working-class share by level of government ...... 123

Figure A.1. Distribution of outcome variable responses ...... 154

Figure B.1. Distribution of instrumental variable values at different bandwidths ...... 187

xiv

Acknowledgements

I am grateful to my supervisor, Alan Jacobs, and committee members, Richard Johnston and

Cesi Cruz. Their guidance, insight and encouragement over the years were indispensable. A supervisor and committee can make or break the experience of a doctoral program, and I quickly realized that I had hit the lottery with mine. Thank you to my doctoral cohort, the political science faculty, and the Institute of European Studies’ Doctoral Fellows for stimulating discussions and advice, as well as to our graduate program assistant, Josephine Calazan, and the department more broadly, for shepherding us through the many administrative steps of a doctoral program. Thank you to Nicholas Carnes, who I have never met, but whose research helped motivate me to continue in political science and whose work I aimed to build upon in this dissertation. I am grateful to Grace Lore for blazing a path in her doctoral research on the descriptive and substantive representation of women, which helped inspire part of the structure of the first paper. Thanks to Shannon Hogan for excellent research assistance, Tomi Ihalainen for advice on Finnish translations, and Neil Lloyd for helping me solve a sticky data set problem.

Thank you to Alastair Fraser for your friendship, advice and support. I could not have completed this journey without my friends and family, most of all Molly Henry, Dawn Hemingway and

Peter Ewart. I am lucky to stand together with you in trying make a better world, as we must.

Dedication

To the memory of my great grandfather, Vincent Spies Segur, who taught his children to “never forget the working class.”

And to my family in its many branches, Segur, Carrell, Ewart and Hemingway, who never did.

xvi

Introduction

The unequal representation of class among legislators has long been recognized in political science. Citing an extensive literature on this point, Matthews (1984) notes:

“Almost everywhere legislators are better educated, possess higher status occupations and have more privileged backgrounds than the people they ‘represent’… Few generalizations have been more exhaustively supported by empirical research” (p. 548).

If anything, these patterns appear to have grown starker over time, including a decline in the representation of blue-collar workers in legislative office in many advanced democracies.

Until recently, the unequal descriptive representation of class tended to be characterized in the literature as having little consequence on the substantive outcomes of representation. However, in the past several years, these issues have received renewed attention sparked by the US-focused work of Nicholas Carnes (2013). This renewed interest has come amid the increasing salience of other class-linked phenomena, most prominently the rise of economic inequality across much of the developed world. At the same time, increasing attention has been paid to the inequality of political influence, with growing evidence that policy outcomes disproportionately reflect the views of affluent citizens and less often reflect those on lower incomes where the preferences of these groups diverge.

Alongside these phenomena, a long-running decline in trust in political institutions and political classes has also been observed over the past few decades. In the post-2008 period, we appear to be witnessing the emergence of an increasingly combustible politics of inequality and identity.

Alarmingly, these growing problems of inequality, democratic malaise and instability come at a

time when effective collective action is needed more urgently than ever to address enormous challenges like climate change.

In this context, the papers that follow seek to contribute to our understanding of unequal class representation. First and foremost, this dissertation examines the consequences of unequal class representation quantitatively at the levels of both individual legislators and aggregate policy outcomes. In doing so, it builds on a small number of recent studies showing that legislative class representation does matter, contrary to an earlier wave of literature and assumptions on this question. In addition, the potential causes of and solutions to the underrepresentation of working- class people are examined.

Paper 1 studies the relationship between legislators’ class and their attitudes and self-reported behaviour in advanced democracies, drawing on existing survey data from 15 countries including

73 national and subnational parliaments in Europe and Israel. Using cross-sectional regression models with fixed effects, I find that legislators from business backgrounds are more likely to support income inequality and small government, as well as less likely to consult with labour groups, than those from working-class and other backgrounds. These results are buttressed by analysis of an additional cross-national survey of European legislative candidates’ attitudes, which replicates key findings, along with a battery of robustness tests using a variety of econometric approaches. An exploratory analysis also examines whether the relationship between the descriptive and substantive representation of class is moderated by electoral institutions that create different incentives for legislators. This appears to be the case, with open list and division of powers systems associated with larger class-based differences in legislators’

attitudes and self-reported behaviour. In short, the results show both that class representation matters and that its importance may vary across institutions.

Paper 2 further examines the consequences of class representation and differs from the first paper in some important ways. While the previous paper covers a broad scope of jurisdictions with correlational evidence, paper 2 is able to employ a much stronger research design amenable to making causal inferences, but with the empirical scope confined to a single country. Paper 2 uses an instrumental variables approach to exploit as-if-random variation in close elections.

Specifically, the paper examines the relationship between the share of working-class representatives sitting on Finnish municipal councils and the levels of social spending in those municipalities. The findings show that a higher share of workers on the municipal councils appears to drive higher levels of social spending. Notably, the paper is examining aggregate policy outcomes that result from differences in the class composition of an elected body, whereas the previous paper studied the relationship between legislators’ class and their attitudes and behaviour at an individual level. Paper 2 represents one of the only studies (see also, Carnes

[2013]) that shows the substantive effects of legislators’ class carry through to policy outcomes.

Finally, given low levels of working-class representation in most jurisdictions, Paper 3 explores the possible barriers to workers taking office in the first place, as well as examines potential solutions to increasing their levels of representation. The paper reviews and analyzes the sparse literature on the topic and employs an exploratory analysis of the data sets from the first paper of this dissertation to probe for evidence that can help guide future research in this area. Informed by the literature and exploratory analysis, Paper 3 then considers possible solutions and interventions that could help increase working-class representation. While Carnes (2018) has

examined these questions in some depth in the US case, this paper considers how we would expect barriers and solutions to vary across contexts, particularly since the US is an outlier in terms of low levels of working-class representation and a political economy characterized by extreme inequality and economic insecurity.

Taken as a whole, this dissertation makes several contributions. First, it extends and expands upon the emerging literature on class representation, showing that the class backgrounds of legislators matter both at an individual level and in terms of aggregate policy outcomes. The dissertation shows that the effect of legislators’ class can be seen across a much broader range of advanced democracies than had previously been demonstrated in the literature, and it adds a novel exploratory analysis examining potential institutional moderators of the class effect. In doing so, it also contributes to the broader literature linking descriptive and substantive representation, which has more often addressed the representation of women and racialized people. In addition, the findings speak to the burgeoning literature on economic inequality and the inequality of political influence. Given the skewed class makeup of legislatures in advanced democracies, the findings may be relevant to our understanding of widespread economic and political inequalities that are increasingly salient in many countries.

However modestly, I hope this dissertation can speak to the urgent moment we are in globally, helping to shed light on growing political convulsions in which questions of inequality, class and identity play an important role. In turn, I hope this research can help to inform potential paths out of these difficulties and towards politics that are more representative, democratic, and ultimately capable of grappling with the huge collective challenges that humanity faces today and in the coming decades.

Paper 1 – Does class shape legislators’ approach to inequality and economic policy? A comparative view1

1 Introduction

Legislatures across the developed world have long failed to reflect the class makeup of the societies they represent (Best, 2007; Best & Cotta, 2000; Norris 1997). The affluent tend to be overrepresented in office, and the descriptive representation of class has also seen shifts over time, including a decline in blue-collar workers and a rise in the professionalization of electoral politics (Best, 2007; Best & Cotta 2000; Evans & Tilley, 2017; Gaxie, 2017; Norris 1997). Could unequal class representation help account for widening income and wealth gaps, and moves in the political agenda away from redistributive questions and towards social issues?

This article examines whether the class backgrounds of legislators shape their views and self- reported behaviour relating to inequality and economic policy while in office, covering a range of developed countries. While the skewed descriptive representation of class has been acknowledged in the literature for decades, the substantive consequences have received much less research attention until recently. There is growing evidence that legislators from different class backgrounds display distinct attitudes and behaviour in office.

Carnes (2013) shows stark evidence that legislators’ class backgrounds are of importance in

American politics. US politicians from working-class occupations have significantly more left-

1 A version of this paper has been accepted for publication in the peer-reviewed academic journal, Government and Opposition (Hemingway, 2020). 5

wing economic attitudes and legislative voting records than their counterparts from business and professional occupations (see also, Grumbach, 2015).2 This work has sparked a new wave of research on the topic. O’Grady (2018) demonstrates a relationship between class and the policy positions of Members of Parliament in the UK Labour Party. Rosset (2016) shows that among

Swiss legislative candidates, higher incomes are associated with lower support for redistribution.

In the context of the developing world, Carnes and Lupu (2014) show a similar link between legislators’ class and their attitudes and behaviour in a study of 18 Latin American countries.

These recent findings run contrary to an earlier wave of research that rejected or downplayed the substantive importance of legislators’ class (Matthews, 1984; Norris & Lovenduski, 1995).

Research on whether the class effect is a phenomenon that extends broadly across the developed world is still needed. As O’Grady (2018) notes, we still know little about class-based differences in the attitudes and behaviours of European legislators in particular.3

In this article, I use unique survey data covering 73 national and subnational parliaments in

Europe and Israel to examine how the class backgrounds of legislators relate to their representation of redistributive and economic issues. This adds to the emerging literature a rare quantitative study of multiple developed countries examining the substantive implications of legislators’ class. The results show that business sector legislators are substantially more favourable to income inequality and small government than legislators from working-class and other backgrounds, even after controlling for their party’s ideology. These conclusions are

2 In the US context, Grumbach (2015) also finds a relationship between the class backgrounds of legislators’ parents and their roll-call votes. 3 Another study examining occupational background is Hyytinnen et al. (2018), which finds that Finnish municipal governments with more councillors from the public sector tend to have higher levels of expenditure. The focus of that research, though, is on rent-seeking behaviour rather than class. 6

buttressed by analysis of an additional cross-national survey of European legislative candidates’ attitudes, which replicates the pattern.

In addition, the results show that legislators from different classes report behaving differently in office, with business sector legislators less likely to be in contact with workers’ organizations and trade unions than legislators from working-class backgrounds (specifically, contact in their roles as MPs). In short, the results show that class representation matters across a wide range of advanced democracies, reinforcing and extending the new wave of literature on this topic. An exploratory analysis also examines institutional moderators of the influence of class.

This article contributes to the literature on descriptive representation, which is concerned with how the personal identities of legislators reflect the broader society being represented (Carnes,

2013; Lore, 2016; Mansbridge, 1999; O’Grady, 2018; Pitkin, 1967). The findings help confirm that, across a range of developed countries, class is an important identity among legislators that predicts their attitudes and behaviour in office. The findings also speak to the broader literature on economic inequality, suggesting that shifts in class representation should be considered among the possible causes of rising income and wealth gaps.

In the remainder of the article, the following section introduces the literature on class and descriptive representation and theorizes how legislators’ class would be expected to affect their attitudes and behaviours. I then go on to describe the data and methods, followed by a presentation of the main empirical results and a consideration of the possible institutional moderators. The next section examines and tests alternative explanations, replicates core findings using a second data set and outlines additional robustness checks. The final part concludes and outlines priorities for future research. 7

2 Descriptive and substantive representation of class

In a time of widespread economic and political inequality (Bartels, 2017; Bernauer et al., 2015;

Donnelly & Lefkofridi, 2014; Giger et al., 2012; Gilens & Page, 2014; Hacker & Pierson, 2010;

Piketty, 2014), important questions arise about not only who pressures and influences politicians, but also who the politicians are – that is, “who governs?” (Dahl, 1961; see also, Carnes, 2013).

The unequal descriptive representation of class among legislators has long been recognized as a widespread phenomenon, with legislators tending to come from privileged backgrounds compared to the broader populations in their polities (Best, 2007; Best & Cotta, 2000; Matthews,

1984; Norris 1997). Power elite theorists (Lindblom, 1977; Mills, 1956) once highlighted the social background and networks of politicians, with the implication that these factors would affect policy outcomes. But, as Carnes notes, they “never tested this possibility systematically, and in the absence of any hard evidence, political scientists tended to side with the pluralists”

(2013, p. 11). To the extent that politicians are office-seeking actors pursuing the median voter within the constraints of disciplined political parties, it was plausible to conclude that legislators’ personal backgrounds should be irrelevant to their policy positions.

Among the relatively few to address class representation in the intervening period, Matthews

(1984) reviewed the comparative literature and concluded there was no convincing effect of legislators’ class on representation. Norris and Lovenduski (1995) also found no effect of class representation in the UK context, and Wessels (1997) and Essaiasson and Holmberg (1996) found only very weak links between legislators’ class and attitudes in Germany and Sweden.

This literature was characterized by some shortcomings, typically looking for effects on broad

left-right orientation, rather than focusing on the most class-relevant dimensions of economic and redistributive policy.

Recently, there has been a resurgence of research on the issue, showing that the class backgrounds of legislators’ do, in fact, have important effects on their policy attitudes and behaviour in the cases of the United States and Latin America (Carnes, 2013; Carnes & Lupu

2014; Grumbach, 2015). We also have evidence that legislators’ occupations affect their attitudes and behaviour in the British Labour Party (O’Grady, 2018), and Swiss legislative candidates with lower incomes have more favourable attitudes to redistribution than their higher-income counterparts (Rosset, 2016).

There are good theoretical reasons to expect legislators’ class to matter. A person’s occupational class corresponds to a distinct set of material conditions and interests, including levels of income and employment security (Evans & Tilley, 2017; Meltzer & Richard, 1981; Rehm, 2011). These differences in material conditions can be seen in objective measures like income inequality and employment histories, and, importantly, inequalities are also perceived and observed subjectively by people from different classes (although they often underestimate the magnitude of material inequality compared to the reality; Evans & Tilley 2017). Some literature has suggested a decline in the importance of class relative to other political cleavages and social identities (Eidlin, 2014;

Inglehart, 1997; Savage et al., 2001), but other analyses show that people across a wide range of countries continue to identify themselves and others in class terms (Andersen & Curtis, 2012;

Curtis, 2016; Evans & Tilley, 2017; Hout 2008).

Beyond reflecting their material interests and identities, occupation can influence people’s views by helping to shape their social circles. An individual’s own experiences and views can be 9

reinforced by “repeated interaction with people that share similar backgrounds and material interests” in the workplace (O’Grady, 2018, p. 7). In turn, the workplace can be a bridge to involvement in other organizations with further socializing effects, such as trade unions and professional associations (Manza & Brooks, 2008). There is an extensive literature showing that people from different economic classes do, in fact, have distinct views and preferences, especially on economic policy issues (Campbell et al., 1960; Evans & Tilley, 2017; Gilens, 2012;

Hout, 2008; Iversen & Soskice, 2001; Manza et al.,1995; Page et al., 2013; Rosset, 2013).

Should we expect class-based differences among the general population to carry over to differences in the views and behaviour of legislators? On the one hand, legislators face a unique set of homogenizing pressures in their roles as politicians, including the authority of party leaders, socialization from co-partisans, and the need to court and respond to an electoral base.

However, while partly constrained by voters and parties, legislators do have room to manoeuvre in office (Bawn et al., 2012; Jacobs & Shapiro, 2000), and they often act based on their own personal views (Burden, 2007; Levitt, 1996). As described, some recent research shows the importance of class carrying over to legislators (Carnes, 2013; Carnes & Lupu 2014; O’Grady

2018), notwithstanding the scepticism of an earlier wave of work (Matthews, 1984; Norris &

Lovenduski 1995). Furthermore, while one might expect the influence of a legislator’s prior occupation to fade over long political careers, Carnes (2013) found little evidence for this in the

US.

In empirically studying class-based differences in the attitudes and behaviour of legislators, one important distinction to draw is between intrinsic effects of class (as a lived experience and identity) and the consequences of partisan sorting. For those individuals who become involved in

electoral politics, their pre-existing beliefs, in part shaped by occupational class, will help sort them into left- or right-wing parties. Their social networks, which as discussed above are influenced by their class, will also likely help sort them into parties. For example, given the historical links between many left-wing parties and organized labour, we expect working-class legislators to be more likely to be recruited into parties of the left. Conversely, a businessperson is more likely to be recruited into a right-wing party. Once sorted into a party, participating in it would constitute an additional stage of personal socialization for a legislator, further shaping their political beliefs (as a result of spending time with co-partisans, and adapting oneself to the demands of party leaders, gatekeepers, and the electoral base).

In its theoretical aims, this paper is primarily interested in the effects of class that originate in legislators’ class identity and lived experiences, as opposed to socializing effects that take place after joining a political party. Therefore, the central hypothesis is whether legislators’ attitudes and behaviour on redistributive and economic issues will depend on their class, after controlling for factors like their party’s ideological orientation. Specifically, I expect business sector legislators will exhibit attitudes and behaviour less favourable to solidaristic policies than working-class MPs and those from other occupations. However, insofar as legislators have sorted into parties because of their class-driven pre-existing beliefs, controlling for party type means the party control variable will capture some of the ‘real’ class effect. As a result, the main estimates of the class coefficient presented below, which do control for party type, can be taken as conservative (i.e., as a possible lower bound estimate).

3 Research design

The main data used in this paper are from the PARTIREP research project’s Comparative MP

Survey (Deschouwer et al., 2014) conducted in 2009 and 2010. The survey includes over 2000

MPs across 15 countries and 73 different national and subnational parliaments within them. The countries surveyed are Austria, Belgium, France, Germany, Hungary, Ireland, Israel, Italy,

Netherlands, Norway, Poland, Portugal, Spain, Switzerland, and United Kingdom. Legislators responded to a range of questions about their roles as MPs, as well as ideology and issue attitudes, among other items. Approximately one in four legislators responded to the survey, ranging between a high of 43% in the Netherlands and a low of 12% in Poland. Deschouwer and colleagues (2014) find that the respondents are reasonably representative of the population of legislators surveyed in terms of measures such as gender and whether they are in opposition or government parties. Case selection for the present paper is driven by the aim to investigate the impact of class representation in multiple developed countries, where evidence remains relatively limited to date. Practically, it is shaped by the countries covered in the available Comparative

MP Survey data. Fortunately, the survey covers a wide range of European countries and Israel, and the data set also includes detailed information about the varying features of electoral institutions across the national and subnational jurisdictions. In the exploratory analysis described below, these varying institutional features are considered as potential moderators of the relationship examined in the main models.

3.1 Dependent variables

In this analysis, the main dependent variables are attitudinal and self-reported behavioural measures. The three dependent variables used are the only measures directly relating to economic and redistributive issues in the survey. They are:

1) Inequality attitude: MPs rate agreement (on a 1–5 scale from strongly disagree to

strongly agree) with the statement: “Larger income differences are needed as incentives

for individual effort.”

2) Role of government in the economy attitude: MPs rate agreement (on the same scale) with

the statement: “Government should play a smaller role in the management of the

economy.”

3) Contact with workers’ organizations: MPs answer the question, “In your role as a

Member of Parliament, how often in the last year have you had contact with … workers’

organizations and trade unions” (on a 1–5 scale ranging from “(almost) no contact” to “at

least once a week”). Note that the question specifies that it is about contact in their role

as MPs and so should exclude contact in social or other capacities, if legislators observe

this caveat in their responses.

Both legislators’ attitudes and their behaviour merit attention. Previous research indicates that legislators’ attitudes may be a key link in the causal chain connecting class and policy outcomes.

For example, when US legislators’ opinions were added as a control variable in Carnes’ (2013) regression models, the substantial class-based differences in Congressional voting records disappeared. Furthermore, the practical importance of the attitude measures is reinforced by

legislators’ responses to two relevant questions in the Comparative MP Survey data. First, nearly

70% of legislators in the sample say that if an MP’s personal opinions conflict with those of voters, that MP should follow his or her personal opinion. Similarly, 47% say an MP’s own opinion should also take precedence over their party’s position. Thus, a large share of the surveyed legislators apparently believe that their own opinions are important in their role as

MPs, even trumping those of their voters and parties. At the same time, it should be acknowledged that the variables on attitudes described above each rely on responses to a single question, using a simple five-point scale from strongly disagree to strongly agree, which cannot fully capture the nuance of respondents’ attitudes on these questions.4 In turn, an important limitation of the behaviour measure is that it is self-reported rather than directly observed.

However, respondents remain anonymous in the survey, meaning there is little incentive to misreport intentionally. The findings on this self-reported behaviour measure complement more direct measures in the literature.

Since more experienced and powerful MPs are more likely to influence party positions and policy outcomes (and therefore their attitudes and behaviour may be of more practical significance), I also run additional versions of my main regression models on relevant subsamples of such legislators (e.g., those who have previously held elected office, sponsored at least one bill, or held at least one parliamentary leadership position).

4 The distribution of responses to the outcome variables are shown in Figure A.1 in the Appendix. 14

3.2 Explanatory variables

Legislators’ occupations prior to taking office are used as the operationalization of class, which is the key independent variable. Sociological literature supports defining class in terms of occupation (Manza et al., 1995; Kalmijn & Kraaykamp, 2007; Weeden & Grusky, 2005), which marks distinct perspectives and lived experiences, and this is consistent with recent research on the substantive representation of class (Carnes, 2013; Carnes & Lupu, 2014; O’Grady, 2018).

Empirical work on the US finds that occupational class predicts legislators’ voting behaviour, but family background, education, and income do so more poorly, or not at all (Carnes, 2013).

The data set includes an open-ended text field describing the legislator’s occupation, which

Deschouwer and colleagues (2014, p. 8) explain they have “collected from official sources such as the parliamentary websites and ‘who’s who’ guides” and refers to their occupation prior to election as an MP. I have coded these descriptions into a set of 10 occupational categories, adapting the categorization scheme used by Carnes (2013) and Carnes and Lupu (2014).5 For example, the “business” category includes occupations like business owners, managers, bankers and consultants, and serves as the omitted base category in the regression models described below. The “worker” category includes manual, service industry, clerical and union staff jobs.

The “service-based professional” category includes teachers, nurses, social workers and community organizers. Notably, the findings are robust to multiple checks on the coding scheme,

5 See Table A.1 in the Appendix for a breakdown of occupation observations by country. More details on the coding procedure are described alongside Table A.2 and Table A.3. 15

including replication in a second data set with differently structured occupation data, described in subsection 6.5 below.6

There is a correlation between class and party type, but one that is far from perfect. Workers tend to be found in left parties, but they also have a substantial presence in right parties. Conversely, businesspeople are found in left parties, though they are predominantly in right parties.7 This variation of class within party types provides an opportunity to separate empirically the influence of class from that of party.

Using a similar categorization scheme, Carnes (2013) and Carnes and Lupu (2014) found that business sector legislators had among the most right-wing attitudes and behaviours on redistributive issues, while those from working-class occupations were the most left-wing, followed by service-based professionals.

In terms of controls, a key control variable indicates whether the legislator is a member of a left- leaning political party. This is based on an expert classification of parties in the PARTIREP data set, which includes 12 categories that I have collapsed into a single “left party” dummy.8 As a robustness check, I also used an alternative left party control (based on MPs’ self-reports about their party’s ideological left–right position) both in dummy and continuous form, and the results are substantively similar. Other legislator-level control variables available in the data set include

6 In the secondary data set, this includes running models that use the provided top-level International Standard Classification of Occupations (ISCO) codes. Further robustness checks are discussed below, including models that drop ambiguous or difficult-to-code cases, as well as separating union staff from the main “workers” category. 7 The full breakdown of observations by occupation category and party type can be found in Table A.2. 8 The left party dummy is equal to 1 for the following party types: socialist, communist, social democratic and green; it is equal to 0 for these party types: Christian democratic, conservative, liberal, far-right, ethnic or regionalist, agrarian, single issue, religious, other. 16

age, gender and education. Education is not included as a control variable in the main model specifications but is included in later robustness checks.9 Age and gender are included as the available personal-level controls in the data set and because of their plausible correlation with the explanatory and outcome variables (and neither would be considered post-treatment variables).

Other control variables used include dummies for each country and parliament.

3.3 Model specification

To assess the main relationship between the descriptive and substantive representation of class, I estimate linear regression models of the form:

OutcomeVariableij = β1–10 Occupationij + β11(LeftPartyij) + β12 (Femaleij) + β13 (Ageij) + γk + ϵij where Occupation refers to the occupation category of the legislator i in parliament j (which includes both national and subnational parliaments). LeftParty indicates membership in an identifiably left-leaning political party. Female and Age refer to the gender and age of the politician, respectively. Country fixed effects are represented by γ and ϵ is the error term.

Standard errors account for potential correlations within parliaments (clustered at the parliament level). The vector of individual-level outcome variables includes: attitude on inequality, attitude on government role in the economy, and self-reported contact with workers’ organizations.

9 This education variable has a high number of missing cases and is in any case not statistically significant in the models. The education variable included in the data set distinguishes between three levels: “primary and/or secondary education”, “non-university higher education” and “university”. 17

For each outcome variable, results are shown for three specifications: 1) a simple regression of occupation on the outcome variables without controls (but with country fixed effects and standard errors clustered by parliament); 2) a model with control variables (the baseline specification); and 3) a model with controls and that also weights observations to equalize the influence of countries and regional/national parliaments.10 In each model, the business occupation category is the omitted reference category for the independent variable.11 Note again that controlling for party type means that the coefficient for legislators’ class should be taken as a lower bound estimate, since politicians with different occupations already tend to sort into different types of parties.

Because the dependent variable is strictly speaking ordered rather than continuous, ordered logistic regression models have also been run as a robustness check. Additional specifications, including a multilevel model with legislators nested within parliaments, were run and give substantively similar results.12

4 Empirical analysis

How do legislators’ economic classes relate to their attitudes and self-reported behaviour in the surveyed parliaments? The results show that legislators’ classes empirically predict (at statistically significant levels) their attitudes on inequality and the role of government in the

10 Since, for example, Switzerland and its many regional canton legislatures are overrepresented in the data set. This issue is explored further in the robustness checks below. 11 There are good theoretical reasons to shine a light on economic elites (Hacker and Pierson, 2010), and more recent work (Page et al., 2013) shows that they have clearly distinct economic policy preferences. 12 More on robustness checks below. 18

economy, as well as their levels of contact with workers’ organizations. Moreover, the coefficient sizes are substantively important. For example, the coefficient size for being a legislator from a business background compared to a worker is in some cases up to 50% that of being in a right- versus left-wing party. The coefficient sizes are remarkable, given that interparty competition is where class conflict is typically thought to play out.

Working-class legislators and service-based professionals (which includes occupations like teachers, nurses, social workers) are the categories with the largest coefficients and are distinct from business-class legislators at statistically significant levels across both attitude measures.

Only working-class legislators are distinctive from their business counterparts on both the attitude and behavioural measures.

4.1 Legislators’ attitudes

Table 1.1 shows the results for each model seeking to examine the relationship between occupational class and legislators’ attitudes (separately for inequality and government intervention in the economy). As described above, these attitude variables are on a 1–5 scale, with higher values corresponding to more right-wing attitudes.

Each model shows substantively and statistically significant (p < 0.05) differences in the attitudes of business sector legislators and at least three other occupational class backgrounds.

All occupation coefficients across the models trend in the expected direction, with business sector legislators displaying attitudes on economic issues to the right of their counterparts from other backgrounds.

Table 1.1. Regression models on legislators’ attitudes

Dependent variable Inequality Smaller government role in the economy Model (1) (2) (3) (4) (5) (6) No controls Controls Controls, No controls Controls Controls, weighted weighted Business – – – – – –

Technical -0.282* -0.132 -0.068 -0.321** -0.168 -0.104 professional (0.108) (0.090) (0.0807) (0.112) (0.105) (0.124)

Farmer -0.0860 -0.263 -0.316 -0.191 -0.408** -0.456* (0.178) (0.179) (0.203) (0.148) (0.149) (0.197)

Lawyer -0.308* -0.192+ -0.191 -0.219+ -0.097 -0.137 (0.121) (0.107) (0.115) (0.123) (0.102) (0.122)

Other white-collar -0.617** -0.384** -0.303** -0.275+ -0.055 -0.030 (0.125) (0.103) (0.102) (0.161) (0.132) (0.137)

Politics/law -0.446** -0.238* -0.164 -0.419** -0.198* -0.247* enforcement (0.108) (0.098) (0.106) (0.102) (0.084) (0.108)

Civil service -0.607** -0.217+ -0.221+ -0.628** -0.197+ -0.170 (0.132) (0.121) (0.113) (0.098) (0.106) (0.117)

Service-based -0.858** -0.449** -0.337** -0.731** -0.271** -0.237* professional (0.102) (0.081) (0.078) (0.113) (0.084) (0.101)

Worker -0.695** -0.305** -0.241** -0.675** -0.251** -0.127 (0.110) (0.089) (0.088) (0.116) (0.079) (0.081)

Left party -1.158** -1.068** -1.284** -1.169** (0.127) (0.144) (0.136) (0.145)

Observations 1891 1817 1817 1894 1820 1820 Notes: Standard errors in parentheses. Higher coefficient indicates more right-wing attitude. The “no occupation information” category, country dummies and coefficients for controls (except party) are not shown here for ease of presentation. See Appendix Table A.4 for full regression results. +p < 0.10; *p < 0.05; **p < 0.01.

Results from Models 2 and 5 (with controls, unweighted) are depicted in Figure 1.1. These can be considered baseline specifications for the two attitudinal outcome variables. Since my interest is in detecting the impact of legislators’ class at an individual level (as opposed to drawing inferences about a representative set of parliaments), the unweighted models are appropriate baselines.

Figure 1.1. Estimated class-based differences in legislators’ attitudes

Note: Lower scores correspond to more economically left-wing attitudes. Dots are model coefficients and lines are 95% confidence intervals. Business sector is the omitted reference category for occupation.

First, consider the inequality attitude outcome variable. The baseline Model 2 results show negative and statistically significant (p < 0.05) coefficients for four categories, meaning each of these occupation categories displays more anti-inequality attitudes compared to legislators from 21

the business sector. The largest differences for this outcome variable are between legislators in the business sector, on the one hand, and service-based professionals, workers and “other white- collar” on the other hand.

One way to think about the substantive importance of these differences is by comparing it to the coefficient for being in a left party. This is a natural point of comparison, since political parties function as the key political aggregators of class conflict. Left-party legislators are generally expected to display greater concern about inequality and more openness to government intervention in the economy. Indeed, such economic questions are usually considered to be the most basic characteristic of the left–right political spectrum.

The results show that occupations predict differences in legislator attitudes that are on a comparable scale to differences associated with party identity. For example, a service-based professional displays significantly more opposition to inequality compared to a business sector legislator, and the coefficient size is about 40% of being in a left party. In terms of inequality attitudes, the coefficient for being a worker versus a business sector legislator is over 25% that of being in a left party.

In Model 3 (weighted), the differences between business sector legislators and the same three occupation categories remain substantively and statistically significant. This model provides assurance that the results are robust to levelling out the influence of the different parliaments in the sample.

Next, consider the second attitude outcome variable, which relates to government intervention in the economy. In Model 5 (controls, unweighted), workers and service-based professionals (as

well as farmers and those in politics and law enforcement) are substantively and significantly (p

< 0.05) more favourable to government’s role in the economy than business sector legislators.

Compared against the coefficient for party type on this attitude, the observed differences by occupational class are substantial. Left-party legislators are, as expected, more favourable to government intervention in the economy than legislators from other parties. The coefficient size of being a worker or service-based professional (as opposed to from the business sector) is about

20% that of being in a left party. For farmers, the coefficient is even larger at over 30% that of being in a left party. In Model 6 (weighted), much the same pattern of coefficients holds, though the coefficient for the worker category is reduced and loses significance.

Overall, the observed differences by legislators’ class are somewhat more pronounced regarding their attitudes on inequality than their attitudes on government’s role in the economy. This makes theoretical sense, since the issue of economic inequality is arguably more intrinsically bound up with class than government’s role in the economy (though both are clearly linked).

The results presented can be taken as lower bound estimates of the coefficient for class in an important sense. To the extent that prospective legislators sort into left and right parties because they hold beliefs shaped by their class backgrounds, the models with party controls would tend to underestimate the coefficient for class (with some of the influence of class appearing in the left party coefficient). Therefore, I also report results for Models 1 and 4, which exclude controls.

These models show much larger coefficient sizes for most occupations compared to the business sector (coefficients more than double in the case of workers) and would represent an upper bound estimate of the relationship between legislators’ class and their attitudes.

The results of Models 1 through 6 bring to bear new evidence of an empirical link between the descriptive and substantive representation of class. They are consistent with the theory that the occupational class of legislators affects their economic policy attitudes across a broad set of developed countries.

To summarize, business sector legislators tend to be more favourable to inequality and less favourable to government intervention in the economy than other occupational backgrounds and particularly compared to workers and service-based professionals. These coefficient sizes are comparable to, though smaller than, those of party type. These findings are consistent with the

US-based research of Carnes (2013, ch. 4), which showed that legislators from business backgrounds had among the most right-wing attitudes, and workers and service-based professionals had the most left-wing attitudes.

4.2 Legislators’ behaviour

The Comparative MP Survey also includes a behavioural dependent variable of interest: self- reported levels of contact with workers’ organizations and trade unions. Note that, as mentioned above, the question specifies that this refers only to contact in their role as MPs (and so should exclude contact in social or other capacities). This allows us to evaluate the prediction that legislators’ class will help shape their behaviour as representatives. Given the well-recognized role of unions in pressuring governments to adopt solidaristic and pro-worker policies (Korpi,

2006; Hooghe & Oser, 2016), contact with these organizations is of substantive importance.

In Model 8 (with controls), there is a significant relationship between legislators’ class and their contact with workers’ organizations (see Table 1.2). Specifically, business sector legislators are much less likely than workers (the only significantly different category) to report higher levels of contact with these organizations. The coefficient for party type again provides a helpful point of comparison for gauging the relative importance of class. Given the well-established links between many social democratic parties and the labour movement, we expect and observe that legislators from left parties have more contact with workers’ organizations than those from other types of parties.

The magnitude of the difference between business legislators and workers is large, amounting to a little more than 50% the coefficient for being in a left party compared to another party type. In the model with the observations weighted to equalize countries and regional/national parliaments

(Model 9), the coefficient is just under 50% of that for party, while in the no-controls specification (Model 7), the coefficient increases substantially relative to its size in the baseline model.

Table 1.2. Regression models on legislators’ contact with workers’ organizations

(7) (8) (9) No controls Controls Controls, weighted Business – – –

Technical 0.127 0.036 0.065 professional (0.083) (0.083) (0.109)

Farmer -0.059 0.015 -0.027 (0.137) (0.136) (0.188)

Lawyer 0.140 0.101 0.133 (0.096) (0.091) (0.112)

Other white-collar 0.267* 0.156 0.074 (0.130) (0.128) (0.155)

Politics/law 0.267* 0.159 0.222+ enforcement (0.102) (0.104) (0.112)

Civil service 0.216+ 0.046 0.129 (0.110) (0.104) (0.094)

Service-based 0.310** 0.099 0.102 professional (0.096) (0.083) (0.093)

Worker 0.579** 0.383** 0.245* (0.121) (0.111) (0.121)

Left party 0.610** 0.502** (0.073) (0.083)

Observations 1946 1874 1874 Notes: Standard errors in parentheses. Higher coefficient indicates more contact with workers’ groups. See Appendix Table A.5 for full regression results. +p < 0.10; *p < 0.05; **p < 0.01.

In short, the results show a significant class-based difference in legislators’ contact with workers’ organizations and trade unions, with working-class legislators making higher levels of contact than their business-class counterparts. When we examined attitudinal outcomes in the previous section, service-based professionals were at least as different as workers from their counterparts in business, but here it is only workers who appear to be distinct.

Thus, for this behavioural measure, the main occupational ‘fault line’ for contact with workers’ organizations and trade unions is between workers and legislators from other occupational categories. Indeed, supplementary analysis with workers as the omitted reference category shows that workers are significantly different from five other occupational categories by this contact measure.13 Working-class legislators are therefore the only category consistently distinct from business-class legislators across all three of the attitudinal and behavioural outcome variables.

5 Electoral institutions as moderators of the influence of legislators’ class

As a cross-national, multi-parliament data set, the Comparative MP Survey also offers an opportunity to examine potential moderators of the influence of legislators’ class, including electoral institutions. A recent study of the link between gender and the representation of women’s issues, which focused on the moderating effects of electoral institutions, offers a useful framework for examining institutional moderators in the context of class (Lore, 2016). Lore found that institutional features like open party lists, local nominations, and division of powers

13 See Appendix Table A.6. This is using the baseline unweighted specification with controls. Note that the results are also robust to dropping those cases coded as workers because of their status as staff members of unions (see robustness checks for more on this point). 27

systems heighten the differences between men and women in the substantive representation of women’s issues.

According to Lore’s (2016) theory of “incentives for unincorporated representation,” each of these three institutional characteristics increases the incentives for competition between individual legislators within parties. The converse of these features – closed lists, centralized nominations, and the fusion of powers – are expected to dampen individual-level differences.

Open party lists and local nomination battles straightforwardly imply greater competition between prospective legislators of the same party. Perhaps less obvious is how division of powers systems, unlike fusion of powers systems, increase intra-party competition between legislators. As Lore (2016) outlines, division of powers systems are characterized by weaker party discipline relative to parliamentary systems with the fusion of powers, which means more opportunities for differentiation between legislators in the same party.

This framework is an elaboration of Carey and Shugart’s (1995) theory of personal versus party votes. Because of increased intra-party competition, there is more opportunity and incentive for legislators to differentiate themselves from their fellow partisans. In Lore’s (2016) framework, this holds particularly for issue dimensions that are not already well-incorporated into inter-party competition.

Class-relevant economic issues, however, are likely to be better incorporated into competition between parties than gender policy issues. Political parties are usually understood to act as key political aggregators of class conflict, and economic issues are constitutive of how we conceptualize left and right-wing political parties. Therefore, we may be less likely to observe a class effect being moderated by institutions than a gender effect. 28

On the other hand, in an era of frequent agreement on liberal economic policies among parties of both the left and right, economic issues may not be as well incorporated into party competition as they once were. Thus, there is reason to believe that institutional features like open party lists might indeed magnify the effects of legislators’ class on their attitudes and behaviour.

One might object that such incentives could condition legislators’ behaviour in office, but that they shouldn’t affect their attitudes and preferences. However, part of adapting behaviour to a strategic context can include adapting one’s privately-held attitudes. Indeed, the fact that people’s attitudes often follow from their behaviour is a robust finding in the psychological literature (Olson & Stone, 2005).

In terms of institutional moderators, three indicator variables in the Comparative MP Survey data set are used in the following analyses. The first variable indicates whether a jurisdiction uses an open list voting system.14 The second indicates whether the party nomination for that MP took place at the local level (as opposed the regional or national level). The third variable indicates whether the jurisdiction is characterized by a division of powers system (presidential or semi- presidential) or a fusion of powers system (parliamentary).

The model specification in this institutional analysis is similar to the baseline models in the main analyses, with the addition of an institutional indicator variable that is interacted with the ten occupation indicator variables, as follows:

14 This variable is defined only for the proportional systems in the data set, which makes up the large majority of observations (over 80%). 29

Outcomeij = β1-10(Occupationij) + β11(Institutionj) + β12-21(Occupationij)x(Institutionj)

+ β22(Controlsij) + γk + ϵij

where Institutionj is a dummy variable indicating the presence of one of the institutions (open lists, local nominations or division of powers) in parliament j and Controlsij represents the same vector of controls as in the models presented in the preceding section.

The results show that the relationship between legislators’ class and their attitudes and behaviour does vary with institutional incentives. First, there is a statistically and substantively significant interaction between legislators’ occupation and open list electoral systems in terms of attitudes towards inequality (using a Wald test, p < 0.01).15 As shown in the marginal effects (see Figure

1.2, first panel), under open party lists (compared to closed lists) there tend to be larger differences in inequality attitudes between business sector professionals and other occupational categories, including workers and service-based professionals.16 In absolute terms, the larger difference between classes under open lists is composed of a combination of a shift to the right by business legislators and a shift to the left by workers and service-based professionals.

Specifically, business sector legislators are more favorable to inequality than legislators from other occupations. The magnitude of these class coefficients under open lists is even larger than those seen in the earlier models without moderators. For example, under open lists, the

15 This is a test of the joint significance of the interaction terms. Full regression results are in Table A.15. 16 Note that I focus here on models with controls, unweighted. Since the purpose of these models is to measure the interactive effect of institutional differences, weighting by jurisdiction would be inappropriate. 30

difference between a business legislator and worker is over 60% the size of the coefficient for being in a left party compared to another type of party.

Figure 1.2. Class-based differences in inequality attitudes, by institution

Note: Lower scores correspond to more economically left-wing attitudes. Dots are marginal effect coefficients and lines are 95% confidence intervals. Business is the omitted reference category for occupation.

Similarly, the interaction of occupation with the division of powers variable is significant (Wald test, p < 0.01). The differences between business sector legislators and other occupational categories tends to be larger under the division of powers (presidential or semi-presidential systems) compared to the fusion of powers (parliamentary systems), as shown in the second panel of Figure 1.2. For example, under the division of powers, business legislators are more

favorable to inequality than workers, and the coefficient size is about 60% that of being in a left party.

In contrast, there is no significant interaction between occupation and the institutional characteristic of having local nominations. Local nominations were also the weakest institutional interaction in Lore’s (2016) analysis of gender representation, suggesting this may be a less potent moderator of intra-party differences in legislators’ attitudes and behaviour.17

For the behavioural dependent variable (levels of contact with workers’ organizations and trade unions), the pattern of results is similar, but more muted. Again, legislators’ class interacts significantly with open lists and division of powers (Wald test, p < 0.05), but not local nomination.18 Under open lists and the division of powers, business legislators have substantially less contact with workers’ organizations and trade unions than worker legislators (see Figure

1.3), and in both cases the differences are more than 90% of the coefficient size of being in a left party. In contrast, under closed lists and the fusion of powers, the marginal effects show no significant differences between business sector and legislators from other occupation categories.

17 For simplicity, I’ve focused here on the inequality attitude dependent variable, for which I observe the strongest coefficient in the main, non-interactive models. When I run these interactive models with the government intervention dependent variable, the results are substantively similar. The interaction effect remains significant for open lists and division of powers (Wald test, p < 0.05) and is not significant for local nominations. The pattern of the marginal effects is similar but less pronounced. These results are available on request. 18 Full regression results in Table A.16. 32

Figure 1.3. Class-based differences in contact with workers’ groups, by institution

Note: Higher coefficient indicates more contact with workers’ groups. Dots are marginal effect coefficients and lines are 95% confidence intervals. Business is the omitted reference category for occupation.

Broadly speaking, these findings suggest that electoral institutions play a role in moderating the theorized effect of class representation. These institutions would appear to magnify individual- level differences between legislators by increasing incentives for intra-party competition.

6 Alternative explanations and additional analysis

To what extent should we conclude that the observed relationship between legislators’ economic class background and their policy attitudes and behaviour is causal? In this section I evaluate some potential alternative explanations and present additional analysis.

6.1 Political culture of a country or region

One concern we might have is that some jurisdictions have, overall, a more economically left- wing political culture than others. In such jurisdictions, voters may be more likely to elect legislators of certain class backgrounds, such as workers (and less likely to elect legislators of other backgrounds, such as the business sector). If so, workers in the sample would be more likely to ‘show up’ in left-wing jurisdictions. Thus, these worker legislators may exhibit more economically left-wing attitudes and behaviour simply as a reflection of their jurisdiction’s political culture. This could lead to the observation of a spurious relationship between class and these measures.

The analyses reported above account for political culture in a few ways. The main models include a “left party” variable, so there is a clear control for political ideology. The main models with controls also include dummy variables for each country that control for characteristics specific to that jurisdiction, such as a left-wing political culture. Furthermore, as an additional check, when country dummies are replaced with parliament dummies, the observed differences by class remain.19

In addition, we can control for a parliament’s ‘political culture’ by taking the average class attitudes of all legislators within the parliament (using an index variable averaging attitudes on the inequality and government intervention in the economy questions). When this new variable is added as an additional control in the main models, the observed differences by class still hold for

19 Results not reported, but available on request. 34

all three dependent variables.20 The differences by class also hold if controls are included for the parliament’s average of legislators’ self-assessed position on the left–right spectrum, as well as the parliament’s average for a variable rating how important legislators believe it is to “represent employees” in their role as parliamentarians.

Finally, in the Comparative MP Survey legislators also assess how they believe the electorate is positioned on the left–right ideological spectrum. When we take the average of this assessment of the electorate’s left–right position (by all legislators in a parliament) and use it as a control in the main model, the relationship between legislators’ class and their attitudes and behaviour still holds.21

6.2 Constituents’ preferences at the district level

The previous subsection dealt with controlling for potential confounds due to political culture at the country and regional levels. In addition, within electoral districts it could be the case that legislators of a certain class tend to be elected by voters with a distinct set of policy preferences.

Legislators may then reflect those voters’ preferences in their attitudes and behaviour. For example, a left-wing district may be more likely both to elect a worker and to demand left-wing policies. Thus, we might be concerned that constituents’ preferences may be a confounding

20 Models described throughout this section are variations on the main models for the three dependent variables (Models 2, 5 and 8), unless otherwise stated. Regression results from this paragraph are in Appendix Table A.7. 21 Regression results are in Table A.8. 35

variable, driving both our independent variable (legislators’ class) and our dependent variables

(legislators’ attitudes and behaviour).

Since the data set does not include information on the characteristics of voters within districts

(and legislators are anonymized), it is not possible to test this possibility directly. However, most of the observations are in proportional representation systems, which are characterized by weaker links between constituencies and individual legislators. When the sample is explicitly limited to legislators in proportional systems and excludes those elected in single-member districts, the relationship between legislators’ class and their attitudes and behaviour persists. In addition, when I limit the sample to legislatures with higher district magnitudes (greater than 15, for example), thereby loosening the link between constituency characteristics and individual legislators, the observed differences by class persist on all three dependent variables.22

Furthermore, suppose constituents’ preferences were acting as a confound driving both the independent variable (occupation) and outcome variables (attitudes and behaviour). We should then reflect on why, for example, left-wing voters would tend to elect more workers in the first place. If voters see legislators’ class as a credible signal of the policies they will represent, then this is precisely consistent with the idea that class has a bearing on legislators’ policy attitudes and behaviour.

22 Regression results from this paragraph are in the Appendix Table A.9. 36

6.3 Alternative outcome variables

We still may be concerned at the possibility of a spurious relationship between legislators’ occupation and their attitudes and behaviour, due to confounds not already addressed above.

Another way of checking whether the observed relationships are spurious is to swap out the dependent variables for a set of ‘placebo’ outcome variables. These are variables that we would not theoretically expect to be as strongly driven by legislators’ class as the core economic issues addressed in the main outcome variables, which relate directly to class (though some differences would not be unexpected).

Three additional non-class-focused “political issues” questions were put to legislators in the

Comparative MP Survey, directly alongside the two questions that serve as the main outcome variables. They ask MPs to rate their agreement (on a 1–5 scale) with the following statements:

• “People who break the law should be given stiffer sentences.”

• “Immigrants should be required to adapt to the customs of our country.”

• “Government should make sure that films and magazines uphold moral standards.”

For all three of these questions, distinctly fewer occupation categories exhibited a significant difference than they did for the main attitude dependent variables.23 For example, working-class legislators are not significantly different from those from the business sector on any of these questions. The weaker class link between legislators’ occupations and these alternative outcome variables is consistent with expectations and strengthens confidence in the main results. There

23 Full regression results from this subsection can be found in Table A.10 of the Appendix. 37

was a significant difference between business sector professionals and at least one other occupation category for each of these questions. The presence of some differences is not surprising, given that classes do think differently about social issues such as these. We would also expect between one and two of the coefficients across these three alternative outcome variables to be significant by chance.

In addition, I can examine an alternative dependent variable that I do expect to be strongly related to class (i.e., not another placebo variable): legislators’ self-placements on the ideological left–right spectrum (on a scale from 1 to 10). Consistent with expectations, workers and service- based professionals (indeed those in all but one occupation category) clearly place themselves further left on the ideological spectrum than legislators from the business sector. Notably, this class-based difference in ideology is observed despite a left party dummy variable still being included in the model as a control.

6.4 Subsample of experienced and influential MPs

Another concern we might have about the reported results is that the attitudes and behaviour of many individual MPs may not be reflected in actual policy outcomes or party positions (for example, their influence may be blunted by party discipline). If so, the results may be of less practical importance. While we cannot directly test for this, we can observe whether the relationship between legislators’ class and their attitudes and behaviour holds among a subsample of relatively experienced and influential MPs, who we may expect to exert greater influence on policy outcomes and party positions.

To examine this possibility, analyses have been run on three different subsamples of MPs: those who have previously held elected office, those who have sponsored at least one bill, and those who hold at least one parliamentary leadership position (such as committee chair). Despite much smaller sample sizes, the observed differences by legislators’ class persists for both the attitude dependent variables (inequality and size of government) and the behaviour variable (contact with trade unions). This holds for the three subsamples of MPs across the three dependent variables, with a loss of significance in only two cases (one of which remains significant at the p < 0.1 level).24

6.5 Testing using a second data set: Comparative Candidate Survey

As a further test of the relationships observed in the Comparative MP Survey, a second data set of legislative candidates was used: the Comparative Candidate Survey (CCS 2016). The CCS analysis conducted includes candidates and their occupations and attitudes for seven European countries for parliamentary elections between 2005 and 2013.25 Like the Comparative MP

Survey, CCS includes questions on candidates’ attitudes on economic issues, as well as relevant control variables like age, gender, education, and the left-right ideological placement of the candidate and party.

A particularly useful feature of the CCS data set is that the occupation variable comes pre-coded using a variation on the International Standard Classification of Occupations (ISCO) system. In

24 Full regression results are available in Table A.11 of the Appendix. 25 The countries included are Germany, Greece, Italy, Norway, Portugal, Switzerland and the UK. 39

contrast, the Comparative MP Survey provided raw, open-ended descriptions of the legislators’ occupations, which I had to interpret and code from scratch.

To analyse the CCS data, I set rules to assign the standardized ISCO occupation codes from the

CCS data to a set of categories similar to my main occupation categorization scheme (to yield an analysis as comparable as possible). Because the ISCO codes in CCS are at a relatively high level of abstraction, it was not possible to map onto exactly the same categories. But it is nonetheless possible to compare business sector candidates to a set of occupation categories similar to those in the Comparative MP Survey analysis.

The CCS data replicate the finding above that legislators’ class predicts their attitudes about inequality and government intervention in the economy (see Figure 1.4). This helps provide assurance that the main findings are not a result of idiosyncratic features of the manual coding work on the Comparative MP Survey data.

Figure 1.4. Estimated class-based differences in candidate attitudes (CCS data)

Note: Lower scores correspond to more economically left-wing attitudes. Dots are marginal effect coefficients and lines are 95% confidence intervals. Business sector is the omitted reference category for occupation.

The CCS asked respondents questions on attitudes towards both inequality and government’s role in the economy (though with somewhat different wording).26 The model specification used in the analyses shown is the same as for the main data set, with the exception that the education control variable is included in the CCS specifications.27

26 The inequality question was worded as follows: “Income and wealth should be redistributed towards ordinary people,” with a five-point agreement scale. The government intervention question was worded: “Politics should abstain from intervening in the economy.” 27 Age, gender and education variables are included in the main regression models shown in Figure 1.4 and Figure 1.5. Including these controls leads to around 1,000 observations being dropped due to missing data. When alternative models are run excluding these controls (including only the key party type control), the same pattern of results holds. These additional results are available on request. 41

The results are shown in Figure 1.4 and confirm that candidates in the business sector held significantly more right-wing attitudes on inequality than most other occupation groups.28 The largest difference, between the business sector and trades/skilled manual workers, is 28% of the coefficient size of being in a left party. Those candidates in the business sector were also significantly less favourable to government intervention in the economy than clerks and trades/skilled manual workers, and the coefficient size was 22% that of being in a left party. The coefficient size and the number of significant categories was somewhat more muted for this second dependent variable than for inequality attitudes (a pattern that we also saw in the main data set).

The CCS also includes two additional class-relevant attitude questions. One is on globalization:

“Globalization should be promoted,” and the other is on social security: “Providing a stable network of social security should be the prime goal of government.”

There is an even more stark relationship between candidates’ class and their responses on the globalization question. Business sector candidates are statistically and substantively more pro- globalization than all other occupation categories (except those designated as politicians; see

Figure 1.5). Strikingly, being in a left party seems to have no significant influence on candidates’ attitudes towards globalization, which may reflect the broad pro-globalization consensus in major parties in Western countries in recent decades. In the CCS data, the issue of globalization is a cleavage that appears to be not well incorporated into party politics.

28 Full regression results for the models in this section are available in Appendix Table A.12. In the CCS analysis, the “left party” variable is based on the average of respondents’ self-placement on the left–right spectrum for each party. Controls used are party type, education, age, gender and country dummies. 42

Figure 1.5. Estimated class-based differences in candidate attitudes (CCS data)

On the other additional question, candidates in the business sector are statistically and substantively less favourable to social security as a “prime goal” of government than clerks and trades/skilled manual workers. The largest difference (between business sector and trades/skilled manual workers) amounts to about half of the coefficient size of being in a left party.

Finally, I ran a different set of models in which I categorized occupations only according to their original top-level ISCO codes (rather than mapping the codes into categories similar to my coding scheme for the main data set). Using this ‘unadjusted’ occupation coding scheme, the

relationship between legislators’ class and their attitudes also remains significant and substantial.29

6.6 Additional robustness checks

The analyses presented above holds up under a variety of specifications. Nevertheless, one concern we might have is that the results are sensitive to the estimation methods used. To address this possibility, a range of additional alternative specifications for each of the three dependent variables can be considered. The models that follow are modifications of the main specification (including controls) for each dependent variable, unless otherwise indicated. Full regression results described in this section are available upon request (and in certain cases, where specifically indicated, they are already included in Appendix A).

When clustering standard errors by country instead of by parliament, the relationship between legislators’ class and their attitudes and behaviour remains significant. The observed differences by class also hold when using a multilevel model with legislators nested within parliaments

(instead of a regression with standard errors clustered by parliament), as well as nesting legislators within parliamentary party groups rather than parliaments (though in this case significance is lost for the government intervention variable). The results also hold when using ordered logit instead of Ordinary Least Squares regression (which is relevant since strictly speaking the dependent variable scales are ordered rather than continuous).30

29 Results for CCS models with this alternative occupation scheme are available upon request. 30 Regression results for the logit specification can be found in Table A.13. 44

Another concern we might have is that legislators’ education levels are not included as a control in the main analyses. Education is a potentially theoretically important control, given the likely relationship between occupation and education. It was excluded in the main analyses because it is not itself statistically significant, and because including it leads to the loss of nearly 300 cases due to missing data for this variable. When education is included as a control, though, the relationship between legislators’ class and their attitudes and behaviour remains substantively similar for all three dependent variables.

One particularly important control variable in my models is the “left party” dummy variable.

This variable is constructed from expert assessments of the party families to which each party belongs (e.g., “Liberal”, “Socialist”, “Communist”, among others). As an alternative variable to control for party ideology, we can use a “left party” dummy constructed from a legislator’s self- reported assessment of their party’s ideological location on the left-right spectrum. We can also use the self-reported party ideology control as a continuous variable rather than a left dummy. In both cases, the models run show relevant coefficients for legislators’ class remain significant for the attitude and behaviour outcome variables.

In addition, the findings also hold up if we drop a set of observations that I flagged as “difficult or ambiguous to code” during the process of categorizing occupation descriptions in the data.

Furthermore, a large subset of the Comparative MP Survey sample are legislators from

Switzerland (approximately one quarter of the sample). To address this issue, in the weighted models presented in Table 1.1 and Table 1.2, observation weights were used to equalize the influence of countries and regional/national parliaments. As a further robustness check on this issue, models were run where the Swiss observations are excluded entirely. The pattern of 45

legislators’ class affecting their attitudes and behaviour holds up with observations from

Switzerland excluded.

Finally, in interpreting the results for the behavioural dependent variable (contact with workers’ organizations), recall that trade union staff are coded in the “workers” occupation category

(consistent with the coding practice of Carnes [2013]). A total of 191 MPs are coded as workers and 37 of these are classified as such because they are trade union staff.31 As an additional check,

I have separated out the union staff subset of workers in a supplementary set of models. Even removing union staff, the coefficients for worker-legislators show a higher level of contact with workers’ organizations and trade unions than legislators in the business sector (and the differences are statistically significant except in the weighted model).32

7 Conclusion

The politics of inequality and identity are erupting across Europe and most Western democracies.

No longer only a preoccupation of the political left, economic inequality is now a mainstream concern, propelled in recent years by protest movements like Occupy Wall Street and researchers like Thomas Piketty (2014). Even institutions like the International Monetary Fund view it as a threat to economic growth and political stability. Evidence has also mounted of class-based inequalities in political influence, and the causes and mechanisms of this phenomenon need to be

31 This could be a concern because trade union staff will of course have more contact with trade unions. However, the survey question stipulates that it refers only to the contact that legislators have specifically in their roles as MPs. 32 Results can be found in Table A.14. These models also show the coefficients for union staff compared to business legislators are even higher than those for worker-legislators. 46

more fully explained. While these emerging themes are undoubtedly interrelated, more research needs to be done to understand how they overlap and interact, and the implications for democratic institutions. As both an economic phenomenon and an element of social identity, the role of class in political representation may be a key to advancing these lines of inquiry.

This paper offers quantitative evidence across multiple developed countries that the class backgrounds of legislators matters. Taking this analysis together with recent work on the US

(Carnes, 2013; Grumbach, 2015), Switzerland (Rosset, 2016) and the British Labour Party

(O’Grady, 2018), there is clear evidence showing the substantive importance of class representation in advanced democracies, as well as parts of the developing world (Carnes &

Lupu, 2014). These findings help put to rest the conclusions of an earlier wave of work

(Matthews, 1984; Norris & Lovenduski, 1995) that downplayed the substantive effects of class representation, but often used inadequate dependent variables that did not examine the most relevant economic and redistributive policy issues.

While this paper focuses on the relationship between legislators’ class and their individual views and self-reported actions, more comparative research is still needed to examine the link to overall redistributive and economic policy outcomes. Carnes (2013) finds such a link to social spending outcomes in his analysis of US states and municipalities, but it cannot be assumed that class- based differences observed in individual-level representation will always translate into distinct policy outcomes in this way. Research on this question of aggregate outcomes would help clarify whether shifting class representation may be among the causes of increasingly salient economic and political inequality in many developed countries (Bartels, 2017; Bernauer et al., 2015;

Donnelly & Lefkofridi, 2014; Giger et al., 2012; Gilens & Page, 2014; Hacker & Pierson, 2010).

To be sure, many other relevant forces are at play alongside descriptive representation, such as class-based biases in electoral participation (Brady et al., 1995), class-biased perceptions in economic voting (Bartels, 2008; Hicks et al., 2016), as well as resource-intensive activities like campaign contributions and lobbying (Ansolabehere et al., 2003; Stratman, 2005) that are part of a broader “politics of organized combat” highlighted by Hacker and Pierson (2010) in their framework of “Winner-Take-All Politics.”

Further research is needed on factors that may moderate the influence of class, including the role of electoral institutions, building on the exploratory analysis presented above. Other possible moderators include polarization and the extent to which class issues are well-incorporated into party politics. Along the same lines, O’Grady (2018) suggests that effects of legislators’ class are more likely to appear when there’s divergence between the preferences of working-class legislators and party leaders.

In addition, the mechanisms through which class matters require more investigation. Because legislators from different classes sort into parties based on class-driven pre-existing beliefs, the measured coefficients for ‘party’ will be composed of the true party effect, as well as a portion of the class effect. Using a party control variable is necessary since parties are themselves sites of political socialization and party discipline, but this means the main empirical models likely underestimate the coefficients for class. Additional research is needed to clarify and better understand what is happening at the different steps in the causal chain.

Finally, another important set of questions follows from this line of inquiry: what drives the skewed descriptive representation of class in the first place, and what can be done about it

(Carnes, 2018)? Work on these issues beyond the US context is particularly needed. At a time 48

when economic and political inequalities are increasingly stark, the class characteristics of legislators deserve further attention and examination. Indeed, in light of the growing popular distrust of political classes, and the political upheavals that are unfolding in many advanced democracies, addressing these matters is urgent.

Paper 2 – The effect of class representation on social spending in Finnish municipalities

1 Introduction

This paper examines the effect of the class makeup of Finnish municipal councils on the levels of social spending in those municipalities. The analysis builds on a growing literature that considers the descriptive characteristics of elected representatives, including their gender, racialized status, and more recently class, along with the substantive consequences of descriptive representation, both at the level of individual legislators and in terms of aggregate outcomes (Carnes, 2013;

Carnes & Lupu, 2014; Grumbach, 2015; O’Grady, 2018; Rosset, 2016). In an era of high economic inequality, inequalities of political influence, and growing political and social instability, questions of class representation and its consequences are increasingly urgent. The importance of working-class representation is of particular interest, as this group tends to be underrepresented in descriptive terms while also displaying distinct policy preferences, particularly on economic issues.

The previous paper in this dissertation examined the relationship between the class of European legislators and their attitudes and behaviour in office across a range of countries (using the

PARTIREP and Comparative Candidate Survey datasets). I found that legislators from business backgrounds are more likely to support income inequality and small government, as well as less likely to consult with labour groups, than those from working-class and other backgrounds.

In both papers, my interest is the consequences of class representation, particularly as they relate to class-relevant policy issues such as inequality and social expenditure. However, there are some important differences between these analyses. The previous paper took individual legislators as the unit of analysis, with their individual class background (prior occupation) being the explanatory variable and their attitudes and behaviour being the outcome variables.

In contrast, in the present paper municipalities are the unit of analysis. The share of working- class representatives on council (specifically, as discussed below, the share in the largest party) is the explanatory variable. Furthermore, in the previous analyses, we could follow outcomes only as far as individual attitudes and behaviours of legislators, but here we can go further by testing effects of class representation on aggregate policy outcomes. Indeed, this is one of the only studies I am aware of that examines the effect of the descriptive representation of class on policy outcomes (Carnes [2013] is another).33 The dependent variable is the level of social spending in the municipality, and my hypothesis is that councils with more working-class representation will have higher levels of social spending.

Finnish municipalities are well suited to examining these issues for a few reasons. Finland systematically collects information about the occupations of its municipal councillors (and candidates), providing an appropriate measure of class representation. Municipal governments in

Finland also have an unusual level of devolved power including setting levels of social spending,

33 Hyytinen and colleagues (2018) examine whether higher shares of municipal employees on councils lead to higher municipal spending levels, but while this study makes use of councillors’ occupations, it is addressing the question of potential rent-seeking by public employees rather than the effect of their class. 51

providing a meaningful outcome variable. With more than 300 municipalities and data available for multiple election cycles, there is enough variation for a quantitative analysis.

Finally, and perhaps most importantly, while the previous paper relied on a fixed-effects regression model (as well as multilevel models, among other robustness checks), the identification strategy for the present analysis uses an instrumental variables approach adapted from Hyytinen and colleagues (2018) that exploits as-good-as-random variation in close elections. This approach, described below, should make the findings less vulnerable to potential confounds resulting from omitted variables.

In short, the main results suggest that the share of workers in the largest party on Finnish municipal councils does indeed affect the levels of social spending in that municipality. One additional worker on an average sized largest party caucus corresponds to about a 1.4% increase in social spending. The results are robust to a range of validity tests, robustness checks and specifications. On the other hand, the outcome of a lagged social spending placebo test raises some concerns about the quality of the instrumental variable.

This paper contributes one of the first studies outside of the US (Carnes, 2013) that directly examines the effect of legislators’ class on aggregate policy outcomes, as opposed to individual attitudes or behaviours in office. It also extends and reinforces the emerging literature linking the descriptive of class to substantive representation (Carnes, 2013; Carnes & Lupu, 2014;

Grumbach, 2015; O’Grady, 2018; Rosset, 2016), which together have been a corrective to a previous wave of literature that tended to downplay the substantive importance of this link

(Matthews, 1985; Norris & Lovenduski, 1995). In addition, these findings contribute to the broader literature on descriptive representation, helping to establish that class, alongside gender 52

and racial background, is among the personal identities of legislators that appear to relate to the representation they provide (Crowder-Meyer, 2013; Lawless & Fox 2004; Lore 2016;

Sanbonmatsu, 2006). Finally, like the previous paper, the findings here speak to the broader literatures on economic and political inequality, raising the possibility that the descriptive representation of class should be considered among the possible explanations of increasingly salient economic and political inequality in many countries (Bartels, 2017; Bernauer et al., 2015;

Donnelly & Lefkofridi, 2014; Giger et al., 2012; Gilens & Page, 2014; Hacker & Pierson, 2010).

2 Linking the descriptive representation of class to substantive policy

outcomes

As economic inequality has increased in much of the Western world, concerns about this issue have risen to prominence since the global financial crisis of 2008. In turn, over the past decade a growing literature has emerged addressing inequalities of political influence. As discussed in the previous paper, the question of “who governs” (Dahl, 1961) had largely fallen off the research agenda as an avenue to understand these phenomena, until a recent resurgence of literature linking the descriptive and substantive representation of class. That paper summarized and added to this new wave of research, which demonstrates that the class backgrounds of legislators do matter to substantive representation, adducing evidence largely in terms of individual level attitudes and behaviours of legislators. The previous paper also outlines theoretical reasons to expect the class backgrounds of legislators to matter in this way, noting that legislators do have

room to act on their own views in office, notwithstanding electoral imperatives and other constraints that they face.34

A further question arises: what effect do these individual-level class effects end up having on actual policy outcomes? For example, would more working-class representation lead to higher levels of social spending? Apart from Carnes’ (2013) analysis of state and municipal social spending in the US, the recent revival in the literature has focused on effects of class at the level of the individual legislator. In this paper, I am examining the relationship between the class composition of an elected body and aggregate policy outcomes in the jurisdiction represented. I examine the effect of real-world variation in working-class representation in Finnish municipalities with levels of social spending.

This raises the question of the mechanisms by which we would expect the class-driven attitudes and behaviours of legislators to translate into effects on policy outcomes. In the case of Finnish municipalities, we are talking about elected bodies of up to 85 councillors. As described in more detail below, the empirical strategy deployed in this paper specifically looks for the effect of the class background of the marginally-elected councillor within the largest party on council (i.e., the lowest vote-getter on the party’s open list). This marginal councillor could affect outcomes by influencing the substance of the policy debate within their party (Laver and Shepsle, 1990; as noted by Hyytinen and colleagues, 2018, in their analysis of rent-seeking behaviour in Finnish municipal councils), and, in turn, the largest party on council is the most likely party to shape

34 See Paper 1 for the literature review and theoretical discussion, which is forgone here to avoid undue repetition. 54

policy outcomes. Regardless of their persuasive powers within their party caucus, this marginally-elected councillor may also at times represent the pivotal vote on council.

The focus on the marginally-elected councillor on a party list is intrinsic to the instrumental variables identification strategy, which is described in detail below. One consequence of this approach is that it represents a conservative test of the influence of a typical councillor’s class on policy outcomes, since we may expect the lowest vote-getter to be relatively less influential compared to other councillors (Hyytinen et al., 2019). In the analysis that follows, within-party influence is perhaps the more plausible mechanism for the class background of a marginally- elected councillor to affect policy outcomes. Particularly given Finland’s quite large municipal councils, the marginally-elected councillor on a party list would only infrequently also happen to be the pivotal vote on council.

If within-party influence is the mechanism at work, we would also expect the effect on policy outcomes to be strongest where the party represented is powerful and in an agenda-setting position (Hyytinen et al. 2018). Hence, the analysis focuses on the marginally-elected councillor within the largest party. In the data set analysed, the average seat share of the largest party is

49% (for the next largest party, it’s only 24%), suggesting that this largest party is typically in a powerful position on council. At the same time, this primary approach is supplemented by additional analyses that examine the marginally-elected councillor within the next largest party, as well as across the whole council. These secondary analyses find that while the class effect on social spending is significant within the largest party, the effect disappears if we look within the next largest party instead. This is consistent with intra-party influence (within the most powerful

party) being the key mechanism by which the marginally-elected councillor affects social spending.

In addition, as Hyytinen and colleagues (2018) note, on smaller rather than larger municipal councils, the marginally-elected councillor will tend to constitute a larger proportion of both the overall council and their party caucus. Therefore, the marginally-elected councillor is more likely to be a pivotal vote on smaller councils rather than larger ones, and we would expect to observe the strongest effect on smaller councils, just as those authors found in their investigation of occupation and rent-seeking behaviour on Finnish councils.

3 Empirical strategy

3.1 Data and variables

The data used in this analysis correspond to four Finnish municipal election cycles from 2000 to

2012. The Finnish councils vary considerably in size, ranging from 8 to 85 representatives (with a median of 27). The unit of analysis is the municipality, and there are over 400 municipalities in the first election cycle (shrinking to less than 300 by the final election cycle as a result of mergers of municipalities).

The explanatory variable relates to the level of working-class representation on each municipal council. Specifically, it is the share of workers in the elected caucus of the largest party on the council (likely to be the most important and influential). The class background of the councillors is coded based on a data set of unstructured, open-ended descriptions of their occupations (that appear to be self-reports), which was obtained from the Finnish Ministry of Justice’s elections information service. Using the Google and Bing translation services, I machine translated the

descriptions from Finnish to English and coded them as working class or not. This entailed coding nearly 20,000 unique occupation descriptions for over 135,000 Finnish municipal candidates.35 As in the previous paper in this dissertation, the working-class category includes manual, service industry, clerical and union staff jobs.36

The outcome variable, social spending, is derived from municipal expenditure data from the

Statistics Finland agency, and it includes categories such as child care, education and social services.37 Statistics Finland distinguishes between health spending and social spending, and I have followed that practice here (health spending is not part of the main dependent variable used here). A range of control variables are used in certain models, where indicated. These controls correspond to factors other than the main explanatory variable that may be expected to affect levels of the outcome variable (social spending in municipalities). The controls are also factors that might plausibly have some impact on the explanatory variable (the share of workers on council), and none of them would be considered post-treatment variables. Controls include population and its square, the share of the population under 15 and over 64 years of age, the seat shares of the different political parties, and the total vote share of working-class candidates in each council election. The instrumental variables identification strategy described below is

35 The coding work was completed with the excellent research assistance of Shannon Hogan. I also sought the advice of my Finnish friend, Tomi Ihalainen, to help resolve cases of ambiguous translations. 36 The unstructured nature of the occupation descriptions, along with the machine translation, meant coding was challenging. The Finnish government has more precise administrative data on the occupational backgrounds of Finnish politicians, but accessing this data is costly, requires special permission, and can only be accessed when physically in Finland. In the future, it may be possible to re-run this analysis using this less “noisy” data source for the explanatory variable. This would be desirable since the noise is likely reducing the precision of the estimates presented, rendering the tests of statistical significance used here more conservative than they would otherwise be. 37 These are net expenditures (operating expense minus operating income), which is the measure Statistics Finland recommends using for best comparability across jurisdictions. The full set of expenditure categories available are shown in Appendix Table B.1 and the social spending items included in the outcome variable are indicated. 57

designed to eliminate omitted variable bias, but including models with these control variables adds another layer of protection against confounds, as well as helps to reduce the standard errors of the model estimates.

3.2 Identification strategy and model

The empirical strategy closely follows Hyytinen and colleagues (2018) in using an instrumental variables approach that exploits close election results in the open list proportional representation system used in Finnish municipal elections.38

In the Finnish open list system, electors must cast their votes for a single candidate, and each vote counts as both a vote for the party list and a vote for that candidate within the party list. Seat totals for each party on council are determined by the total vote share of the party in the election.

The winning candidates within the party lists are chosen based on the candidates’ individual vote totals.

Sometimes two (or more) candidates within a party list are tied for votes and competing for the final council seat allocated to that party. In such cases, the seat is allocated by lottery. If the candidates involved in the lottery are from different class backgrounds (e.g., working class or not), then we have a source of random variation in the class makeup of the party caucus (and the overall council).

38 I was originally conducting a more straightforward fixed effects OLS analysis of this data set when the paper by Hyytinen and colleagues (2018) was brought to my attention. This prior paper analyzes a very similar Finnish data set, but answers a different question relating to rent-seeking by public employees. Given the advantages of their instrumental variables identification strategy compared to OLS, I reworked this paper to adopt their approach. 58

To isolate this variation, we can construct variables that correspond to: (1) the “expected” number of workers elected from a party list, assuming any ties are allocated randomly; and (2) the actual number of workers elected from that party based on the actual outcomes of any lotteries.

The difference between the actual and expected number of workers on a party list instruments for the share of workers in the party’s elected caucus. For each municipality in my analysis, I use the worker share and its instrument (for the largest party represented on council).

Of course, the number of exact vote ties and lotteries is relatively modest. But if we assume there is some random variation in the vote totals that candidates receive, we can treat a broader set of close elections as effective lotteries. In other words, for an election where the vote totals of the winner and runner-up are sufficiently close, we may interpret the actual outcome as essentially determined by chance. At an extreme, it is easy to imagine interpreting the outcome as virtually random if there is a one vote margin in an election with thousands of votes. At a less stringent level, we could adopt the assumption that any race for the final seat within a party list that is decided by up to 0.4% of the party’s total votes is a “close race” and is allocated as-good-as- randomly. By examining close races only within party lists, rather than between parties, any effects of party affiliation do not confound the results.

Just as described above, we can then calculate an “expected” number of workers in the largest party’s elected caucus (based on the assumption that any “close races” are allocated randomly).

By subtracting this expected value from the actual number of workers elected in the party’s caucus, we get the instrumental variable for this party’s worker share at the vote bandwidth of

0.4%. 59

A wider vote bandwidth classifies more races as close and increases the variance in the instrumental variable, allowing for more statistical power in the main regression analyses. But as the bandwidth increases, the assumption that the outcomes of these close races are as-good-as- random becomes more tenuous. Following Hyytinen and colleagues (2018) in their analysis of a similar set of Finnish municipal elections, I use the 0.4% bandwidth in my main analysis. They acknowledge this choice of main bandwidth is somewhat ad hoc, and they examine ten smaller bandwidths as robustness checks. In this paper, I examine an even larger set of robustness bandwidths, testing 20 bandwidths (ten smaller and ten larger than the main bandwidth), as described below in the robustness section.

I will now explain the empirical strategy in more formal terms. Since this empirical approach closely parallels Hyytinen et al. (2018), I will borrow their notation. The main model can be described as follows:

Ymt = β1 Mmt + βcontrols X`mt + umt (1)

Ymt is the outcome variable (per capita social spending) where m indexes the municipalities and t indexes the four electoral cycles. Mmt is the share of workers in largest party caucus on each municipal council, X`mt corresponds to the control variables, and umt is the error term. β1 is the main statistic of interest, corresponding to the effect of worker share on the outcome variable.

To derive the instrument for Mmt, we begin by determining what Hyytinen and colleagues (2018) call the number of “pivotal votes” for each party list (in each municipality in a given election cycle). This is the average of the smallest number of votes received by an elected candidate and the largest number of votes received by a non-elected candidate. For each candidate i in party p

we can then calculate the “vote distance” as their actual number of votes received minus the pivotal vote number for their list.

A candidate is involved in a “close race” if their vote distance is smaller than the chosen bandwidth, ε. For the main analysis, I use ε = 0.4, which means the vote distance for a candidate must be no more than 0.4% of the total party list vote if they are to be considered part of a close race.

Then we can derive the party-level instrument, Tpmt, by taking the difference of two terms: the actual number of workers winning close races in the party list and the expected number:

푁푝 푁 푁푝 ∑ 푝 퐶 푊 푖 푖푝푚푡 푖푝푚푡 푇푝푚푡 = [∑ 퐶푖푝푚푡퐷푖푝푚푡푊푖푝푚푡] − [ 푁 ∑ 퐶푖푝푚푡퐷푖푝푚푡] ∑ 푝 퐶 푖 푖 푖푝푚푡 푖

Here 퐶푖푝푚푡 is equal to one if the candidate is involved in close race, 푊푖푝푚푡 is equal to one if the candidate is working-class, and 퐷푖푝푚푡 is equal to one if the candidate is successfully elected (and each variable is equal to zero otherwise).

Finally, we get our instrument Tmt for each municipality by taking the Tpmt for the largest party on council, multiplying it by 100, and dividing it by the number of total number of seats held by that

39 party. Tmt will be positive if there are more workers elected in the largest party than expected by chance, and negative if there are fewer (and zero if there is no difference between these terms).

39 If two or more parties are tied for largest, then to derive Tmt, the Tpmt value for each of these parties is summed, and the divisor used is the sum of the total number of seats held by each of these parties. 61

The instrument appears to be well constructed. The values of Tmt are distributed symmetrically

40 around zero across a wide range of different bandwidths. As expected, as the instrument Tmt rises, so does Mmt (from Equation 1), which is the share of workers present in the largest party’s caucus in the municipality.

One important insight from the close election studies literature should be considered. Marshall

(2019) recently showed how close election regression discontinuity designs (RDDs) that try to estimate the effect of a candidate characteristic differ from standard RDDs. Specifically, these

RDDs actually measure a compound treatment effect, because the treatment in these setups is

“defined by possessing (or not) predetermined characteristic X, conditional on narrowly winning an election” (p. 4). In the present study, that characteristic is whether the candidate is from a working-class occupation. Marshall demonstrates that if the candidate characteristic of interest in a close election RDD also influences the vote share they receive, this can lead to bias in the treatment effect estimates (and the magnitude of the bias would in part reflect the size of the effect on vote share).

The empirical strategy in the present study is not strictly an RDD, but rather an instrumental variables design with close similarities to an RDD. Nevertheless, Marshall’s (2019) logic about a compound treatment effect remains relevant (since no one can be “randomly assigned” to being a worker – only to winning or losing a seat while being a worker or not). However, Marshall’s observation about compound treatments in RDDs would only apply to the extent that the candidate characteristic affected their vote share. The literature is mixed on this issue, with some

40 See Figure B.1 in the Appendix these distributions graphed for each bandwidth. 62

survey experiments finding no effect of a candidate’s working class occupation status on respondents’ vote preferences (Carnes & Lupu, 2016; Sadin, 2016), others finding voters favour working-class candidates (Kevins, 2019) and disfavour high-income ones (Campbell & Cowley,

2013; Wüest & Pontusson, 2018), and yet another finding voter bias against low-skilled (but not skilled) working class candidates (Wüest & Pontusson, 2019). Taken as a whole, this literature does not suggest a large effect of working-class status on voter attitudes in either direction, which mitigates concerns about the compound treatment. Still, given the heterogeneity of the literature findings, we cannot rule out the possibility that the compound treatment issue may constitute a limitation to the identification strategy in the present paper.41

4 Validity tests

One standard way of assessing the validity of the instrumental variable is to examine how similar or different (in other dimensions) municipalities are that have more or fewer working-class councillors than expected by chance (Hyytinen, 2018). Specifically, we can check whether a number of pre-treatment covariates (in the period before the election) are balanced across positive and negative values of the instrument.

To this end, I have compared municipalities with positive and negative instrument values in terms of lagged (previous electoral term) covariates including social spending, municipal

41 Given that it is unclear from the literature in what direction a candidate’s working class status may affect vote share, if at all, it is impractical in this analysis to speculate on and adjust for the role of any “compensating differential” characteristics of workers, as per the strategy discussed by Marshall (2019). 63

demographic characteristics, and council characteristics.42 Since the instrument is constructed to reflect “as-good-as-random” variation in worker representation at time t, we should expect no systematic differences in these pre-treatment covariates at time t-1. In other words, municipalities with more worker-councillors than expected by chance (positive instruments) should have similar covariate values in the previous time period to municipalities with fewer than expected workers (negative instruments).

The covariate balance tests are shown in Table 2.1 with the same bandwidth used in the main regressions below (ε = 0.4). There are no statistically significant differences in the pre-treatment covaraties between municipalities with positive and negative instruments (social spending is lower in municipalities with negative instruments, but not at a statistically significant level).

There are also no statistically significant differences between municipalities with positive and negative instruments in a smaller set of covariates for the current electoral term at time t (i.e., the same term in which the working-class seat share and instrument are derived). The subset of covariates examined in these balance tests are those not directly tied to the outcome variable or the instrument itself (i.e., the spending and worker share variables are not included, because they are expected to vary with the instrument at time t, by hypothesis and by construction).

42 Lags have been adjusted to account for the many municipal mergers that took place during the study period. For example, suppose municipalities X and Y at time t-1 are merged into municipality Z at time t. Then the lagged per capita social spending for Z (relative to time t) is calculated as the summed social spending of X and Y divided by their summed populations (at time t-1). 64

Table 2.1. Covariate balance tests

Pre-treatment balance tests (values for previous electoral term, t-1), ε = 0.4

Tmt > 0 Tmt < 0 N Mean SD N Mean SD Difference Social expenditures 354 1636.774 342.0895 397 1613.158 334.5014 -23.61608 All other expenditures 354 1704.881 365.3623 397 1695.976 351.55 -8.904329 Worker % 248 18.94292 13.04332 265 17.95825 11.88236 -.9846676 Population 354 12728.02 24371.04 397 13319.8 22345.6 591.7853 Young population % 354 12.05355 8.524532 397 11.34953 8.574731 -.7040258 Old population % 354 14.44879 10.3992 397 13.82862 10.57192 -.6201767 Council size 354 23.36723 22.30374 397 23.62217 25.95146 .2549346 Incumbent on council % 248 56.67563 9.280294 265 57.93328 9.132675 1.257647 Women on council % 248 34.21779 7.983295 265 35.22438 8.650582 1.006593

Balance tests for current electoral term (time t), ε = 0.4

Tmt > 0 Tmt < 0 N Mean SD N Mean SD Difference Incumbent on council % 359 57.15173 9.530738 411 57.21855 9.184954 .0668181 Women on council % 359 35.05918 7.740306 411 35.33091 9.064976 .2717253 Council size 359 30.79109 11.13871 411 31.63017 11.93966 .839084 Population 359 13033.81 25027.21 411 13606.78 22924.88 572.9652 Old population % 359 20.91188 5.363395 411 20.93753 5.430658 .0256424 Young population % 359 17.22774 3.823998 411 17.01785 3.691462 -.2098884 Differences are tested using a t-test, clustered at the municipal level. + p < 0.1, * p < 0.05, ** p < 0.01

5 Results

We can proceed to estimate the effect of the share of working-class representatives in the largest party on levels of municipal social spending, which is averaged across the four-year term in office and presented in logged, per capita terms. My main hypothesis is that councils with more working-class representation will have higher levels of social spending.

As a baseline, we begin with OLS specifications (Panel A of Table 2.2). Coefficients for the effect of worker share are neither statistically nor substantively significant, except in the simplest no-controls model (which shows a small but statistically significant negative relationship). In each successive model (1-4), controls are added, including party shares on council, municipality

population and age demographics, and controls for the total vote share of workers in the municipality.43 Of course, OLS specifications remain vulnerable to potential confounds from omitted variables.

Table 2.2. Results for social expenditures: OLS and IV analysis with ε = 0

Panel A: OLS (1) (2) (3) (4) Workers -0.000709* -0.000407 -0.000326 -0.0000375 (0.000345) (0.000330) (0.000326) (0.000354) R2 0.610 0.666 0.673 0.674 Panel B: IV ε = 0 (5) (6) (7) (8) Workers 0.00255 0.00308 0.00239 0.00211 (0.00509) (0.00473) (0.00433) (0.00356) First stage F-stat 3.957 4.138 4.609 10.75 Panel C: Reduced form of IV ε = 0 (9) (10) (11) (12) Workers 0.00194 0.00247 0.00197 0.00189 (0.00340) (0.00322) (0.00320) (0.00310) R2 0.608 0.666 0.673 0.674 Observations 1393 1393 1393 1393 Year dummies Yes Yes Yes Yes Party controls No Yes Yes Yes Municipality controls No No Yes Yes Vote share controls No No No Yes Standard errors in parentheses. + p < 0.1, * p < 0.05, ** p < 0.01

The OLS models can be compared to the most stringent version of an instrumental variables analysis. Table 2.2 (Panel B) shows IV models using the smallest bandwidth of ε = 0. This bandwidth corresponds only to cases of exact ties in votes for the last council position on the

43 Party control variables are the shares of council seats for eight major political parties. Municipality controls are total population and its square, population share above 64 years, and population share below 14 years. Vote share controls are the total share of votes for workers and its square. 66

party list (decided by lottery). Thus, the variation in worker share at this bandwidth should be truly randomly assigned.

Because there are relatively few of these ties, however, the statistical power of the analysis at this bandwidth is weak. Indeed, we see no statistically significant effects of worker share at this bandwidth, and the first-stage F-statistics are very small. Notably, however, the coefficient values correspond closely to coefficients that are statistically significant at wider bandwidths (as we will see below).

Even at this smallest bandwidth, the coefficients for the effect of worker share are relatively stable across the models as additional controls are added (and adding the controls reduces the standard errors). This is consistent with what is expected when using instruments based on “as- good-as-random” variation. Reduced form models show a similar pattern, with somewhat smaller coefficients (Panel C).

In turn, Table 2.3 shows the results of models using the main bandwidth ε = 0.4, including the instrumental variables models (Panel A) and reduced form models (Panel B). The wider bandwidth allows more variation in the instrumental variable, which as expected results in smaller error variance.

The resulting IV models yields similar coefficients to the models in Table 2.2 using the smallest bandwidth, but here they are statistically significant at the p < 0.1 level (in the three models with controls). As controls are added, the coefficients are stable and standard errors are reduced. Note that the first-stage F-statistics are much larger in these models, meaning there is little concern

about a weak instrument. The reduced form models yield similarly significant results with only slightly smaller coefficients.

Table 2.3. Results for social expenditures: IV analysis with ε = 0.4

Panel A: IV ε = 0.4 (1) (2) (3) (4) Workers 0.00180 0.00218+ 0.00199+ 0.00200+ (0.00119) (0.00112) (0.00109) (0.00107) First stage F-stat 65.50 63.22 68.77 104.4 Panel B: Reduced form of IV ε = 0.4 (5) (6) (7) (8) Workers 0.00161 0.00186* 0.00174+ 0.00174+ (0.00102) (0.000920) (0.000921) (0.000918) R2 0.609 0.666 0.674 0.675 Observations 1393 1393 1393 1393 Year dummies Yes Yes Yes Yes Party controls No Yes Yes Yes Municipality controls No No Yes Yes Vote share controls No No No Yes Standard errors in parentheses + p < 0.1, * p < 0.05, ** p < 0.01

In substantive terms, the IV coefficients mean that a 1% increase in worker share in the largest party on council corresponds to about a 0.2% increase in municipal social spending. That means adding one worker to the caucus of a largest party with 14 members (which is the average sized largest party caucus) corresponds to about a 1.4% increase in social spending.44

As a point of comparison, this effect size is similar to the effect on total municipal spending of the share of municipal employees on the whole council in the study cited earlier (Hyytinen et al.,

2018). In that study, the authors were looking for evidence of rent-seeking by employees of the

44 One additional worker in a caucus of 14 represents 7.1% added to the worker share of the caucus. 68

municipality who became councillors, and they found that one additional municipal employee corresponded to approximately a 1% increase in total municipal spending.

Notably, we are examining the last councillor elected (with the lowest number of votes on the largest party’s list). These are councillors who could be considered marginal compared to their higher vote-garnering colleagues, possibly holding less sway and standing on council and in party caucus. Using only this local component of the variation in worker share could be considered a relatively conservative test, which may be yielding estimates smaller than the average effect of a worker on council. In addition, the first non-elected councillor often serves as an “alternate” when an elected councillor is away, meaning they may have some non-zero influence at council. This may dilute the effect of the variation measured, again making for a conservative test (Hyytinen et al., 2018).

6 Robustness

One important robustness check is to determine how sensitive the results are to the choice of bandwidth for what constitutes a close election. Thus, I tested 21 different bandwidths ranging from 0 to 0.8 and increasing in increments of 0.04 (using the full controls models).

The coefficients for worker share remain quite consistent across the bandwidths (at approximately the same magnitude as at the main bandwidth of ε = 0.4), and they are statistically significant at the p < 0.1 level for 16 out of the 21 bandwidths (the four narrowest and one other

are not significant). This is the case for both the IV and reduced form models. Figure 2.1 shows the reduced form coefficients across the range of bandwidths.

Figure 2.1. Effect of worker representation on social spending

Note: Coefficients for worker share are shown with 90% confidence intervals. Main modelled bandwidth is ε = 0.4.

In another robustness test, we find that results are similar when the analysis is run with each calendar year as the unit of analysis (Hyytinen et al., 2018), instead of each election cycle (as in the main analysis, where social spending has been averaged across the four-year terms). The coefficients for the effect of working class councillors in this alternative model are very similar to those in the main model in both magnitude and significance (see Appendix Table B.2; note

this and the following robustness checks are conducted at the main bandwidth ε = 0.4, unless otherwise indicated).

We can also consider some placebo tests, in which we would not expect to see an effect of class.

For example, instead of examining social spending as the dependent variable, we can consider non-social spending as a placebo dependent variable. As expected, there is no statistically significant effect of class on non-social spending at the main bandwidth (coefficients are near zero in the full controls models; see Appendix Table B.2).

Another approach is to use arbitrarily-determined vote thresholds for being elected within the party lists (instead of using the actual number of votes needed), when generating the instrumental variables (Hyytinen et al., 2018). I created such “placebo thresholds” by arbitrarily multiplying the pivotal vote number (described in the identification strategy section) by 1.25. Since the resulting instruments should bear no relationship to the actual vote thresholds (and thus no relationship to the composition of council), we would expect to see no effect of class in the models using them. These models show no significant effect on social spending, with coefficients near zero (see Appendix Table B.2).45

As another placebo dependent variable, we can look for an effect on the lagged social spending

(spending in the election cycle prior to the council’s election). Naturally, we should expect no effect of the as-good-as-random variation in worker representation at time t on social spending at time t-1 (Hyytinen et al., 2018). While the resulting models show no statistically significant

45 Results shown in Table B.2 and Table B.3 are for the full-controls models only. Results for the additional set of models with fewer controls are available on request. 71

effect, the coefficients for lagged social spending are positive and only slightly smaller than the main models looking at same-term spending (see Table 2.4). This similarity in coefficients is concerning and raises the possibility that the instrument is somehow capturing some characteristics of municipalities other than the purely random variation in class representation that should be isolated from the construction of the instrument. Indeed, looking across a range of bandwidths, the coefficients for lagged social spending remain broadly similar to the main results, even at the smallest bandwidth (ε = 0, corresponding to exact ties, where the presumption of random variation is particularly strong). Coefficients are significant for 9 of the 21 bandwidths, compared with 16 of 21 for contemporary social spending. This is puzzling given that the construction of the instrument should isolate the random component of variation in the share of workers.

Table 2.4. Effect on contemporary vs. lagged social spending: IV analysis with ε = 0.4

Panel A: Social spending (as above) (1) (2) (3) (4) Workers 0.00180 0.00218+ 0.00199+ 0.00200+ (0.00119) (0.00112) (0.00109) (0.00107) First stage F-stat 65.50 63.22 68.77 104.4 Panel B: Lagged social spending (5) (6) (7) (8) Workers 0.00138 0.00182 0.00169 0.00170 (0.00123) (0.00117) (0.00114) (0.00112) R2 65.44 62.73 68.87 104.3 Observations 1364 1364 1364 1364 Year dummies Yes Yes Yes Yes Party controls No Yes Yes Yes Municipality controls No No Yes Yes Vote share controls No No No Yes Standard errors in parentheses + p < 0.1, * p < 0.05, ** p < 0.01

The results of this lagged dependent variable placebo test suggest the main results must be interpreted with caution (though the fact that other placebo tests behave as expected is of at least some comfort). I have not been able to determine what is driving this similar (albeit less often statistically insignificant) result for the lagged models. It is possible that it relates in some way to the many municipal mergers that took place over the period studied (the fact of these mergers becomes particularly relevant when working with lagged variables, which cross time periods).

However, I have taken steps to ensure that, in the case of mergers, lagged social spending for a municipality at time t corresponds correctly to the combined per capita spending of its component municipalities that existed separately at time t-1.

7 Channels of influence

As per the main results, worker share within the largest party appears to have a significant effect on social spending. This raises the question of the channel or mechanism through which the marginally-elected worker is affecting this policy outcome, which might be through influence within their party caucus, persuading other council members, or acting as the pivotal vote on council. To shed some light on this question, I examine the effect of worker share within the second largest party, following Hyytinen and colleagues (2018). Interestingly, worker share within the next largest party has no significant effect, with the coefficient dropping to near zero

(see Appendix Table B.3). This is one important reason to suspect that the channel of influence of the marginally-elected worker is within their party, rather than simply as a member of the overall municipal council. In other words, the effect would seem to only appear within the largest – and plausibly the most powerful and agenda-setting – party on the council. In a further analysis, if we construct the instrument to reflect worker share on the council as a whole, rather

than only in the largest party, then the coefficients are similar to the main results, but they are slightly smaller and lose statistical significance (see Appendix Table B.3).46 This is consistent with the idea that the effect within the largest party is coming through, but it is diluted by including the other less influential parties in the analysis.

As described above, we might also expect that the effects of a marginally elected councillor would be larger in smaller councils (Hyytinen et al., 2018), where they would likely carry more weight relative to a smaller party caucus, exerting more influence in intra-party debates or negotiations (and indeed relative to a smaller council as a whole). To examine this possibility, we can divide the sample between municipalities above and below the median council size. As expected, the effect of additional workers is stronger for smaller councils, and in fact the coefficient drops towards zero for the municipalities with above-median council sizes (see Table

2.5). The coefficient for smaller councils means that a 1% increase in worker share in the largest party on council corresponds to a 0.28% increase in municipal social spending. Since the average largest party caucus among these smaller councils is 12, this means adding one worker corresponds to about a 2.3% increase in social spending in this subsample.

Notably, the largest party tends to have a larger seat share in smaller councils (an average 53% seat share at and below the median council size, compared to 40% above it). In other words, the largest parties in this small-council subsample have more power on council than they do in the

46 For clarity of comparison with the main results, if it were statistically significant, the coefficient from Appendix Table B.3 would imply that a 1% increase in the worker share on council is linked to a 0.19% increase in social spending. Concretely, this would mean that adding one additional worker on an average sized council with 30 members would be associated with a 0.64% increase in social spending. By comparison, the main results implied a 1.4% increase in social spending from adding one additional worker to the average sized largest party caucus. 74

larger municipalities. The fact that the class effect appears to be stronger in these small councils is consistent with the theorized mechanism of within-party influence of councillors (particularly within in powerful parties that are able to shape policy outcomes). However, this interpretation should be made with caution, as a smaller council also corresponds to a smaller municipality in population terms. Therefore, it’s possible that characteristics of larger versus smaller municipalities conditions the effect, rather than council size.

Table 2.5. Results for social expenditures by council size

Council <= 27 Council > 27 Panel A: IV ε = 0.4 (1) (2) Workers 0.00280* 0.000296 (0.00135) (0.00155) First stage F-stat 78.20 32.92 Panel B: Reduced form of IV ε = 0.4 (3) (4) Workers 0.00251* 0.000241 (0.00123) (0.00127) R2 0.683 0.710 Observations 890 503 Standard errors in parentheses. Models include year, party, municipality and vote share controls. + p < 0.1, * p < 0.05, ** p < 0.01

8 Conclusion

In sum, the main results suggest that the share of workers in the largest parties on Finnish municipal councils does indeed affect the levels of social spending in that municipality. In an average sized largest party caucus, one additional worker corresponds to about a 1.4% increase in social spending. When the sample is split by council size, we can see a more pronounced effect in smaller councils, and no significant effect in larger ones, where the influence of a single councillor may be more diluted. No effect is observed for the share of workers in the second- largest party on councils, which is suggestive evidence that the channel of influence may be the marginally-elected worker acting within their party, but only yielding changes in policy outcomes when that happens to be the largest and most powerful party on the council.

The instrumental variable is constructed to isolate a component of the variation in worker share that is expected to be as-good-as-randomly allocated. This identification strategy lessens concerns about potential confounds compared to the simpler OLS regression models, and the

results are robust to a range of validity tests, robustness checks and specifications. There is also potential for future analysis using higher-quality administrative data for the occupation variable, which is currently coded based on a relatively “noisy” open-ended text description that is likely reducing the precision of the estimates presented (rendering the significance tests conservative).

However, the outcome of the lagged social spending placebo test does raise some unresolved concerns about the quality of the instrumental variable.

Nevertheless, the findings here provide additional evidence suggesting that the class backgrounds of politicians can affect concrete policy outcomes. As hypothesized, more working- class municipal councillors appear to yield higher levels of social spending. This expands upon the limited existing evidence in the literature of the class effect extending beyond the individual level to aggregate policy outcomes, bringing this finding to a very different context in Finland

(compared to existing evidence in the United States; Carnes, 2013). Of course, this also builds upon the growing evidence of class effects on the individual attitudes and behaviour of legislatures, including the first paper in this dissertation.

As evidence builds that the class of legislators matters, this may also help to explain emerging findings of political inequality, wherein the preferences of more affluent citizens appear more likely to translate into actual government policy. If the working class are poorly represented in descriptive terms, this may help explain their weak representation in substantive terms. More broadly, as class grievances fester and political instability grows in many developed countries, developing a clearer understanding of the dynamics of class representation is necessary.

Paper 3 – A comparative analysis and exploration of barriers to working-

class representation

1 Introduction

Papers 1 and 2 laid out the empirical core of this dissertation, with a focus on analysing the consequences of the unequal descriptive representation of class, including class-based differences in legislators’ attitudes and behaviour as well as in policy outcomes. In particular, candidates and officeholders from business backgrounds appear to hold less redistributive policy attitudes than their counterparts from the working class, and evidence from Finland shows that having a greater share of working-class representatives on municipal councils is associated with higher levels of social spending. These findings help solidify the emerging literature showing that the class backgrounds of legislators do matter across the range of developed countries, and that working-class legislators appear to think and act in a more economically left-wing manner than representatives from other backgrounds.

The present paper engages in analysis of questions further back in the causal chain: relating to the causes of the unequal descriptive representation of class. Given the low levels of working- class representation in most countries, and their distinctive attitudes and behaviour in office, this paper engages in an exploratory analysis of the following questions:

1. What are the key barriers to working-class descriptive representation in legislatures?

2. What types of policies or interventions might be most effective to increase their

representation?

3. How would we expect these barriers and potential interventions to vary across contexts?

To address these questions, the paper begins by considering what we know about baseline levels of worker representation across jurisdictions and outlining the stages of selection that should be considered in understanding how workers are being “filtered out” of the political process. The subsequent sections consider barriers to working-class representation in comparative perspective, reviewing the relatively sparse existing literature and theorizing how these barriers may vary across contexts, as well as weaving in insights from data sets not previously used for this purpose: descriptive statistics from the Comparative Candidate Survey and PARTIREP survey already employed in paper 1. The penultimate section considers what the most effective policies or interventions might be to increase working-class representation, including exploring from limited evidence how the most effective approaches might vary across countries.

2 Baseline levels of working-class representation

It has been long recognized that legislators tend to come from privileged backgrounds compared to the broader populations in their polities. Matthews (1984) cites an extensive early literature on this point, noting that the pattern held at national and subnational levels of government in the

United States and in developed and developing countries around the world. He concluded that:

“Almost everywhere legislators are better educated, possess higher status occupations and

have more privileged backgrounds than the people they ‘represent’… Few generalizations

have been more exhaustively supported by empirical research” (Matthews, 1984, p. 548).

More recent research has continued to recognize the unequal descriptive representation of class as a characteristic of legislatures across countries (Best, 2007; Best & Cotta, 2000; Carnes, 2013;

Carnes & Lupu 2014; Gaxie 2017; Norris 1997; Norris & Lovenduski 1995; Wüest & Pontusson

2018). If anything, the pattern has become starker over time, including a decline in blue-collar workers and a rise in the professionalization of electoral politics (Best, 2007; Best & Cotta 2000;

Evans & Tilley, 2017; Norris 1997; Norris & Lovenduski 1995).

As in the previous papers in this dissertation, this paper adopts an occupation-focused operationalization of class, with a particular interest in the working-class category. While the unequal representation of class may be a near-universal, it does vary in extent across contexts. In the US, Carnes (2018) found that workers made up only 2% of those in Congress, 3% at the state level, and 10% of city councillors, compared to 52% of the population as a whole. While not measured in occupation terms, Wüest and Pontusson (2018) find that lower income earners are underrepresented in Swiss legislatures, but seemingly less severely than in the United States, with 86% of national legislators having incomes above the median Swiss income. Carnes and

Lupu (2014) find that working-class representation varies between 5-20% across Latin American legislatures, while the population shares of workers in these countries ranged between about 65% and 90%. Norris and Lovenduski (1995, p. 112), found that 15% of UK Labour MPs were from manual and clerical/sales occupations (compared to 83% of their voters), along with only 1% of

Conservative MPs (compared to 64% of that party’s voters).

Broadening the picture to a wider set of European countries, in the PARTIREP survey of largely

European legislators analysed in the first paper of this dissertation, 6% of the respondents in

national office and 11% in subnational office were identified as workers.47 In the Comparative

Candidate Survey, workers made up 12% of the subset of successfully elected respondents (and

15% of those not elected). In the Finnish municipal data set analysed in the second paper, 21% of municipal councillors were identified as working-class. There are important measurement differences between these different data sets, and they are not directly comparable. But they illustrate in broad strokes the underrepresentation of workers, as well as some of the apparent variation between countries and at different levels of government. For example, subnational levels of government appear to have somewhat higher levels of working-class representation compared to the national governments.

This paper’s aim of better understanding the causes of unequal class representation can be of use not only in explaining its variation across contexts, but also in pointing the way to policy solutions that could encourage more working-class representation. Indeed, this paper focuses specifically on identifying the barriers and potential solutions to increasing working-class representation. Therefore, more attention is paid to factors that are amenable to intervention, such as the cost of election campaigns and candidate recruitment practices, as opposed to other relevant factors such as long-term declines in the share of manual labourers in the workforce or the relative strength of social democratic parties.

47 Occupation descriptions were categorized from open-ended descriptions using a scheme adapted from Carnes (2013). The worker category includes manual, service industry, clerical and union jobs. The coding scheme used for both the PARTIREP and Comparative Candidate Survey are described further in the first paper of this dissertation. 81

3 Stages of selection and election

The existing literature on the descriptive representation of underrepresented groups is perhaps best developed in the case of gender and minority representation (e.g., Crowder-Meyer, 2013;

Lawless & Fox, 2004; Lore, 2016; Sanbonmatsu, 2006), but recent work has also begun to directly address barriers to working-class representation (Carnes, 2018; Wüest & Pontusson,

2018).

This literature has developed a broad framework for thinking about the different stages of the political process at which barriers to group representation may arise. This paper distinguishes between the following stages: 1) self-selection of individuals to pursue candidacy; 2) party recruitment and nomination of candidates; 3) election of candidates to office. In his examination of working-class representation in the United States, Carnes (2018) adopts a version of this framework that he refers to as the Qualify-Run-Succeed (QRS) model, where “Run” largely corresponds to stages 1 and 2, and “Succeed” refers to stage 3. In Carnes’ model, “Qualify” refers to whether prospective candidates possess characteristics that voters say in polling are important qualities for political representatives, including being hard-working, personable and assertive. A similar set of stages are employed in the literature on gender representation (Lawless

& Fox, 2004; Lore, 2016).

Broadly speaking, the self-selection stage depends on whether prospective candidates are motivated, willing and able to run for office. The decision to run may depend on whether candidates from different groups have an interest in politics and pursuing office, judge that they have a reasonable chance to win, and possess the material resources to cover campaign costs and time off from work to campaign. 82

The party stage refers primarily to the preferences that party elites and nomination decision- makers may have about prospective candidates from different groups. The influence of the party may be informal in the case of the personal networks that party elites draw upon when they decide who to encourage and recruit as candidates, or formal in the case of official nomination contests, which may involve widely varying sets of decision-makers. As in the first stage, the material resources and time possessed by prospective candidates to compete in a nomination contest is again relevant to their likelihood of ultimately becoming candidates.

Finally, at the election stage, if voters prefer candidates with different personal backgrounds, this may help determine the levels of descriptive representation for various groups including the working class. Party elites can also exert influence on outcomes at the election stage by choosing candidates’ positions on party lists in proportional representation systems and shaping which candidates are fielded in single member districts with decent electoral prospects for their party.

Once again, the material resources and time that candidates possess in order to compete in the election are likely to be relevant factors at the election stage, including both personal and party resources. Indeed, resource disadvantages are a definitional characteristic of working-class candidates, and this set of potential barriers to office cascades across all three stages. Therefore, resource disadvantages receive special attention in the analysis that follows.

These same stages can also be cast in terms of the supply of working-class candidates (self- selection) and the demand for such candidates from party decision-makers and voters (Norris &

Lovenduski, 1995). Moreover, each of the stages of selection is linked to the others. For example, party decision-makers would be expected to take into account their perceptions of voter preferences or biases (e.g., regarding candidates from different class backgrounds) and tend to 83

choose candidates with characteristics that they think voters will view favourably.48 At the self- selection stage, prospective candidates may take into account whether they anticipate being welcomed (or opposed) as candidates by party elites, as well as whether they believe voters are biased against their group. Indeed, what appears to be a lack of supply of candidates from a group like workers may in fact stem from demand-side biases of party elites or voters, correctly observed in advance by potential candidates who choose not to bother running at all (Ashe &

Stewart, 2011). In short, there are difficult matters of measurement and inference in studying both supply and demand side explanations of descriptive representation.

Notably, in the literature and indeed in the prospects for new data collection, knowledge is quite uneven across these stages. We know less about the broader pool of candidates than about elected officials, less still about those who compete for nominations but do not win them, and least of all about potential candidates who are not recruited or self-selected into the process of running for office in the first place, who are by definition difficult to identify and study.

System-level and macro-social factors such as electoral institutions and levels of economic inequality, unionization or party polarization may affect how factors operating at each of the stages influence the descriptive representation of class. For example, levels of economic inequality may help determine the magnitude of resource disadvantages typically faced by workers, moderating the effect of this set of barriers. Moderators may exert effects across different stages, or they may primarily influence factors at certain stages. Table 3.1 summarizes

48 Norris and Lovenduski (1995) call this “imputed discrimination.” 84

the three stages and corresponding barriers to working-class descriptive representation examined in this paper, as well as key actors and moderating factors.

Table 3.1. Summary of barriers to working-class descriptive representation by stage

Stage: I. Self-selection II. Party recruitment and III. Election nomination

Key actors: Workers Party elites, nomination Voters, party elites, selectors, workers workers

Barriers: • Qualification, interest in • Class-biased recruitment • Attitudes to workers (relevant actors politics, and ambition to networks (party elites) (voters) in parentheses) run (workers) • Ideological preferences • Party list positions or • Perceived chance to win (party elites, selectors) districts assigned to (workers) • Perception of workers’ workers (party elites) strength as candidates (party elites, selectors)

(Resource disadvantages extend across all three stages)

• Resources and time to • Resources and time for • Resources and time to pursue office (workers) nomination contest campaign effectively (workers) (workers)

Moderators: (Moderator impact on factors, with varying potential roles across stages)

Inequality; union density; campaign costs; electoral finance rules; voter turnout by class; electoral system; party type; polarization; national vs. subnational government

4 Barriers to working-class representation in comparative perspective

What do we know about barriers to the descriptive representation of the working class? This section discusses findings from the modest comparative literature on this question and theorizes how barriers to working-class representation may vary across contexts. By far the best studied case is the United States, which Carnes (2018) recently treated in depth, but outside the US the existing literature is particularly limited. Informed by analysis of the literature, I also probe the 85

comparative data sets used earlier in this dissertation (the Comparative Candidate Survey and

PARTIREP) for clues that could shed further light on barriers to working-class representation and inform approaches to future research.

Given that much of the literature is US-focused, would we expect early lessons from research on this case to extend to other jurisdictions? The US is an exceptional country, and there are certainly reasons to expect the dynamics of class representation to differ in other contexts. For example, compared to other developed countries, American election campaigns are notoriously expensive, economic rights are weak, unionization rates are low, and political parties are weak.

As described above, the US also appears to have a particularly low level of worker representation. Therefore, any extrapolation to other contexts from US-based evidence must be made with caution.

The first pair of subsections examines non-resource barriers to working-class candidates at the election stage, with a focus on voters’ attitudes towards workers. The second pair of subsections explores non-resource barriers to workers operating at the self-selection and party stages, including potential class biases in the recruitment networks and preferences of party elites. The third pair of subsections turns to a key set of disadvantages faced by workers: the material resources and time available to them.

Resource disadvantages are a distinguishing feature of prospective working-class candidates, and the effects of these disadvantages span all three stages discussed above, which is why these factors are the focus of their own subsections. Each of the following pairs of subsections follows a similar structure: first, examining existing literature and outlining theoretical expectations

about barriers, and, second, exploring the data sets used earlier in this dissertation for potential insights relevant to the many questions that remain to be answered in the literature.

The use of new data in this paper should be understood as an exploratory exercise, with the limited aim of probing theoretical observations and comparing with findings in limited existing literature. In general, descriptive statistics and t-tests are presented based on pooling observations across the jurisdictions covered in CCS and PARTIREP surveys (except where otherwise stated).

CCS surveys both elected and unelected candidates in national elections, and the analysis here includes responses from Australia, Belgium, Germany, Greece, Ireland, Italy, Norway, Portugal,

Switzerland and the United Kingdom. Responses to various CCS survey questions are available in only certain countries, and as a result the countries covered and sample sizes change depending on the variable being examined.

PARTIREP surveys only elected legislators, but it contains over 2000 MPs from more jurisdictions than CCS, including 15 countries and 73 national and subnational parliaments, better allowing some comparative hypotheses to be probed. The countries surveyed in

PARTIREP are Austria, Belgium, France, Germany, Hungary, Ireland, Israel, Italy, Netherlands,

Norway, Poland, Portugal, Spain, Switzerland, and United Kingdom. In PARTIREP, which provides open-ended text descriptions of occupations, the “worker” category includes manual, service industry, clerical and union staff occupations. In the CCS data, the occupation variable is provided in the form of International Standard Classification of Occupations (ISCO) codes, and the “worker” category constructed here includes the combined occupation categories of “Trades

and skilled manual” and “Clerks, service and sales.”49 The CCS and PARTIREP data sets are described further in the first paper of this dissertation.

4.1 Election stage and the role of voter attitudes

This section examines the literature on non-resource barriers to working-class representation at the election stage, with a focus on voter attitudes regarding the class backgrounds of politicians.

Key evidence in the literature comes from survey experiments. Evidence from real-world elections on the rates of electoral success of workers is also considered. Possible moderators of workers’ electoral success are examined, including party type, electoral institutions and levels of polarization. This literature-driven discussion is followed by exploratory data analysis in the subsequent section.

Survey experiments on voter attitudes. In the US context, the overall body of evidence does not suggest there is voter bias against working-class candidates that would account for their underrepresentation. Survey experiments about hypothetical candidates suggest that Americans are as willing to vote for a factory worker as they are for a business owner (Carnes & Lupu,

2016), and survey respondents are similarly indifferent between an ambulance driver and a cardiologist (Sadin, 2016). Indeed, in another study, Kevins (2019) finds that respondents are more likely to consider voting for a factory worker than a surgeon. Given these findings, Carnes

(2018) largely dismisses the role of voter bias in considering potential barriers to working-class

49 Note that these two categories were left disaggregated in the earlier paper 1, where the CCS data was used as part of a robustness check on the main results. 88

representation in the US. However, a recent survey experiment from Wüest and Pontusson

(2019) finds that respondents distinguish between “routine” and “skilled” working class candidates, with occupations like hospital cleaner at a disadvantage in terms of vote preference compared to higher-skilled working-class occupations like paramedic, as well as compared to middle class occupations.

The literature covering respondents outside of the US presents a similarly mixed picture. Kevins

(2019) finds that British respondents tend to favour a factory worker candidate compared to a surgeon while Canadian respondents are indifferent between them. Notably, left-wing respondents in particular are likely to favour the factory worker in these two countries. Carnes and Lupu (2016) studied voters in Argentina and Britain along with their US findings described above. In all three countries, participants in the survey experiment were no less willing to vote for a factory worker than a business owner. In another survey experiment, Campbell and Cowley

(2013) find that British respondents favour candidates with average incomes compared to high- income ones, and this preference was stronger among working-class respondents.

In the Swiss context, Wüest and Pontusson (2018) find in a survey experiment that respondents are biased against candidates from routine working-class occupations compared to skilled working-class and middle-class occupations. Specifically, the routine working-class candidate they use in the experiment is a retail worker with a low income, while the skilled working-class candidate has the same occupation but higher vocational training and income.50 However, preferences varied by the income of respondents, as well. Among respondents with lower

50 In their subsequent 2019 study, in contrast, class categories are defined only in occupational terms, instead of the combination of occupation, income and education used in this 2018 study. 89

incomes, the authors find no bias against routine working class candidates, but bias against upper-middle-class candidates. In their later study, Wüest and Pontusson (2019) also found that

British respondents (and Americans, as discussed above) were less likely to favour routine working-class candidates compared to those from other occupations.

In addition, Pederson and colleagues (2019) find that, in these types of experiments, voters’ preferences about candidates’ class can actually reverse themselves when candidates’ policy preferences are revealed. In this study, Danish respondents initially preferred a warehouse assistant to a lawyer as a candidate, but in conditions when both were revealed to have a left- wing policy position (which was popular with respondents) they preferred the lawyer. This would seem to suggest that respondents were using class as a cue for policy position.

Given these heterogenous results, the survey experiment literature does not strongly suggest an overall bias from voters either for or against working-class candidates. In this sense, consistent with Carnes (2018) conclusion, voter bias does not stand out as a prime barrier to working-class representation. At the same time, distinctions emerge from the literature that will need to be systematically integrated and combined into the design of future survey experiments on this question across contexts. These include distinguishing between routine and skilled working-class candidates, between respondents of different classes, incomes and ideologies, as well as consistently building the policy positions of candidates into the experiments.

Evidence from real-world elections. Moving beyond survey experiments, let us now consider evidence from real world elections that may shed some light on voter preferences. Carnes (2018) finds that among members of the US Congress, workers tend to have similar margins of election victory to those from professional occupations. He also adduces evidence that at the national, 90

state and local level, workers make up roughly the same share of officeholders as they do candidates. Along with the survey studies, this is taken as further evidence that voter bias is not primarily responsible for the underrepresentation of workers in office in the US. Of course, this evidence from real-world elections should be interpreted with caution, since it could be that workers who achieve office are discriminated against by voters on the basis of class, but these workers have compensating characteristics that improve their standing. That is, working-class candidates who win nominations may tend to be “higher-quality” on average, compensating for discrimination they may face (see Lawless & Pearson [2008] for an analogous argument in the gender representation literature). This would allow them to be competitive in elections, accounting for the observed vote margins and run-win rates.

Controlling for candidate quality is not a straightforward affair, particularly in the case of class, as Carnes (2018) points out. Typical proxies for candidate quality, such as education or certain professional backgrounds, overlap substantially with class itself. Prior political experience might be a more viable proxy to include as a control in future analyses of workers’ vote shares.

Conversely, workers might be favoured by voters based on their class, but for example tend to gain nominations in more difficult districts, which could also account for the overall pattern of real-world results. Still, based on real-world results combined with the survey experiment findings, Carnes (2018) concludes that voter bias is unlikely to be the driving force behind the underrepresentation of workers in the US.

In contrast to the US case, Wüest and Pontusson (2018) note that the “misrepresentation” of class does seem to be happening more at the election stage than the candidacy stage in Switzerland.

They find that 86% of those successfully elected to the Swiss Parliament have higher than the 91

median income, but this is only true of 59% of the broader pool of candidates for office.51 As we have seen, this differs from the US, where Carnes (2018) finds that working-class candidates typically are not getting to the stage of candidacy to begin with, but that they are roughly as likely as others to be successful when they do run. For example, workers make up about 5% of both candidates and those successfully elected to US state-level office from 2012-14.

What might account for this differing pattern across contexts? Voter bias against routine working-class candidates is one possible explanation for the Swiss pattern of lower-income candidates being less likely to win office. But Wüest and Pontusson’s (2018, 2019) survey experiments found a similar pattern of bias against routine working-class candidates among both

Swiss and US respondents. Perhaps the very small pool of US workers gaining nominations are more likely to be higher-skilled workers than in Switzerland (or are otherwise strong enough candidates to offset possible voter bias against routine working-class candidates). The relative placement of Swiss workers on electoral lists by party elites is another possible explanation of the Swiss pattern, discussed further in subsequent sections.

At the same time, one might also expect bias faced by lower-income candidates to be counterbalanced by the bias found among some voters against upper-class candidates. But as

Wüest and Pontusson (2018) find in the Swiss case, bias against upper-class candidates is displayed only by lower-income voters, and this group has lower voter turnout. Indeed, as the authors note, class-based differences in voter turnout may interact with class-based differences in voter attitudes. That is, if more affluent voters have less favourable attitudes to working-class

51 Note in the case of this real-world candidate data, as opposed to the survey experiment in the same paper described above, they use only income and not occupation as the measure of class. 92

candidates, and they are more likely to actually vote than lower-income individuals, this would tend to decrease working-class representation. Conversely, the potential countervailing effects of bias from low-income voters against upper class candidates would be diminished by poor turnout from low-income groups. Thus, class-based heterogeneity among potential voters in both rates of voter turnout and in attitudes towards a candidate’s class should be investigated in future studies of this kind.

Given the mixed and limited evidence on voter bias in the literature, one hypothesis to probe in the new data sets examined below will be whether workers appear to run and win at the same rate (as they do the US) or at very different rates (as in the case of low-income candidates in

Switzerland).

4.1.1 Moderators

Party type. In terms of conditional effects, the literature described above also suggests some left-right or partisan differences in attitudes to working-class candidates, with voters on the left more likely to favour workers in survey experiments (Campbell & Cowley, 2013; Kevins, 2019).

This raises the question, probed in the data below, of whether workers’ rates of success as candidates is moderated by party type, wherein we might expect more success when running under the banner of a party of the left rather than the right.

Electoral institutions. Another set of factors that we might expect to moderate the relationship between voter attitudes and levels of working-class representation are electoral institutions. In the first paper of this dissertation, I found that there appears to be more substantive differences

(for example, in attitudes on issues like inequality) between legislators of different classes under certain electoral institutions that encourage personal votes: specifically, open list and division of powers systems (see Carey & Shugart, 1995; Lore, 2016). These institutional features tend to provide more incentive for competition between individual legislators within parties, particularly on issues that are not well-incorporated into inter-party competition (Lore, 2016).

Might personal vote incentives also affect levels of working-class representation, by moderating the impact of any voter bias towards workers? That is, whatever biases voters may have towards working-class candidates (positive or negative) could be amplified where personal vote incentives are present and personal characteristics would therefore be weighted more heavily in voting decisions. This possibility is probed below using the PARTIREP data set, albeit without access to data on voter attitudes themselves.

Polarization. Wüest and Pontusson (2018) also offer the suggestion that cross-national differences in polarization might moderate the effect of class on voter preferences. Specifically, they argue that where polarization is high, the salience (and therefore effect) of candidates’ class may be reduced among voters because they are more likely to be driven by ideological and partisan cues. In their Swiss survey experiment, they find that ideological distance between respondents and candidates trumps the observed effect of the candidate’s class, and they take this as suggestive evidence for the polarization hypothesis. Another approach to evaluating this hypothesis, which I consider below, would be to examine cross-national differences in polarization. By Wüest and Pontusson’s (2018) logic, if polarization is high, partisan cues should tend to dominate class ones. Therefore, if voter bias against workers is an important barrier to their representation, all else being equal, we would expect more workers in office in places with

high polarization, with partisan cues swamping the class bias. However, all else is unlikely to be equal: polarization might go hand-in-hand with higher levels of economic inequality, for example (McCarty et al., 2016).

Geographical distribution of workers. Particularly in Single Member Plurality (SMP) systems, the geographical concentration of workers may be important to their levels of representation.

According to Jusko (2017), dispersed geographical distributions of the poor (in relation to electoral districts) undercuts incentives to substantively represent this constituency. We could also imagine that the geographical distribution of workers might plausibly affect levels of working-class descriptive representation in the same way. That is, as more credible representatives of the interests of working-class voters, the electoral prospects of worker- candidates might increase in districts with high concentrations of workers. From a national point of view, in countries where workers tend to be relatively concentrated in certain districts, rather than spread very evenly across districts, they have more potential to be electorally pivotal, which might yield higher levels of working-class descriptive representation (Jusko, 2017; cf., Rodden,

2018). These dynamics may also depend in part on both the size of districts and the overall size of workers as a group. This is not a set of hypotheses that I am able to test with my data but would be worth future consideration, perhaps using census occupation data.

In summation, in the literature we have observed mixed evidence on possible voter bias towards working-class candidates. Survey experiments covering various countries run the gamut on this question, returning null results as well as biases both for and against workers. Key studies also show the need to make further distinctions based on the skill level of workers, the class of respondents, and varying candidate policy positions along with their class. Evidence from real-

world elections suggests that workers in the US gain enough voter support to win office at roughly the same rate as they run (which is to say, a very low rate), while in Switzerland winning races at the election stage appears to constitute the larger barrier to low-income representation, rather than reaching candidacy. The literature also points to potential moderating factors, including party type, electoral institutions, and polarization. In the next section, the PARTIREP and CCS data are examined to shed light on these issues.

4.2 Exploratory data analysis: election stage

What can we learn about from the data sets used earlier in this dissertation about non-resource barriers to working-class representation at the election stage?

Run-win rate. In the CCS data, covering a range of European countries, workers made up a slightly smaller share of those successfully elected (12%) than they did of the pool of unelected candidates (15%), a difference that is statistically significant only at the p < 0.1 level in a t-test

(all of the statistical tests presented are t-tests comparing group means, unless otherwise stated).52 This is at least consistent with the parts of the comparative literature that find voter bias against some low-skilled working-class candidates. Of course, the magnitude of the class-based difference in “win rate” observed here is modest. Furthermore, we cannot conclude from this comparison of means that voter bias is the cause of the smaller worker share among the elected

52 Tables with the full results of this and subsequent t-tests are available in Appendix C, where they can be found in the order presented in the main text. Note that Hungarian observations are not included in the CCS analyses presented, as they only have occupation information for successfully elected candidates, meaning occupation-based comparisons between the candidate pool and the elected pool are not possible. There also appears to be an idiosyncrasy in the original ISCO occupation coding for Hungary in this data set, where fully 48% of elected respondents are identified as workers, and most of these are have the ISCO code for “office clerk.” This is far out of line with other countries in this data set, and it sharply contradicts the 5% worker share I find for Hungary in the PARTIREP data (both data sets refer to the same 2010 election cycle in Hungary). 96

respondents. This could also be, for example, driven by class-based differences in levels of personal financial resources or party support made available to candidates, both possibilities that are discussed further in subsequent sections.

4.2.1 Moderators

Party type. When I divide the CCS sample along left-right lines, I find that working-class candidates in right parties are less likely to win office compared to candidates from other occupational backgrounds (statistically significant at the p < 0.05 level; see Figure 3.1). In contrast, there is no such occupational difference in the rates of running and winning office among the candidates in left parties. In other words, what is happening in the right-wing parties drives the overall difference observed in the rate at which working-class candidates win office in the CCS sample.

Figure 3.1. Worker share in right parties

Note: Worker share among elected and unelected candidates in CCS data set. Lines are 95% confidence intervals.

This is at least broadly consistent with the survey experiments reviewed above, some of which suggested that left party voters tend to be more favourable to workers. Nevertheless, in both left and right parties, we cannot conclude definitively whether voter bias is a barrier to workers winning their elections. After all, the observed patterns of candidacy and representation could also have to do with levels of party support, financial resources, or even an offsetting combination of these factors that varies by party types. For example, it is possible that in left- wing parties voters are indeed biased against working-class candidates (e.g., those from

“routine” rather than “skilled” working class backgrounds), but that this bias is offset by higher levels of union support (leading to the observation that workers run and win at the same rate).

Electoral systems. The possible effect of electoral systems on voter bias and worker representation is another hypothesis we can begin to probe in the PARTIREP data. In this data set, there is no relationship observed between open versus closed list systems (nor division of powers versus fusion of powers systems) and levels of working-class representation in a parliament. Note that this analysis includes variation among both national and subnational parliaments in the PARTIREP data set.

Still, this is a very indirect test, which cannot rule out the possibility that personal vote incentives may amplify any existing voter biases for or against working-class candidates. What is missing in the analysis here is data on voter attitudes towards workers in each of these jurisdictions. With voter attitude data of this kind, we could more directly test for a moderating role of electoral institutions on the relationship between voter attitudes and the descriptive representation of workers.

Polarization. In the PARTIREP data (where there is comparative variation across 15 countries, more than in the CCS), I compared the share of workers among legislators in countries with different levels of polarization. Specifically, I used the rile_polarization variable from the

Comparative Manifesto Project (Volkens et al. 2019), which is a jurisdiction-level measure of dispersion of parties from the mean left-right position, weighted by each party’s vote share.

Dividing the PARTIREP countries into high and low polarization groups, I observed no significant or substantive difference in worker share based on polarization. Of course, the comparative variation is quite limited here (N=15), and the share of workers among the legislators is a measure that may be affected by factors other than voter attitudes. This is also not a direct test of Wüest and Pontusson’s (2018) polarization hypothesis, which generates 99

expectations about a reduced effect of class on voters’ attitudes as polarization rises, rather than expectations about a direct effect of polarization on levels of working-class representation.

Again, data on voter attitudes toward workers across these jurisdictions would be needed to more directly test the hypothesis.

On the whole, the limited comparative evidence tends to suggest that voter attitudes on class may have some role to play in shaping working-class representation, but they seem unlikely to be among the key barriers to workers taking office. There is some evidence in the literature that voters prefer working-class candidates, and other evidence that voters are biased against a subset of workers. Going forward, Wüest and Pontusson’s (2018) open up a potentially important avenue of inquiry into voter attitudes towards low-skilled working-class candidates, concluding that in the Swiss case, “bias against routine working-class candidates also appears to be an important factor in explaining the descriptive misrepresentation by social class” (p. 27). I have also considered possible moderating factors relating to voters’ class preferences at the election stage and probed the available data where possible, including the roles of electoral institutions and geographical distributions of working-class voters, as well as voter turnout by class and differences between parties of the left and right.

The evidence we have reviewed on barriers to working-class representation at the election stage, from both the literature and this explanatory data analysis, is summarized in Table 3.2.

100

Table 3.2. Election-stage barriers to working-class representation: summary

Literature Exploratory data analysis

Key Voter bias survey experiments: Run versus win rates: findings - Mixed results, including null, - Workers win less often than they run preference for and against workers (CCS); may suggest voter bias, but - Bias against “routine” but not inconclusive “skilled” working-class candidates - Difference in right parties only (moderator) - Low income and left-leaning voters biased against affluent candidates Electoral institutions (moderator):

Real-world elections: - No link between institution and levels of working-class representation (PARTIREP) - US workers run and win at same rate - Need voter attitudes for more decisive test - Low income Swiss candidates less likely to win their election races Polarization (moderator):

Moderators to consider: - No link between polarization and levels of working-class representation (PARTIREP) - Party type, since left voters more - Need voter attitudes for more decisive test likely to prefer workers in surveys - Electoral institutions, polarization

4.3 Self-selection and party recruitment and nomination factors

This section focuses on the stages of candidate self-selection, recruitment and nomination as potential barriers to working-class representation. We begin by considering what the literature can tell us about workers’ interest in and self-perceived qualification for office, and then examine limited evidence on the attitudes of party elites and gatekeepers towards workers.

Unions are considered as potential moderators of workers’ likelihood of success at these stages, followed by a discussion of other possible moderators including party type and the scope of the nomination selectorate. The subsequent section then probes the CCS and PARTIREP data sets for further evidence on these matters.

101

Qualification and interest in office. In the US context, one of Carnes’ (2018) central findings is that workers are simply much less likely than those from other occupational classes to run for office in the first place, and they run at rates far lower than their share of the population. He rejects the possibility that this can be explained by a lack of interest or qualification on the part of workers. In support of this conclusion, he presents surveys showing that workers do not appear to perceive themselves as less qualified for public office or express less interest in the job

(Carnes, 2018). Noting the subjective nature of concepts like “candidate quality,” Carnes gathers a set of characteristics that the public identifies in surveys as key qualifications for political candidates, which include being assertive, personable, honest and a hard worker. He then finds that US workers are as likely as those from other occupations to say they have these qualities.

In a next step, Carnes (2018) identifies the subset of respondents who say they themselves have at least four of six characteristics desirable for politicians, and he classifies these people as having “high potential” to be political candidates. He finds that among this “high potential” group, working-class respondents are just as likely as professionals to say that they feel qualified to run for office and to do the job of an elected official, and they are also just as likely to express interest in running for office. In contrast to these findings on class, the literature on gender representation identify a self-perceived lack of qualification among women (compared to men with the same objective qualifications) as one important barrier to women running for office

(Lawless & Fox, 2013). Indeed, this is a gender-based finding that is replicated in Carnes own data when broken down by gender.

While Carnes (2018) concludes that workers are as interested in office as others and see themselves as qualified, aspects of these findings are in tension with other literature that finds

102

class differences in political interest and participation (Brady et al., 1995; Gallego, 2007; Walsh et al., 2004). These studies use different operationalizations of class than Carnes and tend not to directly address the issue of running for office, but they do suggest there is reason for interpreting his findings with some caution. Still, Carnes’ (2018) replication of findings from the gender literature add to the force of his findings on class, particularly since the same methodologies are used. Overall, we should keep an open mind about political interest and ambition as potential barriers to workers pursuing office, particularly outside of the US in the absence of evidence in other jurisdictions.

A related question is whether workers judge their chances of successful election any less favourably than people from other class backgrounds. This would be consistent with workers being less likely to pursue a nomination, even if they are equally interested and see themselves as qualified. This possibility is examined in the data analysis below (in the literature, I am not aware of existing evidence on this question with respect to class).

Recruitment and selection by party elites. Another key set of non-resource barriers to workers running for office in the US, as identified by Carnes (2018), relates to the recruitment practices of political parties. In surveys, US party leaders report recruiting relatively few workers and have lower opinions of their potential as candidates compared to those from other occupations (and these attitudes don’t seem to change much in safe compared to marginal seats; Carnes, 2018, ch.

3). Among state legislative candidates, compared to those from professional backgrounds, working-class candidates report receiving less encouragement from incumbent local politicians

(when asked in a survey who had encouraged them to run). Surveys also show that party gatekeepers with lower incomes themselves, or who have personal connections to workers,

103

tended to have more favourable attitudes about the potential of workers as candidates (Carnes,

2018, ch. 4).

One of the few directly relevant observations from the comparative literature on party elites as potential barriers to working-class representation comes from Norris and Lovenduski (1995), though they do not provide the same data as Carnes (2018) on the explicit attitudes of party elites in their UK data. They find that overall the pool of those people applying to be candidates for a party generally had quite similar income compositions to those that the party actually approved as candidates. However, this did not hold among low income earners and workers. In the Labour

Party, those making under £10,000 applied to candidacy at about twice the rate that they were approved (p. 149). While this group accounted for 16% of those applying for candidacy, they made up only 7% of candidates accepted by the party, and none of its MPs. Similarly, examined in occupational terms, manual and clerical/sales workers accounted for about 20% of those applying for candidacy in the Labour party, but were only 11% of those accepted as candidates

(p. 112). The Conservative Party had too few low income and working-class people applying for candidacy in the data to examine the hypothesis of party elite bias in this way.

One possible explanation for this pattern could be that Labour party elites are indeed biased against these lowest income prospective candidates, as Carnes (2018) finds in the US. If so, on what grounds? This could be due to a mismatch in their ideological preferences, since the attitudes of workers tend on average to be to the left of other candidates in their party on economic issues, as we saw in the first paper of this dissertation and the literature. Alternatively, party elites may be reluctant to back or nominate workers if they believe they will be less viable candidates, either due to voter attitudes or a lack of financial resources and fundraising potential

104

to help mount an effective campaign. In the absence of direct evidence about party elites’ attitudes, the exploratory analysis below examines whether working-class candidates are more likely to have prior party connections than their counterparts from other backgrounds, as well as whether these party connections are associated with how likely they are to win office.

4.3.1 Moderators

Unions. The role of unions in encouraging workers to run for office also stands out in Carnes’ account. Workers appear to run and win more often in state legislatures where states have higher unionization rates (Carnes, 2018; Sojourner, 2013). Furthermore, because this relationship holds even when controlling for factors like the levels of political participation and knowledge among workers, Carnes suggests that the relationship may be mediated by facilitating recruitment, noting:

“as important players in electoral politics, moreover, unions often serve as a bridge

between workers and political elites, getting talented blue-collar Americans involved in

the kind of activism that connects them to party leaders, politicians, interest groups and

other elite actors. In some states unions actively partner with candidate recruitment and

training organizations… [and] even run their own candidate recruitment and training

programs” (2018, p. 151-2).

There are some indications from limited evidence that this relationship may hold in other country contexts. Norris and Lovenduski (1995) found that in the United Kingdom Labour Party candidates backed by unions were more likely to be manual workers (though this was a pattern

105

that declined over time). In turn, union backing appeared to increase the likelihood that applicants would be selected for a nomination in the Labour Party.

While Carnes (2018) emphasizes the role of unions in facilitating recruitment, increased levels of political participation and knowledge among unionized workers could also mediate the relationship between unionization and worker representation, increasing workers’ likelihood of self-selecting into politics. Mechanisms aside, in the exploratory data analysis below, we examine whether workers have more representation in places where unions are stronger.

Party type. Viewed through the lens of party type, perhaps not surprisingly, Norris and

Lovenduski’s (1995) data shows more representation of low- and middle-income people, as well as more workers, among the pool of approved Labour Party candidates than in the Conservative

Party (though, as we saw above, workers applied for candidacy in the Labour Party at a higher rate than they were actually approved by party elites). Class representation and the potential biases of party elites may differ by party type. In contrast, Carnes (2013, 2018) finds that levels of working-class representation (measured in terms of occupation, not income) barely differ at all between political parties at both the national and state level. Indeed, party type and ideology are explicitly de-emphasized in his account of barriers to working-class representation in the US.

Carnes (2013) also notes that Democratic congressmen have a lower average net worth than their

Republican counterparts in the United States, but both groups are far wealthier than average

Americans. To shed light on a broader set of countries, possible differences by party type in the data are examined in the exploratory data analysis below.

One other finding in the comparative literature relevant to the role of parties comes from

Gherghina and Chiru (2010), who find that Romanian parties tended to favour wealthier 106

candidates in terms of party list position under the country’s closed-list proportional representation system, in a study of elections to the European Parliament. Although I do not have the data to do so here, a similar line of inquiry for future research would be to further examine the ordering of candidates on party lists across jurisdictions. That is, where parties have discretion over the ballot ordering of their candidates on electoral lists, we can observe the selective effects parties exert (e.g., disadvantaging or privileging working-class candidates). In an open list system, we could also observe voter preferences and more effectively separate out the preferences of the party and voters over the same set of candidates.

Other potential moderators. The potential role of party gatekeepers as barriers to workers is probed in the exploratory data analysis, including examining whether a broader or narrower nomination “selectorate” (Best and Cotta, 2000) might be more favourable to workers reaching candidacy. For example, if party elites have different attitudes towards workers than a broader member-driven selectorate, this could be important. In addition, it has been observed in the literature on the descriptive representation of gender that women face more competitive primaries than men, and that this constitutes one of the barriers to their representation (Lawless

& Pearson, 2008). Whether workers face a similar disadvantage in facing more competitive nomination contests is another possibility that is explored in the data analysis below.

In sum, the literature provides evidence that, at least in the US, workers see themselves as qualified and express similar interest in running for office as those from other occupational backgrounds. Party leaders surveyed in the US also have lower opinions of workers compared to prospective candidates from other occupations and reported recruiting few of them. Evidence from outside the US is much sparser, though we observed that more low-income individuals and

107

workers applied for candidacy than were approved by UK Labour Party. In terms of moderators, union strength in the US is associated with higher worker representation, and a role for unions in facilitating worker representation is also suggested in the UK. We also observed some suggestive evidence of party type as a moderator, as well as raised the potential roles of the nomination selectorate and the intensity of competition. On the whole, the literature is too sparse to say with confidence whether patterns of “filtering out” workers at the nomination stage are widespread, let alone what the specific drivers of this might be.

4.4 Exploratory data analysis: self-selection and recruitment

Informed by the literature, in this section the CCS and PARTIREP data are examined to for clues about barriers to working-class representation at the stages of self-selection, recruitment and nomination.

Perceived chances of winning. Because our data only shine a light on those who have already become candidates, it is particularly difficult to find indicators that speak to the issue of self- selection. As one indirect indicator, in the CCS data we can observe a trend that workers rated their own chances of winning as slightly lower than other candidates, but not at a level that is statistically significant at conventional levels. If workers did see their chances as poorer than others, this could be seen as consistent with the possibility that they would self-select out of pursuing candidacy in the first place. But here the data only refer to those who did choose to pursue candidacy and were successfully nominated, so it would not be possible to draw clear conclusions even if the difference were larger and statistically significant.

108

Personal connections. While the data at hand cannot directly shine light on the attitudes of party gatekeepers towards workers, some indirect evidence can be examined regarding the relationship between party elites and workers’ success. One CCS question relates to whether workers were likely to have party connections, specifically by having previously worked for an MP or for the party itself. As we might expect, winning candidates of all classes were more likely than unsuccessful candidates to have these types of party connections. However, there is no relationship observed between class (workers compared to other occupations) and the likelihood of having such party connections.

4.4.1 Moderators

Party type. In terms of conditional effects by party, in the PARTIREP survey, workers made up a substantially larger share of legislators in left parties (12%) than in other parties (6%; p < 0.01).

In the CCS data, workers also trend towards a larger share of successfully elected candidates in left parties (14%) compared to other parties (10%), but this difference is not statistically significant at conventional levels (p = .13). Among the full pool of candidates in the same data set, workers make up a larger share in left parties (15%) than in other parties (13%), though this result was only significant at the p < 0.1 level. Thus, as expected, we do generally observe a difference between party types in terms of worker representation, with more workers in left parties. The mechanisms driving these differences, including disentangling the possible roles of party elites and of their voters, are much less clear.

Unionization. In the PARTIREP data, we can also observe a relationship between unionization and working-class representation. Splitting the countries into two groups, in the “high” union 109

density countries, about 11% of the legislators responding to the survey were workers, compared to 5% in the “low” union density countries (p < 0.1). Similarly, in the CCS data, we see a 19% worker share in high union density countries compared to 8% in low union density countries (p <

0.1), albeit with a very small sample of countries (N=7). At an individual level, a larger share of working-class candidates (35%) in the CCS sample are union members compared to other occupations (29%), a difference that is statistically significant at the p < 0.01 level. In turn, individual candidates who are union members are also more likely to be successfully elected

(17%), compared to those who are not members (14%; p < 0.05).

There is also some indication that workers tend to get a bigger boost to their election chances from union membership than non-workers. In descriptive terms, 18% of worker-candidates who were union members in the data set were successfully elected, compared to 10% of workers who were not union members (p < 0.01). In contrast, among candidates from other occupations, 17% of those with union membership were successfully elected candidates compared to 15% of non- union members (p < 0.1). However, this pattern should be interpreted with some caution, as when examined as an interaction between class and union membership in a regression, the interaction coefficient is not statistically significant at conventional levels. These patterns are at least consistent with the possible role of unions as recruiters and trainers of working-class candidates, as well as with their providing direct material support in nomination processes or elections. Further comparative research on the role of unions should examine these potential mechanisms more directly, as well as examine the possible moderating effect of different levels of linkages between parties and unions across country contexts.

110

Selectorate size. Two further potential moderators identified in reviewing the literature can be examined in the CCS data. One is the scope or size of the selectorate in party nomination processes, which ranges from “all voters” to only the party leadership making nomination decisions. The size of this selectorate might be relevant if party elites have different attitudes towards workers than members or voters. Indeed, in the CCS data, candidates who were nominated by a broader selectorate (party members or voters) were more likely to be workers

(21%) than those nominated by a narrower group (party conference or leadership; 16% workers, p < 0.01), which is suggestive of the possibility that party elites do constitute a barrier to worker representation.

Nomination competitors. Another moderator is the competitiveness of nomination contests, an issue that arises in the literature on barriers to women’s descriptive representation, as discussed above. In the CCS data, we can observe that working-class candidates reported facing more competitors in nomination contests than their counterparts from other classes (p < 0.01).

However, we do not have information in this data set on the pool of unsuccessful nomination seekers, and thus we cannot determine whether workers faced more contested nominations among this larger group. It is not clear why workers should face more contested nominations than their counterparts, but one possibility would be that party elites are less favourable towards workers and thus less likely to cohere in support around their candidacies when they run.

It is worth reiterating that, as the literature on barriers to working-class representation continues to be developed, it will have to contend with how factors at various stages interact with each other. For example, if there is party gatekeeper bias against workers, this may cascade back to the stage of self-selection and discourage workers from pursuing nominations to begin with. The

111

attitudes of party elites, in turn, will no doubt be influenced by their perceptions of voter preferences. As Ashe and Stewart (2011) point out, this blurs the lines between “demand” and

“supply” side explanations of descriptive representation.

Table 3.3. Self-selection, recruitment barriers to working-class representation: summary

Literature Exploratory data analysis

Key Qualification and interest: Self-assessment of chances: findings - US workers express interest in and see - Workers assessed slightly lower chances selves as qualified for office; but this of winning, but not significant (CCS) is in tension with other findings on interest and participation by class Unions (moderator):

Recruitment and party elites: - More working-class legislators in countries with higher union density (CCS, - Party leaders in US view workers less PARTIREP) favourably, recruit fewer of them - Worker-legislators more likely in unions. - In UK workers applying for candidacy Union members more likely to win (CCS) less likely than other occupations to be approved by party (Labour) Party type (moderator):

Unions’ role: - Working-class legislators more likely to be found in left parties (CCS, - Higher union density associated with PARTIREP) more worker representation in US - Unions back worker-candidates in the Other moderators: UK, increase chances of success - Workers less common among candidates facing a narrower selectorate (CCS) - Workers report facing more competitors in their nomination contests (CCS)

4.5 Disadvantages in material resources and time

This section examines some of the defining characteristics of working-class status: disadvantages in material resources. This set of factors has the potential to pose barriers to working-class representation at each of the stages examined in this paper: self-selection, recruitment and

112

nomination, and election by voters. We first review the relevant literature examining matters such as propensity to seek candidacy by income, resource barriers to running that workers identify in surveys, as well as variation in worker representation across jurisdictions with different campaign costs and inequality. Further evidence is then probed in a final set of exploratory data analyses in the subsequent section.

Evidence from United States. A range of evidence suggests that resource disadvantages are key barriers to working-class representation in the United States. Carnes (2018) finds that working- class people face important practical barriers that discourage them from running for office (a self-selection effect in the framework above). In particular, workers tend to have less job flexibility and economic security to be able to bear the forgone income and risk of taking time off work to run a political campaign. They also often lack the money to personally pay for expensive campaign costs, and they are less likely to have a network of well-heeled friends and colleagues able to contribute money to their campaigns. In short, resource disadvantages may exert direct effects at all stages, including on a worker’s likelihood to pursue candidacy, as well as their ability to mount competitive nomination and election campaigns.

Evidence for these claims in the US comes from a mix of survey research and comparisons across jurisdictions within the country. The National Candidate Study found that political candidates themselves, when asked directly about barriers to working-class people in politics, identified lack of time and money the two largest issues (Carnes, 2018, ch. 3). In another survey, well-qualified working-class people identified “giving up my income or job to run for office” as a worry about running and did so significantly more frequently than professionals (Carnes,

2018). The role of money and time as barriers to working-class representation in a wider set of

113

jurisdictions is considered in the exploratory data analysis below, drawing on questions about campaign budgets and duration of full-time campaigning.

Comparative literature. Given the unusual level of money in US politics, as well as the country’s high inequality and weak economic security, we should be cautious about drawing broader inferences from US evidence alone. There only appear to be a handful of non-US studies in the literature that address the role of material resources and time available as barriers to workers running for and taking office. As discussed above, Wüest and Pontusson (2018) find that in Switzerland, the pool of successfully-elected parliamentarians is more affluent than the typical

Swiss person, with 86% of having incomes above the median Swiss, while the broader pool of candidates (elected and unelected) was much more representative with only about 59% above the median income. Thus, there appears to be a correlation between material resources in the form of personal income and winning elections in Switzerland, while in the US where workers tend not to reach candidacy in the first place.

However, it remains unclear what causes this drop-off in class-based representativeness in the

Swiss case at the election stage. Lack of financial resources might be playing a direct role in the less favourable prospects for lower-income candidates, weakening their ability to mount a competitive campaign. But there are many among other possibilities. The pattern could be accounted for by voter bias against routine working-class candidates observed in the same study

(discussed in a previous section), since Switzerland has an open list proportional representation system where voters can express preferences for individual candidates. Party elites might also tend to place workers in less favourable positions on the party lists presented to voters (either for

114

their own reasons, or anticipating voter bias), since in these Swiss elections parties control the list order and can also present the option to vote for certain candidates twice.

Another comparative data point comes from Norris and Lovenduski’s (1995) study of political recruitment in the United Kingdom, which finds that “financial resources affect the supply of those who come forward to seek a Westminster career” (p. 148). Consistent with the authors’ overall conclusion, the biggest barrier to workers taking office that appears in their data is that lower-income people are not pursuing candidacy in the first place, more closely resembling the

US than the Swiss case. In both the Labour and Conservative parties, lower income individuals are much less represented among parliamentary candidates and MPs than they are in the makeup of the parties’ voters.

The data Norris and Lovenduski (1995) present show that those making under £10,000 per year made up 67% of Labour Party voters, but only 16% of those applying for parliamentary candidacy. In turn, this group made up only 7% of prospective candidates accepted by the party, and none of its MPs. In the Conservative party, this group made up 1% of applicants and candidates and none of its MPs (even though they constituted 42% of the party’s voters). Thus, for the most part, those on lower incomes appear to have been “filtered out” before they even pursued a candidacy. It is difficult to say from such limited data, however, what accounts for the observed pattern of differences across these countries.

In interpreting their findings, Norris and Lovenduski (1995) make use the concept of "brokerage occupations” in a way that is broadly consistent with Carnes’ (2018) finding that money and time are important barriers to the supply of workers running for office. Norris and Lovenduski note that, 115

“Parliamentary careers are facilitated by jobs which combine flexibility over time,

generous vacations, interrupted career paths, professional independence, financial

security, public networks, social status… they minimize the costs and risks of horizontal

mobility from the economic to political marketplace” (1995, p. 110).

The authors found in interviews with MPs that they often explicitly identified these occupations as facilitating their political ambitions, and some chose these career paths in part for this purpose.

4.5.1 Moderators

There are also moderating factors to consider, which could potentially help explain some of the patterns we have observed across a limited set of cases in the literature. For example, we have observed that the class makeup of the UK candidate pool seems to bear closer resemblance to the

US than Switzerland. The UK and US share higher levels of inequality than Switzerland, and we can see in the literature (and in the exploratory analysis below) a correlation between inequality and worker representation, suggesting a possible role for material resources. The US and UK also share a single member district-based electoral system, but in the exploratory data analysis above, we observed no correlation between electoral institutions and worker representation. On the other hand, the UK’s party-controlled nomination contests contrast with the more open US system of primaries, and arguably bear somewhat more similarity to Switzerland’s powerful parties. From these patterns alone, and such a limited set of cases, needless to say we cannot determine the mechanisms at work with any clarity.

116

There is more to say about moderating factors within the US. Comparing across states and congressional districts, Carnes (2018) finds that working-class people tend to become candidates and win office less often in places where campaign costs are higher compared to where they are lower. He also finds that workers are less likely to hold office in cities with more practically burdensome elections (i.e., where there is more ground to cover and more voters to contact), such as in citywide compared to district-based elections. In general, worker representation is higher in subnational governments compared to the federal government in the US. Moreover,

Carnes finds that in US states where economic inequality is higher, workers are less represented among both candidates and officeholders. This could be as a result of sharper resource differentials between worker and other candidates.

While these findings are suggestive, given some of the United States’ exceptional characteristics, we have to be careful about making extrapolations from evidence using this within-country variation. The possible moderating role of campaign costs, tier of government, and inequality levels across jurisdictions are explored further in the exploratory data analysis below.

Finally, it bears emphasizing that the disadvantages of workers in terms of material resources and time may affect their representation at different stages of the political process. They may impact the likelihood of running in the first place, successfully winning a nomination contest, and being elected to office. The sparse literature is at least suggestive of possible effects at each stage. Anticipated barriers at a later stage may also exert effects at earlier stages of the process.

For example, if workers tend to have fewer resources and less time to campaign in a general election (and fewer professional contacts to shore up those resources), then party elites may be less likely to support them for the nomination (over and above the direct material disadvantage

117

workers face in the nomination contest itself). In turn, if workers anticipate that party elites will not consider them viable candidates, they may be less likely to bother pursuing a nomination. All of these possibilities may then be affected by contextual factors such as how resource-intensive political campaigns are in different jurisdictions, as well as how large are the economic disparities between prospective candidates from different class backgrounds.

In brief, we have seen evidence that US workers explicitly identify resources and time as barriers to their running for office. This is at least consistent with the UK study examined, where low- income earners applied for candidacy at a much lower rate than their makeup as a share of the party’s voters. In Switzerland, the pool of parliamentary candidates was relatively representative in terms of income, but low-income candidates were much less likely to actually be elected to office. The causal interpretation of these real-world data, however, is indeterminate. In US states, working-class representation is higher where campaign costs and inequality are lower, though there is little evidence on whether these patterns extend to other country contexts. Working-class representation also appears to be higher at the subnational level compared to the national level in the US, where politics is more resource intensive.

4.6 Exploratory data analysis: resource and time disadvantages

Time spent campaigning. Do working-class candidates have less time available to dedicate to campaigning than those from other occupations? One relevant question in the CCS data relates to how far in advance of elections candidates get their campaigns underway. Workers reported organizing their campaigns later than their counterparts from other occupations (p < 0.01), and they also reported undertaking full-time campaigning later (p < 0.01). These differences hold 118

among both the overall pool of candidates and the subset of those who were successfully elected.

This is at least suggestive of workers having less time available to dedicate to campaigning, which could help explain their lower levels of electoral success (and potentially their lower likelihood of running for office). Of course, there are other possibilities. For example, it could be that the working-class candidates were part of campaigns with less realistic prospects of victory and therefore sensibly dedicated less time to their campaigns. Indeed, we noted in a previous section that working-class candidates in the CCS data trended towards judging their chances of winning as slightly lower than those from other occupations, though this difference was not statistically significant at conventional levels.

Campaign budgets. We also expect working-class candidates to be at a disadvantage in terms of campaign budgets. Does this bear out in the CCS data set? Indeed, the average campaign budget reported by working-class candidates in the survey was lower (€7,583) than candidates from other occupations (€11,378; p < .05). Among the smaller subset of candidates who were successfully elected, the class difference in campaign budgets remained substantial (€14,332 for workers, compared to €20,794 for others; p < .05). The relationship also holds when campaign budget is regressed on the disaggregated occupation categories along with control variables including a left party dummy variable, country dummies, and elected status (with standard errors clustered by country; see Figure 3.2).53 Thus, workers were less resourced than their counterparts from business and most other occupations. The data also show that, not surprisingly, election winners were much better resourced than unelected candidates. Still, we cannot definitively

53 Regression results can be found in Appendix Table C.26. 119

conclude that there is a causal relationship between these resource disadvantages of workers and their levels of representation.

Figure 3.2. Estimated class-based differences in campaign budgets (Euros)

Note: Lower scores correspond to lower campaign budgets compared to the reference category (CCS data). Dots are coefficients and lines are 95% confidence intervals. Business is the omitted reference category for occupation.

Staff team size. Another complementary measure of campaign resources found in the CCS data is the size of a candidate’s campaign staff team. Again, working-class candidates trended towards having smaller campaign teams compared to other occupations (but not reaching statistical significance at conventional levels; p = 0.11), as well as getting fewer campaign staffers provided by the party itself (p < 0.05). Among the smaller subset of elected candidates, these trends hold in both cases but are not statistically significant at conventional levels.

However, the samples available for analyses of the staffing variables are particularly limited,

120

covering only Austria, Germany and United Kingdom for overall campaign staff, and only

Germany for party-provided campaign staff. Perhaps not surprisingly, successfully elected candidates have a larger campaign teams and more staffers from their parties compared to the pool of unelected candidates.

These findings in the CCS data show that working-class candidates had fewer campaign resources in terms of both budget and staff. They do not establish a causal relationship between material resources and working-class representation. However, they are at least consistent with the possibility that workers’ lower levels of material resources are a contributing factor to their lower levels of electoral success. Among the possible alternative explanations, voters and/or party elites might be biased against workers, and this bias could be reflected in both lower levels of electoral support and lower levels of financial support that appear in their campaign budgets.

Distinguishing between these and other mechanisms should be an important goal of future research in this area.

Furthermore, because the CCS data only covers those who have already become legislative candidates, they cannot show us to what extent material resources may be a barrier to workers pursuing candidacies in the first place, which Carnes (2018) identifies as the most important stage at which workers are “screened out” of political representation. More information needs to be collected on the earliest stages self-selection, recruitment, and nomination, and in particular on those potential working-class candidates who are filtered out of the running before reaching the stage of candidacy (most of the currently-available data covers only this stage or later).

121

4.6.1 Moderators

Resource-intensity. We can also look at the resource-intensity of politics as a potential system- level moderator, aggregating the CCS sample by country and comparing the four countries with the highest average campaign budgets and the four countries with the lowest campaign budgets.

However, with only eight country observations this comparison provides little statistical power.

We can at least note that the “high campaign budget” countries trend towards a smaller share of working-class candidates of 14.4% compared to the “low campaign budget” countries with

17.8% (t-test does not approach significance at conventional levels), consistent with expectations.

Level of government. As mentioned above, Carnes (2018) finds that working-class representation is higher at subnational levels of government in the US than federally, taking this to be, in part, a consequence of less materially-demanding elections at lower levels of government. This pattern bears out in the countries covered by the PARTIREP survey, where working-class representation is substantially higher in the regional compared to the national legislatures surveyed (p < 0.05; see Figure 3.3).

122

Figure 3.3. Working-class share by level of government

Note: Worker shares in PARTIREP data set. Dots are group means and lines are 95% confidence intervals.

Income inequality. Carnes (2018) also observed that fewer workers in the US ran for or held office in states with higher levels of income inequality, suggesting that this pattern may be owing to high levels of resource imbalances by class among prospective legislators. I observe a similar trend in the PARTIREP data when splitting the countries into high and low inequality halves, based on the OECD’s P90/P10 ratio.54 The higher inequality countries have a lower share of working-class legislators among those surveyed (6%) than the lower inequality countries (9%), but the pattern is not statistically significant at conventional levels (this is with weak statistical

54 The ratio of incomes at the 90th percentile compared to the 10th percentile in each country. 123

power, N=15). In the CCS data, we see that high inequality countries tend to have a marginally lower share of workers among the successfully elected candidates (13%) compared to lower inequality countries (16%), but this result also does not approach statistical significance at conventional levels and is based on a very small sample (N=7).55

While neither of these underpowered analyses provides statistically significant findings, the trends combined with Carnes’ (2018) prior findings suggest it is worth continuing to examine the possibility that inequality moderates the effect of resource disadvantages on working-class representation. Furthermore, since the effects of resource disadvantages may extend across the stages of self-selection, party recruitment and nomination, and election by voters, the impact of potential moderators like inequality may, too, cascade across these stages. Of course, levels of working-class representation may also be among the determinants of inequality levels, dovetailing with the findings in the first two papers of this dissertation. Indeed, there is reason to imagine that causation might flow in both directions.

In sum, the US and sparse comparative literature, along with the additional data points from the

CCS and PARTIREP surveys, provide some reinforcement to the proposition that material resources and time may be important barriers to working-class representation, even across country contexts. There is also suggestive evidence that system-level moderators of these effects, including inequality and the resource-intensiveness of elections, are worthy of further examination in future research.

55 The same pattern holds among the narrower pool of successfully elected candidates in the CCS data, again not approaching conventional levels of statistical significance. 124

Table 3.4. Resource barriers to working-class representation: summary

Literature Exploratory data analysis

Key Self-assessment of barriers: Time campaigning: findings - Loss of income/job identified by US - Workers get later start on full-time workers as key barriers to running campaigning than candidates from other class backgrounds (CCS) Candidacy rates by income: Campaign costs (individual): - UK study: low-income earners rarely reach candidacy, let alone office - Working-class candidates and legislators - Swiss study: incomes of candidates have lower campaign budgets (CCS) typical of population, but incomes of elected officials well above average Moderators:

Moderators (US only): - More working-class candidates in countries with lower than average campaign budgets - More workers reach office in states (non-significant trend; CCS) where campaign costs are lower - Higher inequality associated with lower - Fewer workers reach office where worker representation (non-significant levels of inequality are higher trend; CCS and PARTIREP) - More workers at subnational than - Higher worker share at subnational than national level (resource-intensive) national level (PARTIREP)

5 Policy solutions and interventions

As we have seen earlier in this dissertation and the emerging literature, the unequal representation of class has observable consequences in the representation of policy priorities.

Like the underrepresentation of women and racialized minorities in most legislatures, the underrepresentation of workers can also be seen as a social problem in itself. Having examined the literature and probed two new data sets, in this section, I consider what the limited body of evidence suggests can be done to actively improve and equalize working-class representation in legislatures.

125

As we have seen, albeit still from limited evidence, workers’ disadvantages in terms of material resources seem to be one of the important barriers they face in running for and winning office.

We have seen indications that their access to resources and time may leave workers at a disadvantage while competing in election and nomination contests. Resource disadvantages may discourage them from pursuing candidacy to begin with. Therefore, I first examine various approaches to addressing the resource disadvantages faced by potential working-class politicians.

Second, I consider more direct approaches to recruiting workers into politics. In the US case, we have seen evidence that workers are rarely being recruited to run for office, even if they prove to be capable contenders when they do run. While the evidence is limited and more mixed in other jurisdictions, there are certainly indications that workers are failing to self-select or be recruited into candidacy. Practical mechanisms to actively recruit more workers into political life may constitute another key set of potential solutions to address low working-class representation.

I draw in particular on a set of solutions considered by Carnes (2018) in the US context, critically examining how these might travel to other jurisdictions and proposing other possible solutions.

5.1 Addressing resource disadvantages

To address resource disadvantages faced by workers, one avenue to pursue would be large, structural change in a country’s political economy to achieve a significant redistribution of income and wealth. In the US context, Carnes (2018) identifies the prospects as dim for significant redistribution of economic resources, at least anytime soon, given the institutional barriers such policies would face, and he dismisses the redistributive approach to increasing

126

worker representation as a “long shot.” That may be a plausible assessment, but it is notable that aggressive redistributive proposals are rapidly moving into the political mainstream even in the

United States. The types of wealth taxes recently proposed by Bernie Sanders and Elizabeth

Warren in the 2020 Democratic primary enjoy overwhelming public support, including majorities of Republicans (Casselman & Tankersley, 2019). Similar redistributive proposals also enjoy strong support in public polling in other countries including Canada and the United

Kingdom. Across the developed world, the bounds of economic debate seem to be in substantial flux, and the range of politically realistic redistributive policies could change quickly.

Therefore, as a means of alleviating the resource disadvantages that seem to help keep workers out of office, redistribution should not be dismissed. In countries with fewer institutional barriers to change than the United States, this might be particularly true. Other broad-based reforms that strengthen the economic security of workers might also enhance their ability to bear the costs and risks of pursuing public office. These could include changes like higher minimum wages, stronger protections from dismissal, shorter working hours and longer holidays, more union representation, as well as (particularly in the US context) guaranteed like health care not tied to an employer.

We have already seen that US states with lower levels of inequality tend to have more worker representation (Carnes, 2018), and there is some modest but suggestive evidence of a similar pattern in other contexts in the PARTIREP and CCS data sets. In terms of increasing worker representation, we might expect the efficacy of incremental redistribution and stronger economic security to vary comparatively. For example, from a baseline of low inequality and strong economic security for workers, further enhancements in these areas may have a weaker effect on

127

promoting their representation. Conversely, in country contexts with high inequality and weak economic security, enhancements in these areas might be expected to have a stronger effect on worker representation by easing what amount to more severe resource disadvantages faced by workers aspiring to office.

But there is also a kind of circularity to this redistributive approach to enhancing worker representation. If the descriptive representation of workers matters in large part because of the substantive positions they tend to take on redistributive issues (as per the first two papers in this dissertation), then we are essentially talking about simultaneously addressing both the descriptive and substantive halves of the equation. Indeed, evidence that fewer workers are found in office where inequality is higher could be indicative of causation in either or both directions (or, of course, represent a spurious correlation). That is, inequality may pose a barrier to working-class representation by intensifying workers’ resource disadvantages in seeking office; and/or low working-class representation may shape economic policy in a way that ultimately increases inequality. While the limited evidence must be interpreted with caution, it is consistent with the possibility that incremental advances in redistribution may enhance worker representation, which could in turn increase redistribution in a self-reinforcing process.

Public financing of elections is a more intermediate-level policy approach to address resource imbalances and increase worker representation. If a lack of campaign funds is a barrier to working-class candidates, then providing public funds, or indeed regulating and limiting overall spending levels, may plausibly help level the playing field. In the US context, though, Carnes

(2018) shows only a quite small positive relationship between various heterogenous public campaign financing schemes at the state level and the rates of worker representation. He argues

128

that, while public financing may ease fundraising pressures, it does not ease the personal financial costs and risks workers face when they interrupt their jobs to run for office. In other words, it does not reduce the overall difficulty of running a campaign that workers report they face in surveys of state legislative candidates. Notably, party recruiters also do not report recruiting more workers in states with public financing of elections. As Carnes notes, a weakness of this approach is that public campaign financing is not targeted specifically at workers, but rather it benefits all candidates, though surely the relative benefits are different.

However, from a comparative view, the US is an extreme case in terms of the cost of running elections and the weakness of regulation of campaign spending and donations, which means that private money can still swamp all but massive infusions of public financing. Therefore, it may be a particularly hard test for the potential of public financing of elections to improve worker representation. In jurisdictions where private money plays less of an outsized role, it is possible that public financing may have more of a realistic chance to help equalize opportunities for perspective working-class candidates.

In other words, public financing of a given magnitude may have a larger proportional impact where elections are cheaper and private political donations are more tightly regulated. Under these conditions, public financing could help create a spending floor that helps ensure that resource-disadvantaged candidates reach a threshold of viability. We have already seen above indications in the CCS data that workers had smaller campaign budgets than their counterparts from other occupations, and that successful election winners tended to have larger budgets. If public financing helped reduce campaign budget disparities in these lower cost contexts, it is plausible that it could help increase working-class representation more substantially than seen in

129

the US research. Even in the US, it is possible that more aggressive forms of public financing might get traction, and these are possibilities that should be explored in future research comparing the effects of public financing across jurisdictions.

Another approach identified by Carnes (2018) in the US context is the idea of boosting working- class representation by increasing legislators’ levels of pay. If money is a barrier to workers running for office, then it seems reasonable to expect that ensuring the job itself pays a good salary might help. However, he finds no such relationship among US states, and indeed some indication that working-class representation is lower in states with higher levels of legislator pay.

This would seem to be because legislator pay does not alleviate the costs and risks of running and is only helpful after one succeeds in taking office. Still, it might be the case that some minimum level of legislative pay is needed to make running for office viable for workers, but that pay is not a binding constraint where other barriers to running for office already abound.

Indeed, it seems likely that a combination of interventions at different stages of the political process may be required to effectively address resource barriers to working-class representation.

The issue of legislative pay and class representation should be examined further, particularly in a comparative context where there is currently little evidence to draw upon.

A further and complementary approach would be providing workers with targeted campaign seed money. Carnes (2018) identifies seed money for working-class candidates as one of the most promising interventions in the US context, suggesting efforts of the kind that EMILY’s List engages in for women. The logic is straightforward: seed money provides early funding to help promising campaigns get off the ground and is meant to help attract additional donors with this early show of strength. Seed money can also be attractive to party recruiters concerned with

130

candidates’ fundraising abilities. In the same spirit, newer organizations in the US like Justice

Democrats and Brand New Congress provide funding and volunteer support to the largely working-class candidates they endorse.

The practicality of the seed money approach may vary substantially across context. For example, in jurisdictions where only individuals (and not organizations) can donate to political candidates, this approach may be more difficult to implement. However, in such cases, organizations could serve a similar purpose by using other tactics to draw the attention of individual donors and party elites to promising candidates, such as developing a slate of endorsed candidates. The payoff of the seed money approach would also depend in part on the resource-intensity of political campaigns, including the state of campaign finance regulations, and thus vary across jurisdictions. Where campaign costs are high and large amounts of private money are already flowing through the political system, seed money targeted at workers may be particularly effective by helping to level the playing field. Candidates that align with the interests of well- heeled donors may already be, in effect, accessing “seed money” from them in these types of systems. In contrast, where tighter regulation of private political financing is possible, regulation may be a more desirable starting point and seed money for workers a “second-best” option.

A related approach to addressing the resource disadvantages of prospective working-class candidates is “political scholarships” specifically targeted to this group (Carnes, 2018). The key distinction here is that scholarships would provide money to ease the personal expenses of working-class candidates. This would provide some financial cushion against the lost income and economic risk that may be necessary to mount an effective political campaign. A similar idea is to grant unemployment insurance benefits for the period of a campaign to those pursuing

131

public office, as recently proposed in the Canadian context (Nayler, 2019). In theory, this type of reform could be targeted to potential working-class candidates by income-testing it. A complementary proposal would be changes to labour laws that require employers to allow employees unpaid (or even paid) leave from work to pursue public office, reducing employment risk for workers. Again, we would expect the effectiveness of these reforms to vary across contexts. For example, where economic security is weak and inequality is high, political scholarships might be particularly promising as a way to bolster workers’ effective ability to run for office without untenable loss of income or risk of job loss.

A key advantage of the seed money and political scholarship approaches proposed by Carnes

(2018) is that they are (or could be) targeted specifically to prospective working-class candidates.

There are also complementary with other approaches like public financing of elections or redistribution more broadly, and a combination of efforts would likely be necessary to address resource barriers to working-class representation in legislatures.

5.2 Recruiting workers to candidacy and office

Recruitment and training programs specifically designed to encourage working-class candidates to run for office are an approach to increasing worker representation emphasized by Carnes

(2018) in the US context. This follows from his conclusion that a key barrier to working-class representation is that workers tend not to self-select or get recruited to run for office in the first place, even though they seem to perform similarly well in US elections when they are advanced as candidates. He adduces some systematic and anecdotal evidence that these types of training and recruitment programs are effective, while also arguing that they are relatively easy to set up. 132

For example, he finds a statistically significant increase in worker representation in New Jersey after the opening of the New Jersey Labor Candidates School in 1997, noting that labour groups in other states such as Maine and Nevada also set up programs aimed at emulating the New

Jersey model.

The Pipeline program of the Working Families Party is also identified as employing a similar candidate recruitment function (Carnes 2018). In the past few years, Justice Democrats and

Brand New Congress have also emerged as organizations in the US with the aim of electing candidates from the “working class” and “regular working people,” according to their websites.

Justice Democrats famously helped Alexandria Ocasio-Cortez win the Democratic primary and

House of Representatives seat in New York. Carnes even set up a candidate recruitment and training program himself, and he found that its working-class participants went on to run for office at the same rate as other participants. The key, from this view, appears to be to get workers in the door in the first place. Since the main requirement to set up a successful program is to have ongoing contact with working-class people (and some modest financial resources to cover costs), organizations like unions are well-positioned to launch these types of programs.

Broockman (2014) ran an experiment providing evidence that personalized candidate recruitment efforts do have an effect in encouraging recipients to run. Using thousands of emails to advocacy organization members, he found that receiving a recruitment request personalized to the individual significantly increased the interest they expressed in running for office. Broockman’s survey of the literature also found that candidates frequently cite recruitment efforts and encouragement as among the top reasons that influenced their decision to run for office.

133

An area that requires further investigation is how the effectiveness of civil society recruitment and training efforts may differ across jurisdictions and contexts. For example, how successful might such efforts to encourage workers to run for office be in jurisdictions with selection processes that are less open than US primaries? Training programs might have potential in single member district systems where local party selection processes are relatively open, even if member-based, but less likely to succeed where nominations are more subject to backroom negotiations or driven by the central party. They may also be more likely to succeed at the municipal level, where political parties play much less of a role in certain jurisdictions like

Canada. Where party insiders play a large role in determining candidacies such as closed list proportional systems, a different approach may be necessary.

Another consideration is whether parties could be induced to systematize efforts to recruit working-class candidates in recruitment and nomination efforts. For example, the New

Democratic Party in British Columbia has set a type of gender quota in its candidate nomination practices when incumbent legislators step aside (Schreck, 2011), and a similar class-based screen could be considered. While quotas for the recruitment of working-class candidates may not be a realistic goal at a governmental level, it may be within reach as a matter of political party policy.

Although, if class-based recruitment was made a prominent enough issue to win an explicit policy within a party, this might be enough to significantly improve recruitment practices even in the absence of a strict quota. This would depend in part on whether party elites are ideologically at odds with workers, judge their electoral prospects as poor, or simply fail to recruit them because of lack of contact with workers in their personal networks.

134

Quotas and other efforts to recruit workers may have different effects across electoral systems, such as in closed versus open list proportional representation systems where voters get their say more directly. This may in turn depend on whether voters (or party elites) are biased in favour or against (or are indifferent to) the class backgrounds of candidates, which remains an open question in the literature. The interdependencies and moderating effects relating to these solutions are potentially complex.

Relatedly, Carnes (2018) identifies increasing unionization rates as another “long shot” solution to facilitate recruitment and increase worker representation in government. As discussed in previous sections, unions may play a role in offsetting the resource disadvantages of workers pursuing office by supporting their candidacies, as well as by recruiting and training them to run

(not to mention also by helping to raise wages and promote redistribution). Like significant economic redistribution, revitalizing the labour movement would be difficult and represent a substantial level of social reorganization in most societies. Notably, there could another be a self- reinforcing dynamic at play, wherein if unionization helps facilitate stronger working-class representation among legislators, this might then help create a political environment conducive to further unionization (see Bartolini, 2000).

In the US case, Carnes (2018) argues that unionization is too sweeping a goal to focus on particularly if the intention is to increase worker representation in the short-term. Yet increasing unionization seems no less plausible than broader efforts at redistribution to reduce the resource disadvantages of workers. Like the case of redistribution, the potential efficacy of trying to increase unionization to enhance worker representation would also depend on current levels of union strength in a given country. That is, an increase in unionization of a given magnitude may

135

have a larger impact when starting from a lower base than in places where unions already play a strong role. The role of unions in effectively bringing workers into candidacy would also likely depend on the level of institutional and personal linkages between unions and political parties in each jurisdiction, particularly where nomination contests are relatively closed.

In addition, there are two solutions that Carnes (2018) dismisses nearly without comment in the

US context: explicit quotas for working-class representatives in legislatures and the idea of

Athenian-style lotteries to select legislators. Both indeed seem far-fetched and unlikely to be a realistic solution in almost any country anytime soon. Having said that, the use of deliberative mini-publics employing random recruitment, such as citizens’ assemblies, might be considered as more tractable means to encourage a different but substantive form of political participation by workers (Goodin & Dryzek, 2006). Citizens’ assemblies have been used to address issues as diverse as electoral reform, reproductive rights, and contentious local planning debates. If the more regular use of deliberative mini-publics allowed for ongoing, deeper political participation by some workers, this could conceivably help recruit more to run for legislative office, as well as normalizing worker participation in difficult questions of governance. Of course, since participation is voluntary and potentially time-consuming, mini-publics may suffer from their own difficulties in recruiting workers, even when the initial invitations are distributed randomly.

This tendency could be offset with measures like participation stipends or using sortition to explicitly ensure proportionate representation by economic class in the assemblies.

Finally, through not specific to recruitment, one other large, society-level measure to increase worker representation in legislatures would be to increase the typically lower voter turnout levels of low-income people. We have seen some evidence that low-income voters in particular are

136

biased against affluent candidates (Wüest & Pontusson, 2018), so increasing their turnout may be a relative advantage to working class candidates. This solution might be most immediately relevant in contexts in which workers are running for office but not winning (as appears to be the case in Switzerland), but such a change could also encourage more workers to run in the first place and increase the likelihood of party elites to recruit and support them. Of course, as with all of the potential structural solutions identified in this section, increasing voter turnout among any demographic would be no small feat.

6 Conclusion

From the limited existing evidence, we have attempted to glean some insights and observe patterns regarding barriers to working class representation at different stages of selection and election. This has also included an exploration of potential interventions to increase the representation of workers, as well as how barriers and effective interventions may vary across countries. This paper reviewed and analyzed the limited literature to date, while bringing some new data to bear in the descriptive statistics presented from the Comparative Candidate Survey and PARTIREP. Disadvantages in terms of material resources would appear to be among the key factors discouraging them from running for office, though much more evidence is needed particularly outside of the US. Another important factor may be a disconnect between workers and party elites, who often do not have many workers in their typical recruitment networks and appear to view workers as less viable candidates, at least based on evidence in the US context.

137

Efforts at redistribution, increased unionization, financial supports for workers, and targeted recruitment and training efforts emerge as among the most promising interventions to increase worker representation, though this is based on limited evidence. Comparative research on these issues remains particularly sparse, but we have reviewed some of the possible ways the factors at play may vary across jurisdictions (which is to say, likely with considerable complexity). Levels of inequality and the cost of elections, as well as party and electoral institutions, emerge as key moderators to explore in future research on barriers to working-class representation.

Understanding and addressing the barriers to working-class representation could be of real consequence. As the first two papers in this dissertation show, the class backgrounds of legislators help shape their attitudes and behaviour in office regarding economic and redistributive issues. Moreover, just as there is a normative imperative to equalize the descriptive representation of gender and race, there is arguably such an imperative with respect to the representation of the working class, which has long been severely underrepresented in numerical terms in legislatures around the world. Matters of class identity and inequality appear to be a significant point of contention the volatile contemporary politics of advanced democracies, which should add urgency to pursuing this line of inquiry.

138

Conclusion

In political science, it has long been recognized that the class composition of legislatures does not reflect that of the broader society, and few working-class people tend to hold public office.

However, until a recent resurgence of research, the literature had largely settled on the idea that this unequal descriptive representation was of limited importance to the substance of representation that legislators provided. Building on recent work, this dissertation has examined the consequences of unequal class representation across numerous advanced democracies and using multiple measures at different levels of analysis, while also taking stock of barriers to working-class representation and ways to address them.

Analyzing a range of national and subnational legislatures in 15 advanced democracies, Paper 1 found that legislators from business backgrounds were more favourable to inequality and small government than those from working-class and other backgrounds. They were also less likely than working-class legislators to report meeting with labour groups in their role as MPs. An exploratory analysis showed that these class-based differences between legislators appear to be moderated by electoral institutions, with larger differences seen where institutions provided incentives for more individual rather than party-based competition, such as open list proportional representation and division of powers systems.

One limitation to the analyses in Paper 1 is that they do not strictly allow for making causal inferences, being based on observational data and cross-sectional regression models. However, the findings were robust to a wide range of empirical specifications and key findings were replicated in a second data set. Another limitation is that the outcome measures used in this paper were confined to the individual level, using the handful of survey questions on legislators’ 139

attitudes and self-reported behaviour available in the data sets. Despite these limitations, the paper reinforces and extends (across new jurisdictions) an emerging finding in the literature that the class backgrounds of legislators do affect the attitudes and orientations they bring to their roles as representatives, in contrast with an earlier wave of literature and assumptions on this question. The paper’s institutional analyses also suggest avenues to be explored in future research on the unequal representation of class, particularly regarding the observed moderating effects of electoral systems and potentially other institutional factors.

Paper 2 complements the first paper with an analysis that extends findings to the realm of aggregate policy outcomes and overcomes some of the prior paper’s limitations in terms of causal inference. This paper uses an instrumental variables approach to exploit close election results in examining the relationship between the class makeup of Finnish municipal councils and the level of social spending undertaken by these municipalities. The findings show a positive relationship between the share of workers on the municipal council and the level of social spending. Along with Carnes (2013), this appears to be one of the only studies that examines the effect of the descriptive representation of class on policy outcomes. In terms of limitations, the results hold up under a range of robustness tests, though one such test raises some concerns, as discussed in the paper. In addition, the data on the occupational class backgrounds of councillors is based on relatively imprecise open-ended Finnish language descriptions. However, if anything, this makes for a hard test of the hypotheses, since a “noisy” independent variable would tend to mask the effect that is nonetheless observed in the results. The use of more precise administrative data in future research on this question may help bring the effect into sharper focus. While the

140

findings are limited to municipal governments in one specific country and institutional context, they dovetail with those in other jurisdictions from Paper 1 and the emerging literature.

Given growing evidence that unequal class representation is both widespread and has substantive consequences, Paper 3 steps back further in the causal chain and seeks to explore key barriers to working-class people reaching office in the first place. The paper reviews the sparse and predominantly US-focused literature on this question, considers ways in which we might expect barriers to vary across contexts, probes existing data sets for clues, and examines solutions and interventions that have the potential to increase working-class descriptive representation. Any inferences drawn from the literature reviewed and the exploratory analysis are necessarily provisional, based on a limited body of evidence.

Nevertheless, consistent with the US-focused research of Carnes (2018), resource disadvantages faced by prospective working-class candidates emerge as an important set of barriers, as do class-biased recruitment practices of political party elites. Such barriers could potentially be addressed by large-scale, systemic solutions like reducing economic inequality and increasing the strength of unions, which are fertile ground for cultivating working-class candidates.

Targeted interventions like public funding for candidates and training programs that aim to bring workers into the political process may hold more immediate promise. The applicability of these

(again, tentative) findings is likely to vary across contexts in potentially complex ways, with inequality, the cost of elections, and electoral institutions among the societal-level moderating variables that may shape barriers to working-class representation.

Indeed, a wide range of more granular research is needed on various aspects of these initial findings. There are many ready opportunities for research in this area, including more fine-tuned 141

voter survey experiments, as well as analysis of open party lists of candidates to help distinguish between the preferences of voters and party elites with respect to class. The available data on the class background of candidates and legislators is limited and the few data sets that exist tend to use measures that are not readily comparable with each other, so additional data collection will be crucial. We also remain largely in the dark about events at the earliest stages of the political process, in which individuals may self-select out of pursuing office before they ever engage with a party or electoral institution.

In a time of simmering political unrest and significant collective challenges globally, there is a growing urgency to study the issues addressed in this dissertation. The politics of inequality, class and identity appear to be salient elements of the present situation. Inequality is itself a source of economic and social ill, as well as perhaps being among the drivers of political instability. Therefore, to the extent that the unequal descriptive representation of class plays a role in producing economic inequality, better understanding it is important. It is also worth considering whether the more equal descriptive representation of working-class people in legislatures may be an important (and perhaps even necessary) condition to overcome genuine democratic deficits and reconstitute the legitimacy of and popular confidence in democratic institutions. While these are speculations, better understanding these issues is imperative given the existential crises that happen to coincide with this moment in human history. Foremost among them is the looming crisis of climate change, which requires that we learn with great speed to more effectively marshal our capacities for democratic collective action.

142

References

Ashe, J., & Stewart, K. (2011). Legislative recruitment: Using diagnostic testing to explain

underrepresentation. Party Politics, 18(5), 687–707.

Andersen, R., & Curtis, J. (2012). The polarizing effect of economic inequality on class

identification: Evidence from 44 countries. Research in Social Stratification and Mobility,

30(1), 129–141. https://doi.org/10.1016/j.rssm.2012.01.002

Ansolabehere, S., de Figueiredo, J. M., & Snyder, J. M. (2003). Why is There so Little Money in

U.S. Politics? Journal of Economic Perspectives, 17(1), 105–130.

https://doi.org/10.1257/089533003321164976

Bartels, L. (2008). Unequal Democracy. Princeton: Princeton University Press.

Bartels, L. M. (2017). Political Inequality in Affluent Democracies: The Social Welfare Deficit.

(Center for the Study of Democratic Institutions #5-2017). Nashville: Vanderbilt

University.

Bartolini, S. (2000). The political mobilization of the European left, 1960-1980: The class

cleavage. Cambridge: Cambridge University Press.

Bawn, K., Cohen, M., Karol, D., Masket, S., Noel, H., & Zaller, J. (2012). A theory of political

parties: Groups, policy demands and nominations in American politics. Perspectives on

Politics, 10(3), 571-597. https://doi.org/10.1017/S1537592712001624

Bernauer, J., Giger, N., & Rosset, J. (2015). Mind the gap: Do proportional electoral systems

foster a more equal representation of women and men, poor and rich? International 143

Political Science Review, 36(1), 78–98. https://doi.org/10.1177/0192512113498830

Best, H. (2007). New challenges, new elites? Changes in the recruitment and career patterns of

European representative elites. Comparative Sociology, 6(1), 85–113.

https://doi.org/10.1163/156913307X187414

Best, H., & Cotta, M. (Eds.). (2000). Parliamentary Representatives in Europe 1848–2000:

Legislative Recruitment and Careers in Eleven European Countries. Oxford: Oxford

University Press.

Brady, H. E., Verba, S., & Schlozman, K. L. (1995). Beyond SES: A resource model of political

participation. American Political Science Review, 89(2), 271-294.

https://doi.org/10.2307/2082425

Broockman, D. E. (2014). Mobilizing Candidates: Political Actors Strategically Shape the

Candidate Pool with Personal Appeals. Journal of Experimental Political Science, 1(2),

104–119. https://doi.org/10.1017/xps.2014.3

Burden, B. C. (2007). Personal roots of representation. Princeton: Princeton University Press.

Campbell, A., Converse, P., Miller, W., & Stokes, D. (1960). The American Voter. New York:

Wiley.

Campbell, R., & Cowley, P. (2013). Rich Man, Poor Man, Politician Man: Wealth Effects in a

Candidate Biography Survey Experiment. The British Journal of Politics & International

Relations, 16(1), 56–74.

144

Carey, J. M., & Shugart, M. S. (1995). Incentives to cultivate a personal vote: a rank ordering of

electoral formulas. Electoral Studies, 14(4), 417-439. https://doi.org/10.1016/0261-

3794(94)00035-2

Carnes, N. (2013). White-Collar Government: The Hidden Role of Class in Economic Policy

Making. Chicago: University of Chicago Press.

Carnes, N. (2018). The Cash Ceiling: Why Only the Rich Run for Office--and What We Can Do

about It. Princeton: Princeton University Press.

Carnes, N., & Lupu, N. (2014). Rethinking the Comparative Perspective on Class and

Representation: Evidence from Latin America. American Journal of Political Science,

59(1), 1-18. https://doi.org/10.1111/ajps.12112

Carnes, N., & Lupu, N. (2016). Do voters dislike working-class candidates? Voter biases in the

descriptive underrepresentation of the working class. American Political Science Review,

110(4), 832-844

Casselman, B., & Tankersley, J. (2019, November 29). Warren Wealth Tax Has Wide Support,

Except Among One Group. New York Times.

CCS 2016. Comparative Candidates Survey Module I - 2005-2013 [Dataset - cumulative file].

Distributed by FORS, Lausanne, 2016.

Crowder-Meyer, M. (2013). Gendered Recruitment without Trying: How Local Party Recruiters

Affect Women’s Representation. Politics & Gender, 9(04), 390–413.

145

Curtis, J. (2016). Social mobility and class identity: The role of economic conditions in 33

societies, 1999-2009. European Sociological Review, 32(1), 108–121.

https://doi.org/10.1093/esr/jcv077

Dahl, R. (1961). Who Governs? New Haven: Yale University Press.

Deschouwer, K., Depauw, S., & André, A. (2014). Representing the People in Parliaments. In:

K. Deschouwer & S. Depauw. (Eds.) Representing the People: A Survey Among Members

of Statewide and Substate Parliaments. Oxford: Oxford University Press.

Donnelly, M., & Lefkofridi, Z. (2014). Economic and political inequality in modern

democracies: Differential responsiveness to the policy preferences of economic classes

(MWP Red Number Series 2014/13). Fiesole: European University Institute.

Eidlin, B. (2014). Class formation and class identity: Birth, death, and possibilities for renewal.

Sociology Compass, 8(8), 1045–1062. https://doi.org/10.1111/soc4.12197

Essaiasson, P. & Holmberg, S. (1996). Representation from above: Members of Parliament and

representative democracy in Sweden. Aldershot: Dartmouth.

Evans, G., & Tilley, J. (2017). The new politics of class: The political exclusion of the British

working class. Oxford: Oxford University Press.

Gallego, A. (2007). Unequal Political Participation in Europe. International Journal of

Sociology, 37(4), 10–25.

Gaxie, D. (2017). Political and social backgrounds of political elites. In: H. Best & J. Higley.

(Eds.), The Palgrave handbook of political elites. London: Palgrave. 146

Giger, N., Rosset, J., & Bernauer, J. (2012). The Poor Political Representation of the Poor in a

Comparative Perspective. Representation, 48(1), 47-61.

https://doi.org/10.1080/00344893.2012.653238

Gilens, M. (2012). Affluence and Influence: Economic Inequality and Political Power in

America. Princeton: Princeton University Press.

Gilens, M., & Page, B. I. (2014). Testing theories of American politics: Elites, interest groups,

and average citizens. Perspectives on politics, 12(3), 564-581.

https://doi.org/10.1017/S1537592714001595

Goodin, R. E., & Dryzek, J. S. (2006). Deliberative impacts: the macro-political uptake of mini-

publics. Politics & society, 34(2), 219-244.

Grumbach, J. M. (2015). Does the American Dream Matter for Members of Congress? Social-

Class Backgrounds and Roll-Call Votes. Political Research Quarterly, 68(2), 306–323.

https://doi.org/10.1177/1065912915572902

Hacker, J. S., & Pierson, P. (2010). Winner-Take-All Politics: Public Policy, Political

Organization, and the Precipitous Rise of Top Incomes in the United States. Politics &

Society, 38(2), 152–204. https://doi.org/10.1177/0032329210365042

Hemingway, A. (2020). Does Class Shape Legislators’ Approach to Inequality and Economic

Policy? A Comparative View. Government and Opposition, 1-24.

147

Hicks, T., Jacobs, A. M., & Matthews, J. S. (2016). Inequality and Electoral Accountability:

Class-Biased Economic Voting in Comparative Perspective. The Journal of Politics,

78(4), 1076–1093. https://doi.org/10.1086/686157

Hooghe, M., & Oser, J. (2016). Trade union density and social expenditure: a longitudinal

analysis of policy feedback effects in OECD countries, 1980–2010. Journal of European

Public Policy, 23(10), 1520-1542.

Hout, M. (2008). How class works: Objective and subjective aspects of class since the 1970s. In

A. Lareau & D. Conley (Eds.), Social class: How does it work? (pp. 25). New York:

Russell Sage Foundation.

Hyytinen, A., Meriläinen, J., Saarimaa, T., Toivanen, O., & Tukiainen, J. (2018). Public

employees as politicians: evidence from close elections. American Political Science

Review, 112(1), 68-81. https://doi.org/10.1017/S0003055417000284

Inglehart, R. (1997). Modernization and postmodernization: Cultural, economic, and political

change in 43 societies. Princeton: Princeton University Press.

Iversen, T., & Soskice, D. (2001). An asset theory of social policy preferences. American

Political Science Review, 95(4), 875-893. https://doi.org/10.1017/S0003055400400079

Jacobs, L. R., & Shapiro, R. Y. (2000). Politicians don't pander: Political manipulation and the

loss of democratic responsiveness. Chicago: University of Chicago Press.

Jusko, K. L. (2017). Who speaks for the poor? Electoral geography, party entry, and

representation. Cambridge: Cambridge University Press.

148

Kalmijn, M., & Kraaykamp, G. (2007). Social stratification and attitudes: a comparative analysis

of the effects of class and education in Europe. The British Journal of Sociology, 58(4),

547. https://doi.org/10.1111/j.1468-4446.2007.00166.x

Kevins, A. (2019). Race, class, or both? Responses to candidate characteristics in Canada, the

UK, and the US. Politics, Groups, and Identities, 1-22.

Korpi, W. (2006). Power Resources and Employer-Centered Approaches in Explanations of

Welfare States and Varieties of Capitalism: Protagonists, Consenters, and Antagonists.

World Politics, 58(02), 167–206. https://doi.org/10.1353/wp.2006.0026

Lawless, J. L., & Fox, R. L. (2013). Girls Just Wanna Not Run: The Gender Gap in Young

Americans’ Political Ambition. Women & Politics Institute.

Lawless, J. L., & Pearson, K. (2008). The primary reason for women’s underrepresentation?

Reevaluating the conventional wisdom. Journal of Politics, 70(1), 67–82.

Lindblom, C. (1977). Politics and Markets: The World's Economic Systems. New York: Basic

Books.

Lore, G. (2016). Women’s Descriptive and Substantive Political Representation: The Role of

Political Institutions. PhD thesis. University of British Columbia.

Levitt, S. D. (1996). How do senators vote? Disentangling the role of voter preferences, party

affiliation, and senator ideology. The American Economic Review, 86(3), 425-441.

Mansbridge, J. (1999). Should Blacks Represent Blacks and Women Represent Women? A

Contingent “Yes.” Journal of Politics, 61(3), 628–657. https://doi.org/10.2307/2647821

149

Manza, J., & Brooks, C. (2008). Class and politics. In A. Lareau & D. Conley (Eds.), Social

class: How does it work? (pp. 201). New York: Russell Sage Foundation.

Manza, J., Hout, M., & Brooks, C. (1995). Class Voting in Capitalist Democracies Since World

War II: Dealignment, Realignment, or Trendless Fluctuation? Annual Review of

Sociology, 21, 137–162. https://doi.org/10.1146/annurev.so.21.080195.001033

Marshall, J. (2019). When can close election RDDs identify the effects of winning politician

characteristics? (Unpublished paper).

Matthews, D. R. (1984). Legislative Recruitment and Legislative Careers. Legislative Studies

Quarterly, 9(4), 547.

McCarty, N., Poole, K. T., & Rosenthal, H. (2016). Polarized America: The dance of ideology

and unequal riches. Cambridge: MIT Press.

Meltzer, A. H., & Richard, S. F. (1981). A rational theory of the size of government. Journal of

Political Economy, 89(5), 914-927. https://doi.org/10.1086/261013

Mills, C. W. (1956). The Power Elite. New York: Oxford.

Nayler, T. (2019, July 16). How we can help women in politics overcome the ‘cash ceiling’.

Ottawa Citizen. https://ottawacitizen.com/opinion/columnists/nayler-how-we-can-help-

women-in-politics-overcome-the-cash-ceiling

Norris, P. (Ed.). (1997). Passages to power: Legislative recruitment in advanced democracies.

Cambridge: Cambridge University Press.

150

Norris, P., & Lovenduski, J. (1995). Political recruitment: Gender, race and class in the British

Parliament. Cambridge: Cambridge University Press.

O’Grady, T. (2018). Careerists Versus Coal-Miners: Welfare Reforms and the Substantive

Representation of Social Groups in the British Labour Party. Comparative Political

Studies, 52(4), 544-578. https://doi.org/10.1177/0010414018784065

Olson, J. M., & Stone, J. (2005). The Influence of Behavior on Attitudes. In D. Albarricin, B. T.

Johnson, & M. P. Zann (Eds.) The Handbook of Attitudes, (pp. 223–271). Mahwah, NJ:

Erlbaum.

Page, B. I., Bartels, L. M., & Seawright, J. (2013). Democracy and the policy preferences of

wealthy Americans. Perspectives on Politics, 11(1), 51-73.

https://doi.org/10.1017/S153759271200360X

Pedersen, R. T., Dahlgaard, J. O., & Citi, M. (2019). Voter reactions to candidate background

characteristics depend on candidate policy positions. Electoral Studies, 61.

https://doi.org/10.1016/j.electstud.2019.102066

Piketty, T. (2014). Capital in the twenty-first century. Cambridge, MA: Harvard University

Press.

Pitkin, H. F. (1967). The concept of representation. Oakland: University of California Press.

Rehm, P. (2011). Risk inequality and the Polarized American electorate. British Journal of

Political Science, 41(2), 363–387. https://doi.org/10.1017/S0007123410000384

151

Rodden, J. (2018). Why Cities Lose: Political Geography and the Representation of the Left.

New York: Basic Books.

Rosset, J. (2013). Are the Policy Preferences of Relatively Poor Citizens Under-represented in

the Swiss Parliament? The Journal of Legislative Studies, 19(4), 490–504.

https://doi.org/10.1080/13572334.2013.812363

Rosset, J. (2016). Economic inequality and political representation in Switzerland. New York:

Springer.

Sadin, M. L. (2016). Campaigning With Class: The Effect of Candidate Wealth on Voters’

Evaluations. (Unpublished paper).

Sanbonmatsu, K. (2006). Where women run: Gender and party in the American states. Ann

Arbor: University of Michigan Press.

Savage, M., Bagnall, G., & Longhurst, B. (2001). Ordinary, ambivalent and defensive: Class

identities in the Northwest of England. Sociology, 35(4), 875-892.

https://doi.org/10.1177/0038038501035004005

Schreck, D. (2011, January 5). What’s ‘ridiculous’ about NDP’s gender balance policy? The

Tyee. https://thetyee.ca/Opinion/2011/01/05/NDPGenderBalance/

Sojourner, A. (2013). Do Unions Promote Members’ Electoral Office Holding-Evidence from

Correlates of State Legislatures’ Occupational Shares. ILRReview, 66(2), 467-486.

Volkens, A., Krause, W., Lehmann, P., Matthieß, T., Merz, N., Regel, S., & Weßels, B. (2019).

The Manifesto Data Collection. Manifesto Project (MRG/CMP/MARPOR). Version

152

2019b. Berlin: Wissenschaftszentrum Berlin für Sozialforschung (WZB).

https://doi.org/10.25522/manifesto.mpds.2019b

Walsh, K. C., Jennings, M. K., & Stoker, L. (2004). The effects of social class identification on

participatory orientations towards government. British Journal of Political Science,

34(3), 469–495.

Weeden, K., & Grusky, D. (2005). The Case for a New Class Map. American Journal of

Sociology, 111(1), 141–212. https://doi.org/10.1086/428815

Wessels, B. (1997). Germany. In: P. Norris (Ed.), Passages to power: Legislative recruitment in

advanced democracies. Cambridge: Cambridge University Press.

Wüest, R., & Pontusson, J. (2018). Descriptive representation by social class: do voter

preferences matter? (Unequal Democracies Working Paper #1). Geneva: University of

Geneva.

Wüest, R., & Pontusson, J. (2019, August 23). Voter Preferences for Candidates with Different

Social Backgrounds and Programmatic Commitments [Paper presentation]. Annual

Meeting of the American Political Science Association, Washington, DC.

153

Appendix A: Supporting materials to Paper 1

Figure A.1. Distribution of outcome variable responses

Note: Each outcome variable in the figure is on a scale from 1-5 as described in the main text.

154

Table A.1. Occupation categories by country

AUT BEL FRA GER HUN IRE ISR ITA NET NOR POL POR SPA SWI UNK

Business 40 19 12 15 23 9 4 19 14 9 2 9 16 141 18 Technical professional 15 16 14 39 10 9 5 26 2 2 26 20 24 76 11

Farmer 22 1 4 0 6 4 0 0 1 0 0 0 2 31 2

Lawyer 10 19 7 18 2 2 7 11 2 1 9 21 33 26 10 Other white collar 3 5 2 60 0 1 0 3 0 11 0 5 4 46 9

Politics 31 38 3 21 24 1 1 10 28 1 4 10 2 57 5

Civil service 11 11 13 4 1 0 0 11 11 0 3 6 32 2 2 Service- based professional 37 40 16 33 25 7 7 24 6 11 10 34 50 121 36

Worker 40 13 1 32 5 1 1 15 1 8 1 7 7 49 10

No info 2 0 2 27 2 0 0 2 0 2 0 3 2 35 0

Total 211 162 74 249 98 34 25 121 65 45 55 115 172 584 103

155

Table A.2. Occupation categories by party type

Final Occupation Initial Coding Right / Left Party Total Category Categories Other Party 1. Business Businessperson - top 257 89 346 management

Businessperson - middle management

Other business / entrepreneur 2. Technical professional Technical 177 111 288 professional 3. Farmer Farmer 62 10 72 4. Lawyer Lawyer 120 57 177 5. Other white collar Other white collar 74 74 148 6. Politics / law Military / law 135 97 232 enforcement enforcement

Politician 7. Civil service Civil servant (no 40 66 106 other identification) 8. Service-based University-professor 174 275 449 professional School teacher Other service-based professional 9. Worker Worker 74 114 188

Employee (no other identification) 10. No occupation info Unemployed 38 38 76

Student

Retired (and no previous occupation given)

No previous occupation

Other Unknown/blank Note: I initially coded the open-end, raw text occupation field in the PARTIREP data set into 21 categories shown in the second column of this table, and then combined them into the 10 categories shown in the first column. These 10 categories, as well as how example occupations map onto them, are adapted from the coding scheme outline by Carnes (2013) and Carnes and Lupu (2014).

156

Table A.3. Occupational categories (reproduced and adapted from Carnes and Lupu 2014)

Broad occupational category Narrow occupational category

Businessperson Associate Director / CEO Business owner / manager Farmer, Farm owner / manager Banker Contractor Salesman Business representative Technical professional [labelled Accountant / Economist “private-sector professional in Actor Carnes and Lupu (2014)] Advertising Architect / Urban Planner Author Consultant Doctor / Dentist / Vet Engineer Hospital Administrator Journalist / Publisher Medical Office Manager Mortician Pharmacist Professional Athlete Radio and Television Notary Public Military / law enforcement Military Law Enforcement Lawyer Lawyer Politician Political Consultant Political Party Officer Pub Policy Analyst Public Relations / Lobbyist Judge Mayor Government Attorney

157

Service-based professional NGO / Charity Organizer College Administrator College Professor Education Admin. Guidance Councilor High School Admin. Librarian Minister / Priest Sec. School Teacher Social Worker Other educator Nurse Community organizer Worker Laborer Service industry worker Union officer, staff member Other Note: This table is adapted from Carnes and Lupu (2014) and provided some guidance in the coding process. For observations where the open-ended PARTIREP occupation text fields were ambiguous but suggested a certain category, I flagged the observation as “uncertain.” As a robustness check, I ran versions of the main models with the uncertain cases dropped, as discussed in the robustness section.

158

Table A.4. Regression models relating legislators’ attitudes (full results including controls)

(1) (2) (3) (4) (5) (6) Inequality Inequality Inequality Govt Govt Govt (no controls) (controls) (weighted) economic economic economic intervention intervention intervention (no controls) (controls) (weighted) Business 0 0 0 0 0 0 (.) (.) (.) (.) (.) (.)

Technical -0.282* -0.132 -0.0677 -0.321** -0.168 -0.104 professional (0.108) (0.0900) (0.0807) (0.112) (0.105) (0.124)

Farmer -0.0860 -0.263 -0.316 -0.191 -0.408** -0.456* (0.178) (0.179) (0.203) (0.148) (0.149) (0.197)

Lawyer -0.308* -0.192+ -0.191 -0.219+ -0.0966 -0.137 (0.121) (0.107) (0.115) (0.123) (0.102) (0.122)

Other white -0.617** -0.384** -0.303** -0.275+ -0.0550 -0.0303 collar (0.125) (0.103) (0.102) (0.161) (0.132) (0.137)

Politics / law -0.446** -0.238* -0.164 -0.419** -0.198* -0.247* enforcement (0.108) (0.0979) (0.106) (0.102) (0.0844) (0.108)

Civil service -0.607** -0.217+ -0.221+ -0.628** -0.197+ -0.170 (0.132) (0.121) (0.113) (0.0977) (0.106) (0.117)

Service-based -0.858** -0.449** -0.337** -0.731** -0.271** -0.237* professional (0.102) (0.0812) (0.0776) (0.113) (0.0835) (0.101)

Worker -0.695** -0.305** -0.241** -0.675** -0.251** -0.127 (0.110) (0.0886) (0.0876) (0.116) (0.0787) (0.0808)

No occupation -0.303+ -0.0822 0.0739 -0.149 0.127 0.331 info (0.172) (0.164) (0.319) (0.183) (0.174) (0.212)

AUT 0 0 0 0 0 0 (.) (.) (.) (.) (.) (.)

BEL -0.206+ -0.322** -0.417** -0.108 -0.242+ -0.358* (0.114) (0.111) (0.120) (0.127) (0.130) (0.152)

FRA -0.403* -0.378* -0.492** -0.408** -0.363+ -0.471+ (0.186) (0.155) (0.179) (0.151) (0.206) (0.242)

GER -0.111 0.0178 -0.129 0.0970 0.248+ 0.0783 (0.159) (0.139) (0.147) (0.116) (0.126) (0.156)

HUN 0.215* 0.230* 0.116 0.0949 0.0899 -0.0308 (0.0980) (0.0983) (0.104) (0.0989) (0.115) (0.140)

IRE -0.0802 -0.268** -0.349** -0.362** -0.652** -0.758** (0.0982) (0.0964) (0.101) (0.1000) (0.117) (0.141) 159

ISR 0.540** 0.330** 0.225+ -0.290* -0.563** -0.678** (0.105) (0.107) (0.120) (0.111) (0.130) (0.164)

ITA -0.452** -0.488** -0.586** 0.263* 0.209 0.0645 (0.122) (0.126) (0.130) (0.129) (0.133) (0.160)

NET -0.325** -0.328** -0.422** 0.132 0.151 0.0607 (0.103) (0.0996) (0.105) (0.103) (0.109) (0.131)

NOR -0.161 -0.149 -0.279* -0.291** -0.273* -0.430** (0.106) (0.107) (0.123) (0.109) (0.125) (0.157)

POL 0.676** 0.344** 0.257* 0.639** 0.135 0.0358 (0.109) (0.115) (0.123) (0.118) (0.138) (0.165)

POR -0.385* -0.364+ -0.400+ 0.329* 0.350* 0.250 (0.191) (0.195) (0.213) (0.138) (0.164) (0.193)

SPA 0.123 0.319 0.162 -0.480** -0.234 -0.407* (0.186) (0.230) (0.211) (0.126) (0.166) (0.188)

SWI -0.196+ -0.250* -0.339** 0.151 0.103 0.0307 (0.117) (0.117) (0.102) (0.119) (0.126) (0.135)

UNK -0.491** -0.569** -0.704** -0.120 -0.212 -0.357* (0.106) (0.133) (0.146) (0.109) (0.144) (0.174)

Other Party 0 0 0 0 (.) (.) (.) (.)

Left Party -1.158** -1.068** -1.284** -1.169** (0.127) (0.144) (0.136) (0.145)

male 0 0 0 0 (.) (.) (.) (.)

female -0.0973 -0.0583 -0.171** -0.151** (0.0614) (0.0671) (0.0598) (0.0522)

age 0.00663** 0.00902** 0.00874** 0.0103** (0.00229) (0.00272) (0.00240) (0.00302)

Constant 3.088** 3.055** 2.921** 3.026** 2.940** 2.900** (0.118) (0.164) (0.173) (0.114) (0.163) (0.215) Observations 1891 1817 1817 1894 1820 1820 Standard errors in parentheses + p < .10, * p < .05, ** p < .01

160

Table A.5. Regression models on legislators' contact with workers' organizations (full results) (1) (2) (3) No controls Controls Controls, weighted Business 0 0 0 (.) (.) (.)

Technical professional 0.127 0.0357 0.0653 (0.0827) (0.0827) (0.109)

Farmer -0.0588 0.0151 -0.0272 (0.137) (0.136) (0.188)

Lawyer 0.140 0.101 0.133 (0.0956) (0.0907) (0.112)

Other white collar 0.267* 0.156 0.0745 (0.130) (0.128) (0.155)

Politics / law 0.267* 0.159 0.222+ enforcement (0.102) (0.104) (0.112)

Civil service 0.216+ 0.0461 0.129 (0.110) (0.104) (0.0945)

Service-based 0.310** 0.0987 0.102 professional (0.0956) (0.0832) (0.0933)

Worker 0.579** 0.383** 0.245* (0.121) (0.111) (0.121)

No occupation info 0.224 0.0873 0.103 (0.153) (0.140) (0.151)

AUT 0 0 0 (.) (.) (.)

BEL -0.215 -0.139 -0.201+ (0.159) (0.137) (0.118)

FRA 0.0452 0.0226 -0.0559 (0.121) (0.110) (0.0835)

GER 0.189 0.127 0.104 (0.139) (0.127) (0.107)

HUN -0.161 -0.154 -0.224** (0.118) (0.107) (0.0841)

IRE 0.0563 0.204+ 0.115 (0.114) (0.109) (0.0882) 161

ISR -0.222+ -0.0790 -0.176+ (0.123) (0.114) (0.0940)

ITA 0.612** 0.638** 0.555** (0.119) (0.115) (0.0946)

NET 0.0153 0.00522 -0.0951 (0.126) (0.117) (0.0946)

NOR 0.393** 0.384** 0.371** (0.111) (0.100) (0.0817)

POL 0.0869 0.258* 0.133 (0.122) (0.119) (0.106)

POR 0.319 0.285 0.133 (0.224) (0.237) (0.243)

SPA 0.210 0.190 0.0789 (0.187) (0.181) (0.163)

SWI -0.864** -0.832** -0.730** (0.125) (0.120) (0.124)

UNK 0.135 0.175 0.132 (0.117) (0.118) (0.0952)

Other Party 0 0 (.) (.)

Left Party 0.610** 0.502** (0.0725) (0.0829)

age -0.00979** -0.00886** (0.00238) (0.00263)

male 0 0 (.) (.)

female -0.0410 -0.0330 (0.0577) (0.0628)

Constant 2.959** 3.355** 3.408** (0.117) (0.186) (0.193) Observations 1946 1874 1874 Standard errors in parentheses + p < .10, * p < .05, ** p < .01

162

Table A.6. Regression models on legislators' contact with workers’ organizations (workers as omitted reference category). (1) Controls, unweighted Business -0.383** (0.111)

Technical professional -0.347** (0.105)

Farmer -0.368* (0.176)

Lawyer -0.282+ (0.155)

Other white collar -0.227 (0.155)

Politics / law -0.224 enforcement (0.140)

Civil service -0.337* (0.137)

Service-based -0.284* professional (0.110)

Worker 0 (.)

No info -0.296+ (0.169)

Other Party 0 (.)

Left Party 0.610** (0.0725)

age -0.00979** (0.00238)

male 0 (.)

female -0.0410 (0.0577)

AUT 0 (.) 163

BEL -0.139 (0.137)

FRA 0.0226 (0.110)

GER 0.127 (0.127)

HUN -0.154 (0.107)

IRE 0.204+ (0.109)

ISR -0.0790 (0.114)

ITA 0.638** (0.115)

NET 0.00522 (0.117)

NOR 0.384** (0.100)

POL 0.258* (0.119)

POR 0.285 (0.237)

SPA 0.190 (0.181)

SWI -0.832** (0.120)

UNK 0.175 (0.118)

Constant 3.738** (0.187) Observations 1874 Standard errors in parentheses + p < .10, * p < .05, ** p < .01

164

Table A.7. Regression models on legislators’ attitudes and behaviour, with parliamentary political culture controls (1) (2) (3) (4) (5) (6) Inequality, Govt Contact with Inequality, Govt Contact with avg attitude economic trade unions, avg MP economic trade unions, control intervention, avg attitude ideology intervention, avg MP avg attitude control control avg MP ideology control ideology control control Business ------

Technical -0.119 -0.158 -0.0348 -0.126 -0.166 -0.0398 professional (0.0892) (0.106) (0.0831) (0.0898) (0.105) (0.0830)

Farmer -0.232 -0.384* -0.0120 -0.252 -0.403** -0.0260 (0.177) (0.147) (0.137) (0.178) (0.148) (0.136)

Lawyer -0.170 -0.0794 -0.0999 -0.194+ -0.0973 -0.0995 (0.105) (0.102) (0.0907) (0.108) (0.103) (0.0901)

Other white -0.405** -0.0697 -0.157 -0.387** -0.0558 -0.154 collar (0.0961) (0.133) (0.127) (0.100) (0.132) (0.130)

Politics / law -0.227* -0.189* -0.158 -0.234* -0.196* -0.162 enforcement (0.0953) (0.0846) (0.104) (0.0970) (0.0845) (0.104)

Civil service -0.220+ -0.200+ -0.0459 -0.218+ -0.198+ -0.0441 (0.116) (0.108) (0.104) (0.120) (0.106) (0.104)

Service-based -0.434** -0.260** -0.0978 -0.439** -0.268** -0.104 professional (0.0793) (0.0828) (0.0830) (0.0805) (0.0841) (0.0836)

Worker -0.269** -0.223** -0.381** -0.288** -0.244** -0.395** (0.0890) (0.0775) (0.111) (0.0879) (0.0777) (0.111)

No info -0.0897 0.122 -0.0878 -0.0641 0.134 -0.100 (0.164) (0.168) (0.140) (0.167) (0.174) (0.141)

Other Party ------

Left Party -1.125** -1.260** -0.608** -1.147** -1.280** -0.618** (0.127) (0.136) (0.0725) (0.127) (0.137) (0.0722)

male ------

female -0.0846 -0.161** 0.0417 -0.0954 -0.170** 0.0385 (0.0575) (0.0603) (0.0576) (0.0611) (0.0597) (0.0576)

age 0.00713** 0.00912** 0.00982** 0.00677** 0.00879** 0.00966** (0.00224) (0.00238) (0.00238) (0.00229) (0.00240) (0.00235)

AUT ------

165

BEL -0.165+ -0.126 0.149 -0.293** -0.231+ 0.115 (0.0890) (0.104) (0.134) (0.102) (0.129) (0.127)

FRA -0.0481 -0.119 -0.00139 -0.375** -0.362+ -0.0282 (0.0898) (0.227) (0.113) (0.132) (0.211) (0.104)

GER 0.0249 0.253* -0.126 0.0848 0.272* -0.182 (0.0833) (0.103) (0.126) (0.129) (0.118) (0.135)

HUN 0.0757 -0.0247 0.144 0.195* 0.0772 0.182+ (0.0646) (0.0956) (0.118) (0.0968) (0.118) (0.104)

IRE -0.184** -0.590** -0.199+ -0.238** -0.641** -0.228* (0.0665) (0.0920) (0.106) (0.0898) (0.113) (0.106)

ISR 0.273** -0.606** 0.0754 0.278* -0.582** 0.120 (0.0706) (0.104) (0.117) (0.110) (0.135) (0.111)

ITA -0.392** 0.280* -0.633** -0.461** 0.219+ -0.657** (0.0998) (0.111) (0.112) (0.132) (0.130) (0.105)

NET -0.276** 0.188* -0.00217 -0.329** 0.150 -0.00442 (0.0651) (0.0811) (0.116) (0.0929) (0.107) (0.113)

NOR 0.0380 -0.135 -0.373** -0.244* -0.308* -0.308** (0.0725) (0.0922) (0.0964) (0.121) (0.141) (0.105)

POL -0.175+ -0.250+ -0.291+ 0.178 0.0742 -0.126 (0.0988) (0.135) (0.168) (0.150) (0.167) (0.137)

POR -0.316* 0.385** -0.282 -0.296 0.374* -0.340 (0.126) (0.127) (0.233) (0.200) (0.159) (0.224)

SPA 0.466** -0.128 -0.181 0.433+ -0.193 -0.282 (0.137) (0.109) (0.182) (0.234) (0.161) (0.178)

SWI -0.295** 0.0687 0.829** -0.296* 0.0862 0.869** (0.0784) (0.0929) (0.122) (0.117) (0.130) (0.119)

UNK -0.273** 0.00628 -0.156 -0.495** -0.185 -0.234+ (0.0922) (0.102) (0.117) (0.116) (0.132) (0.132)

classAttitude_ 0.778** 0.576** 0.0485 parliamentAvg (0.108) (0.0978) (0.136)

selfLR_parlia 0.113+ 0.0413 -0.0905+ mentAvg (0.0642) (0.0505) (0.0495)

Constant 0.962** 1.389** 2.514** 2.519** 2.745** 3.074** (0.355) (0.309) (0.378) (0.358) (0.263) (0.289) R-squared 0.299 0.324 0.247 0.282 0.314 0.248 N 1817 1820 1874 1817 1820 1874 Standard errors in parentheses + p < .10, * p < .05, ** p < .01 166

Table A.8. Regression models on legislators’ attitudes, with average electorate attitude controls (1) (2) (3) Inequality, Govt Contact with electorate economic trade unions, control intervention, electorate electorate control control Business ------

Technical -0.137 -0.170 -0.0358 professional (0.0900) (0.105) (0.0830)

Farmer -0.275 -0.414** -0.0154 (0.177) (0.147) (0.135)

Lawyer -0.193+ -0.0970 -0.101 (0.107) (0.102) (0.0907)

Other white -0.389** -0.0576 -0.156 collar (0.103) (0.132) (0.128)

Politics / law -0.241* -0.200* -0.159 enforcement (0.0981) (0.0842) (0.104)

Civil service -0.218+ -0.199+ -0.0461 (0.121) (0.105) (0.104)

Service-based -0.452** -0.273** -0.0988 professional (0.0816) (0.0841) (0.0833)

Worker -0.314** -0.256** -0.383** (0.0868) (0.0788) (0.111)

No info -0.0898 0.123 -0.0874 (0.163) (0.173) (0.140)

Other Party ------

Left Party -1.164** -1.287** -0.610** (0.127) (0.136) (0.0731)

male ------

female -0.0995 -0.172** 0.0409 (0.0615) (0.0604) (0.0578)

age 0.00641** 0.00862** 0.00978** (0.00234) (0.00242) (0.00236)

AUT ------167

BEL -0.335** -0.248+ 0.139 (0.116) (0.133) (0.138)

FRA -0.421* -0.385+ -0.0236 (0.160) (0.200) (0.113)

GER -0.0654 0.204 -0.129 (0.145) (0.131) (0.138)

HUN 0.223* 0.0864 0.154 (0.0998) (0.118) (0.107)

IRE -0.265** -0.651** -0.204+ (0.0985) (0.121) (0.108)

ISR 0.403** -0.525** 0.0807 (0.130) (0.145) (0.118)

ITA -0.522** 0.191 -0.639** (0.126) (0.132) (0.118)

NET -0.359** 0.135 -0.00592 (0.100) (0.111) (0.119)

NOR -0.197+ -0.298* -0.385** (0.108) (0.126) (0.105)

POL 0.320** 0.123 -0.259* (0.116) (0.141) (0.120)

POR -0.464* 0.297+ -0.288 (0.210) (0.169) (0.241)

SPA 0.228 -0.282+ -0.192 (0.194) (0.160) (0.187)

SWI -0.239+ 0.109 0.832** (0.120) (0.130) (0.120)

UNK -0.598** -0.227 -0.175 (0.140) (0.147) (0.120)

electorateLRP -0.0830 -0.0435 -0.00188 arliamentAvg (0.0585) (0.0476) (0.0472)

Constant 3.545** 3.196** 2.656** (0.378) (0.313) (0.324) R-squared 0.281 0.314 0.247 N 1817 1820 1874 Standard errors in parentheses + p < .10, * p < .05, ** p < .01

168

Table A.9. Regression models on legislators’ attitudes, with low district magnitudes excluded (1) (2) (3) (4) (5) (6) Inequality, Govt Contact with Inequality, Govt Contact with SMD economic trade unions, DM > 15 economic trade unions, excluded intervention, SMD intervention, DM > 15 SMD excluded DM > 15 excluded Business ------

Technical -0.136 -0.182 -0.00765 -0.208 -0.209 0.00514 professional (0.101) (0.110) (0.0945) (0.134) (0.130) (0.131)

Farmer -0.287 -0.466** 0.0666 -0.102 -0.207 0.376 (0.185) (0.175) (0.126) (0.355) (0.229) (0.252)

Lawyer -0.198+ -0.0897 -0.0415 -0.273+ -0.119 0.0481 (0.118) (0.101) (0.0896) (0.162) (0.129) (0.128)

Other white -0.348** -0.0580 -0.260* -0.269+ -0.0325 -0.0978 collar (0.112) (0.142) (0.129) (0.149) (0.182) (0.161)

Politics / law -0.278* -0.243* -0.245* -0.256* -0.303* -0.191 enforcement (0.118) (0.103) (0.107) (0.127) (0.121) (0.133)

Civil service -0.162 -0.208+ -0.0121 -0.124 -0.267+ -0.0764 (0.134) (0.114) (0.117) (0.149) (0.146) (0.152)

Service-based -0.453** -0.249** -0.0720 -0.409** -0.260* -0.0102 professional (0.0871) (0.0933) (0.0987) (0.122) (0.104) (0.120)

Worker -0.384** -0.303** -0.417** -0.436** -0.505** -0.439* (0.0976) (0.0900) (0.119) (0.137) (0.121) (0.187)

No info -0.0854 0.201 -0.0525 -0.156 0.236 -0.0206 (0.187) (0.202) (0.157) (0.238) (0.221) (0.216)

Other Party ------

Left Party -1.110** -1.320** -0.613** -1.160** -1.387** -0.606** (0.105) (0.116) (0.0892) (0.113) (0.136) (0.124)

male ------

female -0.141* -0.160* 0.0285 -0.139* -0.102 0.0573 (0.0668) (0.0651) (0.0629) (0.0596) (0.0754) (0.0897)

age 0.00512* 0.00720** 0.00893** 0.00493+ 0.0102** 0.00725* (0.00220) (0.00263) (0.00258) (0.00256) (0.00282) (0.00335)

AUT ------169

BEL -0.328** -0.254+ 0.144 -0.302 -0.361 0.104 (0.110) (0.130) (0.140) (0.236) (0.246) (0.144)

FRA -0.599+ 0.0262 0.0241 -0.719* -0.133 -0.283+ (0.313) (0.393) (0.132) (0.295) (0.389) (0.144)

GER -0.108 0.198 -0.0147 -0.108 0.0426 -0.218 (0.128) (0.141) (0.149) (0.249) (0.248) (0.185)

HUN 0.259** 0.0758 0.0982 0.289 0.00828 -0.140 (0.0949) (0.111) (0.108) (0.229) (0.227) (0.122)

IRE -0.269** -0.655** -0.221+ (0.0957) (0.115) (0.111)

ISR 0.327** -0.575** 0.0581 0.333 -0.749** -0.0508 (0.107) (0.129) (0.116) (0.239) (0.248) (0.130)

ITA -0.503** 0.199 -0.659** -0.558* 0.0107 -0.755** (0.127) (0.134) (0.116) (0.250) (0.241) (0.145)

NET -0.337** 0.148 0.0201 -0.363 0.00349 -0.0923 (0.102) (0.107) (0.119) (0.223) (0.210) (0.122)

NOR -0.162 -0.287* -0.361** -0.402 -0.329 -0.459** (0.107) (0.124) (0.100) (0.245) (0.245) (0.133)

POL 0.342** 0.116 -0.283* -0.0761 -0.273 -0.332* (0.111) (0.133) (0.124) (0.236) (0.249) (0.140)

POR -0.389+ 0.331* -0.299 -0.382 0.137 -0.499* (0.195) (0.164) (0.238) (0.282) (0.245) (0.227)

SPA 0.287 -0.249 -0.216 0.244 -0.384 -0.369+ (0.232) (0.167) (0.182) (0.347) (0.293) (0.185)

SWI -0.251* 0.0868 0.823** -0.255 -0.0910 0.627** (0.115) (0.124) (0.121) (0.252) (0.237) (0.140)

UNK -0.817** -0.442** -0.304* (0.111) (0.124) (0.119)

Constant 3.156** 3.057** 2.699** 3.195** 3.081** 2.862** (0.149) (0.171) (0.199) (0.286) (0.262) (0.259) R-squared 0.274 0.318 0.254 0.279 0.349 0.223 N 1539 1542 1589 869 873 906 Standard errors in parentheses + p < .10, * p < .05, ** p < .01

170

Table A.10. Regression models on placebo attitude questions and alternative DV: full results (1) (2) (3) (4) Stiffer Immigrants Govt should Left-right sentences must adapt regulate self- cultural placement morals Business ------

Technical -0.150+ -0.0892 0.117 -0.643** professional (0.0804) (0.0891) (0.0957) (0.197)

Farmer -0.0949 -0.154 0.599** -0.0993 (0.143) (0.152) (0.144) (0.363)

Lawyer -0.551** 0.0131 -0.206+ -0.484* (0.0801) (0.0909) (0.109) (0.192)

Other white -0.140 -0.0146 0.0888 -0.543** collar (0.0878) (0.128) (0.109) (0.182)

Politics / law -0.256** -0.0672 -0.0405 -0.647** enforcement (0.0840) (0.0846) (0.0882) (0.203)

Civil service -0.187 0.00631 0.0673 -0.495+ (0.116) (0.108) (0.137) (0.261)

Service-based -0.316** -0.180* -0.0494 -0.844** professional (0.0714) (0.0883) (0.0921) (0.179)

Worker -0.0414 -0.0764 0.138 -0.763** (0.0989) (0.0935) (0.0895) (0.191)

No info -0.0609 -0.148 0.0465 -0.296 (0.131) (0.153) (0.135) (0.344)

Other Party ------

Left Party -0.925** -0.985** -0.516** -3.471** (0.0798) (0.0842) (0.0645) (0.156)

male ------

female -0.0897 -0.141* 0.208** -0.252** (0.0609) (0.0570) (0.0593) (0.0945)

age 0.00377 0.00798** 0.0103** 0.00520 (0.00252) (0.00211) (0.00344) (0.00419)

AUT ------

171

BEL 0.139+ -0.181+ -0.489** -0.387 (0.0801) (0.103) (0.126) (0.243)

FRA 0.268* -0.159 0.0818 0.249 (0.130) (0.162) (0.278) (0.231)

GER -0.296* -0.618** -0.375* -0.0584 (0.127) (0.133) (0.187) (0.217)

HUN 0.972** 0.260** 0.533** 0.375+ (0.0767) (0.0860) (0.0978) (0.192)

IRE 0.319** -0.422** -0.229* -1.139** (0.0788) (0.0886) (0.104) (0.193)

ISR 0.0550 -0.998** -0.0838 0.0219 (0.0814) (0.0940) (0.102) (0.202)

ITA 0.345* -0.324* -0.341* -0.0401 (0.150) (0.141) (0.141) (0.311)

NET -0.00861 -0.0729 -1.078** -0.0334 (0.0771) (0.0889) (0.0952) (0.196)

NOR 0.169* -0.298** -0.957** 1.003** (0.0775) (0.0918) (0.101) (0.208)

POL 0.197* -0.467** 0.101 0.804** (0.0851) (0.0990) (0.101) (0.204)

POR 0.249* -1.130** -0.942** -0.290 (0.102) (0.0904) (0.108) (0.234)

SPA 0.0275 -0.398* -0.486** -0.265 (0.162) (0.186) (0.158) (0.244)

SWI 0.311** 0.166 -0.145 0.128 (0.0936) (0.108) (0.108) (0.214)

UNK -0.398** -0.489** -0.333* -0.517 (0.106) (0.172) (0.131) (0.413)

Constant 3.629** 3.790** 2.628** 6.420** (0.168) (0.136) (0.224) (0.277) R-squared 0.278 0.340 0.153 0.550 N 1815 1819 1816 1824 Standard errors in parentheses + p < .10, * p < .05, ** p < .01

172

Table A.11. Regression models on legislators’ attitudes and behaviour, subsample of influential MPs (1) (2) (3) (4) (5) (6) (7) (8) (9) Inequality, Inequality, Inequality, Govt Govt Govt Contact with Contact with Contact with leadership previously bill sponsor economic economic economic unions, unions, unions, bill position held office subsample intervention, intervention intervention, leadership previously sponsor subsample subsample leadership bill sponsor position held office subsample position subsample subsample subsample subsample Business ------

Technical 0.0242 -0.0331 -0.229 0.0358 -0.0377 -0.456** -0.243+ -0.160 -0.0526 professional (0.222) (0.130) (0.189) (0.163) (0.123) (0.154) (0.139) (0.123) (0.208)

Farmer -0.276 -0.0743 0.791+ -0.418 -0.343 -0.768+ 0.227 -0.319 -0.0411 (0.322) (0.292) (0.393) (0.294) (0.271) (0.447) (0.221) (0.216) (0.349)

Lawyer -0.476+ -0.122 -0.336+ -0.497* -0.143 -0.459* -0.115 -0.109 -0.115 (0.258) (0.185) (0.180) (0.216) (0.172) (0.185) (0.208) (0.130) (0.121)

Other white -0.342* -0.212 -0.116 -0.128 -0.0172 -0.173 -0.291 -0.102 0.178 collar (0.150) (0.133) (0.233) (0.193) (0.145) (0.150) (0.244) (0.147) (0.309)

Politics / law -0.210 -0.186+ -0.364** -0.0753 -0.252* -0.458** -0.170 -0.180 -0.352+ enforcement (0.170) (0.0972) (0.117) (0.162) (0.0999) (0.149) (0.189) (0.119) (0.197)

Civil service -0.843** -0.138 -0.477 -0.212 -0.0964 -0.605** -0.128 -0.172 -0.258 (0.183) (0.164) (0.282) (0.249) (0.132) (0.144) (0.197) (0.161) (0.180)

Service- -0.356* -0.394** -0.424** -0.350* -0.290* -0.375** -0.0801 -0.229+ -0.0772 based (0.139) (0.118) (0.141) (0.175) (0.110) (0.134) (0.184) (0.129) (0.159) professional

Worker -0.422** -0.329** -0.558** -0.413* -0.217* -0.149 -0.404* -0.282+ -0.258 (0.135) (0.123) (0.173) (0.178) (0.103) (0.170) (0.197) (0.159) (0.243)

No info 0.263 0.0289 0.621 0.406 0.247 0.622 -0.00481 0.0413 0.0574 (0.260) (0.208) (0.602) (0.292) (0.251) (0.372) (0.215) (0.208) (0.273)

Other Party ------

Left Party -1.404** -1.226** -0.838** -1.403** -1.391** -0.986** -0.633** -0.619** -0.350* (0.142) (0.179) (0.149) (0.128) (0.144) (0.120) (0.0961) (0.0691) (0.136)

male ------

female 0.00389 -0.105 -0.0616 -0.0310 -0.165* -0.131+ 0.190 0.0799 -0.0480 (0.122) (0.0786) (0.101) (0.119) (0.0804) (0.0745) (0.130) (0.0704) (0.118)

age 0.00701 0.00711* 0.00478 0.0107 0.0140** 0.0153** 0.00715 0.0173** 0.0105+ (0.00508 (0.00344 (0.00319 (0.00664 (0.00436 (0.00418 (0.00478 (0.00340 (0.00532 ) ) ) ) ) ) ) ) )

173

AUT ------

BEL -0.438** -0.318* -0.749** -0.0636 -0.202 -0.914** 0.367+ -0.0610 0.294** (0.142) (0.137) (0.0414) (0.168) (0.173) (0.0578) (0.205) (0.144) (0.0930)

FRA -0.198 -0.458* -0.453** -0.203 -0.361 -1.060** 0.430* -0.0714 0.0334 (0.185) (0.199) (0.0741) (0.235) (0.220) (0.0436) (0.169) (0.165) (0.0575)

GER 0.202 0.0779 -0.667** 0.398* 0.442* -0.561** 0.441+ -0.151 0.432** (0.146) (0.152) (0.161) (0.190) (0.180) (0.149) (0.231) (0.140) (0.131)

HUN 0.892** 0.196 -0.411** 0.559** 0.252+ -0.983** 1.051** -0.130 0.462** (0.124) (0.135) (0.0411) (0.157) (0.148) (0.0752) (0.167) (0.122) (0.0502)

IRE -0.561** -0.177 -0.733** -0.665** 0.423* -0.210+ (0.116) (0.110) (0.150) (0.138) (0.169) (0.125)

ISR 0.311* 0.263+ -0.0815+ -0.0998 -0.750** -1.328** 0.553** -0.483** 0.287** (0.128) (0.143) (0.0426) (0.182) (0.183) (0.0609) (0.190) (0.148) (0.0451)

ITA -0.426* -0.512** -0.952** 0.365* 0.220 -0.454** -0.147 -0.672** -0.508** (0.176) (0.171) (0.0905) (0.171) (0.175) (0.0826) (0.213) (0.124) (0.113)

NET 0.0351 -0.489** -0.813** 0.407** 0.221 -0.568** 0.676** 0.0422 0.275** (0.103) (0.135) (0.0449) (0.140) (0.152) (0.0595) (0.162) (0.133) (0.0492)

NOR -0.251* -0.222 -0.214* -0.174 -0.146 -0.388** -0.228 -0.594** -0.100 (0.119) (0.143) (0.0876) (0.158) (0.169) (0.0840) (0.166) (0.115) (0.0981)

POL 0.316** 0.190 0.0301 -0.244 0.146 -0.323** -0.911** -0.493** -0.0838 (0.108) (0.146) (0.0620) (0.158) (0.175) (0.0387) (0.169) (0.131) (0.0661)

POR 0.125 -0.850** 0.576** -0.286** -0.206 -0.187 (0.167) (0.166) (0.165) (0.0525) (0.301) (0.202)

SPA 0.804+ 0.370 -0.205 -0.170 0.0569 -0.247 (0.448) (0.297) (0.236) (0.216) (0.204) (0.191)

SWI -0.423+ -0.169 0.231 0.0927 1.158** 0.717** (0.213) (0.158) (0.169) (0.166) (0.223) (0.160)

UNK -0.460** -0.429* -1.223** 0.431 -0.135 -1.062** 0.0550 -0.167 -0.0565 (0.138) (0.177) (0.126) (0.346) (0.180) (0.0379) (0.271) (0.151) (0.101)

Constant 3.084** 3.042** 3.510** 2.715** 2.623** 3.259** 2.383** 2.342** 2.399** (0.313) (0.240) (0.203) (0.424) (0.291) (0.238) (0.342) (0.222) (0.360) R-squared 0.388 0.290 0.306 0.370 0.354 0.256 0.222 0.221 0.0957 N 433 907 504 432 911 508 442 935 514 Standard errors in parentheses + p < .10, * p < .05, ** p < .01

174

Table A.12. Regression models on candidates’ attitudes (CCS data) (1) (2) (3) (4) Inequality Intervention in Social Security Globalization economy Business sector 0 0 0 0 (.) (.) (.) (.)

Technical professionals -0.0695 -0.0282 0.0207 -0.219* (0.0376) (0.0319) (0.0228) (0.0613)

Agriculture, fisheries 0.135+ 0.0283 0.0661 -0.736** (0.0576) (0.0425) (0.123) (0.0988)

Lower professionals -0.262* -0.0463 -0.155 -0.395* (0.0727) (0.0590) (0.0825) (0.136)

Politics/military -0.0802 -0.0761 -0.0619 -0.0109 (0.115) (0.0574) (0.116) (0.107)

Teachers & university -0.351* -0.157+ -0.136 -0.331* profs (0.115) (0.0701) (0.0809) (0.0878)

Clerks, service & sales -0.162* -0.0886* -0.160** -0.455* (0.0463) (0.0352) (0.0384) (0.117)

Trades & skilled -0.379** -0.184* -0.276* -0.476* manual (0.0688) (0.0728) (0.0873) (0.147)

No info -0.213* -0.0615 -0.120* -0.372* (0.0780) (0.0438) (0.0401) (0.0963)

Left party=0 0 0 0 0 (.) (.) (.) (.)

Left party=1 -1.343** -0.837** -0.573** -0.107 (0.203) (0.162) (0.135) (0.161)

Switzerland 0 0 0 0 (.) (.) (.) (.)

Germany 0.122* 0.0929** -0.0120 (0.0413) (0.0221) (0.0249)

Greece -0.460** -0.279** -0.850** -0.244** (0.0193) (0.0268) (0.00797) (0.0197)

Portugal -0.445** -0.496** -0.236** 0.215** (0.0159) (0.0142) (0.0149) (0.0325)

Norway -0.163* -0.524** -0.0680 0.619** (0.0479) (0.0215) (0.0419) (0.0527)

175

Italy -0.743** -0.409** -0.578** -0.172** (0.0287) (0.0255) (0.00710) (0.0264)

UK -0.122* -0.468** 0.625** -0.203** (0.0340) (0.0428) (0.0161) (0.0175)

education=1 -0.284+ -0.319* -0.103 -0.144 (0.145) (0.0966) (0.171) (0.756)

education=2 -0.388+ -0.144 -0.278* -0.232 (0.199) (0.309) (0.0957) (0.411)

education=3 -0.0964 0.292 -0.278* -0.350** (0.0958) (0.273) (0.0884) (0.0841)

education=4 0.0369 0.169 -0.154** -0.336* (0.0751) (0.106) (0.0322) (0.0906)

education=5 0.117 0.0770 -0.0863 -0.142 (0.0725) (0.0512) (0.0586) (0.0945)

education=6 -0.143+ 0.0555 -0.116** -0.0544 (0.0709) (0.107) (0.0305) (0.164)

education=7 0 0 0 0 (.) (.) (.) (.)

education=8 -0.0797* 0.0263 -0.104** -0.0881+ (0.0299) (0.0355) (0.0194) (0.0427)

education=9 0.0125 0.0789* 0.0350 0.0383 (0.0695) (0.0316) (0.0352) (0.124)

age -0.00180 0.0000119 -0.00864** -0.00761+ (0.00185) (0.00154) (0.00186) (0.00353)

female -0.0881 -0.0883 -0.143+ 0.0183 (0.0853) (0.0561) (0.0599) (0.0426)

Constant 3.720** 2.969** 3.156** 3.697** (0.130) (0.165) (0.138) (0.193) Observations 6570 6497 6529 4720 Standard errors in parentheses + p < .10, * p < .05, ** p < .01

176

Table A.13. Regression models on legislators’ attitudes and behaviour, using ordered logit (1) (2) (3) Inequality Govt Contact with economic unions intervention

Business ------

Technical -0.216 -0.310+ 0.0713 professional (0.160) (0.161) (0.157)

Farmer -0.488+ -0.672** 0.0519 (0.257) (0.251) (0.246)

Lawyer -0.324+ -0.155 0.178 (0.190) (0.189) (0.185)

Other white -0.675** -0.113 0.257 collar (0.202) (0.203) (0.195)

Politics / law -0.438* -0.386* 0.319+ enforcement (0.170) (0.170) (0.167)

Civil service -0.391+ -0.322 0.0523 (0.229) (0.226) (0.221)

Service-based -0.796** -0.510** 0.228 professional (0.150) (0.147) (0.145)

Worker -0.496** -0.527** 0.728** (0.184) (0.184) (0.182)

No info -0.194 0.239 0.177 (0.277) (0.280) (0.256)

Other Party ------

Left Party -1.984** -2.272** 1.059** (0.103) (0.107) (0.0936)

male ------

female -0.215* -0.325** -0.0611 (0.101) (0.100) (0.0976)

age 0.0117* 0.0170** -0.0176** (0.00454) (0.00460) (0.00444)

AUT ------

BEL -0.482* -0.341+ -0.206 (0.204) (0.205) (0.205)

177

FRA -0.801** -0.668* 0.0126 (0.280) (0.275) (0.256)

GER 0.0453 0.555** 0.237 (0.189) (0.190) (0.184)

HUN 0.406+ 0.350 -0.266 (0.240) (0.247) (0.229)

IRE -0.599+ -1.093** 0.305 (0.351) (0.359) (0.333)

ISR 0.454 -0.770* -0.108 (0.409) (0.390) (0.431)

ITA -0.686** 0.556* 1.176** (0.217) (0.219) (0.216)

NET -0.487+ 0.377 0.0308 (0.260) (0.259) (0.269)

NOR -0.392 -0.536+ 0.644* (0.308) (0.313) (0.303)

POL 0.466 0.270 0.440 (0.301) (0.302) (0.322)

POR -0.496* 0.695** 0.537* (0.233) (0.235) (0.236)

SPA 0.608** -0.313 0.308 (0.212) (0.214) (0.211)

SWI -0.468** 0.217 -1.504** (0.164) (0.164) (0.164)

UNK -0.953** -0.233 0.264 (0.245) (0.239) (0.231) / cut1 -2.207** -2.120** -2.843** (0.307) (0.309) (0.304)

cut2 -0.609* -0.135 -1.481** (0.302) (0.304) (0.297)

cut3 0.300 0.670* 0.0989 (0.301) (0.304) (0.295)

cut4 2.417** 2.747** 2.050** (0.313) (0.314) (0.301) Observations 1817 1820 1874 Standard errors in parentheses + p < .10, * p < .05, ** p < .01

178

Table A.14. Regression models on legislators' contact with workers’ organizations, union staff separated (1) (2) (3) No controls Controls Controls, weighted Business 0 0 0 (.) (.) (.)

Technical professional 0.128 0.0392 0.0664 (0.0829) (0.0830) (0.109)

Farmer -0.0647 0.00918 -0.0272 (0.137) (0.136) (0.188)

Lawyer 0.143 0.104 0.135 (0.0955) (0.0908) (0.112)

Other white collar 0.265* 0.159 0.0730 (0.129) (0.127) (0.154)

Politics / law 0.272** 0.166 0.229* enforcement (0.102) (0.105) (0.113)

Civil service 0.219+ 0.0542 0.135 (0.111) (0.104) (0.0954)

Service-based 0.313** 0.107 0.108 professional (0.0961) (0.0840) (0.0943)

Worker 0.397** 0.251* 0.116 (0.124) (0.113) (0.131)

No info 0.219 0.0860 0.0960 (0.152) (0.139) (0.150)

Union staff 1.349** 0.965** 0.687** (0.273) (0.258) (0.219)

AUT 0 0 0 (.) (.) (.)

BEL -0.280+ -0.190 -0.232+ (0.168) (0.145) (0.123)

FRA 0.0252 0.0110 -0.0586 (0.119) (0.107) (0.0789)

GER 0.173 0.118 0.107 (0.134) (0.123) (0.101)

HUN -0.175 -0.161 -0.223** (0.115) (0.105) (0.0783) 179

IRE 0.0402 0.192+ 0.112 (0.111) (0.107) (0.0832)

ISR -0.245* -0.0957 -0.182* (0.119) (0.112) (0.0897)

ITA 0.580** 0.616** 0.547** (0.115) (0.113) (0.0899)

NET -0.0210 -0.0226 -0.111 (0.123) (0.116) (0.0909)

NOR 0.406** 0.395** 0.391** (0.107) (0.0972) (0.0754)

POL 0.0675 0.236* 0.122 (0.118) (0.117) (0.102)

POR 0.308 0.278 0.138 (0.215) (0.229) (0.232)

SPA 0.195 0.183 0.0803 (0.188) (0.181) (0.162)

SWI -0.885** -0.849** -0.745** (0.120) (0.117) (0.115)

UNK 0.0798 0.134 0.111 (0.120) (0.123) (0.0985)

Other Party 0 0 (.) (.)

Left Party 0.588** 0.486** (0.0716) (0.0847)

age -0.0101** -0.00918** (0.00236) (0.00263)

male 0 0 (.) (.)

female -0.0353 -0.0337 (0.0563) (0.0622)

Constant 2.980** 3.391** 3.437** (0.114) (0.183) (0.191) Observations 1946 1874 1874 Standard errors in parentheses + p < .10, * p < .05, ** p < .01

180

Table A.15. The relationship between class and inequality attitudes, conditioned by institution (1) (1) (1) Open Division of Local lists powers nominations Technical 0.0460 0.0840 0.0807 professional (0.126) (0.101) (0.141)

Farmer -0.128 -0.0291 -0.238 (0.267) (0.213) (0.223)

Lawyer -0.260+ -0.262* 0.00626 (0.140) (0.125) (0.181)

Other white collar -0.215 -0.207+ -0.227 (0.131) (0.111) (0.190)

Politics / law -0.219 -0.125 -0.0431 enforcement (0.133) (0.106) (0.156)

Civil service -0.0536 -0.0622 0.0857 (0.132) (0.127) (0.160)

Service-based -0.222* -0.265** -0.220+ professional (0.101) (0.0892) (0.126)

Worker -0.174 -0.113 -0.0979 (0.109) (0.0924) (0.172)

No info 0.246 0.156 0.0391 (0.289) (0.218) (0.323)

Open list 0.237 Division of powers 0.00968 Local nomination 0.397 (0.173) (0.257) (0.303)

Technical -0.378+ Technical professional -0.481** Technical -0.293 professional # Open # Division of powers professional # list Local nomination (0.214) (0.181) (0.202)

Farmer # Open list -0.346 Farmer # Division of -0.563 Farmer # Local -0.0772 powers nomination (0.392) (0.341) (0.312)

Lawyer # Open list 0.459* Lawyer # Division of 0.373+ Lawyer # Local -0.244 powers nomination (0.219) (0.193) (0.249)

Other white collar # -0.244 Other white collar # -0.406+ Other white collar # -0.247 Open list Division of powers Local nomination (0.240) (0.218) (0.253) 181

Politics / law -0.0955 Politics / law -0.215 Politics / law -0.274 enforcement # Open enforcement # Division enforcement # list of powers Local nomination (0.244) (0.221) (0.191)

Civil service # Open -0.187 Civil service # Division -0.433 Civil service # -0.507+ list of powers Local nomination (0.429) (0.336) (0.257)

Service-based - Service-based -0.463** Service-based -0.335+ professional # Open 0.555** professional # Division professional # list of powers Local nomination (0.188) (0.173) (0.176)

Worker # Open list -0.513* Worker # Division of -0.581** Worker # Local -0.276 powers nomination (0.194) (0.182) (0.213)

No info # Open list -0.657+ No info # Division of -0.555+ No info # Local -0.303 powers nomination (0.362) (0.311) (0.367)

Left Party - Left Party -1.159** Left Party -1.185** 1.097** (0.107) (0.128) (0.119) female -0.150* female -0.0996 female -0.0918 (0.0680 (0.0613) (0.0654) ) age 0.0052 age 0.00647** age 0.00646** 8* (0.0022 (0.00224) (0.00233) 1)

BEL -0.304* BEL -0.304* BEL -0.322* (0.122) (0.116) (0.122)

FRA -0.621* FRA -0.186 FRA -0.312* (0.305) (0.285) (0.127)

GER -0.140 GER -0.00228 GER 0.0894 (0.134) (0.140) (0.194)

HUN 0.274** HUN 0.242* HUN 0.327* (0.0946 (0.0986) (0.155) )

ISR 0.350** IRE -0.265** IRE -0.201 (0.114) (0.0950) (0.164)

ITA - ISR 0.372** ISR 0.458 0.477** 182

(0.125) (0.112) (0.287)

NET -0.282* ITA -0.483** ITA -0.370 (0.106) (0.130) (0.279)

NOR -0.183 NET -0.294** NET -0.198 (0.115) (0.0994) (0.275)

POL 0.333* NOR -0.155 NOR -0.167 (0.154) (0.112) (0.116)

POR -0.396* POL 0.624* POL 0.422 (0.180) (0.256) (0.282)

SPA 0.290 POR -0.207 POR -0.439 (0.237) (0.213) (0.266)

SWI -0.224 SPA 0.334 SPA 0.420 (0.160) (0.231) (0.341)

UNK - SWI 0.0455 SWI -0.254+ 0.816** (0.114) (0.249) (0.131)

Constant 3.016** UNK -0.555** UNK -0.589** (0.146) (0.146) (0.135)

Constant 2.929** Constant 2.772** (0.155) (0.355)

Observations 1510 Observations 1817 Observations 1751 Standard errors in parentheses + p < .10, * p < .05, ** p < .01

183

Table A.16. The relationship between class and contact with workers’ organizations, conditioned by institution (1) (1) (1) Open Division of Local lists powers nominations Technical 0.0176 0.0226 0.0631 professional (0.104) (0.0861) (0.149)

Farmer 0.0532 0.0660 0.113 (0.185) (0.207) (0.301)

Lawyer 0.0957 0.0515 -0.0000622 (0.109) (0.115) (0.106)

Other white collar 0.00315 -0.177 0.00655 (0.184) (0.146) (0.173)

Politics / law 0.276+ 0.0273 0.193 enforcement (0.147) (0.132) (0.150)

Civil service - -0.0192 -0.0327 0.00543 (0.145) (0.127) (0.144)

Service-based -0.0411 -0.0386 0.124 professional (0.153) (0.114) (0.167)

Worker 0.316+ 0.256+ 0.370+ (0.167) (0.135) (0.191)

No info -0.0342 0.0258 -0.0239 (0.244) (0.188) (0.274)

Open list -0.0894 Division of powers 0.317+ Local nomination -0.00279 (0.165) (0.172) (0.170)

Technical -0.0966 Technical professional -0.0420 Technical -0.0413 professional # Open # Division of powers professional # list Local nomination (0.197) (0.180) (0.178)

Farmer # Open list -0.153 Farmer # Division of -0.0768 Farmer # Local -0.0933 powers nomination (0.247) (0.262) (0.324)

Lawyer # Open list -0.239 Lawyer # Division of 0.00241 Lawyer # Local 0.166 powers nomination (0.203) (0.183) (0.172)

Other white collar # 0.543* Other white collar # 0.857** Other white collar # 0.257 Open list Division of powers Local nomination 184

(0.239) (0.200) (0.191)

Politics / law -0.107 Politics / law 0.311 Politics / law -0.0729 enforcement # Open enforcement # enforcement # list Division of powers Local nomination (0.216) (0.205) (0.205)

Civil service # Open -0.0255 Civil service # 0.0215 Civil service # 0.189 list Division of powers Local nomination (0.337) (0.265) (0.237)

Service-based 0.255 Service-based 0.362* Service-based -0.0350 professional # Open professional # Division professional # list of powers Local nomination (0.190) (0.158) (0.216)

Worker # Open list 0.248 Worker # Division of 0.392 Worker # Local 0.0883 powers nomination (0.253) (0.257) (0.219)

No info # Open list 0.144 No info # Division of 0.117 No info # Local 0.176 powers nomination (0.320) (0.269) (0.319)

Left Party 0.610** Left Party 0.615** Left Party 0.631** (0.0900 (0.0722) (0.0737) ) female -0.0254 female -0.0486 female -0.0314 (0.0647 (0.0591) (0.0590) ) age - age -0.0100** age -0.00994** 0.00885 ** (0.0026 (0.00248) (0.00251) 6)

BEL -0.127 BEL -0.116 BEL -0.131 (0.139) (0.138) (0.144)

FRA - FRA -0.274 FRA 0.0562 0.00765 (0.128) (0.166) (0.113)

GER 0.0857 GER 0.187 GER 0.151 (0.140) (0.118) (0.131)

HUN -0.113 HUN -0.156 HUN -0.169 (0.114) (0.111) (0.111)

ISR -0.0643 IRE 0.171 IRE 0.210+ (0.123) (0.110) (0.118)

185

ITA 0.667** ISR -0.0895 ISR -0.0500 (0.134) (0.119) (0.142)

NET -0.0481 ITA 0.640** ITA 0.687** (0.127) (0.117) (0.144)

NOR 0.441** NET 0.0145 NET 0.0187 (0.103) (0.124) (0.144)

POL 0.399* NOR 0.450** NOR 0.361** (0.161) (0.102) (0.107)

POR 0.313 POL -0.150 POL 0.282+ (0.230) (0.177) (0.150)

SPA 0.223 POR -0.0388 POR 0.543** (0.184) (0.161) (0.139)

SWI -0.801** SPA 0.190 SPA 0.218 (0.164) (0.182) (0.206)

UNK 0.353** SWI -1.346** SWI -0.833** (0.119) (0.185) (0.127)

Constant 3.323** UNK 0.203 UNK 0.176 (0.209) (0.124) (0.129)

Constant 3.445** Constant 3.335** (0.185) (0.198)

Observations 1560 Observations 1874 Observations 1808 Standard errors in parentheses + p < .10, * p < .05, ** p < .01

186

Appendix B: Supporting materials to Paper 2

Figure B.1. Distribution of instrumental variable values at different bandwidths

Note: The bandwidths begin at ε = 0 and rise in increments of 0.04. Zeros are the most common instrumental value, but they are excluded from these graphs for easier viewing.

187

Table B.1. Statistics Finland municipal expenditure categories, 2016 Statistics Finland Statistics Finland category description Part of social category number spending DV? 110 General administration 212 Child welfare institutional and family care Yes 213 Non-institutional services in child welfare Yes 218 Other non-institutional services for children and families Yes 220 Institutional care for the elderly Yes 221 Housing services for the elderly in around-the-clock care Yes 223 Other services for the elderly Yes 225 Institutional care for the disabled Yes 226 Housing services for the disabled in around-the-clock care Yes 228 Other services for the disabled Yes 231 Home care Yes 238 Services in support of employment Yes 245 Special services for substance abusers Yes 253 Outpatient care in primary health care 254 Oral health care 256 Ward care in primary health care 260 Specialised health care 270 Environmental health care 290 Other social services and health care activities 299 Social services and health care total 302 Early childhood education and care Yes 304 Pre-primary education Yes 305 Primary education Yes 310 Upper secondary education Yes 315 Vocational secondary education Yes 325 Liberal adult education provided by adult education centres Yes 335 Basic arts education Yes 345 Other educational activities Yes 350 Libraries 355 Sports and recreation 360 Youth activities 370 Museum and exhibition activities 375 Theatre, dance and circus activities 385 Music activities 390 Other cultural activities 399 Educational and cultural activities total 410 Community planning 420 Building supervision 440 Environmental protection 188

460 Traffic routes 470 Parks and public areas 480 Fire and rescue activities 520 Relief work services for farmers 535 Premises and leasing services 550 Support services 555 Promotion of economic development 610 Water supply and sewage management 620 Energy supply 625 Waste management 630 Public transport 640 Port activities 660 Agriculture and forestry 690 Other activities 700 Operating activities total Note: The categories available from Statistics Finland differ slightly year-to-year (2016 categories shown).

189

Table B.2. Robustness models: year as unit of analysis, non-social spending, and placebo thresholds Year as unit of Non-social Placebo analysis spending thresholds Panel A: IV ε = 0.4 (1) (2) (3) Workers 0.00198+ -0.000462 -0.00000277 (0.00107) (0.00111) (0.00108) First stage F-stat 105.2 104.4 89.75 Panel B: Reduced form of IV ε = 0.4 (4) (5) (6) Workers 0.00173+ -0.000402 -0.00000247 (0.000915) (0.000973) (0.000972) R2 0.669 0.749 0.675 Observations 5573 1393 1392 Standard errors in parentheses. Models include year, party, municipality and vote share controls. + p < 0.1, * p < 0.05, ** p < 0.01

Table B.3. Results for social expenditures: next largest party and whole council Next largest Whole party council Panel A: IV ε = 0.4 (1) (2) Workers 0.000133 0.00191 (0.000897) (0.00165) First stage F-stat 56.05 380.1 Panel B: Reduced form of IV ε = 0.4 (3) (4) Workers 0.000138 0.00186 (0.000940) (0.00163) R2 0.690 0.674 Observations 1374 1393 Standard errors in parentheses. Models include year, party, municipality and vote share controls. + p < 0.1, * p < 0.05, ** p < 0.01

190

Appendix C: Supporting materials to Paper 3

Table C.1. T-test of working-class share by elected status (CCS)

Unelected N Elected N Diff P-value Workers .1503759 3724 .1212121 660 .0291638+ .0505179 + p < .10, * p < .05, ** p < .01

Table C.2. T-test of working-class share by elected status (CCS; right parties only)

Unelected N Elected N Diff P-value Workers .147651 1341 .097166 247 .050485* .0355208 + p < .10, * p < .05, ** p < .01

Table C.3. T-test of working-class share by elected status (CCS; left parties only)

Unelected N Elected N Diff P-value Workers .1514382 2364 .1365854 410 .0148529 .4360708 + p < .10, * p < .05, ** p < .01

Table C.4. T-test of working-class share by electoral institution (PARTIREP)

Closed list N Open List N Diff P-value Worker .1112661 35 .100558 32 .0107081 .6763233 + p < .10, * p < .05, ** p < .01

Table C.5. T-test of working-class share by electoral institution (PARTIREP)

Fusion N Division N Diff P-value Worker .1128418 43 .0877769 29 .025065 .3098951 + p < .10, * p < .05, ** p < .01

Table C.6. T-test of working-class share by country polarization level (PARTIREP)

Low pol. N High pol. N Diff P-value Worker .0785628 7 .0750263 8 .0035365 .9095911 + p < .10, * p < .05, ** p < .01

Table C.7. T-test of worker share by selectorate scope (CCS)

Broad N Narrow N Diff P-value Workers .2087912 728 .1597908 1721 .0490004** .0034762 + p < .10, * p < .05, ** p < .01

191

Table C.8. T-test of party connections (past employment) by class (CCS)

Non-worker N Worker N Diff P-value Party .1770439 3694 .1843854 602 -.0073415 .6624887 connection Indicates share that previously worked for party or MP. + p < .10, * p < .05, ** p < .01

Table C.9. T-test of level of contestation of nomination by class (CCS)

Non-worker N Worker N Diff P-value Contested 1.581388 3385 1.683656 569 -.1022671** .0057315 nomination Higher number indicates more contested. + p < .10, * p < .05, ** p < .01

Table C.10. T-test of self-assessed chances of winning by class (CCS)

Non-worker N Worker N Diff P-value Chances 1.938637 4563 1.86514 786 .0734969 .1171168 Higher number indicates greater chance. + p < .10, * p < .05, ** p < .01

Table C.11. T-test of working-class share by party type (PARTIREP)

Other N Left N Diff P-value Worker .0642919 1151 .122449 931 -.0581571** 3.98e-06 + p < .10, * p < .05, ** p < .01

Table C.12. T-test of worker share by party type (CCS; elected only)

Other N Left N Diff P-value Workers .097166 247 .1365854 410 -.0394194 .134913 + p < .10, * p < .05, ** p < .01

Table C.13. T-test of worker share by party type (CCS)

Other N Left N Diff P-value Workers .1310924 2380 .1465983 3513 -.0155059+ .0926878 + p < .10, * p < .05, ** p < .01

Table C.14. T-test of working-class share by country union density level (PARTIREP)

Low union N High union N Diff P-value Worker .0515107 8 .1054377 7 -.053927+ .0640628 + p < .10, * p < .05, ** p < .01

Table C.15. T-test of worker share by country union density (CCS)

Low union N High union N Diff P-value Workers .0755953 3 .1862079 4 -.1106126+ .0698026 + p < .10, * p < .05, ** p < .01

192

Table C.16. T-test of union membership by class (CCS)

Not worker N Worker N Diff P-value Union .2944456 4843 .3478774 848 -.0534318** .0017735 member + p < .10, * p < .05, ** p < .01

Table C.17. T-test of elected status by union membership (CCS)

Non-union N Union N Diff P-value Elected .1421438 3004 .1734694 1176 -.0313256* .0109601 status + p < .10, * p < .05, ** p < .01

Table C.18. T-test of elected status by union membership (CCS; workers only)

Non-union N Union N Diff P-value Elected .1004785 418 .1808511 188 -.0803726** .0056646 status + p < .10, * p < .05, ** p < .01

Table C.19. T-test of elected status by union membership (CCS; non-workers)

Non-union N Union N Diff P-value Elected .1488786 2586 .1720648 988 -.0231862+ .0870021 status + p < .10, * p < .05, ** p < .01

Table C.20. T-test of campaign starting time by class (CCS)

Non-worker N Worker N Diff P-value Early start 3.549793 3625 3.834131 627 -.2843377** 7.14e-06 Higher number indicates a later start. + p < .10, * p < .05, ** p < .01

Table C.21. T-test of campaign starting time by class (CCS; elected only)

Non-worker N Worker N Diff P-value Early start 3.317204 372 3.928571 56 -.6113671** .0058082 Higher number indicates a later start. + p < .10, * p < .05, ** p < .01

Table C.22. T-test of full-time campaign starting time by class (CCS)

Non-worker N Worker N Diff P-value Early full- 4.28021 3244 4.428826 562 -.148616** .006372 time Higher number indicates a later start. + p < .10, * p < .05, ** p < .01

193

Table C.23. T-test of full-time campaign starting time by class (CCS; elected only)

Non-worker N Worker N Diff P-value Early full- 4.002688 372 4.528302 53 -.5256137** .0098396 time Higher number indicates a later start. + p < .10, * p < .05, ** p < .01

Table C.24. T-test of campaign budget by class (CCS)

Non-worker N Worker N Diff P-value Campaign 11377.98 2626 7583.094 432 3794.884* .0211132 budget + p < .10, * p < .05, ** p < .01

Table C.25. T-test of campaign budget by class (CCS; elected only)

Non-worker N Worker N Diff P-value Campaign 20793.96 358 14331.91 47 6462.04* .0474333 budget + p < .10, * p < .05, ** p < .01

194

Table C.26. Regression model relating class to campaign budgets (CCS) (1) Campaign budget Business sector ---

Technical professionals (incl -1801.8 nurses) (1607.8)

Agriculture and fisheries -4461.5+ (2267.3)

Lower-level professionals -395.5 (3095.0)

Politics/military 553.1 (2848.3)

Teachers and university -3864.5* professors (1444.0)

Workers (clerks, -4617.6* service/sales, manual (1423.7)

No occupation info -4403.9+ (1831.9)

Left party -4583.0* (1353.2)

Not elected -13273.1** (2978.6)

Elected 1341.4 (4708.4)

No data on elected status ---

Switzerland ---

Germany 2350.9** (544.5)

Ireland 13901.0** (3062.8)

Greece -658.2 (1567.4)

195

Portugal -6376.0** (1403.6)

Norway 149593.1** (2841.5)

Italy -2251.6** (413.9)

Constant 23540.3** (3231.4) Observations 3970 Standard errors in parentheses + p < .10, * p < .05, ** p < .01

Table C.27. T-test of campaign staff size by class (CCS)

Non-worker N Worker N Diff P-value Campaign 12.95358 1605 8.453125 192 4.500458 .1115317 staff + p < .10, * p < .05, ** p < .01

Table C.28. T-test of campaign staff size by class (CCS; elected only)

Non-worker N Worker N Diff P-value Campaign 32.5272 239 16.16667 24 16.36053 .3336424 staff + p < .10, * p < .05, ** p < .01

Table C.29. T-test of party-provided campaign staff size by class (CCS)

Non-worker N Worker N Diff P-value Party staff .7710728 522 .3404255 47 .4306473* .0121567 + p < .10, * p < .05, ** p < .01

Table C.30. T-test of party-provided campaign staff size by class (CCS; elected only)

Non-worker N Worker N Diff P-value Party staff 1.225 140 .75 12 .475 .2113504 + p < .10, * p < .05, ** p < .01

Table C.31. T-test of worker share by country campaign expense (CCS)

Low budget N High budget N Diff P-value Workers .1778654 4 .143928 4 .0339374 .6200249 + p < .10, * p < .05, ** p < .01

Table C.32. T-test of working-class share by elected status (PARTIREP)

National N Subnational N Diff P-value Worker .0630342 936 .1121495 1177 -.0491153** .0000901 + p < .10, * p < .05, ** p < .01

196

Table C.33. T-test of working-class share by country P90/P10 inequality level (PARTIREP)

Low ineq. N High ineq. N Diff P-value Worker .0924919 8 .0586022 7 .0338897 .2647533 + p < .10, * p < .05, ** p < .01

Table C.34. T-test of worker share by country P90/P10 inequality (CCS; elected only)

Low N High N Diff P-value inequality inequality Workers .1573541 3 .1248887 4 .0324654 .6506215 + p < .10, * p < .05, ** p < .01

Table C.35. T-test of worker share by country P90/P10 inequality (CCS)

Low N High N Diff P-value inequality inequality Workers .1730066 5 .1341627 5 .0388438 .4655971 + p < .10, * p < .05, ** p < .01

197