Venice Summer Institute

CESifo, a Munich-based, globe-spanning economic research and policy advice institution Venice Summer Institute 2016 Venice Summer Institute 20 21 July 2016 THE ECONOMICS OF PEACE AND WAR Organisers: Panu Poutvaara & Vesa Kanniainen MILITARY BACKGROUNDS OF POLITICIANS AND PARLIAMENTARY DECISIONS David Stadelmann, Marco Portmann, and Reiner Eichenberger CESifo GmbH Poschingerstr. 5 81679 Munich, Germany Tel.: +49 (0) 89 92 24-1410 Fax: +49 (0) 89 92 24-1409 E-Mail: office@cesifo.de www.cesifo.org/venice

Military Backgrounds of Politicians and Parliamentary Decisions David Stadelmann a,c Reiner Eichenberger b,c Marco Portmann b, c First version comments welcome July 2016 Abstract: Do politicians with a military background decide differently with respect to societal issues? We investigate the voting behavior of politicians in the informative institutional setting of Switzerland with a conscription army and a political system, which allows controlling for voter preferences and party ideology. Our results show that politicians who have served in the military do not differ from those who have not when comparing their voting behavior on issues related to female welfare and welfare of the weak and disabled. However, politicians who served have a higher probability to accept proposals on neutrality and a lower probability to accept proposals linked to international rights and the environment. Due to our setting, we can also analyze differences with respect to military ranks and ideological positions within parties. Keywords: Military, Behavior of politicians, Voting, Voter preferences, Political presentation. JEL Classification: D72, F52, H56 Total word count: 9469 Words (excluding abstract and tables) We thank Ali Sina Önder for inspiring discussions on this first draft. This article is a companion paper to our paper entiteld Military careers of politicians matter for national security policy, Journal of Economic Behavior and Organization, 116, 142-156, 2015. which improves the approach employed therein and extends it to other policy areas instead of only security policy. a University of Bayreuth, Universitätsstraße 30, 95447 Bayreuth, Germany, email: david.stadelmann@uni-bayreuth.de. b University of Fribourg, Bd. de Pérolles 90, 1700 Fribourg, Switzerland. c CREMA Center for Research in Economics, Management, and the Arts, Zurich, Switzerland.

I. INTRODUCTION Numerous political representatives around the world have served in the armed forces because of the widespread prevalence of conscript armies, past conflict situations, and military disputes. This is not only the case for less democratic countries but names like Dwight D. Eisenhower, Charles de Gaulle or Ariel Sharon immediately come to mind when thinking of political leaders with military experience from democratic states. In today s electoral campaigns politicians do not refer to their experience or service in the military as a potentially problematic issue but often portray it as an asset for their political career. 1 It is therefore reasonable to expect that political decisions along numerous societal fields may be shaped by the military experience and the military background of political representatives today. 2 The political science and political economy literature has suggested that the military background of representatives may affect their decisions and opinions as well as voting for or against military issues (see Huntington 1957 and Nordlinger 1977 for seminal contributions or Stadelmann et al. 2015 for a recent analysis). 3 However, the literature has been mostly mute with respect to effects of military background of politicians on other policy fields apart from national security. Thus, while there is some evidence that military careers shape voting on military issues, the effects of military service on policy decisions affecting societal issues at large such as policy proposals related to social policy, neutrality, international rights or the environment have not been systematically analyzed. For many topics, it is not obvious how military background should affect political stance. But for some issues, most notably those affecting specifically female preferences or the weak and disabled, positive and negative stylized societal stereotypes are held about men 1 Take the electoral race between John Kerry and George W. Bush as an example where both served in the armed forces and highlighted that fact during their campaigns, even leading to so called John Kerry military service controversy. 2 Many political commentators still refer to Vladimir Putin s training as a KGB officer when explaining certain of his political choices. 3 See also e.g., Geddes (2003), Holsti (2001), Feaver and Gelpi (2004), Horowitz and Stam (2014). 1

having served in the military. One the one hand they are seen as protectors of the weak and the home country while, one the other hand, views diverge and men with a military service tend to be seen as chauvinists regarding women and socially conservatives. 4 We contribute to shedding light on this relevant but under-researched area by empirically analyzing whether politicians with a military background decide differently on societal issues affecting different voter groups. We thereby complement the previous literature in a new and relevant dimension. A main advantage of our analysis consists in the fact that we can hold constant voter preferences as well as ideological preferences of parties when analyzing politicians behavior. Thus, we provide results related to the voting behavior of politicians who served in the military, which are independent not only of their voters preferences but also of their parties ideological preferences. We believe that there is an explanation that previous literature has abstained from systematically trying to analyze potential links between the military service of politicians and their policy choices. Analyzing such a relationship is an empirical endeavor which is associated with numerous challenges of which the three most important are the following: (1) Politicians are elected by voters and are supposed to represent them but constituents may elect (or not) politicians who served in the military because they see military service as a signal of quality (or not). This implies that voter preferences need to be controlled for when analyzing decisions of politicians in parliament because otherwise their task of representing voters could be correlated with the fact that they served in the military, i.e. we need to distinguish whether decisions of politicians are influenced by their personal background or their task of representation. (2) Ideological preferences reflected by parties on different societal issues are likely to affect individual voting decisions in parliament. At the same time, ideology may also affect whether politicians served in the armed forces and whether they achieved higher military ranks. Thus, controlling for different party 4 For other important societal issues, stereotypes on how politicians who served in the military may vote are less clear and there is little information on how military service may influence voting behavior. 2

ideologies with respect to specific issues is essential to isolate the effect of serving in the military on voting decisions. (3) It is not straightforward to classify political decisions into different societal areas such as issues affecting women, the weak and disabled, neutrality, international rights or the environment. However, it is even more difficult to classify decisions as either pro- or against women, pro- or against the weak and disabled or other societal groups and interests. This is because final decisions in parliaments often represent compromises where different aspects and trade-offs are considered, e.g. certain aspects of a working time reform may be beneficial for women while other elements are less beneficial. Thus, we need an external classification for how specific political decisions affect different groups, which should be independent of politicians choices. More precisely, we need a statement whether an independent societal association considers a certain policy proposal as pro-female, pro-weak and disabled, pro-neutrality, etc. Our setting allows us to tackle all three of these challenges within a common framework and to analyze differences in voting by politicians with and without a military background. Moreover, we can also explore the role of military ranks on the voting behavior of politicians. In Switzerland, constituents regularly reveal their preferences for policies in popular referenda (see Schneider et al. 1981; Hessami 2016 or Stadelmann et al. 2014). Politicians in parliament vote on the same policy proposals with the identical wording, which allows us to account for voter preferences (see also Portmann et al. 2012). We directly observe party ideology with respect to different societal issues because parties give official voting recommendation, i.e. parties recommend to accept or reject a specific policy proposal and we can control for their recommendation. Moreover, we measure the ideological position on a left-right dimension of individual politicians to analyze whether they are relatively right or left within their respective party blocks. Finally, we use official voting recommendations of five independent societal associations to identify legislative proposals which are pro- or against-female, pro- or againstweak and disabled, pro- or against-neutrality, pro- or against-international rights, and pro- or 3

against-environment. For all Swiss politicians, we have data on whether they served in the military or not and their military ranks from 2000 to 2014 and, of course, we have collected other personal characteristics such as age, education, party affiliations, etc. Thus, our setting allows us to analyze how the military service and differences in military ranks are related to the voting behavior of politicians on diverse social issues while controlling for voter preferences and ideological preferences of parties with respect to the identical policy issues at the same time while still holding constant other personal characteristics. Our results reveal new and relevant insights into how military background affects voting behavior: We observe that politicians who have served in the military do not vote any different from those who did not serve with respect to female issues and issues affecting the weak and disabled. However, having served in the military is associated with a higher probability of voting pro-neutrality and a lower probability of voting pro-international rights and pro-environment. These effects are independent of voter preferences and ideological preferences of parties. Other personal characteristics and party affiliations do not systematically influence voting behavior of politicians who served or did not serve in the military. Further refinements show that the personal ideological position that politicians hold within their parties only affects the influence of their military background for environmental policies. Our results remain robust when leaving out politicians from right parties where higher military ranks tend to be observed more frequently. We also explore differences in military ranks as well as gender and find that while our overall results remain robust, different military careers exert a slightly different effect on behavior. In particular, the evidence suggests that simply serving in the military in the soldier ranks does not systematically affect voting on societal issues compared to men who did not serve. Having chosen to serve in higher military ranks is associated with a differential voting pattern, which suggests that motivation for the military affects voting later on while simply serving in the military due to conscription requirements has no systematic effects. 4

The remainder of this paper is structured as follows: Section II relates our contribution to the existing literature. Section III discusses our data and the empirical strategy. We present our empirical findings, robustness tests and refinements in Section IV. Finally, Section V offers concluding remarks. II. RELATED LITERATURE This paper is related to two strands of literature: First, it contributes to the economic literature on the relevance of individual characteristics which may help to explain the behavior of politicians. Apart from pure electoral competition (see Downs 1957), legislative choices and the behavior of politicians may be influenced by personal attributes and personal experience. 5 Such attributes and experience have been found to influence electoral success, redistribution, popularity, rent-seeking, etc. (see, e.g., Armstrong and Graefe 2011; Gagliarducci and Paserman 2012; Kauder and Potrafke 2016). Hayo and Neumeier (2012) show that prime ministers tend to support more the societal class in which they were socialized. Jochimsen and Thomasius (2014) study the development of public deficits and suggest that politicians who gained financial expertise prior to their appointment achieve lower deficits. Other contributions analyze, among other personal attributes, the influence of gender (e.g. Svaleryd 2009, Stadelmann et al. 2014), education (e.g. Ruske 2015), and parenthood (e.g. Washington 2008) on the behavior of politicians. Braendle and Stutzer (2010) find evidence that experience in public service affects the number of submitted interpellations. Ågren et al. (2006) draw on a survey of Swedish voters who indicated their preferences on public spending and show that politicians with similar socioeconomic characteristics as voters exhibit similar preferences for levels of spending. Regarding the military, a number of articles analyze the effect of conscription (see Teigen 2006; Sasson- 5 More generally, the reasons for legislative shirking remain the subject of a vivid and unresolved scientific discourse (see Grofman 2004 or Padovano 2013). 5

Levy 2007; Vasquez III 2005). We contribute to the literature analyzing personal attributes and experience by explicitly addressing the role of past military service on politician behavior while holding constant other personal characteristics. We also analyze effects of conscription on voting behavior later on but our setting also allows us to look at potential selection effects due to motivation for the military as we observe politicians who have served in higher military ranks. Second, our contribution relates to the broader literature on the interaction of the military, security policy issues and political institutions (see, among others, Collier and Hoeffler 2004; 2006; Dunne et al. 2008; Dunning 2011; Gebremedhin and Mavisakalyan 2013). While a few contributions in political sociology turned their attention relatively early to the question of whether an individual s personal military service history affects the willingness to support military action (see Huntington 1957; Nordlinger 1977), the political economy literature has been astonishingly mute with respect to the societal role of military service. Nevertheless, a small a number of articles focus on the role of military socialization for decisions to engage in conflict (see, e.g., Weeks 2012; Horowitz and Stam 2014). Other contributions suggest that military conservatism may contribute to a more cautious behavior with respect to the use of force (see Betts 1991; Gelpi and Feaver 2002). Moreover, hawkish behavior in legislative decisions may be driven by self-selection into the military instead of socialization (see Bachman et al. 2000; Stadelmann et al. 2015). For voting in general party ideology is regarded as a major driving factor (e.g. Poole and Rosenthal 1997; Hix et al. 2003). The same holds for voting on military issues where ideology tends to be associated with political decisions (see Lindsay 1990; Carsey and Rundquist 1999). However, independent of ideology, economic interests play a role too (see Fordham 2008). We contribute to this literature by analyzing the effect of military service more broadly on different societal issues instead of limiting ourselves to military issues only. Our setting also overcomes a general difficulty in the literature regarding the measurement of voter preferences (see, e.g., Lott and Davis 1992, Kau and Rubin 1993, Gerber and Lewis 2004, Lee 6

et al. 2004, Bafumi and Herron 2010). Scholars often rely on demographic variables and past election results as potential proxies for district preferences (e.g., Ardoin and Garand 2003). Our contribution uses referenda to measure revealed voter preferences on the identical societal policy proposals that politicians decide on in parliament. 6 At the same time we also measure party ideology on the same policy proposals. All these elements represent an entirely new way to analyze how politicians with and without a military service decide on different societal issues holding voter preferences, party ideology and other characteristics constant. With the existing literature in mind, the theoretical motivation and background of our study is simple: Politicians react to incentives and their decision are driven by numerous factors including voter preferences and party ideology. Nevertheless, politician have some leeway and may pursue their self-interests and individual beliefs, which are shaped by their own history as well as their socio-economic environment. We explore whether military service represents a relevant factor to explain voting decisions on a different policy issues which affect different societal groups. III. DATA AND EMPIRICAL STRATEGY Switzerland provides an informative institutional setting to compare what politicians do with what their constituents want (see, e.g., Frey 1994, Portmann et al. 2012 or Hessami 2016). In particular, our setting has three distinctive features which allows us to analyze how military background may shape voting behavior of politicians on specific policy issues, controlling for their task of political representation and party ideology. 6 Many authors discuss the effects of referenda on the political process, citizens representation, and citizens well-being (see, e.g., Matsusaka 2005; Osborne and Turner 2010). However, very few scholars consider roll call votes or referenda outcomes as measures for constituents representation (see Hersch and McDougall 1988; Matsusaka 2010) 7

Measuring voter preferences and political decisions A first distinctive feature of our setting is related to the prevalence of referenda on identical policy proposals as politicians decide on in parliament. We focus on individual voting behavior of 516 politicians of the Swiss National Council (lower house of parliament). The National Council s electoral districts are identical to the 26 Swiss cantons for which we observe referendum outcomes. We examine final votes (roll calls) of politicians during their time in parliament. Final votes are considered to be most proximate compared to the adoption of governmental policies (see Krehbiel 1993). The parliamentary services record all votes cast and makes them publicly available. For all members of parliament, we collected information on whether they served in the military, their military ranks, and personal characteristics. As in other democratic countries, parliamentary representatives in Switzerland craft legislative proposals. However, Swiss citizens may challenge all proposed laws and demand a referendum by collecting 50,000 signatures (approximately one percent of registered voters). For all constitutional changes a referendum is mandatory. Initiatives allow citizens to demand a constitutional amendment themselves by collecting 100,000 signatures. Politicians are required to state their preferences on the text of every initiative in a vote but cannot amend it. Thus, we observe how politicians decide in parliament and how their constituents decide in the respective referenda between 2000 and 2014. Importantly, we do not impute any values neither regarding the decisions of politicians nor regarding voter preferences. 7 Portmann (2014) provides an in depth discussion on Swiss referenda and the parliamentary system presents a number of analyses. In general, referenda reflect revealed preferences of constituents as they allow them to compare legislative proposals against the status quo (see Schneider et al. 1981; Portmann et al. 7 We stress this point because in the literature certain values (in particular regarding preferences) are usually approximated or imputed. 8

2012; Stadelmann et al. 2013; Carey and Hix 2013). A direct comparison between final votes and referendum decision has also been performed for the United States (see Brunner et al. 2013). Other branches of the literature draw on survey data or demographic characteristics to approximate voter preferences Ardoin and Garand 2003 or Blais and Bodet 2006 for a review of the literature). 8 We have information on all Swiss parliamentary decisions and referenda from 2000 to 2014 and analyze the results at the subnational (cantonal) level. All referendum decisions are implemented and, thus, entail real policy outcomes and consequences., Measuring party ideology and intra-party position A second distinctive feature of our setting is related to the fact that we can account for ideological preferences of parties for specific policies. Ideology has been identified in the literature as a central driver for political decisions (see e.g. Levitt 1996; Carsey and Rundquist 1999; Hix et al. 2003). With our setting, we do not only observe what voters expect politicians to do but also what their parties expect them to do. In Switzerland, parties proclaim voting recommendations for referenda. Thus, for all parliamentary decisions with subsequent referenda we can directly measure the ideological position of parties by their voting recommendations. Controlling for party ideology is relevant when analyzing the effect of individual characteristics and, in particular, when investigating the effect of past military experience on voting for different societal issues. Ideological preferences reflected by party positions with respect to different societal issues are likely to affect individual voting decisions of politicians in parliament. At the same time, ideology may also affect whether politicians served in the armed forces and whether they achieved higher military ranks. Thus, controlling for party ideology with respect to different policies is relevant to identify potential effects of serving in the military. Our data allows us to do 8 Golder and Stramski (2010) describe different ways of measuring congruence between legislators and voters with surveys. 9

so for individual policy proposals and decisions of politicians. Moreover, our data also permits us to measure the intra party ideology of individual politicians. This intra party ideology measure is based on a NOMINATE score of each politician compared to the average NOMINATE score of her/his party and we construct a variable which indicates whether a politician is relatively rightleaning within the party. 9 Note that this measure is not referendum specific in comparison to our measure of party ideology. Instead, it is specific for every politician and year as is common in the literature. The advantage of this measure is that we can situate each politician within his/her party block, i.e. we can observe whether a person is relatively left or right within the party. We use this control variable to insure that when holding constant voter preferences and party ideology, politicians with a military background are not only relatively right within their parties, i.e. we try to separate a potential effect of right-leaning within party from a potential effect of military service. Identifying preferences for policy proposals of societal organizations and the army system Our setting permits us to observe decisions of politicians, voter preferences and party ideology on precisely the same policy proposals. However, we also need to classify the policy proposals into different fields affecting women, the weak and disabled, neutrality, etc. While this task could be seen as relatively straightforward, we also need to classify policy proposal as to whether the policy proposals extend or restrict the scope within these societal fields, i.e. we need to classify whether proposals are pro-female, pro-weak and disabled, pro-neutrality etc. This is not a trivial task. Parliamentary proposals often present compromises that merge different positions and interests in order to appeal to a majority. Thus, it is rare to observe policies which are entirely and obviously pro- or against-women, or pro- or against-environment. Moreover, as final 9 More precisely, we draw on the NOMINATE scores computed by Michael Hermann from the institute Sotomo. We then calculated the average NOMINATE scores for all members from the same party block and then subtract from the individual score the average score of her/his party block. 10

outcomes become only clear after policies have been implemented for years, in the decision phase arguments in all directions are possible and are actually made. Thus, there is no objective measure to assess a policy proposal. To mitigate classification problems to some extent, we make use of a third distinctive feature of our setting. We explicitly resort to referendum voting recommendations of respected organizations which are regarded as experts in their policy fields. More precisely, we collected the accept and reject voting recommendations of the following five associations: Alliance F for policy proposals pro- and against-females 10, AVIVO for policy proposals pro- and against-weak and disabled 11, AUNS for policy proposals pro-neutrality 12, Amnesty International 13 for policy proposals pro- and against international rights and, the WWF for policy proposals pro- and against-environment 14. Our identification strategy relies on these organizations having vested interests with respect to their policy fields and detailed knowledge in their domains. They are also relatively representative for their respective policy fields and have issued the largest number of recommendations in their domains. Their voting recommendations usually do not overlap such that we can analyze disjoint policy areas which constitutes a further advantage of our setting. Of course, they are not the only associations in their respective policy fields which issue recommendations. We checked the voting recommendation of different alternative organizations such as smaller female associations, the organization for handicapped people (AGILE), the association for democratic solicitors, Greenpeace, etc. and the recommendations they give greatly 10 Alliance F is the largest Swiss female association and regroups over 140 institutional members thereby representing over 400 000 women. The association is active domains such as equality between men and women, family issues and the like. For further information see: http://www.alliancef.ch 11 AVIVO is the Swiss association for widows, invalids, old-age and orphans. It has been established in 1948. The association has made its name by taking positions in favor of solidarity. Its positions figure in newspaper articles. For further information see: http://www.avivo-suisse.ch 12 AUNS is the abbreviation for Association for and independent and neutral Switzerland. It has been established in 1986. It is often regarded as a relatively right-wing organization. However, there is no doubt that it strongly defends Switzerland s independence and neutrality. For further information see: https://auns.ch/ 13 For information on the Swiss section of Amnesty International see: https://www.amnesty.ch/ 14 For information on the Swiss section of the WWF see: https://www.wwf.ch/ 11

overlap with the association we choose. 15 We also note that some high-ranking representatives of the respective groups may act as advisers to parliament and political parties in the elaboration of legislation. Nevertheless, the recommendations of these groups do not target specific representatives as organizations disseminate voting recommendations for referenda only after politicians have made decisions in parliament. More importantly, they do not engage in ranking politicians such that the recommendations do not form a set of strategically chosen, polarized issues as is sometimes argued against when researchers make use of ADA-Americans for Democratic Action Scores for liberal and conservative classifications or other organizations (see Snyder 1992). We recognize that ultimately any use of interest organization to classify specific policies represents a choice. However, we think that it is a novel approach to classify issues beyond the scope of liberal vs. conservative dimension and the vested interests of the organization are helpful, i.e. the organization really want voters to decide in referenda according to their recommendations. As we are not interested in knowing whether pro-female proposals are also welfare enhancing in general, we think it is a worthwhile approach to rely on organizations that can be regarded as specialists in their field. In fact, we effectively restrict degrees of freedom by following their recommendation instead of performing a classification ourselves. Put differently, who, if not the largest female association or the WWF should know whether policy proposals are rather pro- or against women and pro- or against the environment. As we are interested in analyzing how individual politicians with and without experience in the military decide on societal issues, we also briefly present the Swiss army system. Switzerland s army is organized as a militia army of all able-bodied male conscripts between the ages of 19 up to 50 years for specific military functions. 16 While in the 1950s almost all young men without physical disability served in the armed forces, approximately two-thirds of young 15 Our ultimate choice of organizations is also linked to the number of recommendations which is highest for the organizations that we finally rely on. 16 Military service is voluntary for women. 12

Swiss men are judged to be able-bodied. For those considered incapable of military service alternative services, such as civil protection, exist and individuals with a lower hourly burden than those in regular military service are required to pay a military exemption tax as compensation. 17 The military exemption tax ensures that the monetary incentives by trying to be regarded as incapable for service to work instead are fairly low. Detailed information on the Swiss military system provided by Stadelmann et al. (2015) and the Federal Department of Defense, Civil Protection, and Sports. We know for all politicians in our sample whether they have served in the military and their precise military ranks. Empirical strategy Our institutional setting allows us to observe what politicians do, what their voters want and what their party ideology on specific issues is. Moreover, we observe voting recommendations of societal organization which allow as to classify policy proposals and we know whether their interests are pro- or against the issues at stake. We are interested in analyzing whether politicians who served in the military have a higher or a lower probability to accept policy proposals from specific societal fields compared to their counterparts who did not serve in the military while controlling for other factors, in particular voter preferences and party ideology. Our baseline empirical specification is as follows: (1) VoteYesir = αα + ββ1(served in military)i + ββ2(served in military)i*(proposal propolicy field)r+ ββ3(proposal pro-policy field)r + ββ4(voter preferences yes)ir + ββ5(party ideology yes)ir + XXirγγ + εε iiii. 17 Men not serving at all, due to either physical or mental reasons, are required to pay the full military exemption tax on their incomes. 13

MPYesir is an identifier for whether a politician i accepts (identifier equals 1) or rejects the final vote corresponding to referendum r. (Served in military)i indicates whether a politician served in the military. 18 (Proposal pro-policy field)r indicates whether a specific organization recommended to accept or reject the policy proposal to voters in referendum r. As the analyzed policy fields tend to be disjoint, we analyze one field (the recommendation of one organization) after the other. We are particularly interested in the coefficient of the interaction term ββ2 between having served in the military and the voting recommendation of a specific organization reflected by (Proposal pro-policy field)r. As both constitutive terms of the interaction are dichotomous, ββ2 reflects the effect of a military background on parliamentary voting for a change from an anti- to a pro-proposal compared to politicians without a military background. Thereby, it corresponds to the cross-difference of the observed voting outcome minus the crossdifference of the potential voting outcome (see Puhani 2012 for nonlinear models) which permits a direct interpretation of ββ2. By nature, the voting recommendation of specific organizations and the fact that some politicians have served in the military while others have not are both exogenous to the policy proposal at hand such that the interaction term can be interpreted as exogenous (see Nizalova and Murtazashvili 2016). The fact that the constituent variables of the interaction term are exogenous is not sufficient in itself to identify the effect of military experience on voting behavior. As politicians are elected by voters and, in principle, their task should be to represent voter preferences, it might be speculated that voters elect politicians with (or without) military experience to represent their preferences in parliament, i.e. military experience may serve as a characteristic to satisfy voter preferences. Neglecting voter preferences can misattribute their policy preferences to a potential 18 In further analyses, we will also analyze different military ranks. 14

influence of military careers of politicians. 19 Thus, it is necessary but in most situations practically impossible condition on observed voter preferences to disentangle the effect of having served in the military from the task of politicians to represent preferences. Our setting allows us to directly control for voter preferences (Voter preferences yes)ir for specific policy proposals which alleviates a potential omitted variable bias coming from the domain of voter preferences when interpreting the coefficient of the interaction term ββ2. Moreover, we can also control directly for ideological preferences of parties (Party ideology yes)ir. Ideology may be related to voting behavior and also to whether politicians advanced in the military ranks. The systematic control of direct measures for voter preferences and ideology is new in the literature when analyzing specific policy proposals linked to military service. For the rest of the variables in equation (1), we follow the previous literature and include other controls such as personal characteristics, party affiliations, and constituency fixed effects why may also be associated with the voting behavior of politicians. Descriptive statistics for all variables are presented in an online appendix. The influence of these additional factors is reflected by the vector γγ. εε iiii stands for the error term. While we formulated equation (1) in terms of a linear probability model, we will estimate a logit model as the dependent variable is binary. 20 Even though we analyze an actual empirical setting, it is worthwhile to briefly discuss the external validity of our setting and whether our results may generalize. Importantly, politicians in our setting cannot simply follow the revealed voter preferences. As in countries without referenda, Swiss politicians do not have an exact knowledge of their constituents preferences when making decisions in parliament (see Garrett 1999; Brunner et al. 2013). On the other hand, they are most 19 Although the literature recognizes the need to control for constituency preferences when analyzing any type of voting on legislative issues, no previous study has used such a direct measure for revealed preferences on the identical policy proposals voted on by politicians when analyzing voting on societal issues of politicians with a military background. Thus, unless voters only elect politicians but do not affect their policy choices, a control for preferences is relevant (see Lee et al. 2004). 20 We will shift for easier interpretation back to a linear probability model when analyzing more interaction terms at the same time. In any case, both models provide qualitatively identical and quantitatively highly similar results. 15

likely aware of the party ideology on specific issues and they may also be aware of interests of different societal organization. However, this will also be the case in other countries without direct democratic instruments such that again, the results are likely to have some external validity. The Swiss parliament is not specific to have politicians who have served in the military. Politicians in numerous countries have served in the military in the past and decide today on different policies. Of course, we are aware that we cannot provide direct evidence for the generalizability of our results but we believe that our setting is insightful, it can help to explain voting decisions in the past and also provides a useful guide to potential voting behavior for the future. IV. RESULTS Baseline results Our main empirical findings can be found in Table 1 for five different policy areas. We always control for voter preferences (Voter preferences yes) and the ideological preferences of a politician s party (Party preferences yes) in all specification, even numbered specifications also control for other personal characteristics, canton fixed effects and legislature fixed effects. Other personal characteristics include among others the sex of politicians, their education as well as party affiliations. As expected, as the share of voters deciding yes in a referendum increases, so does the likelihood of politicians accepting the policy proposal in parliament. Similarly, if the party recommendation is to accept a proposal, individual politicians are more likely to vote yes too, independent of voter preferences. The focus of our analysis is, however, not on the direct influence of voter preferences or party ideology (see, e.g. Stadelmann et al. 2016) but rather on the effect of military service on voting in different societal policy areas holding constant other factors. < Table 1 here > 16

Specifications (1) and (2) analyze differential voting patterns of politicians who served in the military in comparison to those that did not serve in the military with respect to female issues as identified by the voting recommendation by Alliance F. We do not observe any significant interaction term, which suggests that politicians who served in the military are neither more pro nor more against female issues once controlling for constituents preferences and party ideological preferences. 21 Thus, serving in the military is not associated with any differential voting pattern on female issues, in particular, sometimes stated potential chauvinistic tendencies associated with the army are not relevant for parliamentary voting. A similar pattern is observed when looking at the interaction between military service and pro-weak and disabled policies as identified by the voting recommendations of the Association of weak and disabled citizens (AVIVO) in specifications (3) and (4). Having served in the military does not change the values of politicians regarding disadvantaged groups, at least not when voting patterns on actual issues affecting these groups are analyzed. 22 We observe a significant interaction term when analyzing politicians who served in the military and proposals which are pro-neutrality as identified by the Association for an independent and neutral Switzerland (AUNS) in specifications (5) and (6). This suggests that serving in the military is associated with voting more pro-neutrality independently of what constituents actually want and independently of party ideology. This effect also persists when controlling for other personal characteristics. We calculate discrete effects for the interaction terms (following Ai and Norton 2003 and Puhani 2012) which can be interpreted as percentage changes in the likelihood to vote yes of a politician: Politicians how have served in the military are between 4.5 and 5.0%- 21 If we did not control for voter preferences and party ideology, we would also find an insignificant interaction term. 22 The same holds when using other organizations which tend to represent the weak and disabled. However, when employ the voting recommendations of the Caritas, we observe a negative and significant interaction term. The Caritas does, however, not issue voting recommendations on a narrow field but on topics related to the church, migration and the like. 17

points more likely to support a pro-neutrality proposition compared to politicians who did not serve in the military. We look at policy proposals affecting international rights as identified by Amnesty International in specifications (7) and (8) and we observe a negative interaction term. In contrast to proposals pro-neutrality the negative interaction effects indicates that politicians who served in the military are less likely to vote pro-international rights. The discrete effects are large and point to approximately 11.0-%-points. 23 Finally, politicians who have served in the military service do not tend to support environmental policies as identified by the World Wildlife Fund (WWF) to the same extent as politicians who did not serve as evidenced in specifications (9) and (10). The interaction term between (Served in the military) and (Proposal pro-environment) is negative and significant. The discrete effects suggest that politicians how have served in the military are between 19.7 to 25.4%-points less likely to accept a proposal which is pro-environment than politicians who did not serve. Again, the effect is independent of constituent preferences and the party s ideological position on the issue. 24 In summary, we have no indication that having served in the military systematically affects voting on female issues and issues related to the weak and disabled in society. At the same time there is some evidence that politicians with a military background tend to vote rather for neutrality, they tend to be relatively more against an extension of international rights policies and policies that are classified as pro-environment. 23 We note that the number of referenda with a voting recommendation by Amnesty International is relatively small but this corresponds to the actual parliamentary process, i.e. not many policy proposals are linked to international rights. 24 Employing a limited sample of voting recommendations by Greenpeace yields similar results. 18

Controlling for within-party position and excluding right-wing politicians Next, we look at the robustness of our baseline results. Although we control for differences in ideology by using voting recommendations of parties for different policy proposals, politicians with a military background may be different to their colleagues within parties. In particular, it may be expected that politicians who served in the military are more conservative and right-leaning even if they are members of a center or a left party. Our results so far did not only control for party affiliations but for party ideology on specific proposals in Table 1 and we continue to do so in all our estimates. Exploiting differences in the ideology of politicians of the same parties to explain their individual voting behavior is seldom performed in the literature: Commonly, authors only control for party affiliation (see, e.g. Washington 2008). In specifications (1) to (5) in Table 2 we take account of a measure for intra-party positions of politicians. We use the variable rightleaning within party and interact it with the voting recommendation of our respective organization which issues voting recommendations of analyze the influence of within party ideology on voting. < Table 2 here > Specifications (1) to (5) indicate that our baseline results remain broadly robust even when taking account of the intra-party ideological position of politicians. Right-leaning within party has an insignificant interaction term with pro-female and pro-weak and disabled issues, a positive interaction term with pro-neutrality and a negative interaction term with pro-international rights and pro-environment proposals. Politicians who have served in the military are not more nor less likely to vote against pro-female and pro-weak and disabled policies compared to politicians who did not serve in the military. We also observe that politicians who served in the army are more likely to support pro-neutral policies and they tend to be against pro-environmental policies, holding all other factors and in particular voter preferences and party ideology constant. Only for proposals which are pro-international rights the earlier negative interaction term turns statistically insignificant. Thus, the intra-party ideological position partly matters for voting as expected but it 19

generally does not affect the differences between politicians who served in the military and those who did not serve with the exception of proposals on international rights. We know of no setting which takes account of voter preferences, party ideology and within party position all at the same time and with such direct measures as we employ. Going one step further to rule out the influence of right-wing ideology, we only look at politicians from left and center parties in specifications (6) to (10). This systematically reduces our sample of observations. We do this again to insure that our baseline results are not driven by a right wing fraction only even though we always control for party affiliation and issues specific party ideology. Again, our baseline results remain relatively robust for this subsample. We do not observe any change in the sign of the coefficients and only the interaction term between (Served in the military) and (Proposal pro-female) turns marginally significant while the interaction term between (Served in the military) and (Proposal pro-neutrality) loses significance. Refinements with military ranks and voting with party or government We provide refinements of our analyses in Table 3 and Table 4 with a linear probability model. 25 We always include all constituent terms of the interaction terms. < Table 3 here > Table 3 starts by analyzing military ranks in specifications (1) to (5). There is an inherent appeal in analyzing military ranks: Conscription is compulsory in Switzerland such that all able bodied man have to serve in theory. However, advancing to higher military ranks depends on selection and personal motivation (see Bachman et al. 2000; Stadelmann et al. 2015). Our results show that politicians with past military service decide in certain policy areas differently to those without service. It would be interesting to know whether serving in the military actually changes 25 We employ a linear probability model here to make the coefficients of diverse interaction terms more easily comparable to the reader. Our qualitative and quantitative findings do not change systematically when employing a logit model instead. 20

the behavior of politicians or whether rather personal motivation to advance in the military is related to voting. Numerous politicians in our sample have served at higher ranks than the soldier ranks. To analyze differences between compulsory military service and selection to higher military ranks, we distinguish politicians who chose to advance in the military, namely officers and noncommissioned officers from those who only served as soldiers due to conscription requirements. Thereby, we try to distinguish between motivation and potential socialization effects. Our results show no specific pattern such that both selection and socialization play a role when politicians who served in the military vote in parliament. We find, nonetheless, the following four interesting effects which may require further exploration: (1) Officers tend to be more supportive of pro-female issues than politicians who did not serve in the military. (2) Military ranks and military service in general is almost certainly not related with differential voting patterns on proposal pro-weak and disabled. (3) Noncommissioned officers and soldiers are more pro-neutrality than officers and politicians who did not serve in the military. (4) Officers and noncommissioned officers are less likely to support proposals pro-environment than politicians who have served in the soldier ranks and politicians who have not served at all. In specifications (6) to (10) we interact the policy proposals with another factor, namely a politician s sex. We do this to systematically analyze differences between men and women as women do not have to serve in the military and no women in our sample has a military experience. Again, our overall results do not change systematically but we find four further interesting effects: (1) Female politicians are not different to officers when voting on female issues. In fact, officers and female politicians have a higher probability to support pro-female policies than all other politicians. (2) Female politicians are less likely to support proposals proneutrality. (3) Officers and noncommissioned officers oppose pro-environment policies while females support proposals pro-environment more than men who did not serve in the military. (4) 21