Voting and Participating in Direct Democracies

DISSERTATION of the University of St.Gallen, School of Management, Economics, Law, Social Sciences and International Affairs to obtain the title of Doctor of Philosophy in Economics and Finance

submitted by

Katharina Eva Hofer-Jaronicki

from Germany

Approved on the application of

Prof. Dr. Monika Bütler and Prof. Dr. Patricia Funk

Dissertation no. 4388

difo Druck, Bamberg 2015 The University of St. Gallen, School of Management, Economics, Law, Social Sciences and International Affairs hereby consents to the printing of the pre- sent dissertation, without hereby expressing any opinion on the views herein expressed.

St. Gallen, November 10, 2014 The President:

Prof. Dr. Thomas Bieger

2 Acknowledgements

A dissertation is not only the result of long hours of work. Rather, it thrives upon the interactions and discussions with, as well as input and support from key people along the way. I would like to use this opportunity to express my gratitude to those individuals. First of all, I would like to thank my thesis supervisor Monika Bütler for her support and guidance throughout my dissertation. Her encouragement to work on questions that are of personal interest to me has been very inspiring. My thanks also go to the members of my thesis committee, Reto Föllmi, Pa- tricia Funk and Roland Hodler, for their useful comments that helped further improve my work. In addition, I am grateful to Antonio Merlo for inviting me to spend a year at the University of Pennsylvania. My research has benefitted greatly from our discussions and his input. Moreover, I would like to thank all my colleagues and friends who helped my doctoral studies elapse very quickly and made it such a pleasant experi- ence. Particularly, I am grateful to two former colleagues at the SEW-HSG, Andreas Steinmayr and Martin Huber, for their extremely helpful advice. I would also like to thank my office mate and co-author Christian Marti for long and interesting discussions. My special thanks go to my family. I am deeply grateful to my parents for their unconditional love and support not only during the doctoral studies but throughout my entire life. Last but not least, I would like to thank my husband Tobias who has been of invaluable importance in accomplishing this important milestone in my life.

St. Gallen, November 2014 Katharina Eva Hofer-Jaronicki

i Contents

List of Figures v

List of Tables viii

1 Introduction 1

2 Campaigning in Direct Democracies: Initiative Petition Sign- ing, Voter Turnout, and Acceptance 6 2.1 Introduction ...... 6 2.2 Theoretical Justification and Hypotheses ...... 9 2.2.1 Turnout ...... 11 2.2.2 Acceptance ...... 12 2.3 Institutional Background, Data, and Descriptives ...... 13 2.3.1 Institutional Background ...... 13 2.3.2 Data and Descriptives ...... 14 2.4 Estimation Strategy ...... 17 2.4.1 Turnout ...... 17 2.4.2 Acceptance ...... 18 2.4.3 Fixed Effects and Controls ...... 19 2.5 Baseline Results ...... 21 2.5.1 Turnout ...... 22 2.5.2 Acceptance ...... 26 2.5.3 Controlling for Voter Preferences ...... 30 2.5.4 Probability of Signing an Initiative ...... 35 2.5.5 Additional Robustness Checks ...... 37 2.6 Concluding Remarks ...... 39

ii Appendix 2.A Data Used from Swiss Census (1970, 1980, 1990, 2000) 41 Appendix 2.B Presentations and Acknowledgement ...... 41

3 Ready to Reform: How Popular Initiatives Can Be Successful 43 3.1 Introduction ...... 43 3.2 Institutional Background ...... 48 3.2.1 Main Characteristics ...... 48 3.2.2 Institutional Changes ...... 49 3.3Model...... 51 3.3.1 Model Setup ...... 52 3.3.2 Discussion of Modeling Choices ...... 57 3.3.3 Subgame Perfect Equilibrium ...... 59 3.3.4 Equilibrium Analysis ...... 66 3.4 Data and Empirical Strategy ...... 70 3.4.1 Data ...... 70 3.4.2 Empirical Strategy ...... 75 3.5 Results ...... 79 3.5.1 Probability of Amending the Status Quo ...... 79 3.5.2 Signature Collection ...... 84 3.5.3 Signature Collection Costs ...... 87 3.5.4 Collection Time Constraint and Signature Requirement 89 3.5.5 Voting Rule: Tie-Breaking Question ...... 91 3.6 Concluding Remarks ...... 93 Appendix 3.A Proofs ...... 94 Appendix 3.B Data Appendix ...... 100 Appendix 3.C Coding of Time Periods for Initiatives ...... 102 Appendix 3.D Presentations and Acknowledgement ...... 104

4 Does Female Suffrage Increase Public Support for Govern- ment Spending? Evidence from Swiss Ballots 105 4.1 Introduction ...... 105 4.2 Institutional Setup ...... 110 4.3 Empirical Framework ...... 116 4.4 Data and Estimation Method ...... 120 4.4.1 Data ...... 120 4.4.2 Identifying the Total Gender Effect ...... 123

iii 4.4.3 Identifying the Direct and Indirect Gender Effect ....131 4.5 Results ...... 135 4.5.1 Average Treatment Effect ...... 135 4.5.2 Direct and Indirect Effects ...... 139 4.6 Concluding Remarks ...... 145 Appendix 4.A Estimators of Direct and Indirect Effects ...... 147 Appendix 4.B Propensity Score Histograms ...... 148 Appendix 4.C Propensity Score Estimates ...... 162 Appendix 4.D Direct and Indirect Effects ...... 166 Appendix 4.E Federal Announcements / Bundesblätter ...... 169 Appendix 4.F Presentations and Acknowledgement ...... 170

Bibliography 171

iv List of Figures

1.1 Signatures for Swiss Popular Initiatives at Federal Level .... 2

2.1 Histogram of Turnout, Signatures Per Capita, and Acceptance Rate ...... 15

3.1 Example of Winning Probabilities ...... 53 3.2 High Type Probability over Collection Time ...... 54 3.3 Extensive-Form Game Tree of Initiative Game ...... 56 3.4 Example of Cutoff Counter Proposal ...... 61 3.5 Example of Initiative Type and Winning Probability ...... 63 3.6 Politicians’ Best Strategy ...... 64 3.7 Period Mean Shares of Observed Profiles ...... 80 3.8 Signatures and Collection Time by Observed Profiles ...... 84

4.1 Cantonal Approval Rates for Ballots 1 (15 November 1970) and 2 (6 June 1971) ...... 115 4.2 Mediation Framework ...... 119 4.3 Participation Rate in Parliamentary Elections 1951-1991 ....129 4.4 Histograms of Voter Participation and Acceptance Rate ....135 4.5 Histograms of Propensity Scores by Gender ...... 140 4.6 Histograms of Propensity Scores (Mbasic and Cbasic, Vote 1981) 148 4.7 Histograms of Propensity Scores (Mbasic and Cbasic, Vote 1991) 148 4.8 Histograms of Propensity Scores (Mbasic and Cbasic, Vote 1993) 149 4.9 Histograms of Propensity Scores (Mbasic and Cbasic, Votes 1981, 1991) ...... 149

v 4.10 Histograms of Propensity Scores (Mbasic and Cbasic, Votes 1991, 1993) ...... 150 4.11 Histograms of Propensity Scores (Mbasic and Cbasic, Vote 1993) 150 4.12 Histograms of Propensity Scores (Mextended and Cextended, Votes 1981, 1991) ...... 151 4.13 Histograms of Propensity Scores (Mextended and Cextended, Votes 1991, 1993) ...... 151 4.14 Histograms of Propensity Scores (Mextended and Cextended, Vote 1993) ...... 152 4.15 Histograms of Propensity Scores (Mbasic and Cbasic, Votes 1981, 1991, 1993, Ballot Clustering) ...... 152 4.16 Histograms of Propensity Scores (Mbasic and Cbasic, Votes 1981, 1991, Ballot Clustering) ...... 153 4.17 Histograms of Propensity Scores (Mbasic and Cbasic, Votes 1991, 1993, Ballot Clustering) ...... 153 4.18 Histograms of Propensity Scores (Mextended and Cextended, Votes 1981, 1991, Ballot Clustering) ...... 154 4.19 Histograms of Propensity Scores (Mextended and Cextended, Votes 1991, 1993, Ballot Clustering) ...... 154 4.20 Histograms of Propensity Scores (Mbasic and Cbasic, Votes 1981, 1991, 1993, Canton Known) ...... 155 4.21 Histograms of Propensity Scores (Mbasic and Cbasic, Votes 1981, 1991, Canton Known) ...... 155 4.22 Histograms of Propensity Scores (Mbasic and Cbasic, Votes 1991, 1993, Canton Known) ...... 156 4.23 Histograms of Propensity Scores (Mbasic and Cbasic, Vote 1993, CantonKnown)...... 156 4.24 Histograms of Propensity Scores (Mextended and Cextended, Votes 1981, 1991, Canton Known) ...... 157 4.25 Histograms of Propensity Scores (Mextended and Cextended, Votes 1991, 1993, Canton Known) ...... 157 4.26 Histograms of Propensity Scores (Mextended and Cextended, Vote 1993, Canton Known) ...... 158 4.27 Histograms of Propensity Scores (Mbasic and Cbasic, Votes 1981, 1991, 1993, Canton Clustering) ...... 158

vi 4.28 Histograms of Propensity Scores (Mbasic and Cbasic, Votes 1981, 1991, Canton Clustering) ...... 159 4.29 Histograms of Propensity Scores (Mbasic and Cbasic, Votes 1991, 1993, Canton Clustering) ...... 159 4.30 Histograms of Propensity Scores (Mbasic and Cbasic, Vote 1993, Canton Clustering) ...... 160 4.31 Histograms of Propensity Scores (Mextended and Cextended, Votes 1981, 1991, Canton Clustering) ...... 160 4.32 Histograms of Propensity Scores (Mextended and Cextended, Votes 1991, 1993, Canton Clustering) ...... 161 4.33 Histograms of Propensity Scores (Mextended and Cextended, Vote 1993, Canton Clustering) ...... 161

vii List of Tables

2.1 Descriptives ...... 16 2.2 Descriptives of Control Variables ...... 21 2.3 Effect of Initiative Signing on Voter Turnout I ...... 23 2.4 Effect of Initiative Signing on Voter Turnout II ...... 24 2.5 Effect of Initiative Signing on Acceptance I ...... 27 2.6 Effect of Initiative Signing on Acceptance II ...... 28 2.7 Effect of Initiative Signing on Acceptance - Preferences I .... 31 2.8 Effect of Initiative Signing on Acceptance - Preferences II . . . 34 2.9 Effect of Initiative Signing on Acceptance - Postal ...... 37 2.10 Effect of Initiative Signing on Acceptance - Placebo I ...... 38 2.11 Effect of Initiative Signing on Acceptance - Placebo II ..... 39

3.1 Overview of Main Institutional Changes ...... 50 3.2 Descriptives: Initiatives per Period and Outcomes ...... 71 3.3 Descriptives ...... 74 3.4 Overview of Hypotheses and Relevant Periods ...... 79 3.5 Probability of Amending the Status Quo ...... 82 3.6 Winning Policy by Profile ...... 83 3.7 Effect of Over-Collection on Counter Proposals ...... 86 3.8 Signature Collection Costs ...... 88 3.9 Collection Time Constraint and Signature Requirement .... 90 3.10 Amended Voting Rule: Tie-Breaking Question ...... 92 3.11 Overview of Variables and Data Sources ...... 101

4.1 Chronology of Ballots Concerning the Swiss Federal Tax System112 4.2 Descriptives of Post-Ballot Surveys by Gender ...... 134

viii 4.3 Estimates of the Female Acceptance Rate from Ballot Propo- sitions in 1970 and 1971 ...... 137 4.4 Estimates of the Female Acceptance Rate from Ballot Propo- sitions in 1970 and 1971 ...... 138 4.5 Direct and Indirect Effects I ...... 141 4.6 Direct and Indirect Effects II ...... 142 4.7 Direct and Indirect Effects, Ballot Clusters ...... 144 4.8 Propensity Score Estimates with Mbasic and Cbasic ...... 162 4.9 Propensity Score Estimates with Mextended and Cextended ...164 4.10 Direct and Indirect Effects, Observations with Information about Cantons ...... 166 4.11 Direct and Indirect Effects, Canton Clusters ...... 167 4.12 Direct and Indirect Effects for Respondents who Voted .....168

ix Abstract

This doctoral thesis is composed of three papers related to the topic of voting and participating in direct democracies. The first paper investigates whether petition signing campaigns for pop- ular initiatives constitute a partisan campaigning instrument by revealing potentially relevant information to the signer which increases his benefit from voting or reduces his cost. The analysis is based on the complete sample of Swiss federal initiatives between 1978 and 2000. The results suggest that initiatives collecting many signatures yield higher approval rates at ballot. Results are robust to controlling for various preference measures. Petition signing is, however, not significantly related to turnout, and is dominated by initiative-specific characteristics. The second paper studies how direct democratic popular initiatives can break the politicians’ agenda setting monopoly. The initiative process is mod- eled as a sequential game under uncertainty about the initiative’s winning probability: petitioners collect signatures to qualify the initiative, politicians decide about a political compromise - a counter proposal - then petitioners have the option to withdraw the initiative before the popular vote. We test model predictions based on the data set of all Swiss constitutional initiatives at federal level between 1891 and 2010. Results support model predictions by and large: we find that reform is most likely after counter proposals, and collecting many signatures increases the probability of compromise. In the third paper I challenge the notion that women prefer larger govern- ments than men, which is why extending the franchise to women has led to an increase in government spending in many industrialized countries. I analyze the voting outcomes of two similar Swiss referendum votes concerning the federal government’s authorization to levy taxes. The first ballot took place shortly before the extension of suffrage to women in February 1971, and the other one directly thereafter. Surprisingly, I find that approval for government spending is higher among the male population. By conducting a mediation analysis based on post-ballot surveys after comparable votes in 1981, 1991, and 1993, I find support for a negative gender gap and show that the intrinsic direct effect of being female proves to be the driving force behind the results.

x Zusammenfassung

Die vorliegende Doktorarbeit besteht aus drei wissenschaftlichen Artikeln zum Thema Abstimmungen und Wahlbeteiligung in direkten Demokratien. Im ersten Artikel wird untersucht, inwiefern das Unterschreiben einer Volksinitiative ein Wahlkampfmittel darstellt. Die Fragestellung wird anhand der Abstimmungsergebnisse sämtlicher Schweizer Volksinitiativen zwischen 1978 und 2000 untersucht. Die Ergebnisse zeigen einen signifikant positiven Zusammenhang zwischen der Anzahl Unterschriften einer Volksinitiative so- wie dem Anteil Ja-Stimmen. Allerdings kann kein Zusammenhang mit der Wahlbeteiligung festgestellt werden. Der zweite Artikel beschäftigt sich mit der Frage, ob durch Volksinitiativen eher Reformen herbeigeführt werden können. Wir modellieren den Initiativ- prozess als ein sequentielles Spiel, in dem zunächst Unterschriften gesammelt werden, Politiker einen Gegenvorschlag machen können und Initianten vor der Abstimmung über einen Rückzug der Initiative entscheiden. Wir testen das Modell basierend auf sämtlichen Schweizer Volksinitiativen zwischen 1891 und 2010. Reformen treten am häufigsten nach Gegenvorschlägen auf. Es besteht ausserdem ein positiver Einfluss der Anzahl gesammelter Unterschriften auf die Wahrscheinlichkeit, einen Gegenvorschlag zu erhalten. Im dritten Artikel hinterfrage ich die Auffassung, dass Frauen höhere Staatsausgaben bevorzugen als Männer, was erklären würde, warum Staats- ausgaben nach Einführung des Frauenstimmrechts in verschiedenen Ländern im Durchschnitt deutlich gestiegen sind. Der Effekt des Geschlechts auf Präfe- renzen für Staatsausgaben wird anhand von Abstimmungsergebnissen zweier ähnlicher Schweizer Referenden bezüglich der Kompetenz der Regierung, Bun- dessteuern erheben zu dürfen, geschätzt. Die erste Abstimmung fand kurz vor der Einführung des Frauenstimmrechts 1971 statt und die zweite kurz danach. Erstaunlicherweise befürworten Männer die Vorlagen zur Steuererhe- bung häufiger als Frauen. Eine Mediationsanalyse basierend auf individuellen Abstimmungs-Nachbefragungsdaten nach verwandten Abstimmungen aus den Jahren 1981, 1991 und 1993 bestätigt dieses Ergebnis. Darüber hinaus zeigt sich, dass der direkte Effekt des Geschlechts eine grössere Rolle für die Erklä- rung der Ergebnisse spielt als indirekte Effekte.

xi Chapter 1

Introduction

When voters go to the ballot to elect their political representatives, their choice is influenced by the candidates’ position on policy issues regarded im- portant by the individual voter. As a consequence, some elections evolve solely around particularly salient policies like taxation, minimum wages, or unemployment policies. But they disregard the wealth of other policy areas in the hands of politicians. Voters only get to decide about a limited range of policies. In contrast, countries or states providing the institution of direct democracy reserve their voters the right to decide on single policy issues in addition to electing politicians. This doctoral thesis consists of three papers related to topics in direct democracy, and in particular to voting and participation in a direct democratic setting. They all have in common that they either analyze direct democratic processes, or use direct democratic votes like referendums or initiatives to in- vestigate questions of more general interest. Chapters 2 and 3 are related by the over-arching topic of the signature collection process for popular initia- tives. The last paper in chapter 4 has the introduction of gender preference gaps and female voting rights at its core. The first two papers of my thesis are both concerned with the qualification stage of popular initiatives, however, from two very different perspectives. To successfully qualify an initiative for ballot, petitioners are required to collect a legally stated minimum of signatures. For example, in Switzerland this threshold for constitutional initiatives is presently set at 100,000 signatures,

1 Figure 1.1: Signatures for Swiss Popular Initiatives at Federal Level 400 300 200 100 Valid Signatures in 1,000 Valid 0 1900 1920 1940 1960 1980 2000 Year initiative submitted

Note: The graph shows the number of valid signatures collected in thousands for the Swiss popular initiative at federal level, sorted according to the year of submission. Until 1978 the legally required number of signatures to qualify an initiative was 50,000, and 100,000 thereafter. Data are from the Swiss Federal Chancellery (2013). or in California (U.S.) at 8% of the voting population at the last guberna- torial elections. Since signature collection constitutes a costly activity for petitioners, it is surprising to find that on average considerably more signa- tures than required are collected as can be seen for Switzerland in Figure 1.1.1 While a part might serve as security margin for invalid signatures or reflect coordination failures, the phenomenon “over-collection” is too large to solely reflect these issues. My first two papers offer potential explanations for this phenomenon. While my research is based on Swiss data, initiatives are an important direct democracy instrument commonly used in many U.S. states and also in countries throughout the world as diverse as Hungary, Uruguay, or Georgia (Center for Research on Direct Democracy, 2013). My research is therefore highly relevant to settings beyond the Swiss one. The first paper in chapter 2 investigates whether petition signing cam- paigns for popular initiatives constitute a partisan campaigning instrument by revealing potentially relevant information to the signer, which increases the

1 The average for all qualified initiatives in Switzerland since 1978 is around 127% of the signature requirement, and can be as high as 390%. For a sample of Californian initiatives, Boehmke and Alvarez (2004) find that the collected number of signatures varies between 143% and 187% of the legal requirement.

2 benefit from voting or reduces its cost. The analysis is based on the complete sample of Swiss federal initiatives between 1978 and 2000 with aggregate vot- ing data at cantonal level. The results suggest that initiatives with many sig- natures yield higher approval rates at the polls. Petition signing is, however, not significantly related to turnout, and is dominated by initiative-specific characteristics. Three approaches are pursued to control for voter preferences which potentially could drive both signatures and acceptance rates. First, the fraction of the population voting for parties supporting the initiatives, reflecting political elite mobilization, is controlled for. Next, states particu- larly affected by the initiative are identified. Last, voter preferences measured by acceptance rates from thematically closely related referendum ballots are used. The results prove to be highly robust to the inclusion of these prefer- ence controls. Variation in the introduction of postal voting in Swiss states is exploited to account for the signer’s level of information: the typical signature collection location near places of election becomes less attractive and regular, well-informed voters are less likely to sign initiatives. The link between peti- tion signing and acceptance weakens but remains significantly positive after the introduction of postal voting. This research relates to turnout and voting literature in general, and to campaigning and voter motivation more specif- ically. It extends a small stream of literature analyzing signature collection for initiatives. Chapter 3 of my thesis is co-authored work with Christian Marti and Monika Bütler. We analyze how direct democratic popular initiatives can break the politicians’ agenda setting monopoly and allow for more reforms. The initiative process is modeled as a sequential game under uncertainty: petitioners collect signatures to qualify the initiative and elicit information about the initiative’s winning probability. Politicians decide about a political compromise - a counter proposal - then petitioners have the option to with- draw the initiative before the vote. In our model counter proposals are key to amending the status quo. We explore the likelihood that the status quo is changed by an initiative based on the data set of all Swiss constitutional initiatives at federal level between 1891 and 2010. We test our model by using major institutional changes to the initiative process and compare empirical outcomes to model predictions: lowering the signature collection costs, rising the signature requirement, restricting signature collection time, and changing

3 the voting rules. Counter proposals are quite effective in amending the status quo. We find that reforms are most likely once a far-reaching counter proposal is issued such that the initiative is withdrawn. Moreover, we find a significant effect of collecting more signatures than required to qualify the initiative on the probability of achieving a compromise. In connection with the second paper, jointly with Christian Marti we have compiled an extensive documen- tation of all qualified Swiss initiatives between 1891 and 2010 that we have used in the empirical part of Chapter 3. By making this information available to other researchers, we hope to contribute to inquiries about initiatives, and facilitate future related research. In my third and last paper, I make use of the introduction of female voting rights in Switzerland at federal level level, which led to a doubling in the electorate, to study gender preference gaps. I challenge the notion that women prefer larger governments than men, which is why extending the franchise to women has led to an increase in government spending in many industrialized countries. I estimate the average treatment effect of being female on support for government spending, by analyzing the voting outcomes of two similar Swiss referendum votes concerning the federal government’s authorization to levy income, capital and turnover taxes. The first ballot took place shortly before the extension of suffrage to women in February 1971, and the other one directly thereafter. Based on municipal voting data, I relate the increase in the electorate to the difference in acceptance rates for the two propositions. Surprisingly, I find that approval for government spending is higher among the male population. By conducting a mediation analysis based on post-ballot surveys after comparable votes in 1981, 1991, and 1993, I disentangle the direct gender effect on government spending preferences from the indirect gender effect which runs through important socioeconomic mediators like employment status or education. The intrinsic direct effect of being female proves to be the driving force behind the results while mediators turn out to play a weaker role. My results suggest rethinking the notion that female suffrage caused public spending to increase. Direct democracies are of particular interest to researchers because of their special decision-making process. Not only are policies created by politicians in parliament subject to mandatory or optional public votes. But also the power to propose policies is not concentrated in the politicians’ hands anymore. This

4 has triggered a vast amount of research analyzing the consequences of direct democracy on various outcomes of the political process. Some prominent examples of this research explore the effects of direct democracy on size of government (for example, Matsusaka, 2005; Feld & Kirchgässner, 2001a; Feld & Matsusaka, 2003; Funk & Gathmann, 2011), public debt (Feld & Kirchgäss- ner, 2001b), economic growth (Feld & Savioz, 1997; Freitag & Vatter, 2000), but also more general consequences of direct democratic institutions such as tax morale (Torgler, 2005), or life satisfaction (Frey et al., 2001; Dorn et al., 2007). In this thesis, I contribute to the literature at three different levels. The first paper uses direct democratic votes to explore a phenomenon innate to the initiative process - namely petition signing - on turnout and acceptance behavior. Both have been researched in various environments and are thus of broader interest. The second paper has the functioning of a direct demo- cratic process at its core. Hence, it contributes specifically to understanding mechanisms present in the initiative process, and its effect on policy outcomes. Third, direct democratic votes have the advantage over elections of politicians that they concern a particular policy and not a bundle of policies. They can hence be used to elicit voter preferences, which I do in the last paper. Though this research is not concerned with a direct democratic process per se, it can be used to analyze other interesting questions for which the knowledge of voter preferences is necessary. The results of my thesis demonstrate that not all results in the literature should be taken at face value. This becomes particularly evident in my third paper in which I find results contradicting previous research. It is important to retain scientific curiosity and scrutinize other research. Further, as shown in my second paper, economic models should encompass all relevant elements of the research object in question. In more detail, in the initiative process the equilibrium outcome of the the signature collection phase should be taken into account to gain additional understanding of the equilibrium in the complete initiative game. I expect that this thesis helps advance the research in direct democracy, and lays the ground for future research.

5 Chapter 2

Campaigning in Direct Democracies: Initiative Petition Signing, Voter Turnout, and Acceptance

2.1 Introduction

The main purpose of direct democracy is to provide citizens with political powers beyond the mere election of political representatives. The availability of initiatives and referendums serves as a mean to correct undesirable policy outcomes, or as a threat to politicians already in the early legislative process (Feld & Matsusaka, 2003). A second, less obvious purpose of direct democ- racy is to educate voters to become active citizens (Tolbert & Smith, 2005). The possibility of shaping and influencing policies as well as deciding about single issues awakes the interest of voters. Active participation then leads to better informed and interested voters who ideally become regular voters and responsible citizens.1 A frequently used direct democracy instrument is the voter initiative which

1 Tolbert and Smith (2005) provide an extensive overview of the development of direct democracy and its functions in U.S. history.

6 allows citizens or political minorities to put issues on the political agenda. To qualify an initiative for ballot, the initiating group needs to collect a legally specified amount of signatures to prove that the issue enjoys sufficient sup- port in the population. This qualifying stage of initiatives is at the core of my research. I investigate whether initiative petition signing increases the probability to subsequently turn out on election day and accept the initia- tive at ballot. I hypothesize that by signing an initiative petition signers are exposed to campaigning and receive relevant information about the initiative topic, which increases their awareness of the issue and enhances their benefit from voting. Signers should be more likely to accept the initiative because of positive motivation, or a feeling of moral obligation. These hypotheses are tested with aggregate data from all Swiss popular initiatives at federal level qualified and voted between 1978 and 2000. All data on collected signatures and voting results is at cantonal level,2 which allows to use regional variation in the data. The results show a positive relationship between signatures collected and the initiative’s acceptance rate at ballot. Signature collection displays in- creasing returns to scale over a relevant range of observations. Possible ex- planations are spillover and network effects from talking to family and friends about the initiative. Results remain similar after the inclusion of canton and initiative fixed effects, as well as relevant political and socioeconomic controls. Signatures per capita and voter turnout are also positively related. However, the effect becomes insignificant once initiative fixed effects are controlled for together with canton fixed effects or control variables. Possible explanations are relatively low roll-off rates for federal ballots such that turnout with mul- tiple votes on the same day is almost identical for all propositions (Schmid, 2013). Regarding the significant relation between acceptance rates and signa- tures, the main empirical challenge to campaigning research is the question of causality (Gerber & Green, 2000): if voter preferences drove both the num- ber of signatures and acceptance rates in a canton, a significantly positive regression coefficient would be expected, but not reflect a causal relation. To account for this issue, I develop three controls to proxy voter preferences. I firstly account for elite mobilization by political parties (Kriesi, 1995, 2006).

2 Switzerland has a strong federal structure and is divided into 26 states - the cantons.

7 Voters identify with their preferred parties and look to them for voting cues. Therefore, I control for the fraction of the population that has elected parties issuing a positive voting recommendation. Next, I identify cantons which were particularly affected by the initiative, and thus have a reason to have either positive or negative preferences for the initiative. Last, I use voting results from thematically closely related referendum ballots to proxy voter prefer- ences. The baseline results regarding acceptance prove extremely robust to the inclusion of these preference variables. Moreover, I exploit cantonal variation in the introduction of postal vot- ing to account for the reduced possibility of collecting signatures from regular votes near places of elections. Since regular voters are usually better informed, the introduction of postal voting increases the probability that random citi- zens sign the initiative. Further, the costs of collection increase since it gets more difficult for initiating groups to collect signatures. In line with expecta- tions, the effect of petition signing on acceptance decreases once postal voting has been introduced but remains highly significant. This paper generally relates to investigations in turnout and voting, and gives further insight on why people vote (for overviews see for example Aldrich (1993), Coate and Conlin (2004), Feddersen (2004), Matsusaka and Palda (1999), or Merlo (2006)). It is particularly close to models of voting including the level of the voters’ information about the voting alternatives (e.g., Fed- dersen & Pesendorfer, 1996, 1999; Matsusaka, 1995; Degan & Merlo, 2011). Also, initiative petition signing can be seen as a particular form of face-to- face campaigning to motivate signers to support the initiative. Regarding the link between petition signing and acceptance, this paper further relates to literature about cognitive dissonance (Festinger, 1957): should signers reject the initiative at ballot, they would feel discomfort from two actions obviously contradicting each other. By signing the petition, signers feel morally obliged to cast a positive ballot. My research extends a small literature concerning the link between petition signing and voter turnout by also analyzing accep- tance behavior (Boehmke & Alvarez, 2012; Parry, Smith & Henry, 2012). In addition, my results are more generalizable thanks to a larger sample size, and explicitly addressing the issue of causality. The main advantage of the Swiss setting over data from other countries is the availability of the exact number of signatures for all initiatives over a

8 long time period, which allows generalizable results. Collected signatures are always fully counted. Also by looking exclusively at federal initiatives, it is guaranteed that all cantons are exposed to the same institutional framework, have the same regulation regarding the initiative process, and are thus com- parable. In the U.S. such comparisons between states are virtually impossible since regulation varies from state to state (such as different signature require- ments). By using Swiss data I also overcome the registration problem apparent in the U.S.: while in many states voters need to register before they can vote and possibly even face some registration time restrictions, Swiss voters are automatically registered. The advantage of this regulation is that registration does not require additional effort and does not pose a hurdle to voter par- ticipation. At the same time, the initiative process in Switzerland resembles processes in other direct democracies which makes my findings comparable to other settings. The paper begins with the description of the theoretical foundation and the hypotheses in Section 2.2. In Section 2.3, I give background information on the Swiss initiative process, data, and descriptives. The estimation strategy is described in Section 2.4. In Section 2.5, the results regarding turnout and acceptance are reported. The last section gives a brief discussion of the results and concludes.

2.2 Theoretical Justification and Hypotheses

The process of signature collection to qualify an initiative for public vote can be seen as one - albeit unusual - form of face-to-face campaigning (Parry, Smith & Henry, 2012). Therefore this paper directly relates to research on voter mobilization. This literature finds that turnout is significantly and pos- itively affected by campaigning efforts. Early advances attribute some effect to mobilization through campaign spending (Copeland, 1983; Patterson & Caldeira, 1983; Caldeira, Patterson & Markko, 1985). Subsequent investiga- tions based on field experiments note that the likelihood of voting is increased by face-to-face contact (Gerber & Green, 2000; Green, Gerber & Nickerson, 2003; Niven, 2004), telephone calls from dedicated callers (Nickerson, 2006), and sometimes non-personal messages (Dale & Strauss, 2009). Smith (2001)

9 notes that campaigning can be interpreted as increasing civic duty from vot- ing by creating awareness for the ballot (Riker & Ordeshook, 1968). Gener- ally, two forms of campaigning efforts can be distinguished: non-partisan and partisan efforts (Parry et al., 2008). In the first form, voters are generally motivated to turn out. In contrast, partisan campaigning tries to motivate voters to turn out for a particular candidate. My research contributes to the latter type since citizens are expected to cast votes in favor of the initiative. Connected to the motivation literature mentioned above, there is a small empirical literature emerging which has the motivational effect of initiative petition signing on turnout at its core, and is thus closest to this paper.3 Parry, Smith and Henry (2012) analyze individual voting data matched with signature records from three initiative ballots in Florida as well as Arkansas and find a significantly positive effect of petition signing on turnout only in one of their three models. In particular, campaigning effects are stronger for irregular voters. In a similar vain, Boehmke and Alvarez (2012) conduct their analysis with county-level data from eight Californian initiatives. The results show a positive and significant effect of petition signing on turnout. In ad- dition, they find a positive relationship between petition signing and voter registration, and a negative one between signing and roll-off rates. However, their results are based on a small sample and cannot be generalized to other settings. Similarly, Parry, Smith and Henry (2012) find a positive effect for only one of the three initiatives they analyze. Next, Boehmke and Alvarez (2012) do not control for voting history in the counties. Consequently, their analysis may suffer from endogeneity problems since voting history is an im- portant driver of turnout. Both papers do not provide evidence about the effect of petition signing on acceptance probabilities. For Parry Smith and Henry (2012) this is not feasible since voting records are not publicly available. I therefore extend previous research by addressing this question.

3 Early advances in the analysis of initiative petition signing are scarce, mostly due to dif- ficult data collection work. By drawing two random samples, one from registered voters and one from registered voters who signed a particular initiative, Neiman and Gottdiener (1982) observe that signers show more political interest and knowledge about the initia- tive than non-signers. However, it is beyond the reach of their study to show a causal relationship between signing an initiative and gaining more political knowledge through this channel. By also working with two samples of signers and the general population, Pierce and Lovrich (1982) surprisingly find that signers significantly underreport signing a petition when questioned about it several months after the ballot. They conclude that micro data about petition signing from surveys might be severely biased.

10 2.2.1 Turnout

Signature collection can be interpreted as a campaigning device which pro- vides prospective voters with information or creates awareness of the initiative issue. Signing an initiative petition can thus either activate the voter’s pos- itive predisposition towards the initiative issue, or add information to the undecided voter. Voting models predict awareness, information or factual knowledge to positively impact the probability of turnout. Downs (1957) rec- ognized that information plays a crucial role in the participation and voting decision process of citizens. Even though information is not explicitly included in the standard voting models, it can be interpreted as increasing the voter’s benefit from participating (Smith, 2001). Voters understand the issue better and can evaluate the consequences of a vote more precisely than before. In- formation potentially decreases the costs of voting (Matsusaka, 1995). In a related paper, Degan and Merlo (2011) include information in a model with multiple elections, and show that it does well in explaining voting behavior in the U.S. presidential and congress election in the year 2000. In a series of papers Feddersen and Pesendorfer develop voting models explaining abstention and roll-off without voting costs and including the role of information (Feddersen & Pesendorfer, 1996, 1999). In their 1996 model, agents are either partisans for one of the candidates, or independents prefer- ring one of the candidates depending on the state of the world. Next, voters are either informed or uninformed about the state of the world. In equilib- rium, partisans support their own candidates. Informed independent agents, who by definition know the state of the world, vote for the “correct” candi- date. However, uninformed independent agents have equilibriums in which they are strictly better off abstaining. The intuition is that if they knew the state of the world, they would vote identically as the informed indepen- dent agents. Since they lack this information, they may be better off letting informed independent agents vote for the “right” candidate. The rationale of this model can easily be adapted to the setting in this paper. Voters usually have some innate predisposition when it comes to mak- ing judgements about political issues (Copeland, 1983): they either favor the issue, oppose it, or are indifferent. Partisan agents vote sincerely, and if asked to sign the initiative petition, supporters do so while opponents decline. Inde- pendent agents, however, need information to evaluate the issue at question.

11 Being approached to sign an initiative petition can hence be interpreted as receiving an informative message. When asked by a signature collector, po- tential signers learn the initiative title and text. Though this information is not complete, it should be enough to help agents understand the main fea- tures of the initiative, and create awareness of the issue. If the information is favorable, they sign the initiative. Consequently, their best strategy is also to vote in favor of the initiative. Thus non-partisan agents who sign the ini- tiative should participate in the election and vote in favor of the initiative. Partisan agents should participate and vote sincerely by revealing their true preferences, regardless of signature collection. In the light of this reasoning, the act of signing can be interpreted as an informative action. Based on this understanding, the first hypothesis to be tested is the following:

Hypothesis 1 Signing an initiative petition increases the probability of turn- ing out in the subsequent initiative ballot, everything else held constant.

2.2.2 Acceptance

As described above, voting models with information like in Feddersen and Pesendorfer (1996, 1999), predict that partisan as well as favorably informed independent voters should turn out more likely. In expectation, citizens with a positive predisposition about the initiative and previously indifferent signers should be more likely to accept it at ballot. Alternative theoretical considerations come from psychology. The the- ory of cognitive dissonance predicts that actions might be initiators to form preferences (Festinger, 1957; Mills, 1958). Conducting an action might be the reason to form believes that dictate to act accordingly to the first action also in the future. This contrasts with standard economic models in which preferences usually lead to actions. Related research based on the theory of cognitive dissonance in the voting context is by Mullainathan and Washing- ton (2009). They find evidence that voting for a certain candidate leads to a more favorable opinion about his policies after the election. Beasley and Joslyn (2001) find supporting evidence of a widening evaluating distance be- tween candidates after committing to one of them by voting. In this paper, the act of signing an initiative petition can be interpreted

12 as such an initiating action. Driven by their psychological need for consistent behavior, signers are urged to accept the initiative at ballot. Rejecting the initiative, in contrast, would lead to a feeling of discomfort caused by the clashing actions of signing (supporting) and voting against (rejecting) the initiative. The theory of cognitive dissonance thus predicts that of those who participate in the election, signers should be more likely to accept the initiative. The second hypothesis to be tested in this paper is:

Hypothesis 2 Signing an initiative petition increases the probability of vot- ing in favor of the initiative at ballot, everything else held constant.

2.3 Institutional Background, Data, and Descrip- tives

2.3.1 Institutional Background

Switzerland has particularly strong direct democratic institutions. At federal level, its instruments include the mandatory referendum, optional referendum, and the constitutional initiative. Mandatory referendums take place after the parliament proposed a change to the Swiss constitution such that voters need to agree to the change before it comes into force. For all other federal legis- lation not amending the constitution such as laws, collecting signatures from at least 50,000 citizens leads to an optional referendum. If the referendum is rejected at ballot, the legislation does not come into force. In addition, the 26 Swiss cantons which have many political liberties have their own direct demo- cratic institutions. In this paper, I concentrate exclusively on the initiative at federal level. The initiative at federal level in Switzerland was first established in 1891 and is concerned solely with constitutional changes (Linder, 2007). At the qualifying stage, the initiating petitioners need to collect at least the legally required 100,000 signatures within 18 months. No additional requirement re- garding the signature distribution in Swiss cantons exists. Upon successful completion of the signature collection, government and the two chambers of parliament decide whether to issue a counter proposal or not. The counter

13 proposal is defined as an alternative or compromise to the initiative which would also amend the constitution.4 In case of a counter proposal, it is voted simultaneously with the initiative. If the petitioners withdraw the initiative, only the counter proposal is voted upon. If the initiative or the counter pro- posal are voted upon individually against the status quo, the absolute majority of votes decides whether it comes into force or not. If both, the initiative and the counter proposal, are voted simultaneously, until 1987 voters could accept either one of the alternatives, or reject both. In 1987 this regulation changed and a tie-breaking question was introduced. Voters vote both the initiative and the counter proposal versus the status quo, and choose in the tie-breaking question which of the two they like best. The tie-breaking question is decisive should both of the proposals receive more than 50 percent of the votes.

2.3.2 Data and Descriptives

I use the dataset of all Swiss federal initiatives that have started data collec- tion in 1978 or later, and have been voted no later than 2000. All data are at cantonal level. The reasons to restrict the sample to this time period are threefold. First, the initiative threshold has doubled to 100,000 signatures in 1978, which makes initiatives before and after that date more difficult to compare. Second, in 1978 part of the canton Bern separated from the old canton to create the canton Jura such that the number of cantons increased. Third, there are no comparable socioeconomic controls at cantonal level avail- able outside the time span 1970 to 2000.5 Average income is only available since 1974. There is a total of 68 initiatives in the observation period. In the sampling period 1978-2000, initiative and counter proposal were voted simultaneously three times. Such infrequent cases are likely to stir additional attention and to be prone to strategic voting such that inconsistent voting profiles may occur

4 More precisely, this is called a direct counter proposal. Another option is to issue an indirect counter proposal which is usually a law and thus not at the constitutional level. However, in most cases no ballot takes place after indirect counter proposals which is why none of them are in this sample. In what follows, the term counter proposal is used synonymously with the direct counter proposal at constitutional level which has to be voted by citizens. 5 Control variables are taken from Swiss censuses which are conducted every ten year. Due to a methodological change, the 2010 census is not comparable to the prior ones.

14 Figure 2.1: Histogram of Turnout, Signatures Per Capita, and Acceptance Rate 6 6 5 5 4 4 3 3 Density Density 2 2 1 1 0 0 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 Turnout Yes votes / voters 6 30 5 4 20 3 Density Density 2 10 1 0 0 0 .05 .1 .15 .2 .25 0 .2 .4 .6 .8 1 Valid signatures per capita Yes votes / eligible citizens

Note: Histogram based on cantonal initiative data. The variables are turnout, valid signatures per capita, and acceptance rate measured as either yes votes/ voters or yes votes/eligible citizens.

(Bochsler, 2010). They can have an additional effect on turnout and accep- tance, and voting rules regarding these cases have changed during the sample period. For these reasons, all three initiatives are excluded from the sample. Of the remaining 65 initiatives, 58 ballots have been about the initiative alone, and in the remaining 7 cases the initiating committee withdrew the initiative after the parliament decided to formulate a counter proposal such that only the counter proposal was voted. There exist differences between ballots with either initiatives or counter proposals: while 5 of 7 counter proposals were accepted, only 4 of 58 initiatives received a majority of votes. The reason for this difference lies in the fact that though counter proposals take up the most important issues of the initiative, they compromise on some other points. They are accessible to a larger part of the population than initiatives which usually address minority issues. Switzerland comprises 26 cantons which vary in size and other characteris- tics. The largest canton Zurich was populated by nearly 790,000 eligible vot- ers in the year 2000. In contrast, the canton Appenzell Innerrhoden had only about 10,000 eligible citizens. Population-weighted average turnout for the

15 initiative ballots was 42.9% and varied between 13.8% and 82.4%. Differences in cantonal mean turnouts are large as well: the highest average participation rate prevails in the cantons Schaffhausen6 (68.14%) and Solothurn (50.85%), and the lowest in Vaud (35.66%). The upper left panel of Figure 2.1 shows the distribution of population size-weighted cantonal turnout in the sample. Average population-weighted acceptance defined as the number of yes votes divided by the total number of votes amounts to 38.21%, and varies between 4.98% and 93.12%. Similarly to the turnout data, notable differences between cantons exist: Basel City has the highest average acceptance rate with 44.90%, while the lowest average can be found in Appenzell Innerrhoden with 29.94%. The right panels of Figure 2.1 depict the distribution of the acceptance rate as commonly defined (yes votes/voters) in the upper panel, and defined as yes votes/eligible citizens in the lower right panel. The distribution of voter acceptance is right skewed, which reflects that initiatives usually get rejected at ballot. Between 101,337 and 390,273 valid signatures with a mean of 132,052 have been collected for the 65 initiatives in the sample. On average 2.97% of the Swiss eligible population have signed the initiatives with a standard deviation of 2.19%. Variation between cantons is large. Basel City collects most signatures on average (5.69%) and Appenzell Innerrhoden (1.32%) has the lowest mean. In all estimations, I use the number of valid instead of collected signatures since this is the publicly reported number. Also, the number of invalid signatures is small in the aggregate (mean invalid signatures

Table 2.1: Descriptives Mean Std. Dev. Min Max Turnout (voters/eligible citizens) 0.4287 0.0959 0.1384 0.8244 Acceptance (yes votes/voters) 0.3821 0.1661 0.0498 0.9312 Acceptance (yes votes/eligible citizens) 0.1599 0.0732 0.0178 0.5774 Signatures/eligible citizens 0.0297 0.0217 0.0001 0.2476 Note: 1,690 observations. Summary statistics are weighted according to Swiss eligible population size in the cantons.

6 Schaffhausen poses a special case since it is the only canton with enforced compulsory voting during the observation period. Abstainers have to pay a symbolic fine so that Schaffhausen traditionally has a more active electorate on average (Federal Announce- ment, 2003). Since I include canton fixed effects in the regressions, this should not constitute a problem for the estimates.

16 of collected signatures are 2.50% with a standard deviation of 3.49%) such that the difference between valid and collected signatures in small. The distribution of all cantonal values of signatures per capita is depicted in the lower left panel of Figure 2.1. Descriptives of the main variables are reported in Table 2.1. Data on signatures for initiatives qualified between 1978 and 1998 are hand-collected from the homepage of the Swiss Federal Archive. Since 1999, the signature data are available on the homepage of the Swiss Federal Chancellery. A detailed description of the data, its sources, and how it can be accessed is available in the appendix.

2.4 Estimation Strategy

2.4.1 Turnout

Denote the eligible population in canton c at the point in time when the initia- tive i is voted by eligibleci. The number of collected signatures for initiative i in canton c is signaturesci, and participating voters are denoted by votersci. The main independent variable is defined as the number of valid signatures collected divided by the cantonal eligible population (signatures p.c.ci = signaturesci/eligibleci), which is similar to the variable signatures per capita used by Boehmke and Alvarez (2012). By definition, this variable is con- strained to values between 0 and 1: it takes the value 0 if no one signs the initiative, and the value 1 if the complete eligible population of a canton would sign it. Turnout is defined as the share of eligible citizens who participates, turnoutci = votersci/eligibleci. ci denotes the error term. Equation (2.1) shows the baseline linear estimation equation.

turnoutci = β0 + β1signatures p.c.ci + ci (2.1)

I estimate a weighted least squares model with proportional weights accord- ing to the eligible cantonal population. Weights are necessary for proportional data because they correspond to many more individual observations in large cantons than in small cantons. The coefficient of interest is β1 which is ex- pected to be positive. In a second specification, I also add the squared value

17 of signatures per capita to account for nonlinear effects. In line with campaign spending literature, I expect a negative coefficient for the quadratic term. A main concern is that the analysis potentially suffers from reverse causal- ity. If regular voters were also more likely to sign initiatives, a positive corre- lation between these two variables would be the consequence. But causality could not be established. Generally, voter mobilization has a stronger effect on non-habitual voters because habitual voters participate in an election re- gardless of whether they have been contacted or not (Huckfeldt & Sprague, 1992). Also Parry et al. (2008) find voting history to be a good predictor of turnout. This problem has been widely addressed in the campaigning litera- ture. A remedy is to either conduct field experiments (Gerber & Green, 2000), or to control for the voting history of those being contacted (Parry, Smith & Henry, 2011). I account for cantonal voting history by controlling for turnout in the election for national parliament preceding the initiative ballot. This is a good approximation of voting history for several reasons. First, it is a parliamentary election and thus signature collection cannot play a motivating role for turning out. Second, on election days the parliamentary election is the only federal election taking place. This means that voters deciding to par- ticipate do so because they want to elect their political representatives. Even though there might be cantonal votes on the same day, federal parliamentary elections are likely to be more important elections. The control for voting history is based on voting data from Swiss parliamentary elections starting with the year 1979 and changing every four years. This voting information is available from the Swiss Statistical Office.

2.4.2 Acceptance

The second hypothesis examined in this paper is that signing an initiative pe- tition increases the probability of subsequently accepting it at ballot. Yesci stands for the number of yes votes initiative i receives at ballot in canton c. The standard way to define acceptance would be to divide yes votes by the total number of votes. Instead, I define the acceptance rate as the number of valid yes votes divided by the cantonal eligible population (acceptanceci = yesci/eligibleci). Acceptance thus denotes the part of the eligible population voting in favor of the initiative. Then both the dependent and independent

18 variable have a common denominator, which facilitates the interpretation of the regression results.7 The baseline estimation equation is stated in (2.2). Analogously to the turnout estimation, I conduct a second estimation includ- ing the squared value of signatures per capita. Again, weighted least squares are used for the estimation.

acceptanceci = β0 + β1signatures p.c.ci + ci (2.2)

Similarly to the turnout analysis, reverse causality is an important issue to address. If underlying preferences for a particular initiative are high, a high number of signatures and a high acceptance rate can be expected. A positive coefficient β1 then only reflects the underlying preferences, but not a cam- paigning effect running from initiative petition signing to voting behavior. An ideal control variable would be a cantonal preference measure for each initiative topic to make sure that variation explained by signatures is not predominantly due to underlying preferences. The non-experimental setup does not allow to estimate a causal effect. But I propose several extensions of the baseline specification in Section 2.5 with the goal to control for voter preferences.

2.4.3 Fixed Effects and Controls

As a first extension of the baseline model I include canton and initiative fixed effects.8 Canton fixed effects control for time-invariant unobserved differences between the cantons. Prominent examples for the Swiss cantons are political institutions like strong direct democratic elements, or cultural differences be- tween the mainly German-, French-, or Italian-speaking cantons (e.g., Funk, 2010; Lüchinger, Rosinger & Stutzer, 2007). Initiative fixed effects account for unobserved differences between initiatives with identical impact for all

7 For robustness, I repeat all main regressions using acceptance defined in the standard way as the number of yes votes divided by sum of yes and no votes, acceptanceci = yesci/(yesci + noci). The significance of the results remains unaffected. The results are available from the author on request. 8 An alternative estimation strategy to the weighted linear estimation is to use a grouped logit estimator. While it is a good approach to analyze proportion data, it suffers from the deficiency that fixed effects cannot be included due to the non-linear estimator. Since accounting for fixed effects is important due to the panel structure of the data, I only use the linear estimator.

19 cantons. Among such initiative fixed effects are the state-wide salience of an initiative issue, or campaign efforts at federal level. I also add political and socioeconomic controls to the estimation which are standard to use in the turnout and voting literature. The length of the political process is likely to have an effect on turnout and acceptance. The longer the time between the qualification of the initiative and the respective ballot, the weaker should the campaigning effect be:9 the issue might lose its salience and citizens their interest in the topic.10 The availability of initiatives or referendums at ballot is a motivating factor for voters to turn out but the effect diminishes with an increasing number of ballots (Bowler & Donovan, 1998; Magleby, 1984; Tolbert & Smith, 2005). Similarly to the study of Tolbert and Smith, I control for the number of cantonal issues voted at cantonal level on a particular date in addition to the initiative, and its squared term. The expectation is a positive coefficient for the former and a negative for the latter. I use socioeconomic control variables that have been shown to affect turnout and voting outcomes. Income and education are positive drivers of turnout (Wolfinger & Rosenstone, 1980). Education, age, and unemployment also play an important role in an individual’s ability to understand and process infor- mation which makes them indispensable voting controls (Matsusaka, 1995). Regarding voting propositions with financial issues at stake, income and ed- ucation are relevant predictors of acceptance. For the income control, I take the average taxable income at cantonal level. Education is measured by the share of population older than 15 with tertiary education. I also include the share of old population in the analysis which is measured by the percentage of people 65 years old or older (e.g., Parry et al. (2008) find age to be the second most important driver of turnout in their analysis). The general notion is that older people are more likely to vote. Unemployment is often used in turnout analysis as well. For example, Rosenstone (1982) finds lower voting proba- bilities for unemployed or poor people. I therefore expect unemployment to have a negative effect on turnout and measure it by the unemployment rate

9 During this time, the government and the chambers of parliament discuss the initiative, and decide whether to issue a counter proposal. The maximum duration of the process is fixed by law. However, it is possible to extend the process by several months up to years. (Federal Act on the Federal Assembly, 2002) 10This control is only relevant without initiative fixed effects because it does not vary by canton.

20 of the population older that 15 years. Descriptives are in Table 2.2. The average taxable income is from the Federal Tax Administration. The information on the number of cantonal propositions at ballot comes from the Centre for Research on Direct Democracy. The dates of initiative qualification and ballot were taken from the homepage of the Swiss Federal Chancellery. The control for postal voting comes from Funk (2010). All other controls (population 65 years or older, tertiary education, unemployment) were pro- vided by the Swiss Statistical Office and can be found in the Swiss census. Data on average taxable income are biannual, and census data are compiled every ten years. The relevant censuses are 1970, 1980, 1990, and 2000. To receive yearly data, I linearly interpolate the data for the missing years. Similar to other research using Swiss state data, I rerun all regressions including a canton-specific time trend which is either specified in a linear or a quadratic form (e.g., Hodler, Lüchinger & Stutzer, 2014).

Table 2.2: Descriptives of Control Variables Mean Std. Dev. Min Max Turnout at last parliamentary election 0.4517 0.0705 0.1735 0.7370 Number of cantonal ballots on same day 4.33 2.53 0 21 Days initiative qualification to ballot 1556.0 471.7 370 3184 Counter proposal 0.1069 0.3091 0 1 Year with federal parliamentary election 0.0922 0.2895 0 1 % of old (older than 64) 0.1473 0.0174 0.1055 0.2103 % older than 15 with tertiary education 0.1315 0.0361 0.0444 0.2485 % older than 15 unemployed 0.0169 0.0076 0.0051 0.0407 Average taxable income in CHF 10,000 5.137 1.044 2.877 8.313 Note: 1,690 observations. Summary statistics are weighted according to Swiss eligible population size in the cantons.

2.5 Baseline Results

In all regressions weighted least squares are used for the estimations with weights proportional to the cantonal population size. For robustness, I rerun the regressions unweighted. The results remain qualitatively similar.

21 2.5.1 Turnout

I regress turnout on signatures per capita. For the baseline estimation in the first two columns of Table 2.3, I only use the measure of signatures and no control variables. In specifications (3) to (8), I first add initiative and canton fixed effects one at a time and both at the same time afterwards. The linear effect of initiative signing on voter turnout is positive in all specifications, and the quadratic term is negative as expected. The linear coefficient suggests that collecting signatures from an additional percentage point of the eligible population increases turnout by 0.77 percentage points. The effect is slightly reduced when adding initiative fixed effects, and roughly halved when canton fixed effects are included. At the same time, initiative fixed effects have a high explanatory power for the data since they increase the adjusted R2 by more than 0.5. At first, the effect is significant. However, the coefficient of signatures per capita becomes insignificant as soon as initiative and canton fixed effects are both accounted for. I repeat the above specification this time including a control for voting history as well as political and socioeconomic controls (cf. Table 2.4). Results are very similar but adding initiative fixed effects alone already makes the coefficient of the signature measure insignificant. Hence, including initiative-specific effects in addition to either canton fixed effects or controls, or both, renders the campaigning effect of petition signing on turnout insignificant. I repeat the regressions shown in Tables 2.3 and 2.4 by adding either a canton-specific linear time trend or a quadratic one.11 Adding either one of the time trends reduces the significance of the signatures coefficient even further. Especially in combination with control variables most of the specifications become insignificant. Several explanations exist for an insignificant effect. In Switzerland, typ- ically several federal and cantonal issues are voted on the same election day. Comparing the respective turnout rates, it becomes evident that virtually no roll-off exists in Switzerland: turnout rates for federal ballots on the same day are almost identical. For example, two extremely different initiatives, one on limiting immigration and the other on reducing the number of working hours which were both voted upon on 4 December 1988, had federal turnout rates

11Results are available on request.

22 Table 2.3: Effect of Initiative Signing on Voter Turnout I (1) (2) (3) (4) (5) (6) (7) (8) Signatures 0.770*** 2.006*** 0.721*** 1.614*** 0.342*** 0.960*** 0.125 0.195 p.c. (0.185) (0.394) (0.193) (0.401) (0.115) (0.180) (0.119) (0.199)

Signatures -11.714*** -8.522*** -5.529*** -0.626 2

23 p.c. (2.620) (2.792) (1.295) (1.218)

Controls no no no no no no no no Canton FE no no no no yes yes yes yes Initiative FE no no yes yes no no yes yes Adjusted R2 0.030 0.048 0.544 0.553 0.236 0.239 0.772 0.772 Observations 1,690 1,690 1,690 1,690 1,690 1,690 1,690 1,690 Note: *** p<0.01, ** p<0.05, * p<0.1. The dependent variable turnout is defined as the number of valid votes divided by the number of eligible citizens. Weighted least squares according to Swiss eligible population size. Clustered standard errors at cantonal level. Table 2.4: Effect of Initiative Signing on Voter Turnout II (1) (2) (3) (4) (5) (6) (7) (8) Signatures 0.480*** 0.985*** 0.226 0.249 0.446*** 1.077*** 0.112 0.162 p.c. (0.132) (0.310) (0.148) (0.279) (0.128) (0.230) (0.122) (0.203)

Signatures -4.540* -0.207 -5.616*** -0.439 p.c.2 (2.351) (1.836) (1.848) (1.205)

24 Voting 0.385*** 0.378*** 0.417*** 0.417*** -0.075 -0.090 0.106 0.106 history (0.129) (0.129) (0.133) (0.133) (0.138) (0.139) (0.110) (0.109)

Controls yes yes yes yes yes yes yes yes Canton FE no no no no yes yes yes yes Initiative FE no no yes yes no no yes yes Adjusted R2 0.226 0.228 0.684 0.684 0.308 0.311 0.776 0.775 Observations 1,690 1,690 1,690 1,690 1,690 1,690 1,690 1,690 Note: *** p<0.01, ** p<0.05, * p<0.1. The dependent variable turnout is defined as the num- ber of valid votes divided by the number of eligible citizens. Weighted least squares according to Swiss eligible population size. Clustered standard errors at cantonal level. of 52.84% and 52.86% respectively. This suggests that voters usually vote for most issues once they have decided to participate in the election. Therefore, turnout for two ballot propositions on the same day is very similar. Con- sequently, the number of signatures collected which varies strongly between initiatives would not be a good predictor of turnout. In the sample 65 initia- tives are voted on 39 election days. On 18 of these more than one initiative has been voted on the same day. In total this affects 44 initiatives.12 Initiative fixed effects account for unobserved initiative-specific character- istics. Salience of the initiative topic constitutes one such initiative fixed effect which is an important factor strongly influencing voter turnout in Switzerland (Lüchinger, Rosinger & Stutzer, 2007).13 For example, a highly disputed ini- tiative aiming at abolishing the Swiss army voted on 26 November 1989 had a turnout rate of 69.18%, while for a less salient initiative about the support of public transport only 31.23% of eligible citizens turned out on 3 March 1991.14 This demonstrates the importance of salience for initiative turnout. It may consequently be a better predictor of turnout than the number of signatures collected and be also correlated with the latter.15 Additional evidence comes from Schmid (2013) who finds that a mobilization measure based on petition signatures of the most mobilizing ballot proposition on a particular ballot day has high explanatory power for ballot-day turnout levels. Since fixed effects are necessary to correctly estimate the effect of petition

12Ideally, regressions could be estimated for a subsample of initiatives where the initia- tive was the only federal proposition at ballot. In the sample there are only two such initiatives, therefore this test is not feasible. 13A term indicating the importance of the election is often included in models (cf. Feddersen & Sandroni, 2006). It is high in important elections like presidential election, and low in less important ones like at local level. 14Supporting evidence shows that an initiative also voted upon on 26 November 1989 on velocity limits also had a turnout rate of 69.15%. However, other initiatives concerning traffic and motorways voted on 1 April 1990 had average turnout rates around 41%. Thus, high turnout for the initiative about velocity limits was largely driven by the other attractive initiative on the same election day, and not by the topic itself (Swiss Federal Chancellery, 2013). 15I created a measure of initiative importance by coding a dummy with value 1 if an initia- tive was the most important federal proposition on a particular ballot day (24 initiatives). I also measured if an initiative was the most important proposition for Switzerland on a ballot day based on responses from post-ballot surveys (VOX-surveys, 7 initiatives since 1993). Repeating the regressions for a subsample of important initiatives did not yield significant results. I further created two cantonal measures of importance: (I) affected cantons as described later in Section 2.5.3, (II) from the VOX-surveys, where I define an important initiative for a canton as being significantly above the mean initiative impor- tance (90% confidence). The signature coefficients remain insignificant.

25 signing on turnout, the hypothesis that signing an initiative provides infor- mation to voters and activates citizens to participate in elections has to be rejected when initiative-specific effects are accounted for in combination with canton-specific effects or cantonal control variables. My results are surprising and stand in contrast to the findings of Parry, Smith and Henry (2012) and especially Boehmke and Alvarez (2012). The latter found positive significant effects for eight initiatives based on aggregate data. The former discovered a positive effect by using individual data, but only in one of their three mod- els. My results suggest, that on average over a longer period no correlation between the share of the population that signed an initiative and subsequent turnout exists. Hence, the positive effects in the research mentioned above are likely to stem from initiative-specific effects and need not to hold in general.

2.5.2 Acceptance

The second hypothesis is that the number of valid signatures per capita has a positive effect on the acceptance rate measured as the number of yes votes di- vided by the eligible population size. Table 2.5 shows the results. No controls are included in the baseline regressions, and initiative as well as canton fixed effects are added in alternation. The estimated effect of signatures per capita on acceptance rates is positive and highly significant. Increasing the share of the eligible population which signed the initiative by one percentage point, is related to additional 1.081 percentage points of the eligible population ac- cepting the initiative in the first specification. The linear coefficient decreases slightly when canton fixed effects are included, and more strongly once ini- tiative fixed effects are controlled for as well: the effect decreases to 0.835 percentage points when including canton and initiative fixed effects. Simi- larly as in the above analysis, initiative fixed effects contribute considerably to increasing the adjusted R2, which means that they explain a lot of observed variation in the acceptance rate. In the quadratic specifications, the linear coefficients are positive with values larger than 1, and squared terms display negative effects. Both are highly significant. The size of the coefficients will be interpreted below. Results are robust to the inclusion of control variables, and the coefficients of interest remain highly significant (cf. results in Table 2.6). Importantly,

26 Table 2.5: Effect of Initiative Signing on Acceptance I (1) (2) (3) (4) (5) (6) (7) (8) Signatures 1.081*** 1.837*** 1.107*** 1.791*** 0.873*** 1.371*** 0.835*** 1.229*** p.c. (0.078) (0.186) (0.090) (0.210) (0.073) (0.173) (0.077) (0.163)

Signatures -7.159*** -6.533*** -4.455*** -3.491*** 2

27 p.c. (1.391) (1.282) (1.274) (0.964)

Controls no no no no no no no no Canton FE no no no no yes yes yes yes Initiative FE no no yes yes no no yes yes Adjusted R2 0.102 0.114 0.747 0.757 0.147 0.151 0.804 0.806 Observations 1,690 1,690 1,690 1,690 1,690 1,690 1,690 1,690 Note: *** p<0.01, ** p<0.05, * p<0.1. The dependent variable acceptance is defined as the number of yes votes divided by the number of eligible voters. Weighted least squares according to Swiss eligible population size. Clustered standard errors at cantonal level. Table 2.6: Effect of Initiative Signing on Acceptance II (1) (2) (3) (4) (5) (6) (7) (8) Signatures 0.936*** 1.441*** 0.921*** 1.398*** 0.804*** 1.280*** 0.824*** 1.212*** p.c. (0.060) (0.149) (0.100) (0.200) (0.053) (0.136) (0.078) (0.165)

Signatures -4.565*** -4.305*** -4.233*** -3.433*** 2

28 p.c. (1.171) (1.113) (1.017) (0.998)

Controls yes yes yes yes yes yes yes yes Canton FE no no no no yes yes yes yes Initiative FE no no yes yes no no yes yes Adjusted R2 0.307 0.311 0.768 0.772 0.360 0.363 0.806 0.808 Observations 1,690 1,690 1,690 1,690 1,690 1,690 1,690 1,690 Note: *** p<0.01, ** p<0.05, * p<0.1. The dependent variable acceptance is defined as the number of yes votes divided by the number of eligible voters. Weighted least squares according to Swiss eligible population size. Clustered standard errors at cantonal level. socioeconomic controls like income, education, and unemployment should partly control for differences in preferences between the cantons (specially for economically framed initiatives about pension age, unemployment benefits, or pensions, income and unemployment). In some cases the initiative was withdrawn and only the counter proposal voted upon. Counter proposals constitute compromises by the parliament which take up the main initiative issue but make some concessions. Hence, they are less extreme than initiatives, and appeal to a larger public by being closer to the median voter’s desired policy. For this reason, I restrict the sample to contain only voted initiatives and drop the 7 voted counter proposals - the initiative was withdrawn in all cases - because acceptance is much higher for counter proposals than for initiatives on average. However, results are not affected by this manipulation (results available upon request). This means that voting behavior from counter proposals is not driving the results. Including canton-specific time trends does not affect the significance of the results though the size of the effects is slightly altered. In most cases the coefficients become somewhat larger. The results are available upon request. In what follows, I refer to the results in the specification including both fixed effects and control variables in columns (7) and (8) of Table 2.6. In this specifications the most extensive model is estimated and the R2 is highest, which means that most variation in the data is explained. The linear effect suggests that collecting signatures from one percentage point more of the population is related to 0.824 percentage points of the eligible population accepting the initiative on average. In the quadratic estimation, the coefficient increases to 1.212, and the estimates suggest a significantly negative quadratic effect of -3.433. This means that for all values of signatures per capita below 0.0310, signature collection exerts increasing effects to scale, i.e., the increase in population accepting the initiative goes up by a factor larger than one. Over the range [0, 0.0310], the signature of an additional person increases acceptance by more than one yes vote in the quadratic model. The marginal effect of signing on acceptance turns negative at the value 0.1770 signatures per capita. When looking back at descriptive statistics, the population-size-weighted mean of signatures per capita amounts to 0.0297, and roughly 63% of the observation lie in the interval [0, 0.0310]. Over a relevant range of this variable

29 the marginal effect is hence positive and larger than 1. The existence of spillover effects might be a possible explanation for this result: by talking to family and friends who belong to the personal network, a single signer might motivate a further person to vote in favor of the initiative. Comparing the linear with the quadratic model, results suggest that according to the adjusted R2, the quadratic model has a narrowly better fit, such that it explains more of the variation in the data than the linear one. Hence, it should be the preferred specification.

2.5.3 Controlling for Voter Preferences

Elite Mobilization Elite mobilization constitutes an important factor with large influence on vot- ing results (Kriesi, 1995, 2006). Voters of a particular party look to the party for voting cues because they identify themselves with its political agenda. As elected political representatives, parties reflect their voters’ preferences. For the first preference measure, I thus account for party mobilization. In Switzer- land, all parties and some main interest groups issue voting recommendations to their electorate. These recommendations are publicly communicated. I create a measure of elite mobilization by grouping all parties that have issued a positive voting recommendation together. Next, I take the share of the can- tonal electorate which has voted in favor of these parties at the last election for federal parliament. Suppose only the left party and the green party issue a yes recommendation, and 30% of the voters in a canton supported these two parties in the last national election. Then the control variable takes the value 0.3. Voting recommendations are taken from swissvotes.ch, and party support in national elections is from the Swiss Statistical Office. The elite mobilization variable is distributed between 0 and 1. Its population-size weighted mean is 0.3278 with a standard deviation of 0.2294. Results are reported in columns 1 and 2 of Table 2.7. Including this con- trol does not alter the significant effect of signatures per capita on acceptance. However, the coefficient from the linear model is slightly reduced from 0.824 to 0.763. The results of the quadratic model are similar to the baseline model. The control variable itself has a positive and strongly significant coefficient as expected. If parties favoring the initiative are supported by an additional per-

30 Table 2.7: Effect of Initiative Signing on Acceptance - Preferences I

(1) (2) (3) (4) (5) (6) Signatures 0.763*** 1.135*** 0.832*** 1.235*** 0.775*** 1.162*** p.c. (0.079) (0.168) (0.079) (0.169) (0.079) (0.170)

Signatures -3.278*** -3.555*** -3.404*** p.c.2 (0.998) (1.027) (1.012)

Elite 0.056*** 0.055*** 0.055*** 0.054*** mobilization (0.017) (0.017) (0.017) (0.017)

Pos. affected -0.005 -0.005 -0.007 -0.007 cantons (0.005) (0.005) (0.004) (0.005)

Neg. affected -0.020*** -0.020*** -0.019*** -0.019*** cantons (0.006) (0.007) (0.006) (0.007)

Controls yes yes yes yes yes yes Canton FE yes yes yes yes yes yes Initiative FE yes yes yes yes yes yes Adjusted R2 0.811 0.812 0.808 0.810 0.813 0.815 Observations 1,690 1,690 1,690 1,690 1,690 1,690 Note: *** p<0.01, ** p<0.05, * p<0.1. The dependent variable acceptance is de- fined as the number of yes votes divided by the number of eligible voters. Weighted least squares according to Swiss eligible population size. Clustered standard er- rors at cantonal level. centage point of the cantonal population, additional 0.056 percentage points of the eligible population accept the initiative.

Affected Cantons In a second approach, I account for cantons which are particularly concerned with an initiative and therefore should be more likely to accept or reject it. To this goal, I create two dummy variables. If a canton was supposedly more likely to accept (reject) the initiative, the first (second) dummy variable is coded with the value 1 and zero otherwise. I consulted all communications of the government available through the Federal Chancellery for each initiative in my sample individually. These com- munications are prepared by the government prior to parliamentary debate

31 about the initiative. They contain extensive information on the initiative, its goals, political, economic, and financial consequences. I screened the gov- ernment communications for mention of cantons which might be particularly concerned with the initiative. The cantons were either mentioned explicitly or could be inferred from the communications.16 An example of a positively affected canton is the initiative demanding to allow counter proposals not only for initiatives but also for referendums voted on 24 September 2000. Two cantons, namely Bern and Nidwalden, already had similar provisions at cantonal level. Therefore, they should be more likely to favor such a provision. An example for a negatively affected canton is an initiative asking for the prohibition of animal trials voted on 7 March 1993. Several cantons like Basel Landschaft, Basel Stadt, Vaud and Zurich have a strong pharmaceutical industry relying on animal trials such that they should be less likely to accept the initiative. There are other initiatives like one about the protection of tenants voted on 7 December 1986 for which no especially affected cantons can be found and all cantons are coded with a zero. In total, I identify 60 positively and 51 negatively affected cantons for all initiatives. The former has a mean of 0.0499 while the latter is 0.0493 on average with standard deviations of 0.2177 and 0.2167 respectively. Results are in columns 3 and 4 of Table 2.7. As with the control for elite mobilization, the significance of the baseline results for the effect of signatures per capita on acceptance remains unaffected. This time also the coefficient size is virtually unchanged. The coefficient for negatively affected cantons is significant and, as expected, negative. But the coefficient for positively affected cantons is insignificant. In columns 5 and 6, I control for both elite mobilization and affected cantons, and results remain very similar.

Related Referendums In my third attempt to account for voter preferences, I take an approach similar to Funk and Gathmann (2011), and proxy preferences with old voting results on related issues. I again consult the government communications which contain the information about the article or paragraph of the Swiss

16An alternative measure of importance proposed by Funk (2012) is to use post-ballot responses (VOX-surveys) regarding the question, how important a ballot proposition is perceived for the country. However, this measure is only available since 1993, not all respondents answer the question, and for some cantons there are only few observations per initiative. Therefore, it is not an ideal cantonal measure in the context of this paper.

32 constitution that is about to be altered by the initiative in question. Usu- ally government communications provide information on the history of the initiative and similar ballots concerning the same constitutional article. The best preference controls are voting results of mandatory referendums because voting results from other similar initiatives or optional referendums have a signature collection phase preceding the ballot. Therefore, I would expect the voting results of the two latter forms of ballots to be partly driven by their signature collection process. Mandatory referendums do not require a qualification stage and can consequently serve as preference measure. I identify mandatory referendums concerning the same constitutional ar- ticle or a very similar topic as the initiatives for 37 initiatives in the sample. The reasons that no referendum can be matched are the following: first, the initiative might concern a topic which is regulated by a law and not directly by the constitution. Such issues are typically voted upon in optional refer- endums which have a signature collection themselves. Next, a mandatory referendum with a similar topic might exist. However, sometimes it has been voted too long ago in the past to assume that preferences are time constant. On average, the time difference between voting dates of the initiative and the related referendum amounts to roughly 12 years. Last, some initiatives address issues which have never been on the political agenda before, like the introduction of a national holiday. Consequently, no similar mandatory ref- erendum is available. The mandatory referendums are coded such that they point into the same direction as the corresponding initiative (e.g., more en- vironmental protection, or a more generous pension system). The variable is defined analogously to the dependent variable acceptance as yes votes divided by the number of eligible citizens. It does not change the results qualitatively if the standard definition of acceptance yes votes divided by voters is used. The eligible population-weighted mean is 0.2660 with a standard deviation of 0.0932. For the regressions, I drop the observations from the 28 initiatives for which no applicable mandatory referendum could be found. Estimation results including the voter preference measure are provided in Table 2.8. Since the sample size is reduced for these estimates, I repeat the baseline regressions including controls and both fixed effects in columns 1 and 2. Reducing the sample size from 65 to the selected 37 initiatives has a strong impact on the estimation results. While the main effect remains significant,

33 Table 2.8: Effect of Initiative Signing on Acceptance - Preferences II

(1) (2) (3) (4) (5) (6) Signatures 1.038*** 1.659*** 0.933*** 1.484*** 0.893*** 1.389*** p.c. (0.088) (0.266) (0.102) (0.172) (0.114) (0.201)

Signatures -5.982** -5.339*** -4.759*** p.c.2 (2.323) (1.148) (1.242)

Yes % in 0.047 0.048 0.041 0.042 referendum (0.039) (0.039) (0.037) (0.037)

Elite 0.059*** 0.055*** mobilization (0.021) (0.021)

Pos. affected -0.010 -0.011 cantons (0.007) (0.007)

Neg. affected -0.029*** -0.029*** cantons (0.009) (0.010)

Controls yes yes yes yes yes yes Canton FE yes yes yes yes yes yes Initiative FE yes yes yes yes yes yes Adjusted R2 0.372 0.376 0.779 0.783 0.791 0.794 Observations 949 949 949 949 949 949 Note: *** p<0.01, ** p<0.05, * p<0.1. The dependent variable acceptance is defined as the number of yes votes divided by the number of eligible voters. Weighted least squares according to Swiss eligible population size. Clustered standard errors at cantonal level. Sample size is reduced from 962 (37 initia- tives times 26 cantons) to 949 because there are no voting results from the then non-existent canton Jura before 1978. the coefficient in the linear specification takes on a value larger than one. The adjusted R2 drops by around a half, suggesting that the 28 excluded initiatives have particularly good explanatory power for the data variation. Most likely the 28 excluded initiatives indeed systematically differ from the 37 left in the sample: while related mandatory referendums exist for the latter, this is not true for the former for the reasons explained above. In particular, the excluded initiatives are most likely to affect issues regulated by law and not necessarily be suitable to be included in the constitution.

34 In columns 3 and 4 the control for voter preferences is included. Additional robustness checks where all three control variables elite mobilization, affected cantons, and related referendums are jointly tested, are reported in columns 5 and 6. The relation between signatures per capita and the acceptance rate is unaffected by the inclusion of the variables and is highly significant as in the baseline model. Hence, the main results are robust to the inclusion of various measures of voter preferences which could potentially affect the voting outcome. The control variable for voter preferences itself is positive but not signif- icant. As before, mobilization from political parties significantly increases acceptance, while negatively affected cantons have lower acceptance rates. The coefficient of positively affected cantons remains insignificant. I repeat all regressions reported in Tables 2.7 and 2.8 including either linear or quadratic cantonal time trends. These regressions produce almost identical results as before.

2.5.4 Probability of Signing an Initiative

The total number of signatures collected reflects signatures from initiative partisans signing the initiative, as well as from previously uninformed voters who receive favorable information from the collection campaign. Theoretical models of voting predict that partisan voters are more likely to turn out than uninformed ones. Moreover, the dominant strategy for partisans is to sup- port their own candidate (Feddersen & Pesendorfer, 1996). Potentially, part of the positive relation found between signatures per capita and acceptance stems from partisans. Only part of the correlation, however, would reflect the campaigning effect of signing from previously uninformed voters. Using an institutional change, I therefore control for the randomness of the signature collection process to separate independent from partisan signers. A typical location to gather signatures is near a polling place such that people who have just participated in an election could be asked to sign the ini- tiative. Thus, collection near places of election targets regular and potentially better informed voters with a higher probability than a collection campaign near the train station or a shopping mall. Consequently, collection campaigns near places of election are less random than at other places and thus should

35 be controlled for. A remedy is to exploit cantonal variation in the introduc- tion of postal voting in Switzerland between 1978 and 2005 (Funk, 2010). With the introduction of postal voting, the channel of collecting signatures near places of election has been considerably reduced: voters can return their ballot papers by mail. In Switzerland, citizens often make use of this voting model (Klaus, 2006). Though the option of voting at the booth has not been abolished, the number of voting locations and their opening hours have been considerably reduced (Lüchinger, Rosinger & Stutzer, 2007). Importantly, Hodler, Lüchinger and Stutzer (2014) assert that the timing of the staggered introduction of postal voting in the cantons had no particular reason. Also, no other important institutional changes occurred at the same time. With the availability of postal voting, it becomes more difficult to collect signatures near polling places because part of the population votes by mail. Consequently, it becomes less likely that regular voters who, based on their political knowledge and experience, should be more likely to have a predispo- sition regarding the initiative are asked to sign an initiative petition. If the hypothesis is true that the act of signing increases the citizens’ acceptance probability, then controlling for postal voting and its interaction term with the number of signatures should leave the significance of the signature coeffi- cient unaffected. For the interaction term I expect a negative coefficient. In the baseline regressions the true effect is probably overestimated if part of the signers is more informed regular voters. I regress the following estimation equation.

acceptanceci = β0 + β1signatures p.c.ci + β2postalci × signaturesci

+β3postalci + β4Xci + uc + vi + ci (2.3)

β1 can be interpreted as the effect of petition signing on acceptance for on average more informed citizens (postalci =0). The total effect β1 +β2 reflects the relationship for on average less informed signers (postalci =1). I also estimate a quadratic model which includes signatures per capita squared and its interaction with the postal voting dummy. The results in Table 2.9 show that coefficient β1 remains highly significant in the linear and in the quadratic specification. As expected the interaction term between postal voting and signatures per capita has a negative coeffi-

36 Table 2.9: Effect of Initiative Signing on Acceptance - Postal (1) (2)

Signatures p.c. 0.881*** 1.456*** (0.000) (0.000) Signatures p.c.2 -4.777*** (0.000) Signatures p.c. × postal voting -0.178 -0.649*** (0.173) (0.006) Signatures p.c.2 × postal voting 3.792** (0.018) Postal voting 0.009* 0.018*** (0.092) (0.001) Controls yes yes Canton FE yes yes Initiative FE yes yes Adjusted R2 0.806 0.809 Observations 1,690 1,690 Note: *** p<0.01, ** p<0.05, * p<0.1. The depen- dent variable acceptance is defined as the number of yes votes divided by the number of eligible voters. Weighted least squares according to Swiss eligible population size. Clustered standard errors at cantonal level. cient which is only significant in the quadratic specification in column 2. This means that the relationship between petition signing and acceptance is smaller in cantons with postal voting. The result suggests that petition signing by frequent and consequently more informed voters might indeed play a role. However, though the effect of petition signing on acceptance decreases with postal voting, the significance of the main result remains unaffected. Accep- tance is significantly higher in cantons with the possibility of postal voting. Recall that acceptance is defined as the number of yes votes as a share of the eligible population and therefore encompasses turnout and acceptance decisions. The positive postal coefficient might partly reflect that turnout on average increased in cantons with postal voting as shown by Lüchinger, Rosinger and Stutzer (2007).

2.5.5 Additional Robustness Checks

As additional robustness checks, I conduct several placebo regressions. The main intuition for the placebos is that I relate the measure of signatures to

37 voting results from other initiatives or referendums. I expect the coefficients to be insignificant. First, I assign the signatures of initiatives to the voting results of a manda- tory referendum that has been held on the same day. For this, I take into account that topics should not be related. For example, if an initiative and a referendum about energy policy are held on the same day, it can be expected that cantonal outcomes are similar for both ballots. Therefore, I match ini- tiative signatures with referendums concerning unrelated topics, for example motherhood and migration policy. In total, I can match 19 initiatives with unrelated referendums on the same ballot day. The intuition is that the signa- ture measure should be insignificant because signature collection is matched with the voting results of a different proposition. Indeed, even though the coefficients have the expected sign they are insignificant (Table 2.10). Second, I assign to every initiative the voting result of a thematically related mandatory referendum. I test again whether signatures are a good explanatory variable for the voting results of a distinct but similar ballot which is touching the same subject. If my analysis truly identifies the mo- tivational effect of petition signing on acceptance, in contrast to signatures just reflecting voter preferences, the estimates should be insignificant. I use

Table 2.10: Effect of Initiative Signing on Accep- tance - Placebo I (1) (2)

Signatures per capita 0.036 0.134 (0.104) (0.201) Signatures per capita squared -0.762 (1.030) Controls yes yes Canton fixed effect yes yes Initiative fixed effect yes yes Adjusted R2 0.784 0.784 Number of observations 494 494 Note: *** p<0.01, ** p<0.05, * p<0.1. The depen- dent variable acceptance is defined as the number of yes votes divided by the number of eligible voters, however, not for the initiative ballot but of a differ- ent ballot on the same election day. Weighted least squares according to Swiss eligible population size. Clustered standard errors at cantonal level.

38 Table 2.11: Effect of Initiative Signing on Accep- tance - Placebo II (1) (2)

Signatures per capita -0.240** -0.325 (0.108) (0.336) Signatures per capita squared 0.828 (3.380) Controls yes yes Canton fixed effect yes yes Initiative fixed effect yes yes Adjusted R2 0.665 0.665 Number of observations 949 949 Note: *** p<0.01, ** p<0.05, * p<0.1. The depen- dent variable acceptance is defined as the number of yes votes divided by the number of eligible voters, however, not for the initiative ballot but of a closely related mandatory referendum ballot. Weighted least squares according to Swiss eligible population size. Clustered standard errors at cantonal level. Sample size is reduced from 962 (37 initiatives times 26 can- tons) to 949 because there are no voting results from the then non-existent canton Jura before 1978.

the same 37 initiatives as in Section 2.5.3. As expected, signature coefficients are not significantly positive. For the linear model, the coefficient is negative and significant at the five percent level. In the quadratic specification both coefficients for signatures and signatures squared are negative and insignifi- cant (Table 2.11). The placebos add to the evidence that the analysis does not reflect a random correlation.

2.6 Concluding Remarks

This paper analyzes the qualifying stage of popular initiatives. It extends pre- vious work by exploring the campaigning effect of signing initiative petitions on turnout and voter approval. The main findings are twofold. First, I find no effect of initiative petition signing on the probability of subsequently turning out at the voting booth. Initiative-specific effects like proposition salience are better predictors of turnout. Another explanation is the high intra-day corre- lation of turnout in Switzerland, which potentially does not allow to estimate

39 the relationship correctly. Second, signing an initiative petition is associated with a higher probability of casting approving votes at the subsequent initia- tive ballot. This result proves to be highly robust to the inclusion of various preference measures, socioeconomic and political controls, as well as canton and initiative fixed effects. In the light of my results, initiative signature collection can be interpreted as a partisan campaigning tool for the initiating group: signers receive relevant information and become aware of the initiative issue. In terms of approving votes, it turns out to be worthwhile to run a larger collection campaign and gather additional signatures which increases the citizens’ awareness of the issue and gives them necessary information, especially in ranges where sig- natures display increasing returns to scale. Hence, my results contribute to understanding why initiatives usually collect more signatures than legally re- quired to qualify initiatives for ballot - other than to insure against invalid signatures. Though larger collection campaigns mean higher collection cost, they reap benefits in terms of additional support from the eligible population at ballot. For future research, an analysis based on individual data from post-ballot surveys or field experiments would help to shed further light on the question of causality.

40 Appendix

2.A Data Used from Swiss Census (1970, 1980, 1990, 2000)

Provided by the Swiss Statistical Office, www.bfs.admin.ch

• Total cantonal population

• Population 65 years old or older per canton

• Unemployed population per canton

• Population with tertiary education per canton

Other data

• Variable “mean taxable income” is from the Eidgenössische Steuerver- waltung (Federal Tax Administration) in Bern.

• Number of cantonal ballots on same election day is calculated based on cantonal voting data available online from the Centre for research on direct democracy on www.c2d.ch.

• Turnout in elections for the federal parliament is from the Swiss Statis- tical Office.

• Dates of initiative qualification and ballot to calculate time between initiative qualification and ballot are from the homepage of the Swiss Federal Chancellery.

• Introduction of postal voting is from Funk (2010). Updated for the cantons Ticino and Valais by calling the cantonal administrations.

41 2.B Presentations and Acknowledgement

This paper has been presented at the following conferences and workshops: Sinergia Workshop of the Swiss National Science Foundation (February 2013, St.Gallen, Switzerland), Ph.D. Seminar at the University of St.Gallen (May 2013, St.Gallen, Switzerland), Annual Meeting of the Public Choice Soci- ety (March 2014, Charleston, USA), and the Electoral Integrity Pre-IPSA (International Political Science Association) Workshop (July 2014, Montreal, Canada). For valuable comments, I thank Monika Bütler, Patricia Funk, Christian Marti, Lukas Schmid, and Andreas Steinmayr. I appreciate helpful input by the discussants Florian Chatagny and Patrick Fournier.

42 Chapter 3

Ready to Reform: How Popular Initiatives Can Be Successful

Joint work with Christian Marti and Monika Bütler

3.1 Introduction

Far-reaching political reforms are difficult to accomplish, which is mostly blamed on the so-called status quo bias. Common reasons for the stickiness of policies are uncertainty about reform payoffs (Fernandez & Rodrik, 1991), adjustment costs to new policies (Coate & Morris, 1999), political agenda set- ting power (Romer & Rosenthal, 1978), or inefficient bargaining over reform payoffs (Alesina & Drazen, 1991). Prime examples of status quo bias from the empirical literature are agricultural subsidies (Sowell, 1990), and welfare policies like unemployment benefits or pension age (Hollanders & Vis, 2013). The former is often attributed to strong interest groups, and the latter to voter preferences opposing reform. In a representative democracy the voters’ main channel to influence politics is via the election and potential reelection of politicians. Other possibilities are petitions or protest in the form of demonstrations (e.g., Lohman, 1993,

43 1994). Another form to challenge the prevailing policies designed by elected representatives is through direct democratic votes. Direct democracy is deeply rooted in many U.S. states and Switzerland, and constitutional provisions for direct democracy shape politics in a growing number of countries, mostly at local and state level (Altman, 2010; LeDuc, 2003; Matsusaka, 2005). Two forms of direct democratic instruments can be distinguished based on who is the proposer of the vote. In most countries, it is the decision of the elected au- thorities whether a popular vote, a plebiscitary referendum, should be held or not. Or a mandatory referendum is prescribed by the constitution for certain issues of crucial importance (Center for Direct Democracy, 2014). However, in around half of the U.S. states and in Switzerland, citizens themselves can demand a vote on a policy by collecting signatures (Initiative and Referen- dum Institute, 2014).1 This policy can be either an outcome of the political process (an optional / facultative referendum), or a policy proposal designed by citizens, the popular initiative (Knoepfel et al., 2014). In this paper we deal with the question to what extent the existence of the popular initiative helps to change the status quo, and in how far popular initiatives weaken the politicians’ agenda setting monopoly as argued by Mat- susaka (2005). The initiative can be successful via two channels. First and most intuitive, the initiative can beat the status quo in a popular vote. Then, the initiative replaces the old status quo. The second channel to initiative success works through political compromise: whenever a popular initiative is raised, this could be a hint to the politicians that part of the population desires a policy change. While popular initiatives are frequently raised by strongly motivated minorities, it might well be that some initiatives enjoy broader support since they take up new issues, or preferences of the citi- zens have changed due to exogenous policy shocks. For example, while there were no nuclear power-related initiatives in Switzerland after 1999, signature collection for an initiative favoring a nuclear power phaseout started only 2 months after the Fukushima nuclear disaster in 2011. In such cases, when the initiative is voted against the status quo, citizens might accept the - po- tentially radical - initiative if it comes closer to the median preferred policy than the status quo. If an initiative is adopted, this incurs reputational risk

1 Similar provisions have been introduced in most German states in the last two decades (Association for More Democracy, 2014) and for example in several Eastern European countries after the fall of communism (Center for Direct Democracy, 2014).

44 for office-motivated politicians, and ideologically motivated politicians receive disutility from moving away from their desired policies. Then, proposing a policy compromise might be a prudent alternative for both ideologically and office-motivated politicians to prevent the initiative from winning at the bal- lot. We define initiative success as the ability of a popular initiative to change the status quo, both via popular approval of the initiative or the implemen- tation of a policy compromise. There are formal, institutional provisions for such policy compromises trig- gered by popular initiatives in a number of U.S. states like Washington, Maine, and Colorado, at all state levels in Switzerland and at state and local level in several German states (Center for Direct Democracy, 2014; Initiative and Referendum Institute, 2014; More Democracy, 2014).2 These policy compro- mises are called counter proposals to an initiative (Swiss Federal Chancellery, 2013). For example, in 2005 in Switzerland an association of fishermen de- manded that rivers affected by hydro-energy plants should carry more water. The counter proposal suggested an increase in the amount of water, but by less than was intended by the initiative and not everywhere. To give another example, in 1971 an association of women’s rights activists demanded the le- galization of abortion without any preconditions. This case illustrates that even in a seemingly binary issue (in favor or against abortion), there is space for compromise: the counter proposal stated that abortion should be legalized under certain conditions in the first weeks of pregnancy only.3 We focus on the formalized institutional setting of Switzerland, which allows for a straightforward observation of policy compromises triggered by a popular initiative, to analyze whether initiatives can undermine the status quo bias. Close to the Swiss setting, but also representative for other direct democracies, we develop a sequential initiative game under uncertainty about the initiative’s winning probability. While both petitioners and politicians

2 Moreover, in a number of U.S. states, initiatives are dealt with in an indirect way. This means that the legislature can choose to approve the initiative, and it is only put to the popular vote in case the parliament rejects the initiative. This shares some similarities with a counter proposal as well. The indirect initiative is available in nine states for statutory proposals and in two states (Massachusetts and Mississippi) for constitutional amendments (Initiative and Referendum Institute, 2014). 3 In some U.S. states politicians can put up new laws for popular vote, possibly at the same day as initiatives, as so called referred measures. This comes close to a formal counter proposal (Initiative and Referendum Institute, 2014): the legislature has the possibility to enact laws as alternatives parallel to the initiative process to demonstrate that some action is being taken on the subject.

45 have symmetric information in the model, they are uncertain about the precise position of the median voter and only have incomplete information about the distribution the median position is drawn from. During signature collection to qualify the initiative, information about the initiative’s winning probability is revealed. This partly reduces uncertainty about the median voter position and thus about the initiative’s winning probability. Politicians decide about making a costly political compromise while tak- ing into account their belief about the initiative’s winning probability from the signature collection. After observing the politicians’ decision, petitioners decide whether to withdraw the initiative or not. The incentive to withdraw the initiative is that after a counter proposal the initiative’s probability of winning decreases if it is not withdrawn. Though the initiative forfeits the possibility of winning, the counter proposal’s winning probability and thus the probability of amending the status quo increases. The initiative game has a unique subgame perfect equilibrium. Maximiz- ing their expected payoffs, politicians are more likely to compromise if they belief the initiative to be a real threat to the status quo. They either make a cutoff counter proposal that narrowly makes petitioners withdraw the ini- tiative, such that a vote between the status quo and the counter proposal determines the winner. Or they issue a very small compromise close to the status quo such that all three voting alternatives compete at ballot. When no counter proposal is made, petitioners only withdraw the initiative if cam- paigning costs exceed the expected payoff from having the initiative voted against the status quo. From our model, we derive predictions about the probability that the sta- tus quo is amended. We further make use of four big institutional changes - the introduction of female suffrage in 1971, the doubling of the signature requirement to 100,000 signatures in 1978, a cap to signature collection time at 18 months also in 1978, and a major change in the voting rule to include a tie-breaking question between initiative and counter proposal in 1987 - to test how the model performs in predicting equilibrium outcomes. The empir- ical analysis is based on the complete Swiss dataset of 249 federal initiatives that have completed the initiative process between 1891 and 2010.4 The re- sults confirm the model prediction that the status quo is most likely to be

4 In total, 249 initiatives collected the necessary amount of signatures in due time. However, further 77 initiatives have been raised which did not meet the qualification criteria.

46 amended after a counter proposal. However, if the initiative competes exclu- sively against the status quo at ballot, this rarely leads to change. We also find in accordance with the model that collecting more signatures than legally required to qualify the initiative is related to a higher probability of receiving a counter proposal after which the initiative is withdrawn: petitioners can prove that their initiative is popular and enjoys high popular support. Regarding the model predictions after institutional changes, the model performs well in explaining most of the changes. For example, after the intro- duction of female voting and an accompanying drop in the signature collec- tion costs, data prove the model right that more initiatives should be observed trying to qualify for ballot but these initiatives should have low winning prob- abilities and rarely receive a counter proposal. We extend the research about the status quo bias to direct democratic settings, and add to game theoretical models of direct democracy. Our model shares some similarities with Gerber’s (1996) initiative model which, how- ever, neither includes the signature collection process nor the possibility of a counter proposal. Matsusaka and McCarty (2001) propose a referendum game to analyze the status quo under the threat of a referendum. Related to our research is also the literature on the responsiveness of politics in di- rect democracies. Gerber and Lupia (2010) examine the effect of political competition on responsiveness. Fatke and Freitag (2013) find that popular initiatives serve as a substitute channel for other forms of political protest. Moreover, our research broadens the literature analyzing various effects of sig- natures collected to qualify the initiative: the effect of signature requirements on turnout (Barankay, Sciarini & Trechsel, 2003), the motivational effect of petition signing on political knowledge (Neinman & Gottdiener, 1982), on the probability of turning out (Boehmke & Alvarez, 2012; Parry, Smith & Henry, 2012; Schmid, 2013), and voting in favor of the initiative (Hofer, 2013). The remainder of this paper is organized as follows: in Section 3.2 we describe the Swiss institutional setting and the major institutional changes regarding the initiative process. The model is developed, discussed and solved in Section 3.3. The data and empirical strategy are presented in Section 3.4, and the results follow in Section 3.5. Section 3.6 concludes.

47 3.2 Institutional Background

3.2.1 Main Characteristics

We focus on the Swiss federal initiative at constitutional level and describe its main characteristics. An initiative text is either a fully formulated consti- tutional article, or a general suggestion for a constitutional article that has to be concretized by the legislature. However, since there is more interpretation leeway for the parliament in the this case, general suggestions occur relatively rarely.5 Following Frey (1994), we divide the popular initiative process into three stages: qualification, political, and voting stage. At the qualification stage, petitioners, also referred to as initiative committees, collect signatures. For an initiative to qualify, it has to gather a minimum requirement of sig- natures. In Switzerland, the signature requirement is constant and not a percentage value of the active electorate like in California. Upon successful qualification, at the political stage the two chambers of parliament6 decide whether or not to prepare a counter proposal that takes one of two forms. A direct counter proposal is at the constitutional level and has to be voted by the people. An indirect counter proposal is a law that does not need ratification from the voters. However, it might be subject to an optional referendum if enough signatures are collected against it. In practice this happens very rarely, though, since counter proposals tend to be a compromise of a broad parliamentarian majority and do not face too much opposition. Petitioners may withdraw their initiative anytime during the political stage. This happens frequently after a counter proposal has been made. Initiatives can also be withdrawn for other reasons, such as a lack of funding or if it is evident that the initiative has no popular support (Hofer, 2012). At the ballot stage, there are three different combinations. First, the initiative can be on the ballot without a counter proposal. Second, if the initiative is withdrawn after a counter proposal, only the counter proposal

5 In Switzerland, no federal statutory initiative exists. Cantons and municipalities have their own independent regulations regarding initiatives, but they are not considered in this paper. 6 The Swiss parliament consists of two chambers, one of which (the National Council) is elected to proportionally represent the number of people, similar to the American House of Representatives. The other (Council of States) is elected to represent the cantons (states), similar to the American Senate.

48 is voted at ballot. Third, if there exists a direct counter proposal and the initiative is not withdrawn, it is put to the vote simultaneously with the initiative. In the original voting rule from 1891, it takes a majority of the votes from the participating population for the initiative or the direct counter proposal to come into force. In addition, also a majority of the Swiss cantons has to be in favor of the proposition. The ballot system therefore has a strong federal element: a change to the constitution must win the majority of the votes that are not concentrated in a few populous cantons.

3.2.2 Institutional Changes

The Swiss constitutional initiative at federal level was first introduced in 1891. In its long history, five major institutional changes regarding the initiative process have occurred. This leads to six regulatory periods outlined in Table 3.1. The “Early Period” (1891-1927) is marked by the lack of a formal clause that would have allowed petitioners to withdraw the initiative at any time. The first initiative was withdrawn in 1908, and it remained the only one until 1928. The second period “De Facto Withdrawal” starts in 1928 when withdrawal of initiatives became more frequent after receiving a counter proposal, ar- guably out of political learning (Swiss Federal Chancellery, 2013). While this institutional change did not take place at a formally fixed date, the sudden start of withdrawals after 1928 is a clear change to the political game. The “Formal Withdrawal” period 3 (1951-1970) begins in 1951 when ini- tiative withdrawal was officially legalized conditional on all members of the initiative committee unanimously agreeing to withdraw (Hofer, 2012). This provision was relaxed in 1962 to two thirds of the committee, and further low- ered to the absolute majority in 1978. Since adjustments in the withdrawal regulation in 1962 and 1978 were only minor changes, we do not define them as separate institutional periods but subsume them in period 3. Initiatives have to address single issues by a constitutional requirement. Though this requirement was only formalized in 1962, petitioners have always adhered to it. The new regulation was rather a formalization of the previous common

49 Table 3.1: Overview of Main Institutional Changes Period Years Subject of Reform 1 Early Period 1891-1927 No legal provision regarding ini- tiative withdrawal 2 De Facto Withdrawal 1928-1951 De-facto introduction of initiative withdrawal 3 Formal Withdrawal 1951-1970 Official formalization of possibil- ity to withdraw if petitioners agree unanimously (2/3 major- ity since 1962, absolute major- ity since 1978); parliament has 2 years to deal with initiative 4 Female Voting 1971-1977 Introduction of female voting and political rights like initiative sign- ing 5 Collection Restrictions 1978 -1986 Signature requirement doubled to 100,000, collection time restricted to 18 months, mandatory with- drawal article, extension of par- liamentary discussion period to 4 years 6 Tie-Break 1987-2010 Introduction of tie-breaking ques- tion when both initiative and counter proposal are voted Note: This table gives an overview of the 6 institutional periods, the relevant years and a short description of the main institutional changes defining the period.

practice (Hofer, 2012; Swiss Federal Chancellery, 2013).7 The “Female Voting” period 4 (1971-1977) starts when women were enfran- chised at federal level on 7 February 1971, giving them also further political rights like to sign initiative petitions. Period 5 (1978-1986) begins when new “Collection Restrictions” were in- troduced as a consequence of female suffrage in 1978. First, the signature requirement was increased from 50,000 to 100,000 signatures. Second, while there was initially no restriction regarding signature collection time, a maxi- mum of 18 months was introduced in the same year (Hofer, 2012).

7 There was one exception to this rule: the 1919 initiative for a reduction of the number of foreigners and for the eviction of foreigners dealt with two separate issues. It was subsequently split into two parts by the parliament and treated as two separate initiatives.

50 The sixth period “Tie-Break” (1987-2010) commences when the voting rule regarding the case that initiative and counter proposal are voted simultane- ously was revised. Before 1987 voters had to decide in favor of either the initiative or the counter proposal, or oppose both of them. Since April 1987 it is possible to vote in favor of both initiative and counter proposal plus a third tie-breaking question in which voters declare what alternative they prefer in the event of a tie. If either the initiative or the counter proposal receives an absolute majority of all votes and cantons, it is implemented. If both the ini- tiative and the counter proposal receive a majority, the tie-breaking question is decisive for which one gets adopted. A number of minor institutional changes only mildly affect the initiative process. In 1951, the time the parliament has to deal with an initiative was extended from one to two years, and further augmented to three years for general suggestion and four years for readily applicable constitutional articles in 1978 (Hofer, 2012). In 1978, a part of the canton of Bern separated from the bigger canton to become the 26th Swiss canton Jura. Since popular initiatives require a majority of the cantons for approval, the formation of a new canton affected this requirement. Postal voting was introduced based on cantonal regulation between 1978 and 2005. Last, voting age was reduced from 20 to 18 in 1991. Though these changes might arguably affect the initiative process, the consequences are smaller than from the major changes such that we focus on the latter ones in this paper.

3.3 Model

We model the initiative process as a sequential game under uncertainty in the spirit of the Swiss institutional setup. A main characteristic of the model is that players are uncertain about the initiative’s winning probability. Petition- ers collect signatures to reduce this uncertainty. Afterwards politicians decide whether to propose a compromise. Petitioners have the option to withdraw the initiative before the ballot.

51 3.3.1 Model Setup

Let A be the elected politicians in parliament and B the petitioners. The players are unitary actors8 and information is symmetrical. Both the status quo xq and the initiative xi are located on the real line between zero and one as common in games of political economy (Downs, 1957; Black, 1958). They are exogenously given and observed by both players. Without loss of generality, we assume that xq 0 and upper bound μ<¯ 1. For simplicity, we allow the voter density func- tion to have positive mass at x =0, and mass below 1 at x =1. The winning probability of the status quo against the initiative is denoted by πq(xq,xi) with vote share vq(xq,xi) (Osborne, 2000). The status quo wins against the initiative if it receives at least half the votes, i,e., if the median is closer to the status quo than to the initiative.

1 1 2 (xq + xi) − μ πq(xq,xi)=Prob m ≤ (xq + xi) = 2 μ¯ − μ

It follows that the initiative wins with probability πi(xq,xi)=1−πq(xq,xi)= 1 μ¯− 2 (xq +xi) μ¯−μ . The winning probability of the status quo against the counter proposal is defined analogously. To ensure positive winning probabilities for all policy alternatives, we assume that μ ≤ xq and μ¯ ≥ xi. This reflects the uncertain outcome of the ballot as no policy wins or looses with certainty. It also reflects that potentially a majority of voters either prefers a more extreme policy than the status quo or the initiative. With three policy alternatives at ballot, the initiative wins whenever the median is right to the middle point between the counter proposal and the 1 μ¯− 2 (xc+xi) initiative, with winning probability πi(xq,xc,xi)= μ¯−μ . The counter

8 Cf. Gerber (1996) for a thorough discussion of this assumption.

52 Numerical Example Winning Probabilities: In Figure 3.1 the median voter is distributed between zero and one, xq =0.35,andxi =0.85 with the middle 1 ( + )=06 point 2 xq xi . . The initiative wins whenever the median voter is located in the interval (0.6, 1] represented by the dashed line because it is closer to the initiative than to the status quo. Consequently, the initiative wins with probability ( )= 1−0.6 =040 πi xq,xi 1−0 . .

Figure 3.1: Example of Winning Probabilities 1 ( + ) 2 xq xi

μ =0 xq xi μ¯ =1

Note: Median distribution μ ∼ U[0, 1], status quo xq =0.35, initiative xi =0.85, and their 1 (x + x )=0.6 middle point 2 q i . The initiative wins when the median voter is located in the interval represented by the dashed line, which happens with probability πi(xq,xi)=0.4. proposal wins under two conditions: first, the median must be located in the 1 1 interval 2 (xq + xc), 2 (xc + xi) . Second, the vote share is required to exceed half of the votes.9 The counter proposal’s winning probability is independent of its precise position between status quo and initiative (proof in appendix), b πc(xq,xc,xi)= μ¯−μ . The status quo’s winning probability πq(xq,xc,xi)= 1 2 (xc+xi)−μ−b μ¯−μ follows directly. Players are uncertain about the initiative’s winning probability because they have incomplete information about the median distribution’s lower bound. h They know that the lower bound takes the highest possible value μ = xq with probability p0 ∈ [0, 1]. Else it takes some lower value with equal probability. We refer to the former as high types, and low types to the latter. The players’ = [ ]= ∗ h +(1− ) ∗ 1 h = initial belief is the expected value μ0 E0 μ p0 μ p0 2 μ 1+p0 h 2 μ . To qualify an initiative for ballot, petitioners are required to collect st ≥ s¯ signatures where s¯ denotes the legal requirement, and st is the cumula- tive number of collected signatures at any time t. Time is continuous with t ∈ [0, ∞). The signature collection is a publicly observed common learning process through which petitioners and politicians receive information about

9 Potentially supporters of the initiative and the counter proposal may be cumulatively in the majority. However, if none of them has the absolute majority of voters, the status quo remains in place.

53 the median distribution. During collection, all petitioners continuously re- ceive sl > 0 signatures. High types receive stochastic lump sums of additional signatures sh > 0, which arrive independently over time according to a Pois- son process with intensity λ. During collection, costs of γ continuously accrue. Once collection stops, it cannot be resumed.10 Figure 3.2 shows an example of the probability of being a high type pt as collection time evolves. At collection start at time t =0, the probability that the initiative is a high type is p0. If no lump sum occurs during collection, the belief that the initiative is a high type decreases monotonically with time according to ∂pt/∂t = −λpt(1 − pt) and converges to zero as time goes to infinity (Keller & Rady, 2010). Accordingly, the expected belief about the

Figure 3.2: High Type Probability over Collection Time high type probability pt

1 high type p0

low type with probability (1 − pt)

0 t1 collection time t

Note: The y-axis displays the probability of being a high type pt, and the x-axis collection time. The belief of being a high type pt decreases with collection time if only low numbers of signatures sl are observed. If a lump sum sh occurs e.g. at time t1, the initiative is a high type with certainty and pt =1.

10The signature collection process resembles the optimal experimentation strategy of a single decision maker in Strulovici (2010): the petitioners are a single player deciding whether to continue experimentation (signature collection), or to stop the process. The main difference is that players in Strulovici (2010) receive payoffs during experimentation and also once experimentation has stopped. In our model, petitioners receive one of two possible expected payoffs and then the game stops.

54 = 1+pt h median distribution μt 2 μ decreases with pt. The longer petitioners collect without receiving a lump sum, the smaller players belief is the initia- tive’s winning probability. However, suppose a lump sum sh occurs at time t1. Since only high types receive lump sums, players know that the initiative is a high type with certainty, and beliefs are irreversibly at pt =1. Then players also know that the initiative has the highest possible winning probability. Politicians’ utility depends on the winning policy. The initiative is their A least preferred policy and yields utility Ui =0. Utility increases linearly with A the distance between the winning policy and the initiative: Uc = xi −xc is the A utility if the counter proposal wins, and Uq = xi − xq if the status quo wins. Politicians pay a reputation loss of size r if they make a counter proposal and the status quo wins at ballot. With this assumption we rule out strategic compromises exclusively designed to steal votes from the initiative to help the status quo win, without having any winning chances itself.11 If politicians make a counter proposal, costs c incur. They reflect the politicians’ time and effort to debate the details of the counter proposal in parliament. Similarly, petitioners receive utility from the winning policy. They are B rewarded least if the status quo remains in place, Uq =0, and their utility increases linearly with the distance between the winning policy and the status B B quo. Their utility is Uc = xc −xq if the counter proposal wins, and Ui = xi − xq when the initiative wins. Politicians and petitioners thus have symmetric k utility functions. Petitioners pay campaigning costs κ = μ¯−μ > 0 if they do not withdraw the initiative.12 This reflects that campaigning in favor of the initiative is costly because it takes time, human and monetary resources. The expected campaigning costs amount to E[k]=k>¯ 0 and are publicly known before the signature collection begins. The individual size of k is disclosed only once the initiative is qualified. Figure 3.3 displays the extensive-form game tree, summarizing the following rules:

1. Nature chooses the initiative’s winning probability.

2. Petitioners collect at least s¯ signatures in time t to reduce uncertainty about the median voters position.

11In a citizen-candidate setting, Osborne and Slivinski (1996) show the existence of three- candidate equilibria in which the middle candidate enters just to affect the voting result even though he himself has no chance to win. 12The functional form of campaigning costs κ is chosen for computational ease.

55 Figure 3.3: Extensive-Form Game Tree of Initiative Game 0 high type low type B ... signatures A

no counter proposal counter proposal

B B

56 withdraw ballot ballot withdraw

0 0 0 0

xq xq xi xq xc xi xq xc

A A A A A A A A Uq Uq Ui Uq -c-r Uc -c Ui -c Uq -c-r Uc -c B B B B B B B B Uq Uq -κ Ui -κ Uq -κ Uc -κ Ui -κ Uq Uc

Note: Players 0 nature, A politicians, B petitioners. Counter proposal xc = ∅ or xc ∈ (xq,xi) with cost c. Campaigning cost κ, reputation costs r if status quo wins after a counter proposal, c costs of making a counter proposal. Nature picks the winner among xq (status quo), xc (counter proposal), xi (initiative). Payoffs in brackets. 3. Politicians decide whether to make a counter proposal at costs c, or not. They pay reputation costs r if the status quo wins after a counter proposal.

4. Petitioners decide whether to withdraw the initiative or not. They pay campaigning costs κ if they do not.

5. Voters vote sincerely between the policy alternatives.

6. Both players realize their payoffs. The winning policy is implemented.

3.3.2 Discussion of Modeling Choices

Uncertainty over the initiative’s success probability is a key element of our model as it allows for distinct voting outcomes after the same equilibrium combination of voting alternatives. As an example, suppose that the initiative is voted against the status quo. Then both the initiative and the status quo can win the vote with positive probability. Moreover, we model the impact of information about the initiative’s winning probability from the signature collection on the initiative process. The initiative’s winning probability is a function of the status quo and initiative’s positions on the real line, and the bounds of the median voter distribution. Since players need to know the positions of the status quo and the initiative to make decisions about counter proposals and initiative withdrawal, the link between signature collection and the initiative’s winning probability is chosen to work through μ. In our model, both the status quo and the initiative are exogenous. We assume that petitioners act out of conviction and want to implement the pol- icy they truly believe is best. In the sense of Kartik and McAfee (2007) they have character, and are unpolitical types (Calvert, 1985; Wittman, 1983). The initiative’ position xi thus reflects the petitioners’ preferred policy, max- imizing their utility. Petitioners already derive utility from simply writing the initiative and drawing public attention towards their issue, very much in the sense of Habermas’ (1992) discourse theory. Their bargaining power which gives them a positive expected payoff comes mainly from the possibil- ity of withdrawing the initiative. Moreover, since players are uncertain about the initiative’s winning probability, a nonstrategic positioning of the initiative is viable. An obvious extension of the model would be to allow petitioners

57 choose the initiative point strategically. For example, Gerber (1996) develops a model in which proposing an initiative is costly, and therefore for a given status quo not all initiatives points are proposed. In contrast to Gerber (1996) and also Matsusaka and McCarty (2001), we do not assume that the status quo is set while anticipating the threat of a direct democratic initiative or referendum. We prefer to think of it as old regulation, or a topic that has not been regulated so far. It is commonly argued that office-motivated politicians implement policies preferred by the median voter. However, in models with more than two candidates the median voter’s preferred policy is not necessarily implemented. Moreover, politicians are elected to implement a bundle of policies. Even if the elected politicians’ bundle coincided with the median voter’s preferences on average, this does not have to hold for every single policy in the bundle. Direct democracy, on the other hand, allows voters to see their preferences implemented for single issues (Kirchgässner & Frey, 2012). There are two modeling choices associated with the counter proposal. First, in theory, politicians could design relatively far-reaching counter pro- posals that would not make the initiative withdraw. In this way, the counter proposal could serve to split the pro-reform votes to minimize the initiative’s winning probability. This would considerably increase the status quo’s win- ning probability. However, such counter proposals do not seem to be in line with the tradition of Swiss politics. Moreover, the parliament is legally re- quired to make a positive voting recommendation on behalf of the electorate. This justifies the inclusion of reputation costs if the status quo wins after a counter proposal was made.13 Second, politicians decide whether to make a direct (constitutional) or in- direct (statutory) counter proposal. Only the former has to be voted upon at the same time as the initiative while the latter can be implemented instanta- neously. The decision whether to issue a direct or indirect counter proposal is mostly determined by considerations on the appropriate structure of law. Therefore, it is usually not a strategic decision which form to propose, so that we do not explicitly model this distinction between the two different kinds of

13For example, in 1914 the Council of States wanted to issue a counter proposal against the initiative to introduce proportional representation, which was declined by the National Council since it seemed tactically motivated. Similarly, in 1953 the National Council opposed issuing a counter proposal because it was suspected to be proposed by some Members of Parliament for strategic reasons (Hofer, 2012).

58 counter proposals.14 In addition to issuing a counter proposal, politicians also have the option to support the initiative and issue a positive voting recom- mendation to the voters, which, however, happens very infrequently. While we do not include this as an alternative action of the politicians, it would be a straightforward extension of the model. Regarding the signature collection process, we assume that politicians ob- serve the state of signature collection at each point in time. While signatures are officially verified and counted only after they are filed with the Swiss Fed- eral Chancellery, it happens frequently that the media report about the state of the ongoing signature collection before it has finished.15 Non-public infor- mation channels among politicians most likely lead to even better information about the state of signature collection among members of the political sphere.

3.3.3 Subgame Perfect Equilibrium

We proceed by solving the sequential model recursively from the petitioners’ best withdrawal strategy, and continue with solving for the politician’s best counter proposal strategy given their knowledge about how petitioners react to it. We then present the petitioners’ optimal signatures collection strat- egy given their knowledge about the politicians’ equilibrium counter proposal strategy. We close with a brief description of the petitioners’ entry into the initiative process problem.

14For example, the initiative for more salary benefits for parents was withdrawn in 2006 in favor of an indirect counter proposal which regulated the matter at the statutory level, since such a detailed regulation was not appropriate for a constitutional article. In contrast, the initiative on complementary medicine was withdrawn in favor of a direct (constitutional) counter proposal and voted on in 2009. The topic of complementary medicine and the possibility to include it into mandatory health insurance was a new subject to the constitution, so it was appropriate to have a constitutional article regulating the matter. 15E.g., it was widely published that the initiative to introduce inheritance taxes, started by an ad-hoc committee in 2011, was running behind schedule in signature collection (Schaffner, 2012). Similarly, it was well-known that the conservative-liberal party did not have many signatures for the initiative to reduce bureaucracy at an early stage of collection and finally did not qualify in 2012 (Mäder, 2010). In contrast, the Swiss Popular Party was extremely quick in collecting already more than half the required signatures for an initiative about the expulsion of criminal foreigners, which was broadly covered in the media (Fontana, 2012).

59 Petitioners’ Optimal Withdrawal

Petitioners either withdraw the initiative (w =1), or they do not (w =0). The petitioners’ goal is to maximize their expected payoff over all potential voting outcomes. If a counter proposal is made, the maximization problem can be written as an optimal withdrawal rule:

⎧ B B ⎨⎪ 1 if πq(xq,xc)Uq + πc(xq,xc)Uc ∗ B B B w (xc)= ≥ πq(xq,xc,xi)Uq + πc(xq,xc,xi)Uc + πi(xq,xc,xi)Ui − κ ⎩⎪ 0 else

The optimal decision reflects which of the two following options yields the larger expected payoff: withdrawing the initiative such that the counter pro- posal competes only against the status quo, or not withdrawing such that the initiative competes against both the status quo and the counter proposal. In case of withdrawal, the counter proposal receives all votes that would other- wise have gone to the initiative, and thus the probability of reform increases. Also, campaigning costs do not accrue. However, at the same time petitioners forgo the chance that the initiative wins. Should both payoffs be equal, we assume that the initiative is withdrawn. The optimal withdrawal rule can be written as a cutoff function of the counter proposal as shown in the appendix.

Proposition 1 (Equilibrium Petitioners) There exists a cutoff x¯c such that the initiative is withdrawn for counter proposals greater or equal to the cutoff, and not withdrawn else: ∗ 1 if xc ≥ x¯c w (xc)= 0 if xc < x¯c

Intuitively, petitioners withdraw the initiative if the counter proposal is a satisfactory compromise. Otherwise they prefer to try their chances at ballot. From Proposition 1 (Equilibrium Petitioners) it follows that if politicians do not make a counter proposal, petitioners optimally withdraw the initiative if the campaigning costs exceed the expected payoff difference from having the initiative voted against the status quo and no ballot. If the initiative is withdrawn without counter proposal, the status quo wins unchallenged. B The expected payoff from not withdrawing the initiative is πq(xq,xi)Uq +

60 B πi(xq,xi)Ui − κ. It follows that petitioners prefer to withdraw the initiative B B if πi(xq,xi)(Ui − Uq ) <κ, i.e., if campaigning costs are relatively large compared to expected payoffs.

Numerical Example Cutoff Counter Proposal: Figure 3.4 displays the cutoff counter proposal for the numerical example assuming b =0.24 for the winning interval of a counter proposal without withdrawal, and k =0.1 for campaigning costs. We get x¯c ≈ 0.53 such that petitioners optimally withdraw the initiative for all xc ∈ [0.53, 0.85) as represented by the dashed line, but not below.

Figure 3.4: Example of Cutoff Counter Proposal x¯c

μ =0 xq xi μ¯ =1

Note: Median distribution μ ∼ U[0, 1], status quo xq =0.35, initiative xi =0.85, counter proposal winning interval b =0.24, and campaigning costs k =0.1. The cutoff is at x¯c ≈ 0.53, such that the initiative is withdrawn when the counter proposal is located on dashed line.

Politicians

Politicians either make a counter proposal xc, or refrain from compromising, xc = ∅, taking into account that petitioners withdraw the initiative if the counter proposal is at or above the cutoff x¯c. The politicians’ expected payoff maximization is a two-stage problem. First, they consider which point xc ∈ (xq,xi) to propose optimally if they were obliged to make a counter proposal. Second, they decide whether to make this optimal counter proposal, or make none at all. The first-stage problem is a choice between a vote of the status quo against a counter proposal at or above the cutoff x¯c, and a vote with all three alternatives competing against each other. ⎧ A A ⎨⎪ xc ≥ x¯c if πq(xq,xc)(Uq − r)+πc(xq,xc)Uc ∗ A A A xc = ≥ πq(xq,xc,xi)(Uq − r)+πc(xq,xc,xi)Uc + πi(xq,xc,xi)Ui ⎩⎪ xc < x¯c else

61 We show in the appendix that a cutoff r¯ of reputation costs characterizes their best strategy: Proposition 2 (Counter Proposal Politicians) 16 If politicians were obliged to make a counter proposal, they would optimally choose ∗ x¯c if r ≤ r¯ xc (r)= xq + , → 0 if r > r¯

Depending on the reputation costs, politicians either prefer a small compro-  mise xc = xq + , → 0 close to the status quo after which the initiative is not withdrawn, or the cutoff counter proposal x¯c after which petitioners withdraw the initiative. We assume that politicians propose the cutoff when the equation holds with equality. Politicians’ second-stage optimization problem requires the decision whether to make a counter proposal as laid out in Proposition 2 (Counter Proposal Politicians) and pay costs c, or not to compromise and have a vote between initiative and status quo. Politicians face the tradeoff that not making a counter proposal yields either a high utility from the status quo, or the risk of the initiative winning with lower utility. In contrast, a counter proposal gives positive winning probability to a policy option with medium-sized utility but costs c accrue. The equilibrium strategy depends on cutoffs of the belief regarding the initiative’s winning probability represented by the lower bound of the me- dian distribution. Above the cutoff the initiative is too likely to win and compromising is the better option, and vice versa below the cutoff (proof in appendix). If beliefs equal the cutoff, we assume that politicians make a counter proposal. Proposition 3 (Equilibrium Politicians) If r ≤ r¯ (r>r¯), there exists a  median distribution cutoff μc (μc ) below which no counter proposal is made, and above which a counter proposal is optimally made: ⎧ c ⎨⎪ x¯c if r ≤ r¯ ∧ μ ≥ μ ∗ c xc (r, μ)= xq + , → 0 if r > r¯ ∧ μ ≥ μ ⎩⎪ ∅ else

16 This holds if the status quo wins more likely with xc =¯xc than with xc = xq + ,  → 0. ∗ Otherwise the condition is reversed and xc (r)=¯xc if r>r¯ and xq + ,  → 0 else.

62 It follows that if the cutoff lies in the domain of μ ∈ [0,xq] and the initiative wins with sufficiently high probability, politicians make a counter proposal.17

Numerical Example Median Distribution Cutoff: Keeping all other values constant, we assume that the costs of making a counter proposal are c =0.04, and the reputation costs are r =0.2. Then r

Figure 3.5: Example of Initiative Type and Winning Probability c 1 μ 2 (xq + xi)

0 xq x¯c xi μ¯ =1

Note: Median distribution upper bound μ¯ =1, status quo xq =0.35, initiative xi = 0.85, counter proposal winning probability b =0.24, campaigning costs k =0.2, cutoff for withdrawal at x¯c =0.53, costs of making counter proposals c =0.04, politicians’ reputation costs r =0.2. The initiative gets no counter proposal for beliefs μ ∈ [0, 0.14) (thick line), and a cutoff counter proposal for μ ∈ [0.14, 0.35] (dashed line).

Optimal Signature Collection

Petitioners choose continuously whether to collect signatures, or to stop. They have two goals: first, they want to qualify the initiative for ballot, and thus it is never optimal to stop collection before the legal threshold s¯. Second, they know that collection reveals information about the initiative’s winning probability and affects the politicians’ optimal strategy. The longer they collect, the more certain are the players about the winning probability.

17If the cutoff potentially lies outside the domain of μ, the politicians’ best strategy is independent of μ.

63 The outcome of the signature collection has no impact on the counter c c proposal if the cutoffs μ and μ are outside the domain of μ ∈ [0,xq] because politicians’ optimal strategy is then independent of μ, and they optimally stop collecting at the qualifying threshold. In what follows, we focus on the more interesting case where the politician’s strategy depends on the value of μ. If reputation costs are high, r>r¯, politicians either make the small counter proposal or none. Petitioners are indifferent between the two out- comes (cf. appendix for the proof), and therefore they have no incentive to collect more than the signature requirement. If r ≤ r¯ politicians either pro- pose the cutoff counter proposal or none, and petitioners always prefer the former to the latter (cf. appendix for the proof). They take into account that politicians’ only make the counter proposal if the belief about the initiative’s winning probability is relatively high.

Consider Figure 3.6 with time on the x-axis and belief μt on the y-axis: if the petitioners collect s¯ signatures and beliefs are above μc, they receive the cutoff counter proposal. It follows that petitioners optimally do not over-

Figure 3.6: Politicians’ Best Strategy

lower bound

belief μt

xq

μ0

∗ xc =¯xc

μc ∗ xc = ∅

0 tc collection time t

Note: The y-axis displays petitioners’ belief about the median distribution’s lower bound and the x-axis time. The figure shows the best action played by politicians depending on collection time. If no lump sum occurs and collection stops above μc, politicians make a counter proposal. If not lump sum occurs and collection stops below μc, they make none.

64 collect if either initial belief μ0 is sufficiently high, or they receive a lumps sum and are a high type with certainty. The more interesting case occurs if beliefs fall below μc once the signature requirement is reached and no lump sum has been observed. If they stop collecting, they get no counter proposal, and no further collection costs accrue. Alternatively, they continue at costs γ, and get a cutoff counter proposal with the probability of observing a high lump sum λpt (the product of the probability of being a high type and the lump sum sh arriving). Denote the belief when petitioners are indifferent between these options by μ. For all beliefs above this threshold, beliefs to be a high type are high enough such that continuing collection is more beneficial than stopping, and vice versa. When beliefs equal the threshold, we assume that collection is preferred.

Proposition 4 (Optimal Collection) Petitioners optimally continue col- lection until they qualify the initiative. If reputation costs are small, r ≤ r¯, c the probability cutoff is in the relevant domain (μ ∈ [0,xq]) and beliefs are ≥ sufficiently high (μt μ), petitioners optimally over-collect.

In sum, petitioners over-collect only if they believe to be high types with sufficiently high probability even though no lump sum has occurred.

Entry Decision

For completeness, we briefly discuss the petitioners decision to enter the quali- fication stage. If politicians’ reputation costs are high, r>r¯, petitioners have the same expected payoff regardless of the signature collection, and expect type-weighted costs for collecting the signature requirement.18 If reputation costs are low, r ≤ r¯, petitioners get a type- and lump sum probability-weighted payoff of receiving the cutoff counter proposal or none. Expected collection costs take into account the probability of over-collection. The expected payoff net of expected signature collection costs and ex- pected campaigning costs must be positive for petitioners to start collection.   18 s/¯ (sl+λsh) s/s¯ l In more detail, E[Γ] = p0 0 γdt+(1− p0) 0 γdtwhere (sl + λsh) and sl are the expected number of signatures per period for high and low types respectively, and s/¯ (sl + λsh) and s/s¯ l is the expected time to collect s¯ signatures. Integrating over time yields E[Γ] = γs¯[p0/(sl + λsh)+(1− p0)/sl].

65 3.3.4 Equilibrium Analysis

In this section we first develop the main hypotheses regarding the probability of amending the status quo and the impact of signature collection. Then, we discuss how exogenous changes in some of the model variables affect equilib- rium outcomes to formulate testable implications. In particular, we consider how decreasing signature collection costs, increasing the signature require- ment, a collection time restriction, and a change in the voting rule including a tie-breaking question between the initiative and counter proposal affect the model equilibrium. These changes are of special interest since they correspond to institutional changes described in Section 3.2.2, and can therefore be used to empirically test model predictions. We focus particularly on the number of initiatives, the probability of receiving a counter proposal, and the probability of reforming the status quo.

Probability of Amending the Status Quo and Signature Collection

Proposing an initiative amends the status quo either by directly beating it at the vote, or provoking a successful counter proposal. We have the probability of reform Prob(xq amended)=1− Prob(xq wins) which depends on the status quo’s winning probability:

1 2 (xq + xi) − μ πq(xq,xi)= μ¯ − μ 1 2 (xq +¯xc) − μ πq(xq, x¯c)= μ¯ − μ 1 2 (xc + xi) − μ − b πq(xq,xc,xi)= μ¯ − μ

The cutoff counter proposal inherits all votes from the initiative when it is withdrawn. Moreover, it is closer to the status quo than the initiative, and thus appeals to even more voters. It follows that the status quo is less likely to win in a direct vote against the cutoff counter proposal than against the initiative (πq(xq, x¯c) <πq(xq,xi)). The difference in the status quo’s winning probabilities in the case when the initiative is not withdrawn after the counter proposal and no counter proposal (πq(xq,xc,xi) − πq(xq,xi)) is positive. The

66 status quo is more likely to be defeated when voted against the counter pro- posal and the initiative at the same time than just against the initiative. Intuitively, the compromise appeals to moderate voters and takes away votes from the status quo. This is summarized in the following hypothesis:

Hypothesis 1 (Status Quo) After a counter proposal the status quo is more likely to be amended than when the initiative is voted exclusively against the status quo.

The model shows that signature collection is an important information source when politicians decide whether to propose a counter proposal or not. Only initiatives considered a threat to the status quo receive a counter proposal.

Hypothesis 2 (Signatures) Conditional on not over-collecting, fast collec- tors are more likely to get a counter proposal than slow collectors. If initiatives ∗ over-collect, they are more likely to receive a counter proposal xc =¯xc and withdraw the initiative in case they are a high type. There is no incentive to over-collect if petitioners expect to receive a counter proposal after which they do not withdraw the initiative.

Depending on initial beliefs and the probability of receiving lump sums, po- tentially all high types receive a counter proposal after collecting the required threshold s¯. Over-collection only becomes relevant if signature collection is slow but petitioners believe the odds of being a high type are sufficiently high. However, they may not receive a counter proposal despite of over-collection.

Costs of Signature Collection

When signature collection costs decrease, qualifying an initiative becomes rel- atively more attractive such that the number of initiatives entering the quali- fication stage increases, and the mix of initiatives changes: the new initiatives have relatively low expected payoffs due to low expected winning probabilities of the initiative. Consequently, in expectation decreasing collection costs in- creases the share of low type initiatives that are less likely to receive a counter proposal. From Hypothesis 1 (Status Quo) we know that a lower probability of counter proposal decreases the probability of reform. We should expect the status quo to win more often.

67 Hypothesis 3 (Signature Collection Costs) Lower signature collection costs γ lead to an increase in the number of initiatives entering the qualification stage. In expectation, the share of low type initiatives increases. The proba- bility of receiving a counter proposal and amending the status quo decreases.

Collection Time Restriction

In the model, collection time is not constrained. With a collection time re- striction tmax = T initiatives do not qualify if petitioners are unable to collect the signature requirement sufficiently quickly. In expectation, low types are the slower collectors such that mainly low types with correspondingly low winning probabilities are less likely to qualify for ballot. Since the expected payoff from not qualifying an initiative is negative - the status quo remains and collection costs accrue - we should observe fewer low types entering the qualification stage, and some initiatives not qualifying even though they be- gin collecting. This leads to a higher share of counter proposals and a higher probability of reform.

Hypothesis 4 (Collection Time Restriction) Due to restricted collection time some initiatives do not qualify, and also the overall number of initiatives starting signature collection decreases. On average, the share of counter pro- posals increases, and reform becomes more likely.

Signature Requirement

The signature requirement affects the time petitioners are obliged to collect signatures so the total collection costs increase. Petitioners with low expected payoffs, like low types with low winning probabilities, might therefore choose not to begin with signature collection in the first place. In expectation the number of initiatives decreases and the share of high types increases. Intu- itively, with a higher signature requirement politicians observe the signature collection process for a longer time and wait if a high lump sum is realized. Eventually, they are more certain about the initiative’s type. Since high

68 types are more likely to receive a counter proposal, the share of initiatives with counter proposals increases, and reform becomes more likely.

Hypothesis 5 (Signature Requirement) Increasing the signature require- ment s¯ renders qualification more costly. On average fewer initiatives begin collection, the shares of initiatives receiving a counter proposal as well as the probability of amending the status quo increase.

Both the increase in s¯ and the time restriction predict a reduction in the num- ber of initiatives and higher shares of high type initiatives leading to relatively more counter proposals and more reforms. However, only the collection time restriction predicts some initiatives not to qualify.

Voting Rule: Tie-Breaking Question

When initiative and counter proposal are voted simultaneously and there is a tie-breaking question, voters have to make three choices: xq vs. xi, xq vs. xc, and xc vs. xi (tie-breaking question). If either initiative or counter proposal receives more than half of the votes against the status quo, it wins. In case both receive more than half of the votes versus the status quo, the tie-breaking vote of xc against xi is decisive. In all remaining cases the status quo wins. The new voting rule implies different winning probabilities of the status quo and counter proposal in comparison to the baseline model in case all the policy alternatives are voted simultaneously (cf. proof in the appendix).

Proposition 5 (Tie-Break) With a tie-breaking question, the counter pro- posal wins more likely in a vote against both the status quo and the initiative than before. The status quo is less likely to win, and the initiative wins with the same probability as before.

The main difference to the baseline model is that the counter proposal wins 1 for all realizations of the median voter between the midpoints 2 (xq + xc) and 1 2 (xc + xi) (b was strictly smaller before) at the expense of the status quo. The petitioners’ payoff from receiving a counter proposal below the cutoff increases with its winning probability. By Proposition 1 (Equilibrium Pe- titioners), to make the petitioners indifferent between withdrawing and not

69 withdrawing the initiative, the cutoff counter proposal has to move closer to the initiative to yield a higher utility. However, the further away the cutoff counter proposal from the status quo, the smaller its winning probability. Politicians’ expected payoff from making a counter proposal after which the initiative is not withdrawn increases due to its higher winning probability. The change in the expected payoff from counter proposals after which the initiative is withdrawn is, however, ambiguous. Petitioners’ expected payoff from receiving the cutoff counter proposal increases. But their entry decision also depends on how likely they are to receive one in case they start initiative qualification. The equilibrium effect on the probability of receiving a counter proposal is consequently ambiguous. The testable implications are summarized in the following hypothesis:

Hypothesis 6 (Tie-Break) In a voting regime including a tie-breaking ques- tion between the initiative and the counter proposal, the counter proposal’s winning probability without initiative withdrawal increases while the winning probability of the status quo decreases. If the initiative is withdrawn after the counter proposal, the counter proposal’s winning probability decreases.

3.4 Data and Empirical Strategy

3.4.1 Data

We collected a dataset of all Swiss initiatives at the federal level that have suc- cessfully gathered the legally required number of 50,000 or 100,000 signatures (since 1978) respectively. It covers the entire time span from the introduction of the initiative in 1891 to 2010. In total, our dataset comprises 249 initia- tives at federal level.19 Table 3.2 gives an overview of the number of qualified initiatives and winning policies per institutional period.

19We omit four initiatives that had been declared invalid by the parliament and were not fur- ther discussed thereon. We also omit two initiatives that collected the necessary number of signatures, but were postponed by the government and parliament for an exceptionally long period of time until they became irrelevant, and finally written off by the parliament without further discussion (cf. Swiss Federal Chancellery, 2013). Furthermore, we omit

70 Table 3.2: Descriptives: Initiatives per Period and Outcomes Total Status Counter Total quo proposal Initiative Period Time initiatives wins wins wins 1 1891-1927 23 15 2 6 2 1928-1950 31 18 12 1 3 1951-1970 35 16 19 0 4 1971-1977 26 22 4 0 5 1978-1986 23 13 9 1 6 1987-2010 111 86 15 10 Total 249 170 61 18 Note: This table summarizes the number of initiatives belonging to each institutional period as well as the number of initiatives for which the status quo, the counter proposal or the initiative itself was victo- rious.

There were also 77 popular initiatives that did not accomplish collecting the required number of signatures (Swiss Federal Chancellery, 2013). We collected the titles, the dates of the beginning of signature collection and the date of the official statement of non-qualification for these initiatives. We calculate the number of initiatives successfully submitted per year, the number of initiatives not qualifying per year, and their sum initiatives p.a. We collected information about whether the initiative was put to the bal- lot or withdrawn, whether there was a direct, indirect or no counter proposal at all, and whether the initiative or a counter proposal finally came into force after popular vote.20 Let σpol ∈{CP,nCP} denote the politicians’ de- cision about the counter proposal where CP stands for a counter proposal

three initiatives that were withdrawn after a counter proposal to a similar, but different initiative had been made. The three cases are the initiative “Special justice in cases of urgency (Notrecht und Dringlichkeit)” (withdrawn 1940) and two initiatives on old age insurance ("Initiative for an entirely public-based pension scheme (Für die Einführung einer Volkspension)" (withdrawn in 1974) and “Further expansion of insurances against old age and invalidity (Für einen weiteren Ausbau der Alters-, Hinterlassenen- und Invali- denversicherung)” (withdrawn in 1968). We omit these initiatives because the reaction of both the parliament and the petitioner committee could not be clearly allocated to one initiative and its signatures and collection time, but was mainly driven by other initiatives on similar subjects raised at the same time (Hofer, 2012). 20As explained, indirect counter proposals may be subject to a facultative referendum but this rarely occurs. However, in the few cases in which this happened, we took into account the outcome of the referendum vote.

71 and nCP indicates that none has been made. Analogously, σpet ∈{w, nw} reflects whether petitioners withdrew the initiative (w) or not (nw). From this, we have four mutually exclusive combinations of politicians’ and peti- tioners’ choices which are subsumed in the following observed profiles: σ = (σpol,σpet) ∈{(CPnw), (CPw), (nCP nw), (nCP w)}. The profile frequencies are 27, 54, 145, and 23 respectively. Most frequently politicians make no counter proposal and the initiative is not withdrawn. In Switzerland all signatures are subject to verification in contrast to ran- dom sampling procedures common in many US states. We assembled the number of collected, valid and invalid signatures. As it is of particular impor- tance for the model, let over-collection = signatures − s¯ denote the number of valid signatures that was collected above the legal threshold s¯. Time is the signature collection time measured from collection begin to submitting the signatures. For four initiatives collection time is not available. As all four belong to periods before the collection time restriction, we assume the mean collection time before 1978 for them. If the initiative or a counter proposal was voted upon, we collected data on the winner of the ballot. We recorded all dates concerning the political process of the initiative, which allows us to match the initiatives to the correct institutional periods. In the appendix we provide a detailed description of the rules to allocate the initiatives to the six institutional periods. According to the Swiss Federal Chancellery, several of the withdrawn ini- tiatives without a counter proposal were withdrawn for “other reasons” (as opposed to withdrawal after a counter proposal). We collected complemen- tary data on the initiative from the detailed descriptions of each initiative in Hofer (2012) and Rohner (2012). They provide additional insight into smaller policy concessions by the parliament which were not explicitly registered as counter proposals, but still represent some policy change induced by the ini- tiative. We coded all initiatives of profile (nCP w) that yielded minor policy changes according to Rohner (2012) and Hofer (2012) as having a de facto counter proposal and thus profile (CPw). In total, 10 of the 23 withdrawn initiatives without counter proposal (nCP w) provoked some de facto policy changes. We run our estimates with the officially coded data, but repeat all estimates including the de facto counter proposals. We collected initiative-related control variables. The first control reflects

72 the petitioners’ identity. We distinguish two types of committees according to their experience and resources so that committee =1if a committee is experienced in raising initiatives and can rely on large financial resources. An experienced committee is defined as having already raised at least one initiative before, and inexperienced when raising an initiative for the first time. A committee is considered as powerful if it is formed by a large party (more than 10% vote share in the last parliamentary election)21 or a large, well-established association (many members or large material interest). For example, when the left party qualified its first initiative in 1893, it is coded as inexperienced and powerful. For its next initiative in 1899, it already had gained experience in the initiative process, and is thus coded as experienced and powerful. In total, 46.59% of the committees are experienced, and 37.75% are powerful. Combined, there is a total of 31.33% committees that are both experienced and powerful. As a second control variable we attribute an economic, ideological or state- related topic to each initiative.22 The dummy topic is 1 if the topic is eco- nomic, and value 0 subsumes the other two topics. 42.97% of the initia- tives have economic topics. The remaining ones are split among ideological (48.59%) and state-related topics (8.43%). We code initiative titles according to its wording as provided by the Swiss Federal Chancellery and documented in official Federal Announcements, and the information on the policy context in Hofer (2012). The initiative can be either a fully formulated constitutional article or a general suggestion which is specified by the parliament later on. To capture potential differences between these two categories, we assign a dummy form that is 1 for all initiatives that are a general suggestion, and 0 else. Only 8

21Parties proposing initiatives should not be confused with the unitary actor “politicians” in the model: they constitute only a small part of the parliament and do not hold a majority which would have allowed it to pursue the policy issue through the standard parliamentary procedure. 22For example, the initiative for a tax on capital gains (voted in 2001) is a good example for an initiative that focuses primarily on monetary redistribution of economic assets. On the other hand, the initiative “for mother and child - protection of the unborn child and help to the mother in need” against abortion (voted in 2002) was raised for non- economic, cultural or ideological reasons. A similar example is the initiatives raised against nuclear power plants, such as the initiative “electricity without nuclear energy” (voted in 2003). Finally, initiatives on state-related topics include the initiatives to switch from majoritarian to proportional representation (for example, “proportional election of the National Council” voted in 1918) or the proposal to allow popular referendums in the case of international treaties (voted in 1977).

73 Table 3.3: Descriptives Mean Std. Dev. Min Max over-collection 38.496 51.220 0.038 334.760 time 362.9 188.8 34 1173 speed 0.0476 0.0546 0.0052 0.5344 committee 0.3133 0.4648 0 1 topic 0.4297 0.4960 0 1 form 0.0321 0.1767 0 1 indirect 0.1807 0.3856 0 1 de facto 0.2329 0.4236 0 1 Note: This table provides summary statistics for the complete sample. Over-collection is defined as valid signatures−legal threshold in thousands. Timeis measured in days from collection begin to submitting the signatures. Speed is the number of signatures collected per day. Dummy committee is one if the initiating com- mittee is experienced and powerful. If a topic is econom- ically framed (and not ideologically or state-related), the dummy topic is 1. Form is 1 if the initiative is a gen- eral suggestion. Indirect and defacto counter proposal takes value 1 if an initiative received an indirect or de facto counter proposal respectively. initiatives (3.21%) are in the form of a general suggestion. We also control for whether the counter proposal was a direct (i.e., con- stitutional) or indirect one. The dummy indirect reflects if an initiative had an indirect counter proposal. Of all initiatives that received counter propos- als, they are relatively evenly spread between indirect (51.85%) and direct (48.15%) counter proposals. The direct counter proposals are also evenly spread among profiles (CPw) (55.55%) and (CPnw) (44.44%). However, ini- tiatives tend to be withdrawn after indirect counter proposals as 80.95% are part of profile (CPw) but only 19.05% of profile (CPnw). To sum up, we have initiative-specific control variables committee, topic, form, and indirect.Ta- ble 3.3 provides descriptive statistics, and we give a detailed overview of the data sources in Table 3.11 in the appendix.

74 3.4.2 Empirical Strategy

For the estimation strategy, we rely on the link between observations and their theoretical counterparts. E.g., we know that the initiative is withdrawn after a counter proposal at the cutoff. It follows that if we observe profile (CPw), the cutoff counter proposal was made. Similarly, observing profile (CPnw) means that a small counter proposal was made. Treating the model predictions seriously, means that we assume away other factors influencing the politicians’ choice to propose compromises, and petitioners’ readiness to accept them.  xc = xq + , → 0 is the only counter proposal politicians issue in this model that does not make petitioners withdraw the initiative. Empirically, this does not mean that we should expect all counter proposals without with- drawal to be minor political concessions. Rather, counter proposals have to offer substantial compromise to be considered a real alternative to the status quo. We could adapt the model by requiring a minimal distance δ between ∗ the xq and xc. Politicians would then optimally choose xc = xq + δ if the ∗ expected payoff from doing so was larger than from proposing xc =¯xc. While adding this feature to the model would make it more realistic, empirically it would not be distinguishable from the simpler specification. We index the initiatives by i and denote their institutional period by p.For the empirical specification, we introduce a one-time idiosyncratic signature shock ui with E[ui]=0. It realizes once s¯ signatures are collected and thus generates variations in signatures and collection time of low-type initiatives. Otherwise the model would predict the same collection time for initiatives without lump sums collecting precisely the signature requirement. The first object of interest is the impact of counter proposals on the proba- bility of amending the status quo. The binary dependent variable amendedip is 1 if the status quo was amended and 0 else. The main explanatory variables are dummies for the profiles (CPw) and (CPnw). Profile (nCP nw) serves as reference category. We exclude initiatives of the fourth profile (nCP w) which is a perfect predictor for initiative failure because the status quo then always wins. Vector zip controls for initiative-specific variables as explained in the data section. We also add institutional period fixed effects ξp to account for period-specific changes in the regulation regarding the initiative process.

75 Versions of the following regression equation are estimated:

amendedip = α + β1(CPw)ip + β2(CPnw)ip + β3zip + ξp + ip (3.1)

Coefficients β1 and β2 reflect the impact of receiving a counter proposal on the probability of amending the status quo. α is the intercept and β3 is the vector of coefficients of initiative controls. Since the dependent variable is a dummy variable, a probit estimator is the most appropriate estimation method. Then, ip is a normally distributed error term with zero mean. Overall, the parliament has supported the initiative five times by issuing an official voting recommendation to the citizens asking to accept it at ballot. In all five cases, the initiative was subsequently accepted such that parliament support is a perfect predictor of status quo amendment. For this reason, we also estimate a specification excluding these five observations. Hypothesis 2 (Signatures) postulates that over-collection has a positive effect on the probability of receiving a cutoff counter proposal, but none on the small counter proposal. We use two distinct dummies CPip as dependent variables for the profiles (CPw) and (CPnw). We separate the two profiles because the model makes different predictions for each of them. We use both profiles (nCP w) and (nCP nw) as a reference category. over-collection is the main explanatory variable. We estimate several variants of the following esti- mation equation and we also include period fixed effects ξp in the regressions:

CPip = α + β1over-collectionip + ξp + ip (3.2)

For the estimation of (3.1) and (3.2), we use the complete dataset of quali- fied initiatives. The data encompass almost 120 years such that time effect potentially play a role. Institutional period fixed effects account for the main institutional changes. Moreover, technology improvements and development of the media affect the initiative and signature collection process over time (Matsusaka, 1992). Possibly political developments influence the initiative process as well, so for robustness we include linear and quadratic time trends in the regressions. Though we include various controls and fixed effects in the regressions, we cannot conclusively allege to identify a causal effect. Po- tentially we omit individual initiative characteristics, voter preferences or the political context.

76 To test Hypotheses 3 to 6 about signature collection costs, collection time restriction, signature requirement, and tie-breaking question, we use the fol- lowing four institutional changes corresponding to model parameters. First, when women were enfranchised at federal level on 16 March 1971, they were also given the right to sign initiative petitions such that the pool of poten- tial signers roughly doubled while leaving the signature requirement constant. Therefore, signature collection costs decreased discontinuously on that day.23 This allows us to test Hypothesis 3 (Signature Collection Costs) regarding a decrease in the signature collection cost parameter γ. Two important insti- tutional changes occurred almost simultaneously in 1978: collection time was restricted to 18 months from previously no time constraint, and the signature requirement was increased from 50,000 to 100,000 signatures. We use this second and third major change to the initiative process to test Hypotheses 4 (Collection Time Restriction) and 5 (Signature Requirement) respectively. It is straightforward that the time restriction introduces a cap for the signature collection time at tmax = T to the model, and the raise in the signature re- quirement corresponds to an increase in s¯. Fourth, Hypothesis 6 (Tie-Break) deals with the change of the voting regime to include a tie-breaking question. This exactly mirrors the last major institutional change which can conse- quently be used directly to test this hypothesis. Each hypothesis makes predictions about the effect of the institutional change on the number of initiatives, the probability of receiving a counter proposal, and the probability of amending the status quo. Let changeip denote a binary variable taking value 1 after the institutional change, and 0 before. The first estimation equation regarding the number of initiatives per year with intercept α and error term ip is:

initiativesip = α + βchangeip + ip (3.3)

Similarly to above we use two dummies for the profiles including counter pro- posals (CPw) and (CPnw) to estimate the effect of the institutional change on the probability to receive a counter proposal. Initiatives without counter

23Even if directly after their enfranchisement women were supposedly politically less active than their male counterparts (cf. Lott and Kenny (1999) for some evidence from the USA), the change in the electorate was large enough to expect an impact on the initiative process.

77 proposals serve as the reference category.

CPip = α + β1changeip + ip (3.4)

We estimate the effect of institutional change on the probability of reform with the regression

amendedip = α + β1changeip + ip (3.5)

We estimate the effect of receiving one of the two counter proposals (CPw) or (CPnw) on the probability of amending the status quo while controlling for institutional change. We drop observations with the profile (nCP w) and use (nCP nw) as a reference category. We estimate an equation including the interaction between the profiles and the change dummy. The coefficients β4 and β5 of the interaction terms are of major interest.

amendedip = α + β1(CPw)ip + β2(CPnw)ip + β3 changeip (3.6)

+β4(CPw)ip × changeip + β5(CPnw)ip × changeip + ip

For all estimates, we reduce the sample to initiatives in the two periods adja- cent to the institutional change. This weakens the influence of time trends and minor institutional changes, however, it does not completely exclude them. The coefficients identify the average change in the dependent variable after the institutional change. To estimate a causal effect, we obviously lack an untreated control group. For robustness, we estimate versions with initiatives in smaller, symmetric bins around the institutional change at the price of re- ducing the number of observations. We also run regressions including linear and quadratic time trends. A further problem regarding identification is that two of the major in- stitutional changes, the collection time restriction increase in signature re- quirement, happened simultaneously. Moreover, hypotheses regarding these changes were very similar which makes it even more difficult to disentangle both changes. Potentially, institutional changes are not necessarily exogenous to the initiative process. E.g., the introduction of female voting rights not only reduces signature collection costs, but also changes the composition of the electorate. This might affect the kind of initiatives that come to ballot

78 Table 3.4: Overview of Hypotheses and Relevant Periods H Affected Variables Change Institutional Periods 1 Winning probability - All periods πa(·), a ∈{q, c, i} -

2 Signatures s, - All periods collection time t -

3 Decrease signature Female Voting 3: Formal Withdrawal & collection costs γ 4: Female Voting

4 Maximum collection 18 months cap 4: Female Voting & time T 5: Collection Threshold

5 Signature 100,000 4: Female Voting & requirement s¯ 5: Collection Threshold

6 Voting regime Tie-breaking 5: Collection Threshold & question 6: Tie-Break Note: This table relates the hypothesis and affected model parameter to the institutional change as well as the periods used for the estimation. and voter preferences in general. The tie-breaking question was introduced by referendum vote. Perhaps it reflects a general change in the attitudes to- wards more give-and-take. Arguably the most likely exogenous change to the process was the increase in the signature requirement, since it was a belated adjustment to female franchise. Table 3.4 summarizes the hypotheses and relates them to their institutional periods.

3.5 Results

3.5.1 Probability of Amending the Status Quo

Hypothesis 1 states that change is most likely after a counter proposal. Figure 3.7 shows how the profiles evolve over time measured by the share of all initia- tives per period. With the exception of period 3 (1951-1970), not withdrawn initiatives without a counter proposal make up the largest share of initiatives. It varies roughly between 40 and 90%.

79 For the first three decades there are almost no counter proposal with ini- tiative withdrawal, (CPw). The share increases considerable to roughly 50% of all initiatives by period 3 (1951-1970) and afterwards fluctuates between 10 and 30% per period. Profiles (CPw) and (nCP nw) are almost mirror functions of one another. Thus, most of the tradeoff politicians are facing is between these two profiles. In our model, this corresponds to a case in which reputation costs are low (r ≤ r¯) and politicians consider whether to issue the cutoff counter proposal or none. Comparing the share of counter proposals after which the initiative was withdrawn and the share of initiatives that reformed the status quo (thick solid line) shows a parallel trend and strong correlation of the graphs: after such counter proposals the status quo is extremely unlikely to prevail. The two remaining profiles show little variation over time. Profile (CPnw) makes up between 10 and 20% of all initiatives per period, and shows a slightly increasing trend between 1928 and 1986. The share drops markedly from its

Figure 3.7: Period Mean Shares of Observed Profiles 1 .8 .6 .4 .2 0 1 2 3 4 5 6 Period Mean status quo amended Mean (CPnw) Mean (CPw) Mean (nCPnw)

Note: The figure shows mean shares of observed profiles per institutional period. In each period shares add up to 1. Additionally, the solid line shows the share of initiatives leading to the amendment of the status quo per institutional period. Periods: 1 “Early Period” (1891-1927), 2 “De Facto Withdrawal” (1928-1950), 3 “Formal Withdrawal” (1951-1970), 4 “Female Voting” (1971-1977), 5 “Collection Restrictions” (1978-1986), 6 “Tie-Break” (1987- 2010).

80 20% high to below 10% after 1987. Profile (nCP w) (not displayed) has no time trend, and the share fluctuates between 0 and 20% of all initiatives. This suggests that for profile (nCP w) the institutional setting is not important and rather reflects the size of campaigning costs. The marginal effects from probit regression of reform dummies on counter proposals are reported in Table 3.5. The sample consists of all 226 initiatives with profiles other than (nCP w). In column 8, we also report the results ex- cluding five initiatives that earned a positive voting recommendation from the parliament. In all but the first specification period fixed effects are included. The regression results confirm the tendencies from the descriptive analysis above. In all specifications the dummies for profiles including counter propos- als are positive and highly significant. Having a counter proposal after which the initiative is withdrawn increases the probability of amending the status quo by 50.1 to 53.4 percentage points in columns 1 to 5. The effect of receiving a counter proposal after which the initiative is not withdrawn is smaller reach- ing approximately 35 percentage points in columns 1 to 5. Once we control for indirect counter proposal in columns 6 to 8, the effect from both profiles, and especially from (CPw), decreases considerably (as expected) since most indirect counter proposals are subsumed in profile (CPw). Dropping the five initiatives which earned parliamentary support slightly reduces the marginal effects of counter proposals. Powerful and experienced initiative committees are significantly less likely to amend the status quo (column 3), but the effect becomes insignificant once we drop observations receiving parliamentary support in the last specifica- tion. All of them do not have powerful and experienced committees but were accepted at ballot. Initiatives with an economic topic are on average sig- nificantly less successful in changing the status quo than initiatives dealing with ideological or state-related issues. Potentially, there is less uncertainty about the median voter’s position on economic issues through the availability of other proxies. Descriptive evidence shows that economically-framed initia- tives are less likely to receive a counter proposal, which might partly explain the negative coefficient. The form of the initiative (formulated initiative vs. general suggestion) has no significant effect throughout the specifications. Re- ceiving an indirect counter proposal increases the probability of amending the status quo significantly by 21.2 percentage points. All findings are robust to

81 Table 3.5: Probability of Amending the Status Quo (1) (2) (3) (4) (5) (6) (7) (8) (CPw) 0.505*** 0.510*** 0.507*** 0.534*** 0.509*** 0.390*** 0.410*** 0.377*** (0.028) (0.032) (0.032) (0.034) (0.032) (0.046) (0.046) (0.044) (CPnw) 0.327** 0.361*** 0.355*** 0.375*** 0.359*** 0.260*** 0.271*** 0.258*** (0.048) (0.052) (0.052) (0.052) (0.052) (0.058) (0.058) (0.052) Committee -0.113** -0.086* -0.051 (-0.048) (0.045) (0.042) Topic -0.120** -0.100** -0.096** (0.051) (0.048) (0.047) Form -0.077 -0.113 -0.088 (0.208) (0.231) (0.210) 82 Indirect 0.212*** 0.194*** 0.168*** (0.067) (0.064) (0.056) Pseudo 0.435 0.471 0.491 0.492 0.471 0.507 0.539 0.584 R2 Obs. 226 226 226 226 226 226 226 221 Period no yes yes yes yes yes yes yes FE Note: *** p<0.01, ** p<0.05, * p<0.1. Probit regressions, marginal effects reported. Standard errors in brackets. Dependent variable: dummy is 1 if status quo was amended. Observations with profile (nCP w) are dropped because they predict initiative failure perfectly. In column 8 observations with parliamentary support are dropped. Committee =1for experienced and powerful petitioners. Topic =1for economically framed initiatives. Form =1for general suggestions. Indirect =1if counter proposal made was not at constitutional level. Period fixed effects are based on 6 institutional periods. Table 3.6: Winning Policy by Profile (CPnw) (CPw) (nCPnw) (nCPw) Total Q wins 10 6 131 23 170 (37.04) (11.11) (90.34) (100) (68.27) CP wins 13 48 - - 61 (48.15) (88.89) - - (24.5) I wins 4 - 14 - 18 (14.81) - (9.66) - (7.23) Total 27 54 145 23 249 Note: The table shows the number of initiatives by winning policy and profile. Percentages are in brackets. accounting for de facto counter proposals, linear and quadratic time trends (results available on request). As descriptive evidence, we report the number of initiatives and percent- ages per observed profile by winning policy in Table 3.6. In 68.3% of the cases the status quo prevails. Initiative success mainly runs through counter proposals which are accepted in 24.5% of all cases. The initiative wins with a probability of merely 7.2%. Regarding the winning probabilities by profile, by definition the status quo always stays in place if the initiative withdraws without counter proposal. In total, the status quo remains in 90.3% of the cases if there has been a vote between status quo and initiative only. Reform is much more likely once a counter proposal has been made as the status quo remains in only 37.0% (without withdrawal) and 11.1% (withdrawal) of the cases respectively. Counter proposals are most likely to win (88.9%) if the initiative has been withdrawn, and are consequently the driver of initia- tive success. Counter proposals still have a 48.2% winning probability even if the initiative is not withdrawn. Generally, the initiative is more likely to win when also voted against the counter proposal (14.8%), but has a winning probability of only 9.7% when competing against the status quo alone. We find supportive evidence for Hypothesis 1 (Status Quo) that reform is more likely if a counter proposal has been made. Additionally, we show that reform is more likely if a generous compromise was issued after which the initiative was withdrawn as compared to counter proposals after which the initiative is not withdrawn.

83 3.5.2 Signature Collection

Figure 3.8 depicts over-collection in thousands against collection time in days by profile. Comparing collection outcomes before and after 1978, when the collection time was restricted and the requirement doubled, we observe more variation in time and signatures before 1978. Afterwards over-collection de- creases and collection time is mostly concentrated close to the maximum time of 18 months. Though we observe over-collection in all panels, as expected it is visi-

Figure 3.8: Signatures and Collection Time by Observed Profiles (CPw) before 1978 (CPw) after 1978 1000 1000 500 500 Collection time in days Collection time in days 0 0 0 100 200 300 400 0 100 200 300 400 Over-collection Over-collection

(CPnw) before 1978 (CPnw) after 1978 1000 1000 500 500 Collection time in days Collection time in days 0 0 0 100 200 300 400 0 100 200 300 400 Over-collection Over-collection

(nCPnw) and (nCPw) before 1978 (nCPnw) and (nCPw) after 1978 1500 1000 1000 500 500 Collection time in days Collection time in days 0 0 0 100 200 300 400 0 100 200 300 400 Over-collection Over-collection

Note: Over-collection is defined as valid signatures - legal threshold in thousands. Collec- tion time is measured in days by observed profile.

84 ble best in profile (CPw) combined with fast collection. Initiatives without counter proposals are much more concentrated at low over-collection though the graphs show that over-collecting does not necessarily guarantee a counter proposal. We observe both slow and fast collectors without counter propos- als. While the former is expected, the latter is more surprising: potentially politicians had a low initial belief about the initiative’s winning probability such that regardless of fast collection it was not perceived as a threat to the status quo. Contrary to expectations, there is also some over-collection in profile (CPnw), and collection time is around the average. Table 3.7 shows the regression results for over-collection. Columns 1 to 4 report the results when the dependent variable is a dummy for profile (CPw), and for profile (CPnw) in the other four columns. We include period fixed effects in every other specifications. We include controls for resourceful and experienced committees, and initiatives with economic topics. In columns (3), (4), (7) and (8) we drop initiatives with indirect counter proposals and general suggestions. The effect of over-collection on the probability of observing a counter pro- posal after which the initiative is withdrawn is positive and highly significant, in line with Figure 3.8. The coefficients in the first four columns suggest that over-collecting by 10,000 signatures is associated with an increased probabil- ity of profile (CPw) by 1 to 2 percentage points. Columns 5 to 8 also show a 1 percentage points increase in the probability of observing profile (CPnw). The effect is insignificant once we exclude initiatives with indirect counter proposals and include controls as well as period fixed effects. All results are robust to accounting for de facto counter proposals, linear and quadratic time trends (results on request). To account for the doubling of the signature re- quirement in 1978, we rerun the main regression with two subsamples before and after 1978. We find a highly significant increase of 2 percentage points on the probability of receiving counter proposal (CPw) and no effect on (CPnw) before 1978. After 1978, we do not find consistently significant effects (only the counterparts of columns 1, 4 and 5 are significant). Potentially this can be explained by significantly higher over-collection prior to the increased re- quirement (51.200 and 27.600 on average) and more variation in signatures collected.

85 Table 3.7: Effect of Over-Collection on Counter Proposals

(1) (2) (3) (4) (5) (6) (7) (8) Dep. var. (CPw) (CPw) (CPw) (CPw) (CPnw) (CPnw) (CPnw) (CPnw) Over-collection 0.002*** 0.002*** 0.001*** 0.001*** 0.001** 0.001** 0.001* 0.001 (0.001) (0.001) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Committee -0.044 -0.035 -0.056 -0.041 (0.050) (0.050) (0.049) (0.050) Topic 0.040 0.046 0.072* 0.058 (0.046) (0.046) (0.043) (0.044) 86 Adj. R2 0.075 0.175 0.080 0.114 0.028 0.098 0.061 0.127 Observations 222 222 181 181 195 195 177 177 Period FE no yes no yes no yes no yes Note: *** p<0.01, ** p<0.05, * p<0.1. Probit regressions, marginal effects reported. Standard errors in brackets. Dependent variable is dummy for profile (CPw) in columns 1-4, and profile (CPnw) in columns 5-8. Observations from profile (CPnw) are dropped in columns 1-4, and from profile (CPw) in columns 5-8. Initiatives with indirect counter proposals and general suggestions are dropped in columns 3, 4, 7, and 8. Over-collection is defined as valid signatures - legal thresh- old in thousands. Committee =1for experienced and powerful petitioners. Topic =1for econom- ically framed initiatives. Period fixed effects are based on 6 institutional periods. In sum, over-collection leads to a higher probability of receiving a cut- off counter proposal as predicted in the model, and the effect is particularly strong with the requirement of 50.000 signatures. Figure 3.8 also shows that many initiatives engaging in over-collection do not receive a counter proposal despite their collection efforts. Conversely, most counter proposals are issued for initiatives collecting quickly and over-collecting little. This means that for many initiatives politicians are able to assess the initiative’s chances of success once the official signature requirement is reached. Though the model predicts no incentive to over-collect to induce a counter proposal after which the initiative is not withdrawn, we observe some over- collection in this case. However, the effect is insignificant in several specifi- cations, so the results do not reject the model predictions. Moreover, over- collection may have other reasons than the purely strategic motive to receive a counter proposal. It also serves as insurance for invalid votes, or it may be the result of coordination failures in the collection process as modeled by the idiosyncratic collection shock.

3.5.3 Signature Collection Costs

Hypothesis 3 (Signature Collection Costs) predicts more initiatives when the costs of qualifying an initiative drop. From these initiatives under low qualify- ing costs fewer can be expected to receive counter proposals and to amend the status quo. We compare the institutional periods before and after female suf- frage. In the “Formal Withdrawal” period between 1951 and 1970 we observe 35 initiatives, and 26 in the “Female Voting” period between 1971 and 1977. Respectively, 1 and 4 of them were withdrawn without receiving a counter proposal (profile (nCP w)). Regression results are shown in Table 3.8. After female suffrage, the yearly number of initiatives submitted and qualified increases significantly by 2.4 (column 1). This corresponds to more than double the yearly number of initiatives before the change. The probability to receive a counter proposal satisfactory enough to withdraw the initiative drops significantly by 40.1 per- centage points. However, there is no significant change regarding the share of counter proposals without withdrawal (columns 2 and 3). The highly significant negative coefficient in column 4 reflects the lower

87 Table 3.8: Signature Collection Costs

(1) (2) (3) (4) (5) Dep. var. initiatives p.a. (CPw) (CPnw) amended amended Change 2.350*** -0.401*** -0.060 -0.363*** 0.000 (0.798) (0.093) (0.133) (0.101) (0.095) (CPw) 0.941*** (0.091) (CPw) 0.059 × change (0.179) (CPnw) 0.600*** (0.129) (CPnw) -0.400** × change (0.181) Adj. R2 0.235 0.147 0.004 0.110 0.761 Observations 26 51 41 56 56 Note: *** p<0.01, ** p<0.05, * p<0.1. Results from ordinary least squares regressions (1). Probit regressions, marginal effects reported, pseudo R2 (2-5). Standard errors in brackets. Dependent variables in first row: initiatives p.a. are the average number of initiatives per year in an institutional period. (CPw) and (CPnw) are dummies for the respective profile, amended is one if the status quo was amended. Over-collection is defined as valid signatures - legal threshold in thousands. Estimates are based on observations from periods “Formal Withdrawal” and “Fe- male Voting”. Observations from profile (CPnw) are dropped in column 2, from profile (CPw) in column 3, and from profile (nCP w) in columns 4 and 5. probability of amending the status quo after the institutional change. After (CPnw) the probability of amending the status quo drops by 40 percentage points (column 5). An F-test of joint significance between the dummy for (CPnw), a period dummy and its interaction is significant. In contrast, the probability of amending the status quo after a counter proposal with with- drawal does not change significantly (column 5). There are two main drivers of the lower probability of amending the status quo. The first one is the higher share of initiatives not receiving a counter proposal. The second one is the weak performance of initiatives at ballot. This is in line with the model pre- diction that the higher number of low types leads to fewer counter proposals and consequently to smaller initiative success. All results are robust to accounting for de facto counter proposals. Con- trolling for time trends turns the coefficient of change on the probability of observing (CPnw) significantly negative as predicted by the model. All oth-

88 ers results remain unaffected. After reducing the bandwidth of years around the institutional change to eight, the coefficients of the probability to receive counter proposals turn insignificant due to considerably fewer observations (all results on request.)

3.5.4 Collection Time Constraint and Signature Require- ment

The Hypotheses 4 (Collection Time Restriction) and 5 (Signature Require- ment) predict that some initiatives are unsuccessful in qualifying the initiative, such that the share of high types receiving counter proposals and changing the status quo increases. We focus on the periods “Female Voting” and “Collection Restrictions” ad- jacent to the change. During the period “Female Voting” (1971-1977) there are 26 qualified initiatives, and 23 in the following period “Collection Re- strictions” (1978-1986). In total, there were 4 initiatives withdrawn without counter proposal before the change, and 2 thereafter (profile (nCP w)). Regression results can be found in Table 3.9. The first prominent change in outcomes is that 11 initiatives did not qualify for ballot in the period after the institutional changes in 1978 due to not collecting enough signatures within the restricted time. In contrast, between 1891 and 1978 only two initiatives did not qualify for ballot.24 This corresponds to a significant increase by 0.8 initiatives per year not qualified. The change in the number of qualified initiatives per year is insignificant (columns 1 and 2). The probability of receiving counter proposals (CPw) and (CPnw) in- creases. However, only the former is significant at conventional levels (column 3 and 4). There is a significant increase in the probability of amending the status quo between the two periods by almost 30 percentage points (column 5). There are two drivers of this effect. First, both types of counter pro- posals are at least as successful as in the period preceding the institutional change: the status quo is 40 percentage points more likely to be amended

24These two cases occurred in 1894 (“Initiative for free healthcare and a monopoly on tobacco”, which did not qualify because it collected too few signatures) and in 1922 (“Initiative for a reform of the federal administration, including federal railways” - this initiative did not qualify because of too many invalid signatures). Cf. Hofer (2012) for details.

89 Table 3.9: Collection Time Constraint and Signature Requirement

(1) (2) (3) (4) (5) (6) Dep. var. qualified not quali- (CPw) (CPnw) amended amended p.a. fied p.a. Change 0.200 0.800** 0.234* 0.093 0.280** 0.000 (1.257) (0.338) (0.121) (0.138) (0.116) (0.099) (CPw) 1.000*** (0.148) (CPw) -0.000 × change (0.189) (CPnw) 0.200 (0.121) (CPnw) 0.400** × change (0.177) Adj. R2 -0.108 0.316 0.070 0.010 0.080 0.760 Observations 11 11 39 39 43 43 Note: *** p<0.01, ** p<0.05, * p<0.1. Note: *** p<0.01, ** p<0.05, * p<0.1. Results from ordinary least squares regressions (1-2). Probit regres- sions, marginal effects reported, pseudo R2 (3-6). Standard errors in brack- ets. Dependent variables in first row: (not)qualified p.a. are the average yearly number of initiatives (not) qualified. (CPw) and (CPnw) are dum- mies for the respective profile, amended is one if the status quo was amended. Over-collection is defined as valid signatures - legal threshold in thousands. Estimates are based on observations from periods “Female Voting” and “Col- lection Restrictions”. Observations from profile (CPnw) are dropped in col- umn 3, from profile (CPw) in column 4, and from profile (nCP w) in columns 5and6. after profile (CPnw) in the “Collection Restrictions” period than previously, and is amended with certainty after profile (CPw) (column 6). The latter was already the case before the institutional change so the interaction term in column 7 is insignificant. Both F-tests for the respective profile dummy, pe- riod dummy, and their interaction are highly significant. The second driver of success is that the share of counter proposals per period has increased which leads to higher overall success of initiatives after 1978. When accounting for de facto counter proposals, time trends and initiatives closer to the institu- tional change, we find no significant effects on the probabilities of receiving counter proposals or reforming the status quo. The data support Hypotheses 4 (Collection Time Restriction) regarding the number of initiatives and the probability of reform. However, we find only weak evidence for an increase in the probability of receiving counter

90 proposals. This might be due to concurrent shifts in politics that are not accounted for in the model, or the strong population growth between 1971 and 1978 which further lowered the signature collection costs (Wili, 1982). The timing of the institutional change in 1978 also coincides with the birth of the green movement in Switzerland which raised many initiatives on ecological matters in the subsequent years (for example, on motorway construction, nuclear power plants, cf. Swiss Federal Chancellery (2013)).25

3.5.5 Voting Rule: Tie-Breaking Question

As the last institutional change we analyze the impact of a tie-breaking ques- tion. Hypothesis 6 (Tie-Break) states that counter proposals without ini- tiative withdrawal should win more likely. The counter proposal’s winning probability with initiative withdrawal should decrease. We explore changes between periods “Collection Restrictions” (1978-1986) and “Tie-Break” (1987- 2010). The last period “Tie-Break” is considerably longer than the other periods and thus has more observations (23 vs. 111 qualified, 11 vs. 67 not qualified). Of all initiatives, 2 and 12 belong to profile (nCP w) respectively. Results are shown in Table 3.10. On average the number of initiatives increases and around 2 more initiatives per year do not qualify after 1987 than in the previous period. In the model, this increase corresponds to a rise in the share of low type initiatives trying to qualify for ballot but being unsuccessful. The number of qualified initiatives remains roughly constant (columns 1 and 2). The frequency of counter proposals decreases strongly and significantly after the institutional change. The share of counter proposals after which initiatives are withdrawn (CPw) drops significantly by 19.5 percentage points (column 3). The probability of observing initiatives with profile (CPnw) decreases significantly by 16.3 percentage points (column 4). The coefficient is less significant when accounting for de facto counter proposals and a quadratic time trend. It turns insignificant when we control for a linear time trend. One issue with the estimation is that we use 23 years of initiatives after

25Generally, the early 1980ies were a time of comparably high political turmoil in Switzer- land in the economic aftermath of the oil crisis with rising unemployment numbers and quarrels about the immigration of foreigners competing on the labor market, and in the wake of youth movements such as “Züri brännt” (Zurich burns, cf. Nigg (2001)).

91 Table 3.10: Amended Voting Rule: Tie-Breaking Question

(1) (2) (3) (4) (5) (6) Dep. var. qualified not quali- (CPw) (CPnw) amended amended p.a. fied p.a. Change 0.120 2.040** -0.195** -0.163** -0.202** 0.105 (1.277) (0.879) (0.080) (0.067) (0.097) (0.118) (CPw) 1.000*** (0.169) (CPw) -0.372* × change (0.194) (CPnw) 0.600*** (0.187) (CPnw) 0.045 × change (0.225) Adj. R2 -0.035 0.131 0.045 0.068 0.027 0.461 Observations 30 30 121 112 120 120 Note: *** p<0.01, ** p<0.05, * p<0.1. Note: *** p<0.01, ** p<0.05, * p<0.1. Results from ordinary least squares regressions (1-2). Probit regres- sions, marginal effects reported, pseudo R2 (3-6). Standard errors in brackets. Dependent variables in first row: (not)qualified p.a. are the average yearly number of initiatives (not) qualified. (CPw) and (CPnw) are dummies for the respective profile, amended is one if the status quo was amended. Over-collection is defined as valid signatures - legal threshold in thousands. Estimates are based on observations from periods “Collection Restrictions” and “Tie-Break”. Observations from profile (CPnw) are dropped in column 3, from profile (CPw) in column 4, and from profile (nCP w) in columns 5 and 6.

the institutional change in contrast to just nine beforehand. When using a symmetric number of years around the change, the effect is significant for 9 and 8 years, but insignificant for 7 years. The model prediction regarding these effects was ambiguous. Potentially, the composition of initiatives changed towards more low types who conse- quently received fewer counter proposals. The probability of amending the status quo decreases significantly by 20.2 percentage points (column 5). After profile (CPw) the status quo is signif- icantly less likely to be amended by 37.2 percentage points (column 6) as predicted by the model. We find no significant change in the winning prob- ability of counter proposal after which the initiative is not withdrawn. The reduced probability of inducing reform in the last institutional period therefore

92 has two sources: first, there is an overall drop in the probability of receiving a counter proposal. Second, counter proposals tend to be less successful at ballot.

3.6 Concluding Remarks

We developed a popular initiative model under uncertainty about the initia- tive’s winning probability in which petitioners collect signatures, politicians decide about making a counter proposal and the petitioners decide whether to withdraw the initiative or not. The model can rationalize the number of signatures collected, why counter proposals are made and under which condi- tions the initiative is withdrawn. Moreover, it enables predictions about the probability that the status quo policy is amended, and the initiative is able to break the agenda setting monopoly of the politicians. We checked the model for empirical plausibility by testing theoretical implications, making use of major institutional changes in Switzerland over time. Reaching a counter proposal is the main channel for an initiative to shape politics and to achieve a policy deviation from the status quo. The chance for policy change is higher in the case of far-reaching political compromise that makes the initiative withdraw. The credible threat of a pending popular initiative mobilizes politicians to deviate from the previous status quo to some extent. The decision whether to issue a counter proposal or not is influenced by the perceived popularity of the initiative. Counter proposals after which the initiative is not withdrawn are associated with a higher probability of policy change as well, but to a lesser extent, since in some cases, initiative and counter proposal are competing for votes in favor of a policy change. Our data suggest that when politicians decide whether to make a counter proposal or not, collecting a high number of signatures could play a role in the decision, since more signatures are associated with a higher probability of establishing a counter proposal. We generally find supportive evidence for the mechanisms predicted by the model. For future research, results of cantonal initiatives with variation in the timing of reforms could be used to shed light on the causality of the effects.

93 3.A Proofs

Proof of Winning Probability Counter Proposal

Proof. The proof is laid out in two steps. We first show that for a given xc ¯ 1 1 ¯ there exists an interval [bc, bc] ⊂ [ 2 (xq +xc), 2 (xc +xi)] with length bc = bc −bc in which xc receives more than half of the votes. Second, we show that bc = b for all xc ∈ (xq,xi), and the expected vote share is identical for all xc ∈ (xq,xi). For a given xc ∈ (xq,xi), the vote share for a given realization of the m 1 1 median m is vc (xq,xc,xi)=Fm( 2 (xc + xi)) − Fm( 2 (xq + xc)). By symmetry and single-peakedness of the density function of Fm it is obvious that for a given xc the vote share is maximized when the median sits in the middle of 1 1 ( 2 (xq +xc), 2 (xc +xi)). Let m c denote this point. If xc has πc(xq,xc,xi) > 0, m c 1 then vc (xq,xc,xi) > 2 , the vote share at the vote maximizing m = m c must be strictly larger than one half (if the vote share is below half at this point, the winning probability is zero for all xc). Vote shares for symmetric m m = m c − δ and m = m c + δ are identical. Note further that vc (xq,xc,xi) 1 1 at m = 2 (xq + xc) (and m = 2 (xc + xi)) are at most 0.5 such that xc would loose with certainty for these median realizations. It follows that there exists ¯ a symmetric interval [bc, bc] around m = m that has a vote share larger than 1 2 and thus positive winning probability. 1 1 The interval ( 2 (xq + xc), 2 (xc + xi)) in which realizations of m make xc 1 potentially win has length b< 2 (xi − xq) ∀ xc ∈ (xq,xi). Also vote shares m c m vc (xq,xc,xi)=vc (xq,xc,xi) are identical for all xc ∈ (xq,xi). From this it ¯ follows that bc = b − b = b ∀ xc ∈ (xq,xi), and consequently πc(xq,xc,xi)= ¯ b Prob(b ≤ m ≤ b)= μ¯−μ ∀ xc ∈ (xq,xi) which concludes the proof.

Proof of Proposition 1 (Equilibrium Petitioners)

Proof. The proof proceeds by first establishing that petitioners’ difference in expected payoffs between withdrawing and not withdrawing the initiative af- ter a counter proposal is a quadratic function of xc. Second, we show that the minimum of the parabola is right to xi, such that the function is monotonically decreasing in xc over the relevant range (xq,xi). We then establish conditions, such that petitioners prefer not to withdraw the initiative for counter propos-

94 als close to the status quo and vice versa for counter proposals close to the initiative. Since the payoff difference is monotonically decreasing in xc, this proves the existence of a cutoff counter proposal above which the initiative is withdrawn. Solving the optimal withdrawal problem yields a quadratic function of xc such that petitioners prefer to withdraw the initiative if:

1 2 1 1 1 x − [¯μ + (xi − xq) − b]xc < [b − (xi − xq)]xq +( xi − μ¯)xi + k 2 c 2 2 2 (3.7)

Taking the first derivative with respect to xc and setting it to zero yields ∗ 1 ∗ the minimum of the left-hand side, xc = μ+ 2 (xi−xq)−b>0. Then xi x¯ i. Plugging xc = xq into equation (3.7), yields that the expected payoff from not withdrawing after the counter proposal is larger than from B withdrawing, whenever πi(xq,xi)Ui >κ. This is the condition such that petitioners strictly prefer the ballot to withdrawing the initiative without counter proposal. The reverse holds for plugging xc = xi into equation (3.7) 1 if campaigning costs k>xixq +[2 (xq +xi)−b](xi −xq). Due to monotonicity there exists a cutoff counter proposal x¯c that characterizes the petitioners’ optimal withdrawal strategy.

Proof of Proposition 2 (Counter Proposal Politicians)

Proof. We proceed by first considering two lemmas from which the proof directly follows.

Lemma 1 Politicians never have an incentive to make a more favorable  counter proposal than the cutoff xc > x¯c.

Taking the first derivative of the expected payoff of an initiative that is with- drawn after a counter proposal, PA(xc ≥ x¯c), with respect to xc, yields ∂(·) 1 1 1 1 = [ xc − μ¯ − r] < 0 xc < μ¯ xc ∂xc μ¯−μ 2 2 because 2 . Thus, the higher , the lower is the expected payoff. Thus, politicians never make a counter proposal above the cutoff x¯c.

Lemma 2 If reputation costs r are sufficiently large, there exists a counter  proposal xc = xq +

95  Suppose xc = xq + 0 yields a higher payoff to politicians than xc =¯xc. Then

 A  A   A πq(xq,xc ,xi)(Uq − r)+πc(xq,xc ,xi)Uc (xc )+πi(xq,xc ,xi)Ui A A >πq(xq, x¯c)(Uq − r)+πc(xq, x¯c)Uc (¯xc) (3.8)

 Taking the derivative of the left-hand side with respect to xc , yields a neg- ative number whenever reputation costs r>xi − xq − 2b. We impose this condition since we assume that politicians do not make counter proposals just  stealing votes from the initiative. Thus, the smaller xc (and hence ), the larger the left-hand side of (3.8). It is therefore maximized for → 0. Solving the above inequality (3.8) for , letting it converge to zero, we have to show  that can be positive for the existence of xc . Solving for r yields an inequal- 1 ity depending on the sign of the term b − 2 (xi − x¯c) which depends on the status quo’s winning probabilities when voted against both kinds of counter proposals after which the initiative is (not) withdrawn. It is straightforward 1  to show that b − 2 (xi − x¯c) > 0 if πq(xq, x¯c) >πq(xq,xc ,xi). I.e., the status quo wins more likely when the initiative is withdrawn. Then the solution is:

1 μ¯ − 2 (xi +¯xc) r¯ =(xi − x¯c) 1

 This is the sufficient condition to guarantee the existence of xc = xq + πq(xq,xq + , xi).

96 Proof of Proposition 3 (Equilibrium Politicians)

∗ ∗ Proof. The politicians’ optimal choice between xc =¯xc and xc = ∅ is: ⎧ A A ⎨⎪ ∅ if πq(xq,xi)Uq + πi(xq,xi)Ui ∗ A A xc (μ|r ≤ r¯)= >πq(xq, x¯c)(Uq − r)+πc(xq, x¯c)Uc (¯xc) − c ⎩⎪ x¯c else

∗ ∗ Their optimal choice between xc =¯q + and xc = ∅ is: ⎧ ⎪ ∅ ( ) A + ( ) A ⎪ if πq xq,xi Uq πi xq,xi Ui > ⎨ (  )( A − )+ A(  )+ ∗( | ¯)= πq xq,xc ,xi Uq r βUc xc xc μ r>r  A ⎪ πi(xq,x ,xi)U − c ⎩⎪ c i xq + , → 0 else

Politicians choose whichever action maximizes their expected payoff. We take the differences between the expected payoffs of the respective counter pro- posal and no counter proposal and take derivatives according to μ. We find that the payoff differences are monotony increasing in μ and thus in the ini- tiative’s winning probability: the more likely the initiative wins, the higher is the expected payoff from making a counter proposal compared to making none. This proves that there exists a cutoff μc above which counter proposals are made, and below which they are not. However, the cutoff μc does not nec- essarily have to lie in the domain of μ ∈ [0,xq]. The belief about μ only plays a role for the politician’s best action if the cutoff is located in the domain. This concludes the proof.

Proof of Petitioners’ Preference Ranking

Proof. Petitioners’ payoff from a vote of the initiative against the status quo B taking into account that Uq =0is

∗ B B PB(xc = ∅)=πq(xq,xi)Uq + πi(xq,xi)Ui − κ B = πi(xq,xi)Ui − κ (3.10)

Petitioners’ payoff from a vote of the initiative against the cutoff counter

97 B proposal after which the initiative is withdrawn with Uq =0,is

∗ B B PB(xc =¯xc)=πq(xq, x¯c)Uq + πc(xq, x¯c)Uc (¯xc) B = πc(xq, x¯c)Uc (¯xc) (3.11)

Petitioners’ payoff from a vote of the initiative against the counter proposal B after which the initiative is not withdrawn, taking into account that Uq =0  B  B , xc → xq and Uc (xc ) → Uq =0,is

∗   B  B  PB(xc = xc )=πq(xq,xc ,xi)Uq + πc(xq,xc ,xi)Uc (xc )  B +πi(xq,xc ,xi)Ui − κ  B   B = πc(xq,xc ,xi)Uc (xc )+πi(xq,xc ,xi)Ui − κ B = πi(xq,xq,xi)Ui − κ B = πi(xq,xi)Ui − κ (3.12)

∗  ∗ We get that PB(xc = xc ) − PB(xc = ∅)=0such that petitioners are indifferent between receiving no counter proposal and the counter proposal after which they do not withdraw the initiative. This is very intuitive since the counter proposal is close to the status quo. ∗ ∗ Taking the difference between the payoffs yields PB(xc =¯xc) − PB(xc = 1 B ∅)= 2 (xi − x¯c)(xi − xq)+π(xq, x¯c)Uc (¯xc)+κ>0. Therefore, petitioners are better off if they receive the cutoff counter proposal than if they do not receive a counter proposal. Though the cutoff counter proposal yields lower utility than the initiative, it has a considerably higher winning probability.

Proof of Proposition 5 (Tie-Break)

Proof. Let a), b), c) and d) denote the four possible sincere preference rankings of voters where a) is xq xc xi,b)isxc xq xi,c)is xc xi xq, and d) is xi xc xq. Note that whoever prefers the initiative to the status quo also prefers the counter proposal to the status quo. However, the reverse is not alway true (it is violated by preference ranking xc xq xi).

98 The proof proceeds by checking who wins depending on what preference rankings have the majority of voters. First, if a) is in the majority (m ≤ 1 1 2 (xq + xc)), then vq(xq,xc,xi) ≥ 2 and the status quo wins. Second, if 1 1 a) and b) have jointly the majority (m ∈ ( 2 (xq + xc), 2 (xq + xi))), then 1 1 vi(xq,xi) < 2 and vc(xq,xc) > 2 s.t. the counter proposal wins. Third, 1 1 if c) and d) jointly have the majority (m ∈ ( 2 (xq + xi), 2 (xc + xi))), then 1 1 1 vi(xq,xi) > 2 , vc(xq,xc) > 2 , and vc(xc,xi) > 2 . I.e., the counter proposal 1 wins in the tie-break. Last, if d) is in the majority (m> 2 (xc + xi)), then 1 1 1 vi(xq,xi) > 2 , vc(xq,xc) > 2 , and vi(xc,xi) > 2 . I.e., the initiative wins in the tie-break.

99 3.B Data Appendix

The data source for the main variables is the Swiss Federal Archive which col- lects the Federal Announcements issued by the Swiss Federal Chancellery. All information is available online from the home pages of the Swiss Federal Chan- cellery and the Swiss Federal Archive. The data can be accessed on the follow- ing home pages: for an overview with links to federal announcement of more recent initiatives http : //www.admin.ch/ch/d/pore/vi/vis_2_2_5_1.html (Swiss Federal Chancellery), and for all federal announcements http : //www.amtsdruckschriften.bar.admin.ch/showHierarchyContent.do (Swiss Federal Archive). Some of the data we collected have been assembled independently by swissvotes.ch (a project of the Institute of Political Science at the University of Bern, Switzerland, and the Année Politique Suisse). How- ever, their database only comprises information on initiatives that have been voted on, and the most recent initiatives are not included. For validation of our data collection, we have compared our data with this dataset. For the identity of the initiative committee, we rely on data provided by the Swiss Federal Chancellery (2013), and on complementary information in Hofer (2012). The Swiss Federal Chancellery (2013) recorded some petitioners as committees formed especially for the purpose of raising the initiative, so called “ad-hoc-committees”. With additional context information in Hofer (2012), we are able to allocate 5 of these committees to the groups behind them. If the composition of the committee remains unclear, these committees are coded as inexperienced and not powerful. A detailed overview of all variables, their sources and short descriptions is given in Table 3.11.

100 Table 3.11: Overview of Variables and Data Sources Variable Source(s) Description Counter proposal Swiss Federal Chancellery (2013) Dummy variable whether any formal counter pro- posal was made Direct counter proposal Swiss Federal Chancellery (2013) Dummy variable whether formal direct counter proposal was made Indirect counter proposal Swiss Federal Chancellery (2013) Dummy variable whether formal indirect counter proposal was made De facto counter proposal Hofer (2012) Dummy variable whether informal, related policy compromise was made Time points related to initiative Swiss Federal Chancellery (2013) All time points of initiative process (begin of col- lection, submission, eventual withdrawal or of- ficial statement of non-qualification, voting day,

101 etc.) Voting combinations Swiss Federal Chancellery (2013) Indicates which proposals were voted on, and what the final voting outcome was Institutional conditions Hofer (2012), Swiss Federal Chan- Signature threshold, maximum collection time, fe- cellery (2013) male voting, withdrawal regulations, tie-breaking question Number of (valid and invalid) sig- Swiss Federal Chancellery (2013) Number of signatures collected for all initiatives natures Form of initiative Swiss Federal Chancellery (2013) Indicates whether initiative was a general sugges- tion or a formulated constitutional article Initiative committee Swiss Federal Chancellery (2013), Type, experience and power of initiative commit- Hofer (2012), Rohner (2012) tee Initiative topic Swiss Federal Chancellery (2013), Topic of the issue in question (economic, ideolog- Hofer (2012), Rohner (2012) ical, state order) Voting recommendations for initia- Swiss Federal Chancellery (2013) Recommendations of National and State Council tives on initiative Note: This table provides an overview of the variables used in the empirical part with a short variable description, and the source from which it was retrieved. 3.C Coding of Time Periods for Initiatives

The allocation of initiatives into time periods takes into account the timing of the institutional changes and the relevant time points in the initiative process of each initiative.

1. Period 1 starts with the introduction of the popular initiative in 1891.

2. Period 2 begins when withdrawals of initiatives become a de-facto rou- tine. The relevant time point is the time of withdrawal (30 October 1930) of the initiative that marks the beginning of more frequent with- drawals. All initiatives that passed the political discussion afterwards arguably face a different game setup. No initiatives were being dealt with at this point of time (the next initiative was only started in 1931 and all previous initiatives had already been voted upon).

3. Period 3 begins with the formal legalization of withdrawal clauses on 1 February 1951. The most relevant time point is the start of the sig- nature collection (since legalized withdrawal clauses could be included in the initiative text). No initiatives were going through the signature collection process at this time, so coding is straightforward.

4. Period 4 starts when women were enfranchised to vote and allowed to sign popular initiatives as well on 16 March 1971. Since the main effect on our model comes from the doubled population of potential signers, all initiatives that were still at the signature collection stage at this point of time would have been affected by the change. By coincidence, there were again no initiatives at collection stage at this point in time, so there are no initiatives that were only partly affected by the change, and the assignment into period 4 is clear.

5. Period 5 starts when the signature requirement was doubled (on 27 December 1977) and a maximum time period for signature collection was introduced (on 1 July 1978). The time limit of 18 months was only binding for initiatives handed in 18 months after the new law was in force. Thus, the two reforms came into force shortly after each other, but were not effective from exactly the same date. There was only one initiative which already had to collect 100,000 signatures but did not

102 face the limit of 18 months. However, there was a de-facto limit for this initiative as well, as the old regulation phased out at some point of time. The initiative took slightly more than two years to collect the signatures. 16 days after the signatures were handed in, the old regulation phased out definitely. Thus, there was a de-facto limit of 26 months collection time (instead of 18 months afterwards), whereas there was no time limit at all for earlier initiatives. For that reason and because the signature requirement was already enhanced, we code the initiative as obeying to the new regulation. Apart from this special case, coding is straightforward.

6. The last period 6 starts when the tie-breaking question between counter proposal and initiative was introduced by popular vote on 5 April 1987. The relevant time point for an initiative to be affected by the new reg- ulation is its date of ballot.

103 3.D Presentations and Acknowledgement

This paper has been presented at the following conferences and workshops: Meeting of the European Public Choice Society (April 2013, Zürich, Switzer- land), Meeting of the Australasian Public Choice Society (Singapore, 2013), Singeria Workshop of the Swiss National Science Foundation (September 2014, Zürich, Switzerland), as well as seminars at the University of Pennsylvania (April 2014, Philadelphia, USA) and the University of St.Gallen (March 2014, St.Gallen, Switzerland). For valuable comments, we thank Urs Birchler, An- tonio Merlo, Adam Pigon, and Andreas Steinmayr.

104 Chapter 4

Does Female Suffrage Increase Public Support for Government Spending? Evidence from Swiss Ballots

4.1 Introduction

Several contributions found a positive relation between female enfranchise- ment and a subsequent rise in government spending as well as revenue point- ing towards the existence of stronger female preferences for large governments: analyzing historical data from the U.S., Lott and Kenny (1999) find that the introduction of female suffrage raised government spending and revenue as gradually more women made use of their voting rights. Following Lott and Kenny (1999), Aidt and Dallal (2008) confirm their results for six western European countries for which the long-run effects are significantly larger than the short-run effects. Similarly, Abrams and Settle (1999) find that the in- troduction of female suffrage in Switzerland increased government spending on welfare issues by 28%, and also total government spending grew by about 12%. Aidt et al. (2006) confirm these findings for Europe. Bertocchi (2011)

105 provides empirical support that allowing women to vote increases government spending, however, only in non-Catholic countries in which the cost of dis- enfranchisement is relatively high.1 This leads to the conclusion that women prefer larger governments than men. Female risk aversion with the increased need for insurance from the state, or the breakdown of the family leading to higher divorce rates constitute some of the most commonly provided explana- tions for this gender gap in taxation preferences (Edlund & Pande, 2002). In marriage, husbands tend to earn more and transfer income to their wives who specialize in household production and care for the children (Becker, 1974). While income differences and specialization are internalized in marriage, the possibility of divorce, however, makes women more vulnerable economically since they might be rendered with a low income to care solely for the children. However, the literature does not convincingly show where gender gaps orig- inate from: are they due to “being female” or can they partly be explained by observable differences between men and women? For example, Meltzer and Richard (1981) famously hypothesized that enfranchising new constituents such that the median voter is poorer than before, increases demand for redis- tribution. Husted and Kenny (1997) exploit the repeal of literacy tests and poll taxes in the U.S. which hitherto prevented poor and foreign men from vot- ing. They find a 15% increase in welfare spending from enfranchising poorer men. Consequently, one would expect that extending suffrage to women, who on average have lower incomes than men, should have a positive effect on redistributive spending. Not controlling for socioeconomic gender gaps might falsely attribute some of the effects to the fact that women can vote, instead of having enfranchised a new group which is on average poorer than the former electorate.2 Also, the existent literature mostly relies on analyses of represen- tative democracies. A potential mechanism through which female voting could increase government spending is through politicians’ behavior. These either change their policies, or women elect new, more spending-friendly politicians.

1 In contrast, Stutzer and Kienast (2005) who use the variation in the timing of female suffrage in Swiss cantons, the 26 states in Switzerland, find that surprisingly enfranchis- ing women decreased government expenditures at cantonal level. They conclude that the negative effect might stem from the existence of direct democracy instruments in Swiss cantons for which previous research shows that they are likely to lead to smaller government size (Feld and Matsusaka (2003) provide some evidence). 2 In Switzerland, women earned 51% of the male hourly wage in 1930, 66% in 1971 and 67% in 1995 which shows how big the gender wage gap was despite its tendency to decrease over time (Swiss Economic and Social History Online Database).

106 To understand whether the relationship between higher expenditures and fe- male voting is causal or influenced by a third force, like more liberal thinking, the analysis of elections and politicians’ behavior would be required. Yet, the literature is relatively silent on the mechanism which would lead to higher expenditures.3 Literature confirms that women vote more often in favor of higher expen- diture, but the outcomes also depend on the issue the money is spent for. For example, Aidt et al. (2006)) find that female voting rights increased spending on health, welfare and education. Miller (2008) documents rising levels of public health expenditure to enhance child welfare that can be attributed to the enfranchisement of women. While there exist clear-cut predictions and empirical evidence for gender preference gaps on several spending categories, explaining why women would prefer larger governments per se turns out to be more difficult. This paper provides a direct way of analyzing gender preference gaps for government spending. Instead of deriving preference gaps indirectly via the development of government expenditures around the time of female enfran- chisement, this paper relies on results from referendum votes. The merit of examining direct democratic ballots lies in the fact that voters make observ- able choices and reveal their preferences in this way. I analyze the voting outcomes of referendums laying down the constitutional basis for the Swiss government to levy income, capital and goods turnover taxes. They are a measure of preferences for the federal government’s spending. Without pop- ular approval at the ballot, the Swiss government would be deprived of its authorization to levy federal taxes which are crucial for financing government expenditures. While taxation of income and consumption is commonly found all over the world, it is a Swiss particularity that voters even nowadays need to accept it’s legislative basis in a referendum every few years. Therefore, over time a large number of comparable votes on the federal financial system exists. The goal of the analysis is twofold: first, I estimate the size of the gender preference gap for government spending, i.e., the average treatment effect of

3 Lott and Kenny (1999) also look at the politicians’ voting behavior in the U.S. senate and find that after the introduction of female suffrage politicians voted more liberally. However, they do not show that women were more likely to vote for liberal politicians and did so because they desired higher government spending.

107 being female on voting yes in one of these referendums. For this, I analyze the voting outcomes of two very similar referendum ballots in Switzerland concerning federal taxation of which one took place shortly before the exten- sion of suffrage to women in 1971, and the other directly thereafter. The first proposition in November 1970 with a men-only suffrage was rejected at ballot. But the second proposition, which took place 7 months later with universal suffrage, was accepted. Since the two ballots took place under two distinct suffrage regimes, changes in voting outcomes can be directly attributed to changes in the electorate after accounting for the differences between the two propositions. The analysis is based on data from 2,143 Swiss municipalities. To isolate female approval for government spending, a similar reasoning to Lott and Kenny (1999) is employed. How much of the increase in voter ap- proval for government spending can be explained by female voting depends crucially on the intensity with which women made use of their voting rights. I therefore estimate the impact of the change in the number of voters on the change in the number of yes votes. I refine the estimates taking into account the share of women in municipal populations and canton fixed effects. More- over, I take into account that the two ballot propositions, though very similar, are not identical but differ in so far that the second proposition included a time limit. This means that even in case of acceptance the second proposi- tion would have required a new ballot after 10 years while the first one did not. Traditionally, permanent federal financial orders have been rejected in Switzerland, which suggests that the inclusion of a time limit is an important factor influencing voter decisions. I utilize voting results from a similar ballot in 1963 under the males-only suffrage to proxy for the difference in the content of the two ballots which might have led to some men changing their voting behavior between the two ballots. I also provide extensive evidence for the validation of this approximation. The results show that men were significantly more likely to favor taxation and thus government spending than women. These results contradict the notion that women are per se more likely to support large governments. The goal of the second part of the analysis is to disentangle the total aver- age treatment effect of gender on support behavior into its direct and indirect component. As argued above, being female has an indirect effect on support for government spending which is mediated by socioeconomic variables like

108 income and employment that are generally known to differ across men and women. Moreover, being female has a direct channel influencing support be- havior which is intrinsically female. To decompose both potential channels of how gender can influence voting behavior in the referendums, I use a me- diation framework to non-parametrically estimate the (average) direct and indirect effects of being female. Estimates are based on inverse propensity score weighting, which allows to control for potentially confounding factors influencing both mediators and the outcome variable (Huber, 2013, 2014; Imai, Keele & Yamamoto, 2010). I conduct the mediation analysis based on individual-level data from ran- domized post-ballot surveys after three comparable referendum votes about the federal fiscal order in 1981, 1991 and 1993. Since they are available for both voting and non-voting respondents answering the hypothetical question how they would have voted, selection into voting problems can be safely ig- nored. The results confirm that women are on average less likely to support the taxation propositions. The mediation analysis reveals a strong negative direct channel of being female on support for government spending. The in- direct effect is also negative, however, less significant than the direct gender effect. Consequently, mediating socioeconomic variables with negative gender gaps like full-time employment or having a high school degree play a smaller role for female preferences for government spending than direct effects. This paper adds to the existing literature on the effects of franchise exten- sion on government spending and revenue as well as gender preference gaps. The main innovation in my approach is to directly analyze the outcomes of ballots instead of relating suffrage to government spending. With the ex- ception of Funk and Gathmann (2012) who explore gender preference gaps for different spending categories by utilizing post-ballot polls in Switzerland, literature has so far only analyzed the effect of female suffrage on the size of state expenditure. This approach, however, is imperfect since voters only elect politicians who finally decide about policies. By analyzing outcomes of referendum ballots, I provide evidence for how voters decide directly on tax- ation and consequently government spending. I complement literature which emphasizes the importance of distinguishing between spending items when it comes to analyzing gender preference gaps. While women might be more likely to care for redistributive spending as can be inferred from Meltzer and

109 Richard (1981), they might be indifferent or even opposed to other spending categories. The remainder of the paper is organized as follows. Section 4.2 provides information on the institutional setting this paper is based on. I present the econometric framework in Section 4.3. Data and the estimation method are described in Section 4.4 where I also discuss the identifying assumptions. The results are presented in section 4.5, and the concluding remarks in Section 4.6.

4.2 Institutional Setup

Beginning with the foundation of the Swiss state in 1848, duties were the main revenue source at federal level.4 It took until the First World War, collapsing international trade and growing state expenditure before an income tax was introduced. But income was only taxed in times of need such as during the war, or when budgetary problems got out of hand in the 1930ies. In 1941 the Wehrsteuer (defense tax, an income and capital tax; referred to as direct federal tax in what follows) was introduced to finance growing military expenditure. After the Second World War, the direct federal tax remained in place to finance other state expenditure like the social security system but also new spending fields like education and culture. In addition, a goods turnover tax (Warenumsatzsteuer) on goods but not on services, resembling a value-added tax, was introduced also in 1941 (Stockar, 2007). However, both taxes lacked a constitutional basis, and were a product of an increased need of state revenue during war and emergency times. Besides revenues from duties, the goods turnover tax and the direct federal tax were the most important revenue sources for the Swiss government. In the 1960ies, roughly 10 to 15 percent of revenues came from the direct federal tax, and around 25 percent from the goods turnover tax. Revenues from duties then dropped by 10 percentage points. (Swiss Statistical Office, 1973) The main reason for the decline was the increasing international integration and the general trend to reducing duties in connection with the World Trade

4 Information about the history of the Swiss Federal Tax are from Grütter (1968). Oechslin (1967) gives an overview of the overall development of the Swiss tax system.

110 Organization’s rounds (Federal Announcement 1969 II, p.754). The lack of a permanent constitutional basis for levying federal taxes left some uncertainty about how to finance growing government expenditure. The main items of expenditure at federal level were defense and the social security system which together accounted for nearly 50 percent of total expenses. Other growing and new expenditure categories were infrastructure and energy, as well as culture and sports. Agricultural expenditure remained relatively stable at around 10 percent of total expenditure (Swiss Statistical Office, 1974). A proposition to allow the state collecting a direct federal tax as well as the goods turnover tax on a constitutional basis without time limitations was issued in 1953. Because it involved amending a constitutional article, the issue was subject to a mandatory referendum5, of which the outcome is binding in Switzerland (Linder, 2007). Since public finances are a core element of a state, a wealth of similar ballots concerning the federal government’s admission to file taxes exists. Table 4.1 gives an overview of all relevant ballots between 1953 and 2004. Even nowadays, it remains a Swiss particularity that citizens have to approve the federal financial order. Without acceptance, the federal government would not have the right to levy federal taxes. For a referendum to be successful, the majority of voters and a majority of cantons is required. In 1971 there were 19 cantons and 6 half cantons who’s votes counted as a 1/2 vote. The 1953 proposition was rejected. Only one year later, a similar proposi- tion to include the federal right of levying income, capital and goods turnover taxes in the constitution but with a time limit of four years was put to the vote, and eventually approved by the male voting population. It was fol- lowed by another temporary financial order from 1959 to 1964. The time limit forced the government to prepare new legislation regarding the financial order in 1962. Essentially, it was an extension of the old provision for another 10 years with some minor changes (Federal Announcement 1962 I, p.997)6. Again, the proposition was accepted at ballot.

5 In Switzerland, all changes to the constitution have to be approved by the voters. When the change is proposed by the parliament, this requires a referendum vote. In contrast, changes proposed by (groups of) citizens are called initiatives. 6 All federal announcements (Bundesblatt) are collected by the Swiss Federal Archive (Schweizerisches Bundesarchiv) and published by the Federal Chancellery (Bundeskan- zlei). A detailed list and possibility of online access is described in the online references section in the appendix.

111 Table 4.1: Chronology of Ballots Concerning the Swiss Federal Tax System Ballot date Time limit Decision % yes votes Accepting cantons 06.12.1953 unlimited rejected 42.0 3 24.10.1954 1955 - 1958 approved 70.0 21 11.05.1958 1959 - 1964 approved 54.6 17 1/2 08.12.1963 1964 - 1974 approved 77.6 22 15.11.1970 unlimited rejected 55.4 10 06.06.1971 1972-1982 approved 72.7 22 12.06.1977 unlimited rejected 40.5 1 20.05.1979 unlimited rejected 34.6 0 29.11.1981 1982-1994 approved 69.0 23 02.06.1991 unlimited rejected 45.6 2 1/2 28.11.1993 1994-2006 approved 66.7 22 28.11.2004 2006-2020 approved 73.8 22 For approval, the referendum needs more than half of total votes and at least 13 accepting cantons. In 1971, 19 cantons are “full” cantons while 6 cantons count only as “half” cantons. Data about acceptance are available on the homepage of the Swiss Federal Chancellery, http://www.bk.admin.ch. The time limits are from Federal Announcements published by the Swiss Federal Archive. See appendix for information on how to access the Federal Announcements. Votes in bold are used in empirical part.

The first of the two ballots at the core of the analysis took place on 15 November 1970. The second referendum took place with a new electorate on 6 June 1971: Switzerland was the last European country to grant women voting rights at federal level on 7 February 1971. It came into force on 16 March 1971. Swiss women were demanding suffrage more intensively in the aftermath of both world wars when democratization was spreading all across Europe. They also received support from male politicians who recognized that the women’s position in society had changed to a more active role in public live and private employment (Ruckstuhl, 1986). However, female suffrage in Switzerland could only be brought about by a constitutional amendment, which required the male population to hold a vote on extending the franchise. While at a first ballot in 1959 female suffrage was rejected with 66.9% of the male votes,7 a second run in 1971 saw the majority of voters and majority

7 Only three francophone cantons, Geneva (60.0%), Neuenburg (52.2%), and Waadt (51.3%) had a majority favoring universal suffrage. They were also the first three cantons to introduce universal cantonal suffrage.

112 of cantons accepting the constitutional amendment. The next paragraphs describe the propositions on government spending and voting results in more detail.

Ballot proposition 1: 15 November 1970 Facing a big budget deficit and the urgent need to ensure government revenue for the next years, the government and parliament proposed to discard the time limit and the maximum taxes from the constitution in the “federal en- actment about the amendment of the federation’s financial order”8 (Federal Announcement 1969 II, p.749). The new ballot proposition had to take into account that regardless of the good economy federal expenditures were expected to exceed revenues by a large amount. Therefore, income, capital and goods turnover taxes had to be increased and old rebates reduced. In more detail, the tax burden would be shifted from the direct income tax to the indirect goods turnover tax such that revenue from the goods turnover tax would increase considerably and revenue from income taxes would stay roughly constant. The proposition wanted to increase the goods turnover tax for retailers from 3.6 to 4 percent, and for wholesalers from 5.4 to 6 percent. The income tax set in progressively at an annual income of 8,500 Swiss Francs after deductions (7,700 Swiss Francs before). It allowed for deductions for married couples (2,500 Swiss Francs), children under 18 years and dependents (1,200 Swiss Francs) (Federal An- nouncement 1970 II, p.3). Regarding the income tax, high income households would be worse off with the new regulation than low income households be- cause of a more progressive system. Married couples or families with many children would be better off than with the old regulation. The government argued that an increase in goods turnover taxes to gen- erate state revenue was the preferable revenue source: it was not a typical consumption tax because of various exemptions for goods of daily use like food. It mainly taxed investment goods purchased by firms and the govern- ment, in addition to goods like alcohol, tobacco, and clothing purchased by households (Federal Announcement 1969 II, p.778). However, there seemed to be a general uncertainty about who would carry the burden of the higher goods turnover tax. Presumably the biggest load would be paid by enterprises.

8 Original title in German is “Bundesbeschluss über die Änderung der Finanzordnung des Bundes”.

113 Critics of the proposition mostly pointed to an unsatisfactory regulation concerning the Swiss cantons (Année Politique Suisse, 2012). In particular it lacked a clear division of revenue and expenditures between the federal government and the cantons because direct income taxes were an important revenue source for cantons and municipalities (Federal Announcement 1969 II, p.773). All major parties, associations and unions recommended their voters to accept the proposition. Exceptions were the small Liberal Party of Switzer- land (LPS), and the Labor Party (PdA) who opposed the proposition for not being progressive enough (Année Politique Suisse, 2012). These almost unan- imously positive voting recommendations indicate the importance of the issue at stake. On 15 November 1970 the Swiss voters - which was the male eligible population at that point - rejected the proposition in a mandatory referen- dum. Though 55.4% of the voters were in favor the proposition, it failed to accomplish a majority of cantons: in 13 of 22 cantons the approval rate was below 50 percent. The rejecting cantons were mainly concentrated in rural, German-speaking areas.

Ballot proposition 2: 6 June 1971 The Swiss government immediately prepared a new proposition9 because it urgently needed more revenue sources to finance growing state expenditure (Federal Announcement 1970 II, p.1581). In the major parts, the new propo- sition was identical to the old proposition, but it had the following changes. The biggest change included a time limit of 10 years (Federal Announcement 1971 I, p.487). This meant that in case of approval at the polls, the federal financial order had to be voted upon again in 1980 at the latest. As a further change, income tax ceilings of 9.5 percent for natural persons and 8 percent for legal persons were included. The income tax schedule became slightly more progressive and started to tax individuals at incomes after deductions of 9,000 Swiss Francs. These measures were taken to account for price inflation. It is important to note that the only essential change between the first and the second ballot proposition was the inclusion of the time limit. Comparing the precise wording of both legislative texts shows that they are almost identical

9 “Federal enactment about the continuation of the federation’s financial order”. Origi- nal title in German is “Bundesbeschluss über die Weiterführung der Finanzordnung des Bundes”

114 in all paragraphs.10 Consequently, if a man changes voting behavior between the ballots, content-wise the only obvious reason can be the time limitation of the second proposition. As in the first proposition, the parties and associations almost unanimously asked the voters to accept the proposition in their voting recommendations. Only the Labor Party (PdA), the Swiss Evangelic Party (EVP), and the Alliance of Independents (LdU) were opposed to the proposition because it disregarded deductions for working wives and was not progressive enough (Année Politique Suisse, 2012). This time with universal suffrage the ballot proposition concerning the Federal Tax System was accepted by a majority of votes (72.7%) and cantons. Figure 4.1 shows the approving and rejecting cantons for both ballots. The maps are based on swissvotes.ch.11

Figure 4.1: Cantonal Approval Rates for Ballots 1 (15 November 1970) and 2 (6 June 1971)

Note: Accepting (white) and rejecting (grey) cantons. Based on swissvotes.ch

Ballot propositions in 1981, 1991, and 1993 The votes in 1981, 1991, and 1993 are used in the empirical part to decompose the total gender preference gap into its direct and indirect components. The fiscal order approved by voters in 1971 was about to phase out in 1982. Both referendums in 1977 and 1979 which tried to change the tax system from the goods turnover tax as explained above to a value added tax (VAT) similar to those in other European countries were rejected at ballot. The financial order voted in 1981 and limited to the years 1982 to 1994 was therefore essentially a continuation of the old financial order from 1971. Minor changes included reliefs in the direct income tax which had to be compensated by increases

10The comparison is available from the author on request. 11A project of the Institute of Political Science at the University of Bern, Switzerland, and the Année Politique Suisse.

115 in the goods turnover tax (Federal Announcement 1981 I, p.20). In 1991 government and parliament tried again to switch from goods turnover taxes to the VAT. Again, the proposition was rejected at ballot. It was argued that the proposition might have been a too complex package, which led to the rejection. Two years later a new financial order in a less complex proposition finally brought about the change to the VAT system, and secured the fiscal fundament for the federal state until 2006 (Année Politique Suisse, 2012).

4.3 Empirical Framework

Two questions guide the setup of the econometric model. The first question evolves around the size and the sign of the total gender preference gap for government spending. I aim at estimating which gender has higher approval for government expenditures. The second question focuses on the underlying causes of the gender preference gap: what part of the gap is due to being female, and what part can be explained by other factors like socioeconomic differences between men and women? The question that arises is about the correct definition of the counterfactual. It is not enough to compare women to men but it is also required to account for factors like employment status and education that distinguish women from men. In the subsequent analysis, I propose a framework that allows to identify the size and causes of gender preference gaps for government spending. A population of n is divided into two groups G, with the realizations g =1 (female) and g =0(male). For each individual, outcome Y is observed which takes the value 1 if an individual is a supporter of government spending, and 0 else. The average treatment effect (ATE) then reflects the total effect of be- ing female on the outcome variable when averaging over the total population. Gender affects the outcome Y via two different channels. First, being female has an immediate impact on the probability of supporting larger governments which is henceforth referred to as the “direct” effect. Second, gender has a mediated effect through other variables (mediators) that thereafter affect the outcome variable which is commonly termed the “indirect” effect of the group variable. Write these mediating variables as a vector of observables M where th Mk, k ∈{1, ..., K} denotes the k element of the vector. M(g) is a func-

116 tion of gender. In the jargon of the mediation literature M(g) “lies on the causal path” of gender to support behavior, where gender marks the start of the causal chain (Baron & Kenny, 1986; Imai & Yamamoto, 2013). For example, employment status is an important factor influencing preferences for government spending. However, employment itself is a function of gender since women are less likely to work and more likely to stay at home to care for children than men. Thus, part of the effect of gender on support for gov- ernment spending is mediated by employment status. The outcome variable has therefore to be written as a function of both gender and the mediating variable, Y (g,M(g)). The total effect τ of gender on the probability of supporting government spending is the ATE. It can be written in the following way:

τ = E[Y (1,M(1))] − E[Y (0,M(0))] (4.1)

The ATE is the difference in expected outcomes between men and women. In fact, the ATE is the average difference in support for government spending by gender, i.e. the total gender preference gap. This is the first object of interest in this paper. Positive values of (4.1) would point to stronger female than male preferences for government spending as suggested by the literature cited above. For negative values the opposite would be true. The second goal is to decompose the total effect into the direct effect of gender on Y , and the mediated effect. The total effect can be written as the sum of indirect η(g) and direct δ(g) effects where the effects are:

η(g)=E[Y (g, M(1))] − E[Y (g, M(0))] (4.2) δ(g)=E[Y (1,M(g))] − E[Y (0,M(g))] (4.3)

The total effect is decomposed into: (4.2) the indirect effect η(g) which comes from the difference in expected values when evaluating mediators for both groups while keeping gender constant at g, and (4.3) the direct effect δ(g) which shows the effect of varying gender on the difference in expected out- comes when mediators are kept at their values for g. Note that outcome Y (g,M(1 − g)) is never observed because each individual can only be ob- served in either one of the groups G but never how the individual would have acted if he was of the other gender. Decomposition of the total effect is thus

117 based on the potential outcome framework (e.g., Rubin, 2004). Since individ- ual treatment effects cannot be estimated, the analysis relies on population averages: the above equations (4.2) and (4.3) already denote population av- erage indirect and direct effects. Also note, that after a simple manipulation the total effect can be written as the sum of indirect and direct effects:

τ = E[Y (1,M(1))] − E[Y (1,M(0))] + E[Y (1,M(0))] − E[Y (0,M(0))] = η(1) + δ(0) (4.4) = E[Y (0,M(1))] − E[Y (0,M(0))] + E[Y (1,M(1))] − E[Y (0,M(1))] = η(0) + δ(1) (4.5)

For identification, it is required that G is independent (i.e., assignment into groups G is random), and mediators M are exogenous when conditioning on G (Huber, 2014). While it can be argued that the assignment into gender is random or non-manipulable by other factors (at least in the context of this paper), independence of mediators is a relatively strong assumption. This means that, conditional on G, the error term is not allowed to impact me- diators M and the outcome Y at the same time. This assumption is easily violated and would lead to biased estimates. For example, being of pension age affects government spending preferences but at the same time also reduces the probability of being employed which is a gender mediator as argued above. The solution is to replace the independence assumption by a set of condi- tional independence assumptions. For this, I introduce a further vector of j observables C =[C1, ..., CJ ], j ∈{1., , , .J}. They have a confounding effect on G, M, and Y which means that they influence some or all of the three variables. Pension age would be such a confounding factor by the above ar- gumentation. η(g) and δ(g) are then correctly identified under a sequential ignorability assumption (e.g., Huber, 2013; Imai, Keele & Yamamoto, 2010).

Assumption 1 (Sequential Ignorability) 1.1 {Y (g,m),M(g)}⊥G|C ∀ g,g ∈{0, 1} 1.2 {Y (g,m)}⊥M|G = g,C = c ∀ g,g ∈{0, 1} 1.3 P (G = g|M = m, C = c) > 0 ∀ g ∈{0, 1}

Assumption 1.1 implies that once conditioning on confounders C no other confounders exist which would either impact gender G and mediators M at

118 the same time, or gender and outcome Y , or both. Ignorability thus means that besides confounders C all other variables can be ignored, or that all con- founders must be observed. According to Assumption 1.2, after conditioning on gender G and confounders C, no variables should have an effect on the mediators and the outcome Y . In more detail, the assumption demands that how mediators impact the outcome variable is not confounded once control- ling for gender and confounders. For example, if pension age was truly the only confounding variable, the effect of employment status on supporting gov- ernment spending is unconfounded after conditioning on gender and pension age. The last Assumption 1.3 is a common support assumption demanding enough comparable observations for both groups g =1and g =0in order to have comparable units for both groups. This means that there should be enough individuals in the sample that are similar regarding all mediators and confounders but differ by gender. If employment status and pension age were the only mediators and confounders, all feasible combinations of the two vari- able should be observed in the data for men as well as women (e.g., employed and below pension age, unemployed and below pension age, etc.). A graphical representation of the mediation framework can be found in Figure 4.2. The solid lines represent direct effects while the dashed lines visualize the indirect gender effect.

Figure 4.2: Mediation Framework G Y

M

C

Note: G are groups, M are mediators, C are confounders, and Y the outcome. Solid lines represent direct effects, and dashed ones indirect or mediated effects.

119 4.4 Data and Estimation Method

The biggest challenge in the literature concerning gender preference gaps for government spending is to correctly identify the mechanism of female suffrage on government expenditure and revenue. This is crucial since observed rises in government expenditure over time can be due to other causes than female enfranchisement: changes in social security, economic crises, changes in the tax base, and various other reasons not necessarily connected with female voting. I overcome this problem by examining voting in referendums which has several attractive features that makes it particularly interesting for the analysis (cf. also Deacon and Shapiro (1975) for a discussion of using referenda to elicit voter preferences). First, voting outcomes are real choices and not just political outcomes for which the exact cause is unknown. Second, political information about the issues at ballot is available from brochures which are sent to Swiss households before ballots since the 1950ies (Rohner, 2012), and the media report extensively on important political issues. Therefore, the Swiss are likely to make mostly informed decisions when voting. Moreover, referendums have been introduced in 1874 (Linder, 2007), so they have become institutions and are strongly positioned in the Swiss society. In what follows, I first describe the data. Second, I explain the estima- tion strategy and the identifying assumptions to estimate the total gender effect based on municipal voting data. Last, I show the estimators for the di- rect and indirect effect with the empirical specification of the mediating and confounding variables.

4.4.1 Data

To identify the gender preference gap for government spending and the role socioeconomic mediators play in explaining it, I use two different datasets. The first data set is from the ballot propositions in 1970 and 1971 that allows me to draw conclusions about gender taxation preferences by relating changes in the electorate to changes in voting outcomes. The second dataset contains individual data from post-ballot surveys of votes about the federal financial order which took place at later points in time (1981, 1991, and 1993). The three surveys allow me to investigate socioeconomic drivers of government spending preferences, and to disentangle them from intrinsically female effects.

120 Municipal Data For the estimation of the ATE, I collected a dataset of 2,143 Swiss municipal- ities with voting information for the relevant ballots on November 15, 1970, June 6, 1971, and December 8, 1963. They include the number of yes and no votes, valid votes and eligible citizens. Data from the three cantons Aargau, Freiburg, and Ticino are not available at municipal level. Instead I include the data from voting districts which comprise several municipalities each for these three cantons adding 26 voting districts to the dataset.12 For the canton Geneva, data are missing for the vote in 1963. Therefore, it is excluded from the analysis. All voting data come from the Political Atlas of Switzerland pro- vided by the Swiss Statistical Office. In addition, I merge voting data with a set of demographic variables from the Swiss census of 1970 also published by the Swiss Statistical Office. Since voting data come from two ballots with a time difference of 7 months, municipal mutations need to be taken into account, because several munic- ipalities merged during this time. Therefore, I adjust the voting data from ballot 1 such that they are comparable to ballot 2. I do the same for the census data which means that I sum the data from municipalities which have merged between 1970 and 1971. Information of municipal mergers comes from the online register of municipal mutations provided by the Swiss Statistical Office. There are two special cases in the cantons Bern and Thurgau. The first one in the canton Bern is that voting results from very small municipal- ities are counted and reported in some larger nearby municipality. Second, in Thurgau several municipalities which are available separately in the census data together form a political municipality with different administrative tasks. Voting data are reported for the latter only. I account for both special cases by adjusting the census data accordingly such that they are comparable.

Post-Ballot Surveys Post-ballot surveys are conducted shortly after all referendum and initiative ballots at national level in Switzerland since 1981. The project is called VOXit, and the data are being published by the Swiss foundation for research

12I have contacted the cantonal archives of the three cantons in question. For only 20 municipalities in the canton Freiburg complete voting data required for the estimation exist, so using district data is the only way to include data from these cantons in the regressions.

121 in social sciences.13 Randomly chosen respondents answer a questionnaire by telephone. Among the information included are the voting behavior and vari- ous socioeconomic controls as well as contextual information. The advantage of these polls is that until the end of 1999 voters as well as eligible citizens who did not go to the polls answered the questions. Importantly, they include the hypothetical answer of the nonparticipating respondents to the question of how they would have decided if they had voted. This allows me to conduct an analysis of voters and nonvoters by gender, and overcomes potential selection into voting problems. As becomes clear from above, the Swiss have to approve the federal fi- nancial order, i.e. the federal government’s competency to levy federal taxes, by popular vote. This has not changed until now since all propositions with- out a time limit have always been rejected so far. The mediation analysis is based on all three votes regarding the federal financial order between 1981 and 1999. These are the ballots voted on 29 November 1981, 2 June 1991, and 28 November 1993 which is the latest ballot concerning the federal finan- cial order before 1999, and thus contains the answers of non-voters. Though tax rates and deductions have of course changed since the ballot propositions in 1971, the matter is in fact identical to the propositions analyzed above. The propositions of 1981 and 1993 include time limits for the federal financial order until 1994 (Federal Announcement 1981 II, p.561) and 2006 (Federal Announcement 2003, p.1540) respectively. The 1991 proposition does not have a time limit. While it might be of concern that women have not yet grown accustomed to their voting rights in 1971 and might have hesitated to participate, for the later ballots female voting rights were already well estab- lished. Moreover, any potentially strategic male voting behavior stemming from the introduction of female suffrage should have ceased to exist by then. I use all observations for which the participation and voting decision are available. All observations where according to the survey the respondent submitted an empty vote are dropped in order to follow the official rule to calculate voting results. I drop observations where the respondent claims that he turned out for the vote but there is an answer for the voting behavior of non- participants in the data set (and vice versa for non-participants where there is information on the behavior of participants for that individual). These are

13Data are available online on the following homepage: http://nesstar.sidos.ch/webview/index.jsp

122 only 11 and 17 cases respectively and most likely the result of data mistakes. One typical concern about using surveys to elicit voter preferences is po- tential survey bias: either respondents misrepresent their voting behavior, or they choose not to participate in the survey conditional on their characteris- tics. To account for this issue, Funk (2012) proposes to compare the official voting results with the share of survey respondents claiming they have voted “yes”. Subtracting the official results from the survey results based on the vot- ing population only, yields a difference of 10.37 (1981), -1.84 (1991), and 1.64 (1993) percentage points between the two. The 1991 and 1993 values confirm Funk’s (2012) result that on average no significant survey bias occurs in votes concerning federal finances. However, the first value suggests the existence of a survey bias and points to problems of accuracy with the survey results from 1981.

4.4.2 Identifying the Total Gender Effect

Estimation Strategy

The first step in my empirical strategy is to estimate the sign and size of the gender preference gap. Theoretically, I could provide a simple regres- sion of gender on acceptance behavior using post-ballot surveys. However, this evidence would only be suggestive and not necessarily constitute a causal effect. To estimate the ATE in equation (4.1), I therefore analyze two refer- endum ballots on two very similar propositions concerning approval for fed- eral taxation with different suffrage regimes: the first ballot was under a males-only rule and the second with universal suffrage. Recall, that the ATE, τ = E[Y (1,M(1))] − E[Y (0,M(0))] = E[Y (1)] − E[Y (0)], is the difference between female and male support for government spending. By conducting the analysis based on referendum votes, the outcome Y takes the value one if a voter votes “yes”, and zero else. Hence, the empirical counterpart of E[Y (1)] is the female acceptance rate and E[Y (0)] the male acceptance rate. Then, the ATE is simply the difference between gender acceptance rates. Since the observed total acceptance rate can be written as the gender-weighted sum of female and male acceptance rates, it is enough to estimate the acceptance rate of women. The calculation of the ATE follows immediately.

123 In my empirical strategy to estimate the female acceptance rate, I follow the idea of Lott and Kenny (1999) who recognize that the effect of female suffrage on voting outcomes depends on how intensely women make use of their voting rights. The intuition for the empirical strategy is that changes in voting outcomes can be explained by changes in the electorate’s composition. The main independent variable in the analysis is thus the change in the num- ber of voters between the two ballots and the dependent variable the change in the approving votes. What makes the analysis more complicated is the fact that the ballots are not entirely identical. As explained above, the main difference between the propositions is a time limit of ten years in the second proposition. I have noted before that propositions regarding the federal financial order including time limits have also been approved by the male voting population, like in 1954, 1958, and 1963. This means that not only women are expected to vote in favor of the proposition but also some men should change their minds and vote yes instead of no. To account for the fact that the second proposition is less extreme than the first, I use voting results of the related vote in 1963 which also included a time limit and was accepted by a large margin (cf. Table 4.1). The true number of men who have switched from voting no to voting yes in municipality m, Δmenm, is approximated by a variable Δmen m calculated from the old voting results, and the error term m. Then the true number of men changing their voting behavior can be written in the following way.

Δmenm =Δmen m + m (4.6) votersm1970 = yesm1963 ∗ − yesm1970 + m (4.7) votersm1963

Δmenm is the difference between the number of yes votes in the years 1963 and 1970, normalized by the growth in the number of voters during the seven year difference. In this manner, I can proxy the change in male approval rates when propositions include a time limit or not. The error term m reflects that this variable is only approximated, and voter preferences in municipalities might have slightly changed between 1963 and 1970. The goal is to estimate the female acceptance rate for the second ballot proposition, acceptancef . For notational ease, denote the year 1970 by t =1, and 1971 by t =2. Define the change in the number of yes votes between the

124 ballots in municipality m as Δyesm ≡ yesm2 − yesm1, and the change in the number of voters as Δvotersm ≡ votersm2 − votersm1. The female acceptance rate can be written in the following form:

Δyesm Δmenm acceptancef = − (4.8) Δvotersm Δvotersm

Female acceptance is the change in the number of yes votes relative to the change in the electorate, net of the change in male voting behavior where Δmenm is defined in (4.7). Equation (4.8) can be easily transformed to

Δyesm − Δmenm = acceptancef ∗ Δvotersm (4.9)

In the last step, I use equation (4.7) to account for the approximation of men changing their voting behavior. This leads directly to the baseline estimation equation.

Δyesm − Δmen m = β1Δvotersm + m (4.10)

 Under the exogeneity assumption E(Δvotersmm)=0, the coefficient β1 of a linear regression thus identifies the female acceptance rate acceptancef which is the main object of interest here. Note, that (4.10) is in fact a first difference equation where men m2 = Δmen m, men m1 =0, m2 = m, and m1 =0. Therefore, by definition estimating (4.10) with least squares should be equivalent to estimating a fixed effects model of the following form where υm are municipal fixed effects.

yesmt − menmt = βFEvotersmt + υm + mt,t=1, 2 (4.11)

The equivalence of both coefficients β1 = βFE only holds under the assump-  tion of strict exogeneity, E(Δvotersmm)=0(Wooldridge, 2010). Hence, I run both regression (4.10) and (4.11) to explore if exogeneity poses a problem or not. Typically, for estimations using data from federal states, it is common to include canton fixed effects in the regression. I therefore run a specification including canton fixed effects ξc. They account for cantonal, time-invariant factors between both ballots like cultural differences, cantonal female voting

125 rights, or compulsory voting rules.14

Δyesm − Δmen m = β2Δvotersm + ξc + m (4.12)

I further extend the analysis to refine the measure of the female acceptance rate. Intuitively, if there were no women in a municipality, female suffrage should not have any effect on the participation rate. Conversely, if a mu- nicipality was populated by women exclusively, the change in voters would reflect the change in acceptance with certainty. Given that foreigners have no federal voting rights in Switzerland, I calculate the share of Swiss adult women in a municipality, %womenm =(eligiblem2 − eligiblem1)/eligiblem2, and multiply the change in voters Δvotersm with this variable. The validity of the variable %women is implicitly based on the assumption that the Swiss population did not increase between November 1970 and June 1971. While this constitutes a simplification, the population size should not have grown a lot during 7 months, and the variable should be a good approximation. I run the following regression, and a further specification also including canton fixed effects.

Δyesm − Δmen m = β3%womenmΔvotersm + m (4.13)

The female acceptance rate can then be recovered by calculating acceptancef = β3∗%womenm. Even though the population-weighted average share of women in the Swiss adult population amounts to 53.8 percent, it varies between 0 and 72.2 percent. In total, 22 (11) municipalities have shares below (above) the band of 40 to 60 percent. Municipalities with low shares of Swiss adult women are located in only 4 cantons Bern, Graubünden, Waadt, and Wallis, and have only 89 inhabitants on average.15 Municipalities with high shares of Swiss women are located in six different cantons and have 812 Swiss adults on average. Most of the shares are, however, close to 60% and thus not such big outliers. I run different specifications excluding municipalities with extreme shares of female population. In further specifications, I also exclude very large (above

14Note that female suffrage for cantonal votes is independent of federal regulation. Though some cantons introduced female voting close to the federal switch, many did not. 15Bern, Graubünden, Waadt, and Wallis are the four largest cantons in terms of area, and encompass some of the least densely inhabited regions in Switzerland.

126 10,000 Swiss adults) and very small (below 100 Swiss adults) municipalities. Excluding large municipalities technically means that data from the three cantons with district data are excluded, as well as 32 relatively large munici- palities. In this way, potential outliers can be accounted for because only 1,630 Swiss adults live in an average municipality, and 95 percent of observations have less than 5,500 eligible citizens.

Identifying Assumptions

The validity of the above estimations is based on two main identifying assump- tions. They particularly reflect that the estimations are based on aggregate data and not on individual observations. The first one is the following:

Assumption 2 The change in the number of voters between ballots 1 and 2 stems from the female part of the population.

Or put differently: men are not more likely to participate once women are enfranchised. This might be due to a decrease in the marginal benefit to vote when the electorate roughly doubles. Further evidence which confirms that men should be unlikely to increase their participation comes from comparing participation rates of the two similar ballots in 1953 and 1954, both with male suffrage only. For the first ballot in 1953 which did not include a time limit and was highly contested 60.27% of the male eligible population turned out. In contrast, the less contested proposition including a time limit of 4 years in 1954, drew only 46.77% of eligible men to the polls.16 Hence, I am confident that the additional participation is likely to be a lower bound for female participation. I provide further evidence from cantonal votes that the change in the number of voters between the ballots reflects the number of female voters: on the day of the second vote, June 6 1971, 11 cantons have not yet introduced female suffrage for cantonal votes. This means that women in these cantons were allowed to vote on federal issues but not on cantonal ones. Of these 11 16Turnout at a particular election day is influenced by all votes on the ballot list. Luckily, on the ballot day in 1954 there was no other federal vote, so turnout was truly for the vote under investigation. On the ballot day in 1953 there was one additional federal vote about the protection of waters. It received a narrowly smaller turnout than the other vote so it is safe to assume that the vote about the federal financial order was the main reason to turn out on that day.

127 cantons, 5 held cantonal votes on 6 June 1971.17 Consequently, the number of voters who voted on cantonal issues on this day were men with certainty. Owing to this constellation, the number of women voting on the day can be calculated by taking the difference between the number of voters in the federal vote and the highest number of voters in the cantonal vote. I call this the true number of women. Next, I compare this number with my approximated number of female voters, Δvotersc ≡ voters2c − voters1c for each of the five cantons c. On average, the true number of women exceeds my approximated number of female voters by 1.745 percentage points with a standard deviation of 7.794 percentage points. This shows that the change in the number of voters between ballots 1 and 2 reflects female turnout well on average. The standard deviation is relatively large because the values for two cantons are relatively far away from unity. This might be a result of the fact that whoever votes on cantonal issues does not necessarily have to vote on the federal issue and vice versa, so that there might be some roll-off between the votes. I compare federal and cantonal turnout in the five cantons for all ballot days between 1973 and 2010 which gives 287 cantonal ballot day observations after all 5 cantons introduced cantonal female voting rights. Data show that cantonal turnout for these 5 cantons is on average 1.520 percentage points higher than federal turnout with a standard deviation of 1.616. Therefore, the difference between the true and estimated number of female voters could partly be explained by selective abstention between cantonal and federal votes. In sum, this evidence further strengthens the plausibility of Assumption 2. Ballot 2 is the first federal voting date after the introduction of female suffrage on which voters decided on two bills.18 Intuitively, at the day of the introduction of universal suffrage not all women make use of their new rights immediately. It takes time until women grow accustomed to the possibility of voting (e.g., for the U.S. Lott and Kenny (1999) find that the full effect of female suffrage is revealed after 43 years of female voting rights). To visualize the effect of female suffrage on voter participation in Switzerland the number of voters normalized by the Swiss population of age for elections to the Swiss

17This were the cantons Bern, Graubünden, Schwyz, Thurgau, and Uri. Data on cantonal voting results is from the Centre for Research on Direct Democracy available online on www.c2d.ch. 18The other proposition was about the protection of humans and their environment.

128 Figure 4.3: Participation Rate in Parliamentary Elections 1951-1991 60 50 40 Voters/Swiss 20 and older Voters/Swiss 30

1950 1955 1960 1965 1970 1975 1980 1985 1990 Election

Note: Participation rate (voterst / Swiss adult populationt) for parliamentary elections over time. Without (1951-1967) and with female suffrage (since 1971). Data from the Swiss Statistical Office. parliament between 1951 and 1991 is presented in Figure 4.3.19 In this time period, parliament was elected every 4 years. The x-axis shows the election year t before and after the introduction of female suffrage in 1971. The number voterst participationt = of voters normalized by the Swiss adult population adultst is depicted on the y-axis. I take the total number of Swiss people above 20 years old from Swiss censuses in 1950, 1960, and 1970, and interpolate the numbers for the inter-census years. The data are from the Swiss Statistical Office. For the later years, adultst = eligiblet can be directly used from official election data. The fraction of voters as compared to the total adult population was steadily decreasing before the introduction of female suffrage. As expected, the participation rate jumps by more than 25 percentage points in the 1971 election with universal suffrage. However, afterwards the participation rate has a decreasing trend again. This contrasts with the observation of Lott and Kenny (1999) who show that the turnout rate in the U.S. continued increasing many years after the introduction of female suffrage. Thus on average women in Switzerland made use of their voting rights relatively quickly. This is not surprising since female suffrage was introduced relatively late in history. This

19It is preferable to depict turnout for parliament than turnout for referendum votes since participation for referendums varies a lot which might be due to the importance of an issue or campaigning effects.

129 coincides with higher education levels among women than in countries that enfranchised women around the first world war. Also, in some cantons women have received female voting rights for cantonal votes independently of federal voting rights such that they have gathered some voting experience even before 1971. However, the second ballot took place only four months after female en- franchisement and was the first voting date that allowed all Swiss women to vote on federal issues. So while women might have already been more likely to vote in the federal elections at the end of 1971, they might have hesitated to participate in their very first voting opportunity. The participation deci- sion is a selection into voting based on the citizens’ utility from voting. The question is, thus, whether participation is a function of underlying variables which would render the participating female population unrepresentative for tax preferences of the female population. I overcome this problem by car- rying out the second part of my analysis with post-ballot data of votes that took place well after female suffrage was introduced. The post-ballot data are available from both respondents who participated in the vote, and those who did not vote. This allows to draw conclusions about the gender preference gap for taxation for the entire population. The second identifying assumption relates to the change in the ballot propositions by including a time limit restriction in the second one.

Assumption 3 Men who have approved of the first proposition should also be in favor of the second one which includes a time limit and is thus less radical. The inclusion of the time limit in the second ballot proposition makes some men switch from voting no to voting yes.

Based on these assumptions, I have constructed a measure of the number of men changing their voting behavior from rejecting to approving, Δmenm,as stated in equation (4.7) which uses voting data from 1963 and 1970 under the male-only voting regime. The validity of this proxy relies on the assumption that male preferences regarding government spending were relatively time constant between 1963 and 1970 and differences in acceptance are due to the inclusion of a time limit in the 1963 proposition and the lack of it in the latter. To substantiate this claim, I again provide evidence from two comparable ballots on the federal financial order in 1953 and 1954. Recall, the first one had no time limit and was rejected, while the second one had a time limit and was

130 approved by the entirely male electorate. The population-weighted average difference in approval rates for the two propositions was 27.7 percentage points which is substantial, and similar to the difference between 1963 and 1970 amounting to 29.8 percentage points. A t-test of both differences is highly significant. Because preferences between 1953 and 1954 can be assumed time constant, this significant difference in acceptance shows that the inclusion of a time limit is indeed responsible for higher shares of yes votes among the male population. Theoretically, some men might have radical preferences and vote against the second proposition even though they supported the first one to protest and signal dissatisfaction. However, based on the supporting evidence from past ballots that including a time limit on average increases voter support this should seldom be the case.

4.4.3 Identifying the Direct and Indirect Gender Effect

Estimation Strategy

The second step of my analysis is to disentangle the ATE into its two compo- nents, the (average) indirect and direct effect. A standard way of estimating η(g) and δ(g) is through estimating a set of linear equations of the following form (e.g., Baron & Kenny, 1986; Judd & Kenny, 1981):

J = + + + ∈ 1 Mk αMk Gβk Cjγjk υ, for k , ..., K (4.14) j=1 K K J Y = αY + Gξ + Mkθk + GMK μk + Cjλjk + (4.15) k=1 k=1 j=1

αMk and αY are intercepts, βk, γjk, ξ, θk, μk and λjk are coefficients of gender, mediators, their interaction, and confounders, and υ as well as are the error terms. While this is the standard approach in the mediation literature (cf. Blinder (1973) or Oaxaca (1973) for linear wage decompositions, however, without confounding factors), it imposes relatively strong functional form assumptions which might be overly restricted given that my outcome variable is bivariate (Hicks & Tingley, 2011). Huber (2013, 2014) proposes to

131 use a nonparametric model of the following form instead:

Mk = χk(G, C, υ),fork∈ 1, ..., K (4.16) Y = φ(G, M, C, ) (4.17)

This approach is more flexible and appropriate for my analysis since χ and φ are functions that do not need to be specified more precisely. Under the sequential ignorability assumptions 1.1 to 1.3, Huber (2013) shows that the direct and indirect effects are non-parametrically identified. The iden- tification relies on a reweighing mechanism according to propensity scores P (G =1|M,C) and P (G =1|C). From this the direct and indirect effect can be calculated by using sample moments and propensity score estimates.

    Y · G Y · G 1 − P (G =1|M,C) η = E − E · (4.18) P (G =1|C) P (G =1|M,C) 1 − P (G =1|C)     Y · G 1 − P (G =1|M, C) Y · (1 − G) δ = E · − E (4.19) P (G =1|M, C) 1 − P (G =1|C) 1 − P (G =1|C)

For estimation, I use the normalized variants of (4.18) and (4.19) as in Huber (2013) and suggested by Imbens (2004). Their exact form can be found in the appendix. Both propensity scores p(M,C) and p(C) are estimated with probit regressions. The ATE can be estimated with a probit regression of gender on approval while conditioning on confounders which gives τ. From this it is straightfor- ward to calculate the mediated indirect effect for women and men:

η(1) = τ − δ(0) (4.20) η(0) = τ − δ(1) (4.21)

For a detailed derivation of the estimator and its normalized version I refer the reader to Huber (2013).

Empirical Specification

For the estimations, I make use of individual data from post-ballot surveys as described in the data section. I use the pooled set of the responses after three

132 different votes of both voters and non-voters who hypothetically state their voting decision. The dummy Yi takes the value 1 if individual i voted yes or would have voted yes, and the value 0 else. The main variable of interest is the gender dummy Gi which becomes 1 for women, and 0 for men. As shown above, I need to define vectors of mediators and confounders. In addition to voting behavior, the post-ballot surveys contain a wealth of socioeconomic and contextual variables which can be used as mediators and confounders. Some of the variables are available for all three votes and are denoted by Mbasic and Cbasic. Some variables have only been collected for a subset of the three votes and are therefore denoted by Mextended and Cextended respectively. As basic mediators of gender on accepting the voting proposi- tions I consider the following variables: work:full-time is a dummy if the respondent is full-time employed. A second dummy work:part-time denotes if the respondent has a part-time job. Not being full-time or part-time employed means that the respondent is either unemployed, or in pension, in education, or stays at home. The dummy education:high school becomes one if the re- spondent has a high-school degree or higher. The dummy education:vocational is one if the respondent received vocational training. Additional mediators available for a subset of the votes are the following: life standard is a dummy with value one if a respondent rated his or her life standard as high or medium high as compared to medium low and low (available for the votes in 1981 and 1991). Two dummies denote the respondent’s occupation in case he or she is not working full-time. work:pension denotes if a respondent has retired, and work:household reflects if a respondent stays at home to keep house (available for the votes in 1991 and 1993). I also include four income dummies (available for the vote in 1993). They reflect household income on a scale from 0 to 4 (3,001-5,000; 5,001-7,000; 7,001-9,000; >9,000 Swiss Francs per month). The category (<3,001) is left out as reference category. The following variables are used as confounders potentially influencing both the outcome and mediators and/or gender at the same time: age denotes the respondent’s age. The dummies status:married and status:single denote the respective marital status. If the respondent is Catholic this dummy takes the value one.20 urban is one if the respondent lives in a city and not in

20The majority of the Swiss population is either Roman Catholic (46.2% in 1980), or Protestant (45.3% in 1980) (data are from the website of the Swiss Statistical Office www.bfs.admin.ch).

133 Table 4.2: Descriptives of Post-Ballot Surveys by Gender

Mean Mean Variable (women) (men) Difference t-statistic p-value yes 0.566 0.622 -0.056 2.555 0.011

Mediators work:full-time 0.288 0.688 -0.401 19.532 0.000 work:part-time 0.249 0.054 0.195 -12.984 0.000 education:high 0.222 0.308 -0.086 4.369 0.000 school education:vocational 0.496 0.528 -0.032 1.445 0.149 life standard 0.360 0.389 -0.029 1.033 0.302 work:pension 0.141 0.201 -0.060 3.039 0.002 work:household 0.475 0.012 0.463 -25.795 0.000 income:1 0.342 0.332 0.011 -0.300 0.764 income:2 0.227 0.242 -0.015 0.472 0.637 income:3 0.142 0.174 -0.032 1.176 0.240 income:4 0.062 0.105 -0.043 2.084 0.038

Confounders age 44.973 47.196 -2.223 2.932 0.003 status:married 0.629 0.631 -0.002 0.072 0.943 status:single 0.190 0.256 -0.066 3.523 0.000 Catholic 0.419 0.448 -0.029 1.296 0.195 urban 0.622 0.578 0.044 -2.019 0.044 region:West 0.221 0.214 0.007 -0.355 0.723 region:Center 0.251 0.257 -0.006 0.312 0.755 region:Center-West 0.254 0.243 0.011 -0.547 0.584 region:Center-East 0.255 0.250 0.004 -0.228 0.819 housesize:2 0.277 0.343 -0.066 2.4389 0.0149 housesize:3 0.433 0.369 0.064 -2.2355 0.0256 housesize:4 0.114 0.117 -0.003 0.1684 0.8663 Note: T-tests based on data from VOX-surveys no. 161, 421, and 511. Data are available online on http://nesstar.sidos.ch/webview/index.jsp. a rural area. To account for cultural differences between the geographical and linguistic areas in Switzerland, I include the region dummies region:West, region:Center, region:Center-West,andregion:Center-East. The Southern re- gion is left out as reference group. For the votes in 1981 and 1991 I include dummies for the size of the respondent’s household: housesize:2 denotes a two-person household, housesize:3,4 three or four persons, and housesize:5 households with five or more inhabitants. To summarize, the vector of basic gender mediators is Mbasic = {work:full- time, work:part-time, education:high school, education:vocational} which is ex- tended by some of the following mediators for some of the votes Mextended =

134 {life standard, work:pension, work:household, income1, income2, income3, in- come4 }. The vector of basic confounders is Cbasic = {age, status:married, sta- tus:single, Catholic, urban, region:West, region:Center, region:Center-West, region:Center-East} which is extended by Cextended = {housesize:2, house- size:3,4, housesize:5 }. Summary statistics of the variables by gender, the mean difference and t-statistic are reported in Table 4.2.

4.5 Results

4.5.1 Average Treatment Effect

Figure 4.4 depicts the histograms of voters and yes votes in municipalities normalized by the number of adults for both voting regimes. This descriptive evidence show that the distribution of both voters and yes votes made a very similar shift to the right showing that both the number of voters and accepting votes increased significantly between the two votes. Table 4.3 shows the main regression results based on the votes in 1970 and 1971. With the exception of the fixed effects regression, the estimates are conducted using a weighted least squares estimator. Weights are proportional to the inverse of the total number of eligible voters in 1971 to account for heteroscedasticity of the standard error.21 In all regressions I use clustered

Figure 4.4: Histograms of Voter Participation and Acceptance Rate Participation Acceptance 12 12 10 10 8 8 6 6 Density Density 4 4 2 2 0 0 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1

Voters 1970/Adults Voters 1971/Adults Yes votes 1970/Adults Yes votes 1971/Adults

Note: Left panel: histogram of participation rate (voterst / Swiss adult populationt)by voting regime. Right panel: acceptance rate (yes votest / Swiss adult populationt)by voting regime. Municipal data.

21Residual plots after ordinary least squares regressions reveal that the variance of the

135 standard errors according to the 25 Swiss cantons to account for potential serial correlation of the error terms.22 Cantons are the most natural cluster for Swiss municipalities. All estimates are highly significant. In the baseline specification (1) with full sample and without canton fixed effects, the female acceptance rate amounts to 63.9%. Since the official voting result had an acceptance rate of 72.7%, this means that women were less likely to vote yes in the second ballot than men and the gender gap is negative. The second specification is run with a fixed effects estimator. Both first difference and fixed effects estimations lead to very similar regression coefficient. This means that violation of strict exo- geneity is not a problem, and I can continue the rest of the analysis based on first-difference estimates only. The remaining specifications include canton fixed effects. The estimated female acceptance rate slightly increases when canton fixed effects are ac- counted for. In other words, the gender preference gap for taxation becomes slightly smaller. In the remaining specifications (4) to (8), I restrict the sam- ple according to either the number of Swiss adults in the municipality, or the share of women to account for potential outliers. The female acceptance rate is relatively stable over the various specifications. The most likely explana- tion is that estimates are weighted by the inverse of the eligible population. This way, big municipalities like Zurich or Bern which would have biased the results receive lower weights in the regression. For the calculation of the gender wage gap, I first calculate the male ac- ceptance rate by taking the true voting result which has to be the gender turnout-weighted sum of the male and female acceptance rates. Taking the difference of the gender acceptance rates, yields a gender preference gap for taxation of -18.4 percentage points for baseline specification (1). For spec- ification (3) including canton fixed effects the gender gap amounts to -12.9 percentage points, and -11.8 percentage points when excluding very large and very small municipalities in specification (6). The gender preference gap is consistently negative over all specifications. Table 4.4 reports the results when the change in voters is interacted with

residual gets larger as Δvotersm increases. Since this variable is negative for several observations, I preferably use the number of voters in 1971 for weighting. This variable is always positive. 22Note that today’s 26th canton Jura was still part of the bigger canton Bern in 1971.

136 Table 4.3: Estimates of the Female Acceptance Rate from Ballot Propositions in 1970 and 1971

(1) (2) (3) (4) Sample Full sample Full sample Full sample <10,000 Swiss Δvoters 0.639*** 0.640*** 0.666*** 0.667*** (0.023) (0.002) (0.019) (0.027) Gender preference gap - 0.184 - 0.181 -0.129 -0.128 Canton FE No No Yes Yes Estimation Method FD FE FD FD Adjusted R2 0.962 0.985 0.972 0.962 Observations 2,143 4,286 2,143 2,095

(5) (6) (7) (8) <10,000 and <60 and <55 and Sample >100 Swiss >100 Swiss >40% women >45% women Δvoters 0.669*** 0.672*** 0.666*** 0.660*** (0.017) (0.024) (0.019) (0.022) Gender preference gap -0.124 -0.118 -0.130 -0.142 Canton FE Yes Yes Yes Yes Yes Estimation Method FD FD FD FD Adjusted R2 0.975 0.965 0.972 0.968 Observations 1,868 1,820 2,110 1,816 Note: *** p<0.01, ** p<0.05, * p<0.1. First difference (FD) with weighted least squares or fixed effects (FE) estimates. Dependent variable is change in the num- ber of yes votes (FD), or number of yes votes (FE). Clustered standard errors at cantonal level in brackets. For the fixed effects estimator, within R2 is reported. the percentage of Swiss adult women. The recovered female acceptance rate is in the first row of the table, and the estimated coefficient in the second one. Results have similar tendencies as in Table 4.3: including canton fixed effects results in higher estimated female acceptance rates. Again, excluding very large and very small municipalities as well as outliers regarding the share of women has only a small effect on the coefficient. However, in comparison to the baseline results the female acceptance rate when accounting for the share of women drops in all specifications. Instead of a female acceptance rate of 63.9% in baseline specification (1), the acceptance rate drops to 60.6%. Con- sequently, the gender gap increases in comparison to the baseline estimates. The results reject the hypothesis that women are more likely to favor tax- ation and thus government expenditure. At first sight, this is a surprising result because based on findings from Lott and Kenny (1999) the expectation goes into the opposite direction. Similarly, Abrams and Settle (1999) find

137 Table 4.4: Estimates of the Female Acceptance Rate from Ballot Propositions in 1970 and 1971

(9) (10) (11) (12) Sample Full sample Full sample <10,000 Swiss >100 Swiss Female acceptance 0.606 0.621 0.635 0.626 Δvoters ∗ %women 1.198*** 1.228*** 1.257*** 1.228*** (0.038) (0.031) (0.041) (0.030) Gender preference gap -0.244 -0.215 -0.188 -0.204 Canton FE No Yes Yes Yes Estimation Method FD FD FD FD Adjusted R2 0.965 0.973 0.964 0.975 Observations 2,143 2,143 2,095 1,868

(12) (13) (14) <10,000 and <60 and <55 and Sample >100 Swiss >40% women >45% women Female acceptance 0.641 0.623 0.633 Δvoters ∗ %women 1.261*** 1.228*** 1.253*** (0.038) (0.031) (0.039) Gender preference gap -0.173 -0.213 -0.188 Canton FE Yes Yes Yes Estimation Method FD FD FD Adjusted R2 0.967 0.973 0.969 Observations 1,820 2,110 1,816 Note: *** p<0.01, ** p<0.05, * p<0.1. First difference (FD) with weighted least squares. Dependent variable is change in the number of yes votes. Clus- tered standard errors at cantonal level in brackets. that Swiss public spending at federal level rose after enfranchising women in 1971. At second glance, however, there might exist several explanations for these findings. The most important one is that the ballot propositions at questions concerned government expenditure as an aggregate and did not distinguish separate spending issues. Literature examining gender preference gaps cited above suggested that women should be more likely to favor govern- ment spending on items like redistribution or health. Both are items which concern women more directly, either via lower employment rates and wages, or via the care for their children (e.g., Funk & Gathmann, 2012; Miller, 2008). The estimation strategy exploits the institutional change in female voting rights but relies on several assumptions about male voting behavior. E.g., I assume that men turn out with the same probability as before. But potentially they altered their turnout behavior as a reaction to female voting and lower

138 pivot probabilities. Since the estimated gender preference gap is negative, note that overestimating male participation is unproblematic as this would even underestimate the negative preference gap. Since the analysis is based on data from Switzerland which has a strong federal structure, a substitution effect from preferences for federal spending to cantonal spending due to female suffrage is of concern. However, the re- sults of Stutzer and Kienast (2005) suggest that cantonal spending decreased with female suffrage which is evidence against a substitution effect. Hence, preferences captured in my analysis are not specific for federal government expenditure but for government expenditure in general.

4.5.2 Direct and Indirect Effects

The main mediation analysis results based on individual post-ballot responses from the votes in 1981, 1991, and 1993 are reported in Tables 4.5 and 4.6. In Table 4.5, the first column shows the results for the complete sample, while in the remaining specifications the results are reported separately for each of the three ballots. In Table 4.6, I use combinations of the three ballots for which additional mediators and confounders are available as described in the empirical specification. The first three columns display results based on the same set of mediators and confounders that are available for all three ballots (Mbasic and Cbasic), while the other three specifications are based on the same subsets of observations but include extended vectors of mediators and confounders that are available for these subsets. In this way, columns (5)-(7) are directly comparable to columns (8)-(10) because they are based on the same subsets of observations. For the complete sample and all specifications in Table 4.6, I provide the propensity scores regression results of P (G =1|M,C) and P (G =1|C) con- ducted with probit estimates in the appendix (Tables 4.8 and 4.9). Figure 4.5 shows exemplary histograms of propensity scores for both men and women in the total sample (specification (1) in Table 4.5) to validate the common sup- port assumption 1.3. The assumption is not violated and enough comparable observations are available in both groups G. Propensity score histograms by gender for all the other specifications are attached in the appendix. They re- veal that the common support assumption might be violated for specifications

139 Figure 4.5: Histograms of Propensity Scores by Gender g=1 (female) g=0 (male) 6 6 5 5 4 4 3 3 Density Density 2 2 1 1 0 0 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 p(C) p(C)

g=1 (female) g=0 (male) 6 6 5 5 4 4 3 3 Density Density 2 2 1 1 0 0 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 p(M,C) p(M,C)

Note: Based on data from VOX-surveys no. 161, 421, and 511. Propensity scores from probit estimates. p(M, C) propensity scores include mediators and confounders, p(C) only the latter. based on the joint sample of votes from 1991 and 1993 when adding the ex- tended vector of mediators including dummies for retirement, house keeping, and income dummies. Including the measures for house keeping and income evidently reduces the number of comparable observations across gender be- cause men tend to earn more on average (6.2 vs. 10.5% are in the >9,000 income category), and hardly any men stay at home for house keeping whereas a large share of women does (47.5 vs. 1.2%). For this reason, the result in specification (9) in Table 4.6 should be interpreted with caution. Already descriptive statistics revealed that the female acceptance rate in the three ballots under investigation is significant and 5.6 percentage points lower than the male acceptance rate. Regressing gender on the binary outcome variable while controlling for all confounders using a probit estimator, gives similar values of -5.4 percentage points for the total sample, and between -3.2 and -9.7 percentage points in subsamples. The estimates are significant at conventional significance levels. Only in the subsample using data exclusively from the 1981 ballot (as well as in combination with the 1991 ballot), there

140 Table 4.5: Direct and Indirect Effects I

(1) (2) (3) (4) Total effect: τ -0.054** -0.032 -0.097*** -0.060* (0.022) (0.038) (0.037) (0.033)

Direct effect: δ(1) -0.094*** -0.018 -0.102** -0.112*** (0.029) (0.062) (0.052) (0.040) Indirect effect: η(1) -0.041** -0.024 -0.102** -0.002 (0.021) (0.052) (0.045) (0.029)

Direct effect: δ(0) -0.013 -0.008 0.005 -0.057 (0.029) (0.065) (0.057) (0.044) Indirect effect: η(0) 0.040* -0.014 0.006 0.053** (0.022) (0.051) (0.039) (0.026) M,C basic basic basic basic Ballots all 1981 1991 1993 Observations 2,018 535 686 797 Note: *** p<0.01, ** p<0.05, * p<0.1. Standard errors in brackets. Based on data from VOX-surveys no. 161, 421, and 511. Inverse propensity score weighted results. The binary de- pendent variable is 1 if the respondent voted yes, 0 if no. All specifications include mediators and confounders available for all three votes. Standard errors of total effect τ are from probit esti- mates. Standard errors for direct (δ) and indirect effects (η) are based on 799 bootstrap iterations. is no significant difference in the acceptance behavior of men and women.23 Recall, there is a survey bias in the 1981 survey that suggests some prob- lems with the representativeness of the data from this year. These values will henceforth be used as the total effect of being female on support for govern- ment spending τ and are reported in the first row of the results tables. Hence, the total effect as suggested by data from post-ballot surveys is smaller than what was estimated from the 1970/1971 referendums in the previous section. There are two potential explanations. First, the post-ballot surveys are based on data from votes that took place at a later point in time. Any kind of strate- gic voting incentives or hesitation to vote which might have been present in 1970/1971, should most likely have disappeared in the later ballots. Second, though very similar, the ballot propositions are not identical which also might explain part of the variation of the total effect over time. The objects of interest are the direct and indirect effect both evaluated at

23In all specifications I report the marginal effect of being female on voting yes from probit regressions as the total effect τ.

141 Table 4.6: Direct and Indirect Effects II

(5) (6) (7) (8) (9) (10) Total effect: τ -0.064 -0.066** -0.082** -0.062** -0.066** -0.082** (0.029) (0.026) (0.034) (0.029) (0.026) (0.034)

Direct effect: δ(1) -0.061** -0.120*** -0.132*** -0.066 -0.303*** -0.129*** (0.041) (0.035) (0.040) (0.043) (0.112) (0.039) Indirect Effect: η(1) -0.017** -0.036 0.001 -0.081*** -0.009 -0.001 (0.031) (0.024) (0.031) (0.029) (0.026) (0.034)

Direct effect: δ(0) -0.047 -0.029 -0.082* 0.019 -0.057 -0.081* 142 (0.041) (0.034) (0.047) (0.040) (0.038) (0.047) Indirect effect : η(0) -0.003 0.054** 0.050* 0.003 0.237** 0.048* (0.031) (0.026) (0.028) (0.035) (0.109) (0.028) M,C basic basic basic extended extended extended Ballots 1981, 1991 1991, 1993 1993 1981, 1991 1991, 1993 1993 Observations 1,190 1,483 719 1,190 1,483 719 Note: *** p<0.01, ** p<0.05, * p<0.1. Standard errors in brackets. Based on data from VOX-surveys no. 161, 421, and 511. Inverse propensity score weighted results. The binary de- pendent variable is 1 if the respondent voted yes, 0 if no. Columns (5) to (7) include mediators and confounders available for all three votes. Columns (8) to (10) include additional mediators and confounders exclusively available for a subset of votes on top of the basic mediators and confounders. Standard errors of total effect τ are from probit estimates. Standard errors for direct (δ) and indirect effects (η) are based on 799 bootstrap iterations. the value g =1, δ(1) and η(1), for women in the second and third rows of the results tables. The respective values for men are reported in the last two rows for completeness. Standard errors are computed with a bootstrap procedure with 799 iterations. Results suggest that the direct effect of being female on the probability of voting yes in the financial order referendums is -9.4 per- centage points in the full sample in specification (1). Recall, this means that when evaluating mediators at their values for women, the expected differ- ence in acceptance rates between men and women amounts to -9.4 percentage points. The direct effect is highly significant in most specifications. For the significant specifications it varies between -6.1 to -13.2 percentage points in the subsamples. There is one extreme outlier with a value of -30.3 percentage points in specification (9). The common support assumption is most likely violated for this specifica- tion so the size of the effect should not be taken at face value. The size of the direct effect is considerable and leads to the conclusion that a large part of the effect of being female on preferences for government support runs through intrinsically female factors, and directly affects voting behavior. The mediated effect amounts to -4.1 percentage points in the baseline specification (1) but varies strongly between -1.7 and -10.2 percentage points in the other significant specifications. In total, the coefficient is significant in only four out of ten specifications. The negative coefficient reflects that mediators like full-time employment and high-school education, for which sig- nificantly negative gender gaps exist, mediate part of the gender effect and make women less likely to vote yes in these type of referendums. However, the direct gender effect is much stronger than the indirect gender effect. For robustness, I account for potentially serially correlated standard errors either within ballots or within cantons. Table 4.7 shows the mediation results for specifications based on at least two ballots with standard errors clustered at ballot level. The direct gender effect now turns significant in all specifica- tions, and the indirect effect is significant in three out of five specifications. Thus, the significance of the results even improves when accounting for ballot clusters. I repeat the estimates while clustering the standard errors at can- tonal level. Cantons are not reported for 195 observations previously used, so I drop these observations. I also exclude the regional dummies from the set of confounders because they correspond to groups of cantons. First, I rerun the

143 Table 4.7: Direct and Indirect Effects, Ballot Clusters

(1) (2) (3) (4) (5) Total effect: τ -0.054*** -0.064 -0.066*** -0.062 -0.066*** (0.014) (0.043) (0.018) (0.041) (0.018)

Direct effect: δ(1) -0.094*** -0.061** -0.120*** -0.066*** -0.303* (0.027) (0.031) (0.007) (0.024) (0.192) Indirect effect: η(1) -0.041* -0.017** -0.036 -0.081** -0.009 (0.025) (0.033) (0.036) (0.033) (0.036)

Direct effect: δ(0) -0.013 -0.047 -0.029 0.019 -0.057*** (0.021) (0.011) (0.022) (0.014) (0.022) Indirect effect: η(0) 0.040 -0.003 0.054** 0.003 0.237** (0.020) (0.008) (0.021) (0.012) (0.206) M,C basic basic basic extended extended Ballots all 1981, 1991 1991, 1993 1981, 1991 1991, 1993 Observations 2,018 1,190 1,483 1,190 1,483 Note: *** p<0.01, ** p<0.05, * p<0.1. Standard errors in brackets. Based on data from VOX-surveys no. 161, 421, and 511. Inverse propensity score weighted results. The binary dependent variable is 1 if the respondent voted yes, 0 if no. Columns (1) to (3) include mediators and confounders available for all three votes. Columns (4) and (5) include additional mediators and confounders exclusively available for a subset of votes on top of the basic mediators and confounders. Stan- dard errors of total effect τ are from probit estimates. Standard errors for direct (δ) and indirect effects (η) are based on 799 bootstrap iterations with ballot clusters. main estimates from Tables 4.5 and 4.6 with the reduced set of observations for which information on cantons is available (reported in Table 4.10 in the appendix). In this way, I make sure that the significance of the results is not affected by the reduced sample size. Indeed, coefficients change only slightly. Second, I cluster the error terms at cantonal level (reported in Table 4.11). Even though significance is slightly reduced as compared to coefficients with- out canton clusters in Table 4.11, results are very similar, which means that clustering at cantonal level does not have a big effect on the significance of the results. Moreover, I repeat the four baseline estimates as reported in Table 4.5 based on the sample of respondents who turned out on ballot day. This is to account for potential differences in the voting and non-voting population. The estimation results are reported in Table 4.12 in the appendix. They show that the voting population indeed differs from the total population. While the direct gender effect is significant in all but the 1981 specification and of similar size as before, all total effects and indirect female effects are insignificant.

144 In sum, evidence points to the importance of direct gender effects when explaining the gender gap in acceptance behavior. Potential explanations of the strong direct gender effect are unobserved mediators. Though the effect is called direct, it not only refers to observable differences between men and women. Research based on experimental techniques examines gender gaps other than socioeconomic differences which might explain why women could have different preferences for government spending than men (cf. Croson and Gneezy (2009) as well as Shapiro and Mahajan (1986) for literature reviews). Literature documents that women are more risk averse (e.g., Holt & Laurya, 2002, 2005; Schubert et al., 1999) and dislike competition (Gneezy, Niederle & Rustichini, 2003; Niederle & Vesterlund, 2007). Experimental evidence suggests that women are more altruistic, and dislike inequality (Andreoni & Vesterlund, 2001; Selten & Ockenfels, 1998). This evidence might partly reflect the strong direct gender effect.

4.6 Concluding Remarks

The aim of this paper is to provide direct evidence for gender preferences for government expenditure from ballot analysis. This method is preferable to analyzing indirect links between the electorate, politicians, and their subse- quent choice of budgets and policies since the relation between preferences and subsequent voting behavior is much clearer. The analysis of the main effect is based on aggregate voting data around the introduction of female suffrage such that individual voting behavior remains unobserved. However, I argue extensively that my preference measures are likely to reflect gender preference gaps, and also provide evidence from post-ballot surveys of comparable votes. I find that approval for government spending is lower among women than among men. My findings contrast with Lott and Kenny’s (1999) results. It might be that the timing of female voting and the takeoff of government spending correlates in their examination, however, this does not necessarily mean that women display stronger preferences for larger governments than men. The estimation results suggest that the expected gender preference gap does not exist when looking at the total of what governments spend. While the results seem surprising at first, they might stem from the fact that the

145 object of my analysis is the aggregate government expenditure, and potentially reflects a bundle of taxes and ways to finance government spending. However, my findings are compatible with other results from literature which shows that gender preference gaps exist but only for certain spending categories. This suggests that the scope of government might matter more in the analysis of gender preference gaps than the size of government. For example, Abrams and Settle (1999) find particularly strong effects of female suffrage on welfare spending in Switzerland. Similarly, Aidt et al. (2006) show increases in health, welfare, and education spending which are categories typically relevant to women. Also Funk and Gathmann (2012) discover gender preference gaps from individual data for health, environmental issues, defense and welfare spending. This paper also shows that it is important to account for socioeconomic gender differences, mediators, when analyzing gender preference gaps for the size of government. Otherwise some of the voting differences which are due to differences in employment are taken to be caused by gender, and could overestimate female preferences for government. However, I find weaker evi- dence for mediated effects than for direct effects. In more detail, factors being intrinsic to women display a much stronger negative effect on the probability of supporting government spending than socioeconomic variables. After the enfranchisement of women, public support for a larger govern- ment budget from people who are eligible to elect politicians is not larger than before. Regarding the outcome of the vote, female suffrage does not change anything because women are not pivotal in this case. Most likely, the male population alone would also have approved the proposition. Democratic legitimization of government expenditure does not increase per se but only for spending categories which are more important for women than for men. For future research, my results emphasize the need to put effort into un- derstanding for which spending categories gender preference gaps exist and what their determinants are.

146 Appendix

4.A Estimators of Direct and Indirect Effects

To estimate the direct and indirect effects, I estimate normalized versions of (4.18) and (4.19) (Huber, 2013). For the normalization weights are adjusted such that they add up to one for both men and women. For simplification, write p(M,C) ≡ P (G =1|M,C) and p(C) ≡ P (G =1|C), with their their estimated counterparts p(M,C) and p(C). Let i denote the index for each of the N observations. Then the direct effect evaluated at g =1and g =0 respectively is identified by the following equations: N N −1 N N −1 1 1 2 2 δ(1) = YiWi YiWi − YiWi YiWi (4.22) i=1 i=1 i=1 i=1 N N −1 N N −1 3 3 4 4 δ(0) = YiWi YiWi − YiWi YiWi (4.23) i=1 i=1 i=1 i=1

The four weights are defined as:

1 Gi Wi ≡ p(Ci)

2 (1 − Gi)p(Mi,Ci) Wi ≡ (1 − p(Mi,Ci))p(Ci)

3 Gi(1 − p(Mi,Ci)) Wi ≡ p(Mi,Ci)(1 − p(Ci))

4 1 − Gi Wi ≡ 1 − p(Ci) p(Mi,Ci) and p(Ci) are estimated with probit regressions, and the rest of the estimator is based on sample moments from which it is straightforward to calculate both δ(1) and δ(0).

147 4.B Propensity Score Histograms

Propensity Score Histograms of Table 4.5, No Clustering

g=1 (female) g=0 (male) 6 6 5 5 4 4 3 3 Density Density 2 2 1 1 0 0 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 p(C) p(C)

g=1 (female) g=0 (male) 6 6 5 5 4 4 3 3 Density Density 2 2 1 1 0 0 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 p(M,C) p(M,C)

Figure 4.6: Histograms of Propensity Scores (Mbasic and Cbasic, Vote 1981)

g=1 (female) g=0 (male) 6 6 5 5 4 4 3 3 Density Density 2 2 1 1 0 0 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 p(C) p(C)

g=1 (female) g=0 (male) 6 6 5 5 4 4 3 3 Density Density 2 2 1 1 0 0 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 p(M,C) p(M,C)

Figure 4.7: Histograms of Propensity Scores (Mbasic and Cbasic, Vote 1991)

148 1991) iue49 itgaso rpniySoe ( Scores Propensity of Histograms 4.9: Figure Clustering No 4.6, Table of Histograms Score Propensity iue48 itgaso rpniySoe ( Scores Propensity of Histograms 4.8: Figure

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 g=1 (female) g=1 (female) g=1 (female) g=1 (female) .4 .4 .4 .4 p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 .8 .8 .8 .8 149 1 1 1 1

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 M M g=0 (male) g=0 (male) g=0 (male) g=0 (male) basic basic .4 .4 .4 .4 p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 and and .8 .8 .8 .8 C C basic basic 1 1 1 1 oe 1981, Votes , oe1993) Vote , iue41:Hsorm fPoest crs( Scores Propensity of Histograms 4.10: Figure 1993) iue41:Hsorm fPoest crs( Scores Propensity of Histograms 4.11: Figure

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 g=1 (female) g=1 (female) g=1 (female) g=1 (female) .4 .4 .4 .4 p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 .8 .8 .8 .8 150 1 1 1 1

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 M M g=0 (male) g=0 (male) g=0 (male) g=0 (male) .4 .4 .4 .4 basic basic p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 and and .8 .8 .8 .8 C C 1 1 1 1 basic basic oe 1991, Votes , oe1993) Vote , iue41:Hsorm fPoest crs( Scores Propensity of Histograms 4.13: Figure ( Scores Propensity of Histograms 4.12: Figure 91 1993) 1991, 1991) 1981,

Density Density Density Density 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 g=1 (female) g=1 (female) g=1 (female) g=1 (female) .4 .4 .4 .4 p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 .8 .8 .8 .8 151 1 1 1 1

Density Density Density Density 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 M M g=0 (male) g=0 (male) g=0 (male) g=0 (male) extended extended .4 .4 .4 .4 p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 .8 .8 .8 .8 and and 1 1 1 1 C C extended extended Votes , Votes , iue41:Hsorm fPoest crs( Scores Propensity of Histograms 4.14: Figure iue41:Hsorm fPoest crs( Scores Propensity of Histograms 4.15: Figure cluster- Ballot 4.7, ing Table of Histograms Score Propensity 91 93 altClustering) Ballot 1993, 1991, 1993)

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 g=1 (female) g=1 (female) g=1 (female) g=1 (female) .4 .4 .4 .4 p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 .8 .8 .8 .8 152 1 1 1 1

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 M M g=0 (male) g=0 (male) g=0 (male) g=0 (male) .4 .4 .4 .4 basic extended p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 and .8 .8 .8 .8 and C 1 1 1 1 basic C extended oe 1981, Votes , Vote , iue41:Hsorm fPoest crs( Scores Propensity of Histograms 4.17: Figure ( Scores Propensity of Histograms 4.16: Figure 93 altClustering) Ballot 1993, Clustering) Ballot 1991,

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 g=1 (female) g=1 (female) g=1 (female) g=1 (female) .4 .4 .4 .4 p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 .8 .8 .8 .8 153 1 1 1 1

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 M M g=0 (male) g=0 (male) g=0 (male) g=0 (male) .4 .4 .4 .4 basic basic p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 and and .8 .8 .8 .8 C C 1 1 1 1 basic basic oe 1991, Votes , 1981, Votes , iue41:Hsorm fPoest crs( Scores Propensity of Histograms 4.19: Figure ( Scores Propensity of Histograms 4.18: Figure 91 93 altClustering) Ballot 1993, 1991, Clustering) Ballot 1991, 1981,

Density Density Density Density 0 2 4 6 8 10 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 g=1 (female) g=1 (female) g=1 (female) g=1 (female) .4 .4 .4 .4 p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 .8 .8 .8 .8 154 1 1 1 1

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 M M g=0 (male) g=0 (male) g=0 (male) g=0 (male) extended extended .4 .4 .4 .4 p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 .8 .8 .8 .8 and and 1 1 1 1 C C extended extended Votes , Votes , Propensity Score Histograms of Table 4.10, Canton Infor- mation Available, No Clustering

g=1 (female) g=0 (male) 6 6 5 5 4 4 3 3 Density Density 2 2 1 1 0 0 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 p(C) p(C)

g=1 (female) g=0 (male) 6 6 5 5 4 4 3 3 Density Density 2 2 1 1 0 0 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 p(M,C) p(M,C)

Figure 4.20: Histograms of Propensity Scores (Mbasic and Cbasic, Votes 1981, 1991, 1993, Canton Known)

g=1 (female) g=0 (male) 8 8 7 7 6 6 5 5 4 4 Density Density 3 3 2 2 1 1 0 0 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 p(C) p(C)

g=1 (female) g=0 (male) 8 8 7 7 6 6 5 5 4 4 Density Density 3 3 2 2 1 1 0 0 0 .2 .4 .6 .8 1 0 .2 .4 .6 .8 1 p(M,C) p(M,C)

Figure 4.21: Histograms of Propensity Scores (Mbasic and Cbasic, Votes 1981, 1991, Canton Known)

155 atnKnown) Canton iue42:Hsorm fPoest crs( Scores Propensity of Histograms 4.23: Figure ( Scores Propensity of Histograms 4.22: Figure 93 atnKnown) Canton 1993,

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 g=1 (female) g=1 (female) g=1 (female) g=1 (female) .4 .4 .4 .4 p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 .8 .8 .8 .8 156 1 1 1 1

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 M M g=0 (male) g=0 (male) g=0 (male) g=0 (male) .4 .4 .4 .4 basic basic p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 and and .8 .8 .8 .8 C C 1 1 1 1 basic basic oe 1991, Votes , oe1993, Vote , iue42:Hsorm fPoest crs( Scores Propensity of Histograms 4.25: Figure ( Scores Propensity of Histograms 4.24: Figure 91 93 atnKnown) Canton 1993, 1991, Known) Canton 1991, 1981,

Density Density Density Density 0 2 4 6 8 0 5 10 15 20 25 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 g=1 (female) g=1 (female) g=1 (female) g=1 (female) .4 .4 .4 .4 p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 .8 .8 .8 .8 157 1 1 1 1

Density Density Density Density 0 1 2 3 4 5 6 0 5 10 15 20 25 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 M M g=0 (male) g=0 (male) g=0 (male) g=0 (male) extended extended .4 .4 .4 .4 p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 .8 .8 .8 .8 and and 1 1 1 1 C C extended extended Votes , Votes , iue42:Hsorm fPoest crs( Scores Propensity of Histograms 4.26: Figure iue42:Hsorm fPoest crs( Scores Propensity of Histograms 4.27: Figure Clus- Canton 4.11, tering Table of Histograms Score Propensity 91 93 atnClustering) Canton 1993, 1991, 93 atnKnown) Canton 1993,

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 g=1 (female) g=1 (female) g=1 (female) g=1 (female) .4 .4 .4 .4 p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 .8 .8 .8 .8 158 1 1 1 1

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 M M g=0 (male) g=0 (male) g=0 (male) g=0 (male) .4 .4 .4 .4 basic extended p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 and .8 .8 .8 .8 and C 1 1 1 1 basic C extended oe 1981, Votes , Vote , iue42:Hsorm fPoest crs( Scores Propensity of Histograms 4.29: Figure ( Scores Propensity of Histograms 4.28: Figure 93 atnClustering) Canton 1993, Clustering) Canton 1991,

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8 0 0 0 0 .2 .2 .2 .2 g=1 (female) g=1 (female) g=1 (female) g=1 (female) .4 .4 .4 .4 p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 .8 .8 .8 .8 159 1 1 1 1

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8 0 0 0 0 .2 .2 .2 .2 M M g=0 (male) g=0 (male) g=0 (male) g=0 (male) .4 .4 .4 .4 basic basic p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 and and .8 .8 .8 .8 C C 1 1 1 1 basic basic oe 1991, Votes , 1981, Votes , atnClustering) Canton iue43:Hsorm fPoest crs( Scores Propensity of Histograms 4.31: Figure ( Scores Propensity of Histograms 4.30: Figure 91 91 atnClustering) Canton 1991, 1981,

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 g=1 (female) g=1 (female) g=1 (female) g=1 (female) .4 .4 .4 .4 p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 .8 .8 .8 .8 160 1 1 1 1

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 0 0 0 .2 .2 .2 .2 M M g=0 (male) g=0 (male) g=0 (male) g=0 (male) extended .4 .4 .4 .4 basic p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 and .8 .8 .8 .8 and C 1 1 1 1 basic C extended oe1993, Vote , Votes , iue43:Hsorm fPoest crs( Scores Propensity of Histograms 4.33: Figure ( Scores Propensity of Histograms 4.32: Figure 93 atnClustering) Canton 1993, 91 93 atnClustering) Canton 1993, 1991,

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 7 8 9 0 5 10 15 20 25 0 0 0 0 .2 .2 .2 .2 g=1 (female) g=1 (female) g=1 (female) g=1 (female) .4 .4 .4 .4 p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 .8 .8 .8 .8 161 1 1 1 1

Density Density Density Density 0 1 2 3 4 5 6 0 1 2 3 4 5 6 0 1 2 3 4 5 6 7 8 9 0 5 10 15 20 25 0 0 0 0 .2 .2 .2 .2 M M g=0 (male) g=0 (male) g=0 (male) g=0 (male) extended .4 .4 .4 .4 extended p(M,C) p(M,C) p(C) p(C) .6 .6 .6 .6 .8 .8 .8 .8 and and 1 1 1 1 C C extended extended Votes , Vote , 4.C Propensity Score Estimates

Table 4.8: Propensity Score Estimates with Mbasic and Cbasic

(1) (2) (3) (4) (5) (6) (7) (8) Mediators

work:full-time -1.208*** -1.338*** -1.097*** -1.045*** (0.076) (0.104) (0.087) (0.125) work:part 0.328*** 0.275* 0.420*** 0.447*** -time (0.104) (0.145) (0.116) (0.162) education: -0.699*** -0.739*** -0.735*** -0.709*** high school (0.094) (0.127) (0.113) (0.162) education: -0.341*** -0.361*** -0.323*** -0.337** vocational (0.083) (0.104) (0.102) (0.150) 162

Confounders

status:married -0.400*** -0.409*** -0.386*** -0.418*** -0.407*** -0.448*** -0.481*** -0.478*** (0.085) (0.090) (0.112) (0.121) (0.100) (0.105) (0.143) (0.150) status:single -0.792*** -0.747*** -0.764*** -0.714*** -0.837*** -0.804*** -0.863*** -0.776*** (0.109) (0.115) (0.144) (0.153) (0.128) (0.133) (0.183) (0.190) Catholic -0.031 -0.033 -0.045 -0.090 -0.020 -0.000 0.078 0.143 (0.061) (0.066) (0.079) (0.086) (0.071) (0.077) (0.102) (0.112) region:West 0.414** 0.244 0.471* 0.295 0.384** 0.249 0.255 0.116 (0.190) (0.203) (0.282) (0.313) (0.195) (0.208) (0.279) (0.293) region:Center 0.418** 0.292 0.573** 0.392 0.357* 0.281 0.156 0.102 (0.189) (0.201) (0.282) (0.312) (0.193) (0.205) (0.277) (0.290) region:Center 0.402** 0.237 0.467* 0.333 0.416** 0.235 0.309 0.162 -West (0.190) (0.202) (0.282) (0.312) (0.194) (0.206) (0.279) (0.292) region:Center 0.412** 0.289 0.485* 0.356 0.400** 0.301 0.319 0.178 -East (0.189) (0.200) (0.281) (0.311) (0.193) (0.204) (0.278) (0.290) (1) (2) (3) (4) (5) (6) (7) (8) age -0.012*** -0.027*** -0.010*** -0.030*** -0.013*** -0.025*** -0.015*** -0.025*** (0.002) (0.002) (0.003) (0.003) (0.002) (0.003) (0.003) (0.004) urban 0.112* 0.137** 0.073 0.053 0.066 0.130 0.086 0.197* (0.061) (0.067) (0.079) (0.086) (0.073) (0.080) (0.106) (0.116) constant 0.401* 2.149*** 0.232 2.379*** 0.487* 1.994*** 0.783** 2.041*** (0.234) (0.280) (0.338) (0.412) (0.255) (0.308) (0.358) (0.427) 2 Adjusted R 0.026 0.199 0.023 0.209 0.028 0.192 0.034 0.199 Ballots all all ’81, ’91 ’81, ’91 ’91, ’93 ’91, ’93 ’93 ’93 Observations 2,018 2,018 1,190 1,190 1,483 1,483 719 719

Note: *** p<0.01, ** p<0.05, * p<0.1. Standard errors in brackets. Based on data from VOX-surveys no. 161, 421, and 511. Propensity score estimates based on vectors Mbasic and Cbasic exclusively. Dependent variable is dummy being 1 for female, 0 for men. Columns (1), (3), (5) and (7) are P (G =1|C). Columns (2), (4), (6), and (8) are P (G =1|M, C). 163 Table 4.9: Propensity Score Estimates with Mextended and Cextended

(1) (2) (3) (4) (5) (6) Mediators

work:full-time -1.361*** -0.376*** -0.218 (0.106) (0.141) (0.202) work:part-time 0.265* 0.112 0.092 (0.147) (0.153) (0.210) education:high school -0.810*** -0.624*** -0.425** (0.133) (0.124) (0.190) education:vocational -0.385*** -0.352*** -0.227 (0.106) (0.112) (0.173) life standard 0.161* (0.089) work:pension 0.105 0.107

164 (0.204) (0.312) work:household 2.308*** 2.367*** (0.181) (0.243) income:1 -0.268 (3,001-5,000) (0.173) income:2 -0.454** (5,001-7,000) (0.199) income:3 -0.370* (7,001-9,000) (0.222) income:4 -0.576** (>9,001) (0.257)

Confounders

status:married -0.365*** -0.403*** -0.407*** -0.720*** -0.481*** -0.688*** (0.133) (0.142) (0.100) (0.112) (0.143) (0.169) status:single -0.732*** -0.668*** -0.837*** -0.438*** -0.863*** -0.265 (0.146) (0.155) (0.128) (0.138) (0.183) (0.200) Catholic -0.042 -0.087 -0.020 -0.076 0.078 -0.008 (0.079) (0.087) (0.071) (0.084) (0.102) (0.125) (1) (2) (3) (4) (5) (6) region:West 0.466* 0.327 0.384** 0.198 0.255 0.057 (0.283) (0.315) (0.195) (0.217) (0.279) (0.313) region:Center 0.571** 0.415 0.357* 0.169 0.156 0.008 (0.282) (0.313) (0.193) (0.214) (0.277) (0.310) region:Center-West 0.465* 0.356 0.416** 0.150 0.309 0.163 (0.283) (0.313) (0.194) (0.216) (0.279) (0.312) region:Center-East 0.485* 0.381 0.400** 0.138 0.319 -0.017 (0.281) (0.312) (0.193) (0.214) (0.278) (0.312) age -0.009*** -0.029*** -0.013*** -0.012*** -0.015*** -0.011* (0.003) (0.003) (0.002) (0.004) (0.003) (0.006) urban 0.076 0.061 0.066 0.227*** 0.086 0.309** (0.079) (0.087) (0.073) (0.087) (0.106) (0.131) housesize:2 -0.113 -0.139 (0.133) (0.140) housesize:3,4 0.033 0.008 165 (0.142) (0.150) housesize:5 -0.053 -0.015 (0.171) (0.185) constant 0.184 2.309*** 0.487* 0.804** 0.783** 0.809* (0.355) (0.429) (0.255) (0.335) (0.358) (0.478) 2 Adjusted R 0.024 0.212 0.028 0.322 0.034 0.349 Ballots 1981, 1991 1981, 1991 1991, 1993 1991, 1993 1993 1993 Observations 1,190 1,190 1,483 1483 719 719 Note: *** p<0.01, ** p<0.05, * p<0.1. Standard errors in brackets. Based on VOX- surveys no. 161, 421, and 511. Propensity score estimates based on vectors Mextended and Cextended and vectors Mbasic and Cbasic. Dependent variable is dummy being 1 for female, 0 for men. Columns (1), (3), and (5) are P (G =1|C). Columns (2), (4), and (6) are P (G =1|M, C). 4.D Direct and Indirect Effects

Table 4.10: Direct and Indirect Effects, Observations with Information about Cantons

(1) (2) (3) (4) (5) (6) (7) Total τ -0.056** -0.071** -0.064** -0.074** -0.070** -0.069*** -0.074** effect (0.023) (0.032) (0.026) (0.034) (0.032) (0.026) (0.034)

Direct δ(1) -0.095*** -0.059 -0.119*** -0.129*** -0.056 -0.277*** -0.127*** effect (0.030) (0.045) (0.035) (0.044) (0.045) (0.104) (0.044) Indirect η(1) -0.044** -0.099*** -0.036 0.006 -0.107*** -0.007 0.006 effect (0.022) (0.031) (0.025) (0.031) (0.033) (0.026) (0.033)

Direct δ(0) -0.012 0.029 -0.028 -0.080* 0.036 -0.063* -0.079* effect (0.030) (0.042) (0.036) (0.047) (0.044) (0.036) (0.047) 166 Indirect η(0) 0.039* -0.011 0.055** 0.055* -0.015 0.208** 0.053* effect (0.022) (0.033) (0.025) (0.029) (0.035) (0.102) (0.032) M,C basic basic basic basic extended extended extended Ballots all 1981, 1991 1991, 1993 1993 1981, 1991 1991, 1993 1993 Observations 2,018 1,190 1,483 719 1,190 1,483 719 Note: *** p<0.01, ** p<0.05, * p<0.1. Standard errors in brackets. Based on data from VOX- surveys no. 161, 421, and 511. Inverse propensity score weighted results. The binary dependent variable is 1 if the respondent voted yes, 0 if no. The sample is restricted to observations for which the canton is known. Columns (1) to (4) include mediators and confounders available for all three votes. Columns (5) to (7) include additional mediators and confounders exclusively available for a subset of votes on top of the basic mediators and confounders. Standard errors of total effect τ are from probit estimates. Standard errors for direct (δ) and indirect effects (η) are based on 799 boot- strap iterations. Table 4.11: Direct and Indirect Effects, Canton Clusters

(1) (2) (3) (4) (5) (6) (7) Total τ -0.056** -0.071** -0.064** -0.074** -0.070** -0.069*** -0.074** effect (0.024) (0.029) (0.027) (0.035) (0.029) (0.026) (0.035)

Direct δ(1) -0.095*** -0.059 -0.119*** -0.129*** -0.056 -0.277* -0.127*** effect (0.029) (0.047) (0.032) (0.046) (0.049) (0.145) (0.046) Indirect η(1) -0.044** -0.099*** -0.036* 0.006 -0.107*** -0.007 0.006 effect (0.019) (0.031) (0.018) (0.044) (0.034) (0.025) (0.043)

Direct δ(0) -0.012 0.029 -0.028 -0.080 0.036 -0.063 -0.079

167 effect (0.030) (0.038) (0.032) (0.059) (0.038) (0.039) (0.055) Indirect η(0) 0.039* -0.011 0.055** 0.055* -0.015 0.208 0.053 effect (0.022) (0.033) (0.024) (0.032) (0.036) (0.147) (0.033) M,C basic basic basic basic extended extended extended Ballots all 1981, 1991 1991, 1993 1993 1981, 1991 1991, 1993 1993 Observations 2,018 1,190 1,483 719 1,190 1,483 719 Note: *** p<0.01, ** p<0.05, * p<0.1. Standard errors in brackets. Based on data from VOX- surveys no. 161, 421, and 511. Inverse propensity score weighted results. The binary dependent variable is 1 if the respondent voted yes, 0 if no. The sample is restricted to observations for which the canton is known. Columns (1) to (4) include mediators and confounders available for all three votes. Columns (5) to (7) include additional mediators and confounders exclusively available for a subset of votes on top of the basic mediators and confounders. Standard errors of total effect τ are from probit estimates. Standard errors for direct (δ) and indirect effects (η) are based on 799 boot- strap iterations with canton clusters. Table 4.12: Direct and Indirect Effects for Respondents who Voted

(1) (2) (3) (4) Total effect: τ -0.043 -0.022 -0.081 -0.060 (0.027) (0.047) (0.050) (0.038)

Direct effect: δ(1) -0.105*** -0.006 -0.167*** -0.120*** (0.033) (0.075) (0.071) (0.046) Indirect effect: η(1) -0.016 0.055 -0.050 0.007 (0.027) (0.089) (0.054) (0.040)

Direct effect: δ(0) -0.027 -0.029 -0.030 -0.067 (0.038) (0.104) (0.074) (0.055) Indirect effect: η(0) 0.063*** 0.032 0.088* 0.059** (0.022) (0.062) (0.052) (0.030) M,C basic basic basic basic Ballots all 1981 1991 1993 Observations 1,258 292 404 562 Note: *** p<0.01, ** p<0.05, * p<0.1. Standard errors in brackets. Based on data from VOX-surveys no. 161, 421, and 511. Only respondents who turned out to vote are included. In- verse propensity score weighted results. The binary dependent variable is 1 if the respondent voted yes, 0 if no. All specifica- tions include mediators and confounders available for all three votes. Standard errors of total effect τ are from probit estimates. Standard errors for direct (δ) and indirect effects (η) are based on 799 bootstrap iterations.

168 4.E Federal Announcements / Bundesblätter

The federal announcements are accessible online via http : //www.amts druckschriften.bar.admin.ch.

• Federal Announcement 1962 I, pp. 997-1014. Botschaft des Bundesrates an die Bundesversammlung über die Weiterführung der Finanzordnung des Bundes.

• Federal Announcement 1969 II, pp. 749-807. Botschaft des Bundesrates and die Bundesversammlung über die Änderung der Finanzordnung des Bundes.

• Federal Announcement 1970 II, pp. 1-5. Bundesbeschluss über die Änderung der Finanzordnung des Bundes.

• Federal Announcement 1970 II, pp. 1581-1608. Botschaft des Bun- desrates an die Bundesversammlung über die Weiterführung der Finan- zordnung des Bundes.

• Federal Announcement 1971 I, pp. 486-491. Bundesbeschluss über die Weiterführung der Finanzordnung des Bundes.

• Federal Announcement 2003 I, pp. 1531-1565. Botschaft über die neue Finanzordnung.

• Information about mutations of the municipalities are taken from the historical municipality register of the Swiss Statistical Office available online http : //www.bfs.admin.ch/bfs/ portal/de/index/infothek/ nomenklaturen/blank/blank/ gem_liste/02.html

• The Année Politique Suisse (2012) is accessible online (http : //www. anneepolitique.ch/de/aps − online.php) and provides additional back- ground information on ballots.

• Number of voters for cantonal votes is available online from the Centre for research on direct democracy on www.c2d.ch.

169 • Information about municipalities counting votes together in the canton Bern, and political municipalities in the canton Thurgau were received by email from the Swiss Statistical Office. They are available on request.

• Data used from Swiss census (1970): total population

• Voting data are from the Political Atlas of Switzerland of the Swiss Statistical Office. They were retrieved for the following ballots:

– Bundesbeschluss vom 27.09.1963 über die Weiterführung der Fi- nanzordnung des Bundes (Verlängerung der Geltungsdauer von Art.41ter BV und Ermässigung der Wehrsteuer). Ballot on 8 De- cember 1963. – Bundesbeschluss vom 24.06.1970 über die Änderung der Finan- zordnung des Bundes. Ballot on 15 November 1970. – Bundesbeschluss vom 09.10.1970 über die Einführung des Frauen- stimm- und Wahlrechts in eidgenössischen Angelegenheiten. Ballot on 7 February 1971. – Bundesbeschluss vom 11.03.1971 über die Weiterführung der Fi- nanzordnung des Bundes. Ballot on 6 June 1971.

4.F Presentations and Acknowledgement

This paper has been presented at the following conferences and workshops: Meeting of the European Public Choice Society (April 2013, Zürich, Switzer- land), Spring Meeting of Young Economists (May/June 2013, Aarhus, Den- mark), European Political Science Association Annual Meeting (June 2013, Barcelona, Spain), CESifo Venice Summer Institute (July 2013, Venice, Italy), European Economic Association Congress (August 2013, Gothenburg, Swe- den), and Jahrestagung Verein für Socialpolitik (September 2013, Düsseldorf, Germany). For valuable comments, I thank Monika Bütler, Patricia Funk, Martin Hu- ber, and Alois Stutzer. I appreciate helpful input by the following discussants at conferences and workshops: Andreas Bernecker, Krisztina Kis-Katos, and Silke Übelmesser.

170 Bibliography

Abrams, Burton A. and Russell F. Settle, “Women’s Suffrage and the Growth of the Welfare State,” Public Choice, 1999, 100 (3/4), 289–300.

Aidt, Toke S. and Bianca Dallal, “Female voting power: The contribution of women’s suffrage to the growth of social spending in Western Europe (1869-1960),” Public Choice, 2008, 134 (3/4), 391–417.

, Jayasri Dutta, and Elena Loukoianova, “Democracy Comes to Eu- rope: Franchise Extension and Fiscal Outcomes 1830-1938,” European Eco- nomic Review, February 2006, 50 (2), 249–283.

Aldrich, John A., “Rational Choice and Turnout,” American Journal of Political Science, 1993, 37 (1), 246–278.

Alesina, Alberto and Allan Drazen, “Why are Stabilizations Delayed?,” American Economic Review, 1991, 81 (5), 1170–1188.

Altman, David, Direct Democracy Worldwide, Cambridge: Cambridge Uni- versity Press, 2010.

Andreoni, James and Lise Vesterlund, “Which Is the Fair Sex? Gender Differences in Altruism,” Quarterly Journal of Economics, 2001, (Febru- ary), 293–312.

Association for More Democracy, Mehr Demokratie e.V., “www.mehr- demokratie.de,” 2014.

Barankay, Iwan, Pascal Sciarini, and Alexander H. Trechsel, “In- stitutional Openness and the Use of Referendums and Popular Initiatives:

171 Evidence from Swiss Cantons,” Swiss Political Science Review, 2003, 9 (1), 169–199.

Baron, Reuben M. and David A. Kenny, “The Moderator-Mediator Variable Distinction in Social Psychological Research: Conceptual, Strate- gic, and Statistical Considerations,” Journal of Personality and Social Psy- chology, 1986, 51 (6), 1173–1182.

Beasley, Ryan K. and Mark R. Joslyn, “Cognitive Dissonance and Post- Decision Attitude Change in Six Presidential Elections,” Political Psychol- ogy, 2001, 22 (3), 521–540.

Becker, Gary, “A Theory of Marriage,” in Theodore W. Schultz, ed., Eco- nomics of the Family: Marriage, Children, and Human Capital, Chicago: University of Chicago Press, 1974, pp. 299 – 351.

Bertocchi, Graziella, “The Enfranchisement of Women and the Welfare State,” European Economic Review, May 2011, 55 (4), 535–553.

Black, Duncan, The Theory of Committees and Elections, Cambridge: Cam- bridge University Press, 1958.

Blinder, Alan S., “Wage Discrimination: Reduced Form and Structural Estimates,” Journal of Human Resources, 1973, 8, 436–455.

Bochsler, Daniel, “The Marquis de Condorcet Goes to Bern,” Public Choice, 2010, (144), 119–131.

Boehmke, Frederick J. and Michael R. Alvarez, “The Influence of Ini- tiative Signature Gathering Campaigns on Political Participation,” 2012. VTP Working Paper 27.

Bowler, Shaun and Todd Donovan, Demanding Choices: Option, Voting, and Direct Democracy, Ann Arbor: University of Michigan Press, 1998.

Caldeira, Gregory A., Samuel C. Patterson, and Gregory A. Markko, “The Mobilization of Voters in Congressional Elections,” Journal of Politics, 1985, 47 (2), 490–509.

Calvert, Randall L., “Robustness of the multidimensional voting model: candidate motivations, uncertainty , and convergence,” Americal Journal of Political SciencePolitical Science, 1985, 29 (1), 69–95.

172 Center for Direct Democracy, Homepage of the Center for Direct Democ- racy, www.c2d.ch, 2014.

Coaete, Stephen and Michael Conlin, “A Group Rule-Utalitarian Ap- proach to Voter Turnout: Theory and Evidence,” American Economic Re- view, 2004, 64 (5), 1476–1504.

Coate, Stephen and Stephen Morris, “Policy Persistence,” American Economic Review, 1999, 89 (5), 1327–1336.

Copeland, Gary W., “Activating Voters in Congressional Elections,” Polit- ical Behavior, 1983, 5 (4), 391–401.

Croson, Rachel and Uri Gneezy, “Gender Differences in Preferences,” Journal of Economic Literature, May 2009, 47 (2), 448–474.

Dale, Allison and Aaron Strauss, “Don’t Forget to Vote: Text Message Reminders as a Mobilization Tool,” American Journal of Political Science, 2009, 53 (4), 787–804.

Deacon, Robert and Perry Shapiro, “Private Preference for Collective Goods Revealed Through Voting on Referenda,” American Economic Re- view, 1975, 65 (5), 943–955.

Degan, Arianna and Antonio Merlo, “A Structural Model of Turnout and Voting in Multiple Elections,” Journal of the European Economic As- sociation, 2011, 9 (2), 209–245.

Dorn, David, Justina A. V. Fischer, Gebhard Kirchgässner, and Alfonso Sousa-Poza, “Direct Democracy and Life Satisfaction Revisited: New Evidence for Switzerland,” Journal of Happiness Studies, 2007, 9 (2), 227–255.

Downs, Anthony, An Economic Theory of Democracy, New York: Harper and Brothers, 1957.

Edlund, Lena and Rohini Pande, “Why Have Women Become Left-Wing? The Political Gender Gap and the Decline in Marriage,” Quarterly Journal of Economics, 2002, (August), 917–961.

173 Fatke, Matthias and Markus Freitag, “Direct Democracy: Protest Cat- alyst or Protest Alternative?,” Political Behavior, 2013, 35 (2), 237–260.

Feddersen, Thimothy and Alvaro Sandroni, “A Theory of Participation in Elections,” American Economic Review, 2006, 96 (4), 1271–1282.

Feddersen, Timothy, “Rational Choice Theory and the Paradox of Not Voting,” Journal of Economic Perspectives, 2004, 18 (1), 99–112.

Feddersen, Timothy J. and Wolfgang Pesendorfer, “The Swing Voter’s Curse,” The American Economic Review, 1996, 86 (3), 408–424.

and , “Abstention in Elections with Asymmetric Information and Di- verse Preferences,” American Political Science Review, 1999, 93 (2), 381– 398.

Federal Act on the Federal Assembly, “Parlamentsgesetz,” 2002, pp. 1– 62.

Federal Announcement, “Botschaft über die Gewährleistung der Verfas- sung des Kantons Schaffhausen vom 9. April 2003,” 2003, pp. 3347–3357.

Feld, Lars P. and Gebhard Kirchgässner, “Does Direct Democracy Re- duce Public Debt? Evidence from Swiss Municipalities,” Public Choice, 2001, 3/4 (109), 347–370.

and , “The Political Economy of Direct Legislation: Direct Democracy and Local Decision-Making,” Economic Policy, 2001, 16 (33), 329–367.

and John Matsusaka, “Budget Referendums and Government Spending: Evidence from Swiss Cantons,” Journal of Public Economics, 2003, (87), 2703– 2724.

and Marcel R. Savioz, “Direct Democracy Matters for Economic Per- formance: An Empirical Investigation,” Kyklos, November 1997, 50 (4), 507–538.

Fernandez, Raquel and Dani Rodrik, “Resistance to Reform: Status Quo Bias in the Presence of Individual-Specific Uncertainty,” American Economic Review, 1991, 81 (5), 1146–1155.

174 Festinger, Leon, A Theory of Cognitive Dissonance, Stanford, California: Stanford University Press, 1957.

Fontana, Katharina, “Druck für eine strikte Umsetzung der Ausschaffungsinitiative,” Neue Zürcher Zeitung, 2012. http : //www.nzz.ch/aktuell/schweiz/druck − fuer − eine − strikte − umsetzung − der − ausschaffungsinitiative − 1.17913532.

Freitag, Markus and Adrian Vatter, “Direkte Demokratie, Konkordanz und Wirtschaftsleistung: Ein Vergleich der Schweizer Kantone,” Schweiz- erische Zeitschrift für Volkswirtschaft und Statistik, 2000, 136 (4), 579–606.

Frey, Bruno S., “Direct Democracy : Politico-Economic Lessons from Swiss Experience,” The American Economic Review, 1994, 84 (2), 338–342.

, Marcel Kucher, and Alois Stutzer, “Outcome, Process and Power in Direct Democracy - New Econometric Results,” Public Choice, 2001, 107 (3/4), 271–293.

Funk, Patricia, “Social Incentives and Voter Turnout: Evidence from the Swiss Mail Ballot System,” Journal of the European Economic Association, 2010, 8 (5), 1077–1103.

, “How Accurate are Surveyed Preferences for Public Policies? Evidence from a Unique Institutional Setup,” 2012. Barcelona GSE Working Paper 657.

and Christina Gathmann, “Does Direct Democracy Reduce the Size of Government? New Evidence from Historical Data, 1890-2000,” The Eco- nomic Journal, 2011, pp. 1–29.

and , “Gender Gaps in Policy Making: Evidence from Direct Democ- racy,” 2012.

Gerber, Alan S. and Donald P. Green, “The Effects of Canvassing, Telephone Calls, and Direct Mail on Voter Turnout: A Field Experiment,” American Political Science Review, 2000, 94 (3), 653–663.

Gerber, Elisabeth R., “Legislative Response to the Threat of Popular Ini- tiatives,” American Journal of Political Science, 1996, 40 (1), 99–128.

175 and Arthur Lupia, “Campaign Competition and Policy Responsiveness in Direct Legislation Elections,” Political Behavior, 2010, 17 (3), 287–306.

Gneezy, Uri, Muriel Niederle, and Aldo Rustichini, “Performance in Competitive Environments: Gender Differences,” Quarterly Journal of Eco- nomics, 2003, (August), 1049–1074.

Green, Donald P., Alan S. Gerber, and David W. Nickerson, “Getting Out the Vote in Local Elections : Results from Six Door-to-Door Canvassing Experiments,” Political Science, 2003, 65 (4), 1083–1096.

Grütter, Alfred, Die Eidgenössische Wehrsteuer, ihre Entwicklung und Be- deutung, Zürich: Juris Druck + Verlag, 1968.

Habermas, Juergen, Faktizität und Geltung, Frankfurt: Suhrkamp, 1992.

Hicks, Raymond and Dustin Tingley, “Causal Mediation Analysis,” Stata Journal, 2011, 11 (4), 1–15.

Hodler, Roland, Simon Lüchinger, and Alois Stutzer, “The Effects of Voting Costs on the Democratic Process and Public Finances,” American Economic Journal: Economic Policy, Forthcoming, 2014, pp. 1–52.

Hofer, Bruno, Volksinitiativen der Schweiz, One to One, 2012.

Hofer, Katharina E., “Campaigning in Direct Democracies: Initiative Pe- tition Signing, Turnout, and Acceptance,” 2013. University of St.Gallen Economics Working Paper Series, No. 1333.

Hollanders, David and Barbara Vis, “Voters Commitment Problem and Reforms in Welfare Programs,” Public Choice, 2013, 155 (3), 433–448.

Holt, Charles A. and Susan K. Laury, “Risk Aversion and Incentive Effects,” American Economic Review, December 2002, 92 (5), 1644–1655.

and , “Risk Aversion and Incentive Effects : New Data Without Order Effects,” American Economic Review, 2005, 95 (3), 902–904.

Huber, Martin, “Identifying Causal Mechanisms (Primarily) Based on In- verse Probability Weighting,” Journal of Applied Econometrics, 2013.

176 , “Causal Pitfalls in the Decomposition of Wage Gaps,” Journal of Business and Economic Statistics, Forthcoming, 2014, pp. 1–25.

Huckfeldt, Robert and John Sprague, “Political Parties and Electoral Mobilization: Political Structure, Social Structure, and the Party Canvass,” The American Political Science Review, 1992, 86 (1), 70–86.

Husted, Thomas A. and Lawrence W. Kenny, “The Effect of the Ex- pansion of the Voting Franchise on the Size of Government,” 1997, 105 (1), 54–82.

Imai, Kosuke and Teppei Yamamoto, “Identification and Sensitivity Analysis for Multiple Causal Mechanisms: Revisiting Evidence from Fram- ing Experiments,” Political Analysis, 2013, 21, 141–171.

, Luke Keele, and Teppei Yamamoto, “Identification, Inferences and Sensitivity Analysis for Causal Mediation Effects,” Statistical Science, 2010, 25, 51–71.

Imbens, Guido W., “Nonparametric Estimation of Average Treatment Ef- fects Under Exogeneity: A Eeview,” Review of Economics and Statistics, 2004, 86 (1), 4–29.

Initiative and Referendum Institute at the University of South- ern California, Homepage of the Initiative and Referendum Institute, www.iandrinstitute.org, 2014.

Judd, Charles M. and David A. Kenny, “Process Analysis: Estimating Mediation in Treatment Evaluations,” Evaluation Review, 1981, 5, 602–619.

Kartik, Navin and R. Preston McAfee, “Signaling Character in Electoral Competition,” The American Economic Review, 2007, 97 (3), 852–870.

Keller, R. Godfrey and Sven Rady, “Strategic Experimentation with Poisson Bandits,” Theoretical Economics, 2010, 5, 275–311.

Kirchgässner, Gebard and Bruno S. Frey, Demokratische Wirtschaft- spolitik, Vahlen, 2012.

Klaus, Max, Briefliche Stimmabgabe. Analyse der Eidg. Volksabstimmung vom 27. November 2005, Bern: Bundeskanzlei, 2006.

177 Knoepfel, Peter, Yannis Papadopoulos, Pascal Sciarini, Adrian Vat- ter, and Silja Häusermann, Handbuch der Schweizer Politik - Manuel de la politique Suisse, NZZ Libro, 2014.

Kriesi, Hanspeter, “The Political Opportunity Structure of New Social Movements: Its Impact on Their Mobilization,” in J. Craig Jenkins and Bert Klandermans, eds., The Politics of Social Protest, Comparative Perspectives on States and Social Movements, Minneapolis MN: University of Minnesota Press, 1995, pp. 167–198.

, “Role of the Political Elite in Swiss Direct-Democratic Votes,” Party Pol- itics, 2006, 12 (5), 599–622.

Leduc, Lawrence, “Opinion Change and Voting Behavior in Referendums,” European Journal of Political Research, 2003, 41 (6), 711–732.

Linder, Wolf, “Direct Democracy,” in Ulrich Kloeti, Peter Knoepfel, Hanspeter Kriesi, Wolf Linder, Yannis Papadopoulos, and Pascal Scia- rini, eds., Handbook of Swiss Politics, Zürich: Neue Zürcher Zeitung, 2007, pp. 101 – 120.

Lohmann, Susanne, “A Signaling Model of Informative and Manipulative Political Action,” American Political Science Review, 1993, 87 (2), 319–333.

, “Information Aggregation Through Costly Political Action,” American Economic Review, 1994, 84 (3), 518–530.

Lott, John R. and Lawrence W. Kenny, “Did Women’s Suffrage Change the Size and Scope of Government?,” Journal of Political Economy, 1999, 107 (6), 1163–1198.

Lüchinger, Simon, Myra Rosinger, and Alois Stutzer, “The Impact of Postal Voting on Participation: Evidence from Switzerland,” Swiss Political Science Review, 2007, 13 (2), 167–202.

Mäder, Lukas, “Noch kaum Unterschriften für FDP-Initiative,” 20 Minuten, 2010. http : //www.20min.ch/schweiz/news/story/14953477.

Magleby, David, Direct Legislation: Voting on Ballot Propositions in the United States, Baltimore: John Hopkins University Press, 1984.

178 Matsusaka, John G., “Economics of Direct Legislation,” Quarterly Journal of Economics, 1992, 107 (2), 541–571.

, “Explaining Voter Turnout Patterns: An Information Theory,” Public Choice, 1995, 84, 91–117.

, “The Eclipse of Legislatures: Direct Democracy in the 21st Century,” Public Choice, 2005, 124 (1), 157–177.

and Filip Palda, “Voter Turnout: How Much Can We Explain,” Public Choice, 1999, (98), 431–446.

and Nolan M. McCarty, “Political Resource Allocation: Benefits and Costs of Voter Initiatives,” Journal of Law, Economics, & Organizations, 2001, 17 (2), 413–448.

Meltzer, Allan H. and Scott F. Richard, “A Rational Theory of the Size of Government,” Journal of Political Economy, 1981, 89 (5), 914–927.

Merlo, Antonio, “Whither Political Economy? Theories, Facts and Issues,” in Richard Blundell, Whitney Newey, and Torsten Persson, eds., Advances in Economics and Econometrics, Theory and Applications: Ninth World Congress of the Econometric Society, Cambridge, UK: Cambridge Univer- sity Press, 2006, pp. 381–421.

Miller, Grant, “Women’s Suffrage, Political Responsiveness, and Child Sur- vival in American History,” Quarterly Journal of Economics, 2008, (Au- gust), 1287–1327.

Mills, Judson, “Changing in Moral Attitudes Following Temptations,” Jour- nal of Personality, 1958, (26), 517–531.

Mullainathan, Sendhil and Ebonya Washington, “Sticking With Your Vote: Cognitive Dissonance and Political Attitudes,” American Economic Journal: Applied Economics, 2009, 1 (1), 86–111.

Neiman, Max and Mark Gottdiener, “The Relevance of the Qualifying Stage of Initiative Politics: The Case of Petition Signing.,” Social Science Quarterly, 1982, 63 (3), 582–588.

179 Nickerson, David W., “Volunteer Phone Calls Can Increase Turnout: Ev- idence From Eight Field Experiments,” American Politics Research,May 2006, 34 (3), 271–292.

Niederle, Muriel and Lise Vesterlund, “Do Women Shy Away from Com- petition? Do Men Compete Too Much?,” Quarterly Journal of Economics, 2007, (August), 1067–1101.

Nigg, Heinz, Wir wollen alles, und zwar subito! Die Achtziger Jugendun- ruhen in der Schweiz und ihre Folgen, Limmat, 2001.

Niven, David, “The Mobilization Solution? Face-to-Face Contact and Voter Turnout in a Municipal Election,” Journal of Politics, 2004, 66 (3), 868–884.

Oaxaca, Ronald, “Male-Female Wage Differences in Urban Labour Mar- kets,” International Economic Review, 1973, 14, 693–709.

Oechslin, Hanspeter, Die Entwicklung des Bundessteuersystems der Schweiz von 1848 bis 1966, Einsiedeln: Etzel-Druck AG, 1967.

Osborne, Martin J., “Entry-Deterring Policy Differentiation by Electoral Candidates,” Mathematical Social Sciences, 2000, (40), 41–62.

and Al Slivinski, “A Model of Political Competition with Citizen Can- didates,” Quarterly Journal of Economics, 1996, (111), 65–96.

Parry, Janine, Daniel Smith, and Shayne Henry, “The Impact of Peti- tion Signing on Voter Turnout,” Political Behavior, 2012, 34, 117–136.

Patterson, Samuel C. and Gregory A. Caldeira, “Getting Out the Vote: Participation in Gubernatorial Elections,” 1983, 77 (3), 675–689.

Pierce, John C. and Nicholas P. Lovrich, “Survey Measurement of Po- litical Participation: Selective Effects of Recall in Petition Signing,” Social Science, 1982, 63 (1), 164–171.

Rohner, Gabriela, Die Wirksamkeit von Volksinitiativen im Bund 1848- 2010, Zürich: Schulthess, 2012.

Romer, Thomas and Howard Rosenthal, “Political Resource Allocation, Controlled Agendas, and the Status Quo,” Public Choice, 1978, 33 (4), 27–43.

180 Rubin, Donald B., “Direct and Indirect Causal Effects Via Potential Out- comes,” Scandinavian Journal of Statistics, 2004, 31, 161–170.

Ruckstuhl, Lotti, Frauen sprengen Fesseln, Bonstetten: Interfeminas, 1986.

Schaffner, David, “Streit beim Unterschriftensammeln für die Erbschaftssteuer-Initiative, 27.6.2012,” Tages-Anzeiger, 2012.

Schmid, Lukas, “Political Decisions in Multiple Referendums,” 2013. Uni- versity of St.Gallen, mimeo.

Schubert, Renate, Martin Brown, Matthias Gysler, and Hans Wolf- gang Brachinger, “Financial Decision-Making: Are Women Really More Risk-Averse?,” American Economic Review, 1999, 89 (2), 381–385.

Selten, Reinhard and Axel Ockenfels, “An Experimental Solidarity Game,” Journal of Economic Behavior & Organization, March 1998, 34 (4), 517–539.

Shapiro, Robert Y. and Harpreet Mahajan, “Gender Differences in Policy Preferences: A Summary of Trends from the 1960s to the 1980s,” Public Opinion Quarterly, 1986, 50 (1), 42–61.

Smith, Mark A., “The Contingent Effects of Ballot Initiatives and Candi- date Races on Turnout,” American Journal of Political Science, 2001, 45 (3), 700–706.

Sowell, Thomas, Preferential Policies: An International Perspective, New York: William Morrow, 1990.

Strulovici, Bruno, “Learning While Voting: Determinants of Collective Ex- perimentation,” Econometrica, 2010, 78 (3), 933–971.

Stutzer, Alois and Lukas Kienast, “Demokratische Beteiligung und Staat- sausgaben: Die Auswirkungen des Frauenstimmrechts,” Swiss Journal of Economics and Statistics,, 2005, 141 (4), 617–650.

Swiss Federal Chancellery, “Chronology of Popular Initiatives,” Homepage of the Swiss Federal Chancellery www.bk.admin.ch, 2013.

181 Swiss Statistical Office, Finanzen und Steuern von Bund, Kantonen und Gemeinden 1971, Bern: Eidgenössisches Statistisches Amt Publikationsdi- enst, 1973.

, Öffentliche Finanzen der Schweiz 1972, Bern: Eidgenössisches Statistis- ches Amt Publikationsdienst, 1974.

Tolbert, Caroline J. and Daniel A. Smith, “The Educative Effects of Ballot Initiatives on Voter Turnout,” American Politics Research, March 2005, 33 (2), 283–309.

Torgler, Benno, “Tax Morale and Direct Democracy,” European Journal of Political Economy, June 2005, 21 (2), 525–531.

Wili, Heinrich, “Befristung der Unterschriftensammlung und Eröhung der Unterschriftenzahlen bei Volksbegehren: Erste Bilanz ihrer Auswirkungen,” in A. Meier-Hayoz, P. Dutoit, and Saladin. B., eds., Zeitschrift für Schweiz- erisches Recht, Basel: Helbling und Lichtenhahn, 1982, pp. 61–86.

Wittman, Donald, “Candidate Motivation: A Synthesis of Alternative The- ories,” American Political Science ReviewPolitical Science, 1983, 77 (1), 142–157.

Wolfinger, Raymond E. and Steven J. Rosenstone, Who Votes?, New Haven: Yale University Press, 1980.

Wooldridge, Jeffrey M., Econometric Analysis of Cross Section and Panel Data: Second Edition, Cambridge, MA: MIT Press, 2010.

182 Curriculum Vitae Education 2011-2014 Ph.D. in Economics and Finance, University of St.Gallen 2013-2014 Visiting Scholar, University of Pennsylvania, Philadelphia 2011 Doctoral Program at Study Center Gerzensee, Foundation of Swiss National Bank 2009-2011 Master of Arts HSG (M.A. HSG) in Economics, University of St.Gallen 2009 Exchange Semester, University Commerciale Luigi Bocconi, Milan 2006-2009 Bachelor of Arts HSG (B.A. HSG) in Economics, University of St.Gallen

Professional Experience 2011-2014 Research Assistant, Swiss Institute for Empirical Economic Research, University of St.Gallen 2011-2014 Teaching Assistant, Introduction into Micro- /Macroeconomics, University of St.Gallen 2010-2011 Assistant, Chair of Logistics Management, University of St.Gallen

183