What women want: Patricia Funk SITE - Stockholm School of Economics. Christina Gathmann Stanford University. This draft: October 2005

Transcription

1 What women want: Su rage, female voter preferences and the scope of government* Patricia Funk SITE - Stockholm School of Economics Christina Gathmann Stanford University This draft: October 2005 Abstract How women su rage a ected the size of government is poorly understood. While aggregate data suggest that US states with su rage had higher expenditures than the ones without, individual data from elections describe women as more conservative during that period. This paper uses of a unique data set of issue votes to fully describe voter preferences and to directly estimate policy e ects. With surveys on all 195 federal votes held in Switzerland between 1981 and 2003, we nd evidence for gender gaps in multiple dimensions. Secondly, we analyze support for costly projects. Consistent with the gender gaps in preferences, women approve higher expenses in certain areas, but oppose them in others. Overall, we nd larger gender di erences regarding the scope than the size of government. Di erence-in-di erences estimations for the aggregate su rage e ects support the view that scope mattered more than size in the last 50 years. JEL: H10, H50, J16, K00. *Correspondence: Patricia Funk, Stockholm Institute for Transition Economics (SITE), Stockholm School of Economics, Patricia.Funk@hhs.se; Christina Gathmann, Department of Economics and Stanford Center for International Development, Stanford University, cgathman@stanford.edu. We thank Renee Adams, Ulf Axelson, Erik Bergloef, Tore Ellingsen, Daniel Ferreira, Mariassunta Gianetti, Henning Hillmann, Helena Svaleryd and various seminar participants of the Stockholm Institute of Transition Economics, the Stockholm School of Economics, and the University of Uppsala for useful comments and suggestions. We are grateful to Werner Seitz, Magdalena Schneider and Elisabeth Willen from the Swiss Bureau of Statistics, Andreas Ladner, Klaus Armingeon, Hans Hirter and Christian Bolliger from the University of Bern and Francois Loretan from SIDOS for invaluable help in collecting the data. All remaining errors are our own.

2 1 Introduction At the heart of every democracy lies the hope that the preferences of its citizens are best possibly re ected in policy outcomes. 1 A big change in voter preferences might have occurred when women received the right to vote. Since women su rage roughly doubled the electorate, we would expect government to take quite a di erent form, if women have di erent policy preferences than men. 2 Even though granting women the right to vote caused a big change in the electorate in many countries, surprisingly little is known about the nature of female voter preferences and the resulting policy responses. Even worse, research on female voter preferences seems hard to reconcile with the studies on aggregate su rage e ects. For the U.S., Lott & Kenny (1999) found that adoption of women su rage remarkably increased government expenditures between 1870 and Concerning female voters preferences, however, the literature suggests that at least until the presidential election of 1964, women voted more right-wing than men (see e.g. Seltzer, Newman & Leighton (1997), Edlund & Pande (2002)). Since rightist cabinets are generally found to increase government expenditures less than leftist cabinets, 3 these results seem somewhat contradictory. Either, a one-dimensional left-right -wing indicator describes female policy preferences insu ciently, 4 or the aggregate su rage estimates capture more than the transmission of female voter preferences into outcomes such as the size of government. 5 This paper studies the e ects of political participation of women in Switzerland. The fact that 1 While the Downsian (1975) model has a rather optimistic view on that issue (electoral competition forces the policy makers to implement the median voter s position), rent seeking models are more suspicious about the politicians willingness to pursue the voters interests (see e.g. Grossman & Helpman (1996)). 2 Gender has been found to matter in India: female policy makers invested in projects directly relevant to the needs of their own genders. Depending on the regional area, female leaders allocated resources to drinking water and roads di erently than men; see Chattophadhyay & Du o (2004). 3 See Tavares (2004) for an overview of the literature. 4 Note that policy preferences a ect outcomes not only through voting, but also through policy making. 5 The early data set ( ) certainly poses restrictions on the amount of control variables. 1

3 Switzerland is the world leader in the use of direct democracy 6 provides us with a unique data set for studying gender di erences in voting. Alone between 1981 and 2003, 203 federal votes have been held, with survey data available for 195 of these issues. Since the issues span all major policy areas, gender di erences can be described in as many dimensions as necessary. Not only voter preferences, but also the scal consequences of female political participation at the federal level are straightforward to identify. By analyzing a sub-set of issues with implied increases or decreases in government spending, it becomes very obvious whether and in which areas women voted more pro-government than men. The big advantage of our data set is the large amount of issues, we have voting choices upon. Even though the federal micro-data allow to fully describe voting choices over a large span of policy areas, we complement our analysis with an estimation of the reduced-form su rage e ects. Since Switzerland was one of the latest western countries that adopted su rage (on a federal level in 1971, on a Cantonal level between 1959 and 1990), we have an ideal data set for re-estimating the aggregate su rage e ect (with a large number of controls) and to check consistency with the ndings from the federal level. Our main results are the following: (1) The federal survey data display signi cant gender di erences in many political areas. Gender gaps are largest in the areas of environmental policy, energy policy, military policy, social policy and votes involving equalization of women and men. Due to lower turnout of women, female preferences are slightly under-represented in most of the votes. (2) A one-dimensional characterization of voter preferences does not capture the full dimensionality of the gender gap. Most of the gender di erences prevail after controlling for party-ideology. Furthermore, di erences in characteristics (e.g. 6 See Ladner & Braendle (1999). education, employment, proxy for income etc.) 2

4 explain the gender gap in certain social a airs, but not in most other areas. (3) The gender gap in voter preferences translates into scal policy outcomes in a predictable way. In accordance with their preferences, women support issues with a positive e ect on social, environmental and public transport expenditures, and a negative e ect on defense and agricultural expenditures. While women undoubtedly had an e ect on the scope of government (increase in certain types of expenditures and decrease in others), the total size e ect is small. Women were two percent more likely to say yes to votes with a positive e ect on the size of government. (4) Cantonal data support the nding that scope matters more than size. In a di erence-indi erence setting, where the identifying period is between 1959 and 1990, the reduced-form e ect of women su rage is even negative for most types of revenues and expenditures. Only for culture, transport and environmental expenditures, insigni cant positive e ects are found. (5) The main di erence between the Cantonal and federal results concerns social expenditures. While women approved projects with a positive e ect on federal social expenditures (between 1981 and 2003), woman su rage reduced aggregate Cantonal social expenditures between 1959 and The explanation for this di erence seems to lie in the di erent time periods. For social expenditures, in contrast to the other expenditures, we nd robust evidence for a time-varying gender gap. One reason seems to be the latest increase in a sub-group of women (single/divorced and/or employed), which supports certain types of social expenditures. Our article is related to several strands of the literature. As for the political preferences of women and men, there is a substantial literature in political science, which tries to explain the reversal of the gender gap in elections. 7 A recent economic paper on that topic is by Edlund & 7 While women were considered as more right-wing in the postwar era ( traditional gender gap ), there was a process of gender alignment in the 80 s, and in many developed countries, women got more left-wing than men in the 90 s ( modern gender gap ); see e.g. Inglehart & Norris (2000), Inglehard & Norris (2003). A common explanation for this reversal in gender gap is societal modernization, which brought a break-up of traditional family units, a transformation in sex roles and increased labor force participation of women. As for empirical evidence on that subject, Inglehard & Norris (2000) compare men and women s election patterns for a large range of countries. 3

5 Pande (2002), who focus on divorce patterns (and the related income changes) as the main driver of the changing gender gap. 8 In contrast to this substantial literature, which captures political preferences of men and women in a one-dimensional right-left scale, similar research on issue voting is basically absent. 9 Only Berman (1993) looks at gender di erences in issue voting in an indirect way. By focusing on ballots in Arizona in 1914 and 1916, he compares outcome by precinct with respect to the share of women as registered voters. Similar to the literature on the gender gap in elections, Berman (1993) nds that the addition of women to the electorate had signi cant short-term e ects and strengthened conservative components. 10 According to our best knowledge, we are the rst to directly analyze a broad range of voting choices. Therefore, we certainly add to the ndings from opinion polls 11 and electoral preferences. Furthermore, by directly comparing women s and men s voting choices on projects with scal consequences, the impact of female political participation on the size of government is straightforward to identify. In contrast, the e ect of female political participation on the size of government has only been studied with aggregate data so far. 12 Since most of the countries introduced su rage in the be- Consistent with their theory of societal modernization, they generally nd a larger gender gap (i.e. left-scale women - left-scale man) in industrialized countries than in transition countries and within industrialized countries, di erences between age cohorts. As such, these authors advance the political science literature on the gender gap, which focused on the United States and proposed explanations speci c to the U.S. (see e.g. Mueller (1998)). 8 The setting is an economy, where the male income distribution rst-order-stochastically dominates the female one, and where there is positive sorting in the marriage market. Divorce then a ects the taste of redistribution among middle-income voters. Women, who nancially lose from divorce, tend to vote more left, and men, who nancially win, tend to vote more right. Empirical evidence consistent with this hypothesis is provided in Edlund & Pande s (2002) study. 9 The big bulk of research focused on the electoral gender gap (i.e. voting decisions in elections) or related, the partisan gender gap (i.e. gender di erences in party identi cation); see Conover (1988) for a review of the literature. 10 Precisely, he nds that precincts with a relatively large percentage of female (registered) voters were more negative towards a progressive labor package and the abolishment of the death penalty, but more positive towards prohibition. 11 Gender di erences have been discovered in areas involving force, and to a smaller extent in compassion and regulation issues (see e.g. Shapiro & Mahajan (1986) or more recently, an overview by the Center for American Women and Politics (1997)). However, opinion polls include the whole citizenship, and not only the voting population. Furthermore, as Matsusaka & McCarty (2001) demonstrate, statements from opinion polls are bad predictors for later voting behavior. 12 See Lott & Kenny (1999), Abrams & Settle (1999) and Stutzer & Kienast (2004). While Lott 4

6 ginning of the century, the data basis is comparatively poor. Furthermore, reduced-form estimates of women su rage with aggregate data (di erence-in-di erence estimation) help to understand the total impact of female political participation (voting and policy making), but only to the extent that well-known problems such as endogeneity of policy adoption and unobserved trends of early and late adopters are absent. Next to our direct estimates of the scal gender gap (with micro data), we contribute to this strand of literature by re-estimating the aggregate su rage e ect with Swiss panel data. The later time period of adoption bears the advantage of having a larger amount of control variables. In addition, we use Cantonal voting records on all federal issues to carry out robustness tests in a unique way. 13 The structure of the paper is as follows: Section 2 gives an overview of the institutional background in Switzerland. Section 3 compares female and male voter preferences. The policy e ects on the federal level are analyzed in Section 4. Section 5 presents the reduced-form su rage e ect with Cantonal data. Robustness tests are presented in Section 6, and a discussion rounds up in Section 7. 2 Institutional Background When describing the institutions of Switzerland, federalism and direct democracy are certainly the key characteristics. 2.1 Federalism in Switzerland Switzerland has a federalist structure with three levels (federal level, Cantonal level, community & Kenny nd that su rage increased total revenues and expenditures (but not the di erent subcategories), Abrams & Settle (1999) analyze social welfare spending and posit an increase after female political participation. Only Stutzer & Kienast (2004), in parallel work to ours, nd a negative e ect of women su rage on Swiss Cantonal expenditures. 13 The key idea is to directly estimate male ( scal) voter preferences for the Cantons and to check, whether they pose a potential source for omitted variable bias. 5

7 level). While expenses on the federal level account for about 40 percent of total expenditures, the Cantons and communities have also a strong role in providing public goods. For instance, the constitution states that all responsibilities retain with the cantons unless they were explicitly ceded to the federal government. 14 As can be seen from table A1 in the Appendix, there are many areas with shared responsibilities, where Cantons and communities provide a substantial share of public goods next to the federal level. The distribution of revenues between federal, cantonal and local levels is decentralized as well. Cantons have the authority to tax labor and capital income which account for roughly 50 percent of cantonal and local revenues. 15 In contrast, the federal government relies mostly on indirect taxes, the sales tax and several consumption taxes. Since tax revenues are not shared between di erent levels of government or cantons, there is substantial variation in the tax burden across cantons Direct Democracy No other country uses direct-democracy as intensively as Switzerland. Alone on the federal level, more than 450 ballots have been put forward since the country s foundation in 1848 (see Ladner & Braendle (1999: 284)). While citizens may propose changes/additions to the constitution by initiatives, 17 the referendum gives them the control power by voting on actions of the government. Mandatory referenda are hold e.g. on changes in the constitution or the joining of international organizations. In contrast, all laws issued by the government underly the facultative referendum, if eligible voters or 8 Cantons require it within 100 days since the formal resolution. 14 The federal level has the sole responsibility in international relations, defense, customs and currency, atomic energy, media, postal service, telecommunication as well railways and air tra c. Shared responsibilities with the Cantons exist in agricultural policy, civil and criminal law and taxes. In the areas of environmental policy, social security system, roads and industrial and labor regulation, the federal government provides the legal basis while Cantons execute the federal laws. 15 Local jurisdictions can levy a surcharge on Cantonal direct taxes and raise their own property taxes. 16 See for example Feld (2000) for an analysis of tax competition between cantons. 17 A federal vote will occur if more than eligible voters sign. 6

8 Extensive use of direct democracy can also be found on a Cantonal level. The most important direct-democratic elements are again the initiative, and the referendum. However, the 26 Cantons in Switzerland di er widely in their provision and strength of direct-democratic instruments. 18 Generally speaking, direct democracy is stronger in German-speaking parts of Switzerland, and weaker in the French-Speaking part. Direct democracy in its purest form was or still is performed in a couple of Cantons called Landsgemeindekantone (town-meeting Cantons). In these Cantons, the parliament has very little power, but the citizens meet up in town-meetings to decide on laws, elect the judges, set income tax rates and decide on cantonal spending (see Moeckli, 1987). 19 As will be shown later, the (former) town-meeting Cantons, which have the strongest form of direct democracy, were also the ones which gave women the right to vote at the latest. As such, the very direct democratic system in certain Swiss Cantons may actually have inhibited the step towards women su rage and the Long Road to Women Su rage Switzerland was among the last countries in Europe to introduce voting rights for women. At the federal level, su rage was granted in 1971, half a century later than in the United States and many European countries. 20 Since an extension of voting rights requires a change in the constitution, voters have to approve it in a mandatory referendum. 21 The rst federal referendum on the adoption of women su rage was held in Yet, the 18 See Trechsel & Serduelt (1999) for a systematic overview of the Canton s direct-democratic institutions between 1970 and Historically, the function of the parliament has only been to support the citizens with carrying out the issues which have been decided at the town-meeting. The law-making authority lies solely in the hands of the eligible citizens. 20 In Europe, women su rage was adopted by Finland in 1906, Denmark in 1915, Germany, Austria, Poland and Russia (all 1918), Sweden in 1919, Spain and Portugal (both 1931), and Italy in First attempts to introduce women su rage at the cantonal level failed in several Cantons (Neuenburg, Basel, Glarus, Zurich, Geneva and St. Gallen) in 1920/21. In 1929, a petition of citizens supporting women su rage was sent to the parliament but neither the government nor the parliament reacted. 7

9 referendum failed as only 1 in 3 men voted for adoption. In 1971, the referendum passed with almost 2 in 3 men voting in favor of women su rage. Still, in eight cantons out of a total of 26, less than 50 percent of men voted for its introduction. As for the change in support of women su rage, one potential explanation is that the Swiss Government wished to sign the European Human Rights Convention, but only with restrictions ( Vorbehalt ) because women did not have the right to vote Political Preferences of Women and Men The votes that attract most citizens interests are no doubt the federal votes. With decisions on health policy, changes in unemployment insurance, directions in environmental policies, or membership in international organizations being taken on a federal level, top turnout rates of 80 % have been reached. 23 In the following, we will analyze gender di erences in voting on national votes held between 1981 and Even though women received the right to vote on a federal level in 1971, systematic surveys have only been held after The data set is a repeated cross-section survey of between 500 and 1,000 respondents 24 and contains detailed information on the respondent s demographics, economic situation, Canton of residence as well as political values and attitudes. We dropped all respondents under the age of 21, who were not eligible to vote until March of 1991, and under 18 thereafter. 22 See Seitz (2004) for more details on the political struggle for women su rage and the overall representation of women in politics. 23 Between 1950 and 2000, votes on foreign in ltration and the membership in the European Economic Area have reached turnout levels of more than 70 percent. 24 The VOX-surveys started with interviewing around 500 citizens, but then steadily increased the sample up to 1000 in the early 90 s. Not only the number of respondents, but also the number of questions has been enlarged over time. As such, more details about marital status (divorce as a single category compared to a joint category with the widowed prior to 1992) or the household structure have been added. 8

10 3.1 Votes with the largest gender gaps: the top-10 In order to get a feeling for the most gender-sensitive issues, we ranked all the votes according to the gender gap (% yes women -% yes men) in the population. Table 1 displays the 10 votes with the biggest gender gap in the approval rates. insert Table 1 here As can be seen therefrom, big gender gaps occurred in the area of equal rights. Women seem to be more reluctant towards discrimination, no matter whether it concerns gender, race, or opportunities for the disabled. Environmental and energy policy is another area, where gender divides. Women are more pro-environment and less favorable of a nuclear based energy policy. Finally, issues involving risks provoke gender gaps as well. The fact that women voted e.g. against relaxing the speed limit on Swiss national highways is consistent with the view that women are more risk averse than men (see e.g. Dwyer, Gilkeson & List (2002)). At a rst glance, it seems that the most gender-sensitive issues are typical value issues. Attitudes towards the environment, the disabled, the security on streets, or experiments with animals are all issues that entail strong value-statements and hardly a ect a voter s economic position. As such, typical economic issues, with e ects on the income distribution, were not among the most gender-sensitive in the last 20 years. However, whether they matter to a smaller extent will be analyzed subsequently. 3.2 Classi cation of the Votes into di erent Policy Areas We classi ed the votes held between 1981 and 2003 into 7 broad categories, with most of the broad categories having further subcategories. The subcategories are built so that all the votes in a certain 9

11 subcategory pursue the same policy direction. 25 For instance, there were votes on agricultural policy, which either targeted at liberalizing, or at further subsidizing agriculture. Therefore, two separate subcategories Against subsidizing Agriculture and Pro liberal Agriculture were built. In the Appendix, table A2 provides an overview of the titles of the votes, which will be analyzed. As can be seen therefrom, major policy areas ranging from Environmental Policy, Transportation Policy, to Military and Social Policy are covered. For the sake of overview, we focus on the policy areas where we suspect a possible e ect on federal expenses. We also analyzed policy areas with no obvious e ect on the federal budget (e.g. International Politics, Swiss Law & Direct Democracy, Immigration Politics, Gen-Technology), but relegate these results to the footnotes. 3.3 Participation and Voting Outcomes in di erent Policy Areas So far, surveying a representative sample of the Swiss population has displayed gender gaps in several areas. However, to the extent that participation rates between men and women di er, male and female voter preferences receive a di erent weight. The purpose of this section is to estimate the gender gap among the voters in the di erent policy areas. In order to asses, how well female preferences are represented through the voters, we also estimate participation rates and the di erence between the gender gap voters/total population. 26 The model to be estimated is the following: 25 See for a listing of all the votes. The parliamentary debates and arguments of the initiative s supporters are also often included in the comments on initiatives. All comments by the federal council and parliamentary resolutions can be viewed online at 26 Since the questionnaire asked also non-voters, how they would have voted, policy preferences are known for a representative sample of the whole Swiss population. If the gender gap covering the whole population is very di erent from the one found in the voting population, then voters seem to be non-representative. 10

12 Yes ij = + j F emale i + " ij Yes is a binary variable denoting the decision of citizen i, either to participate (turnout model), or to support a vote in subcategory j ( yes =1, no =0). For the moment, we are interested in the pure di erence between women and men s voting choices and only include a dummy for the sex. Later on (Section 3.4), we will investigate more closely why women vote the way they vote (e.g. because of lower education, lower income etc.) and control for various characteristics. On average, the data show a statistically signi cant 1.5 percent di erence of voting yes or no in a federal ballot between men and women. However, there are much larger gender di erences in voting behavior for speci c policy issues. Table 2 reports marginal coe cients of a probit model. As can be seen from the rst row, female voters have signi cantly di erent approval rates in the elds of Environmental & Nuclear Policy, Transportation Politics, Agricultural Policy, Social Security, Culture and Military Policy. 27 Speci cally, women voted more in favor of protecting the environment, the rights of elderly and disabled people, the use of public instead of private transportation, the reduction of military and the preservation of culture (see also Longchamp and Bieri, 2001). In contrast, men were more likely to support a nuclear energy policy, a reform of the unemployment insurance, which included reductions in bene ts, and to relax the speed limit on national highways From the votes in other areas, the largest gender di erences were discovered in the area Equal Rights of Women and Men, where women were 11% more likely to support equalization. Signi cant gender di erences were also found in the areas of Gentechnology, where women were 6% more likely to oppose. Finally, Immigration Policy divided gender, with women being more open towards foreigners. No signi cant gender gaps were found in votes on joining international organizations, concerning directions of drug policy and regulating the housing market. 28 Our ndings match the results from public opinion surveys, which normally cover a subset of 11

13 Women did not only vote di erently, but also participate di erently. As for the total sample, male turnout was 62 % and female turnout was 54 %. For the individual policy areas, women generally had a slightly lower turnout (see column 3), with the turnout gap being lowest for women-speci c topics such as maternity leaves or the legalization of abortion (latter not shown). Therefore, women s preferences are slightly under-represented in most of the issues. Concerning selection into the voting decision, di erences between the gender gap in the voting and the whole population are not that big (compare rows 1 and 2). One exception is the area of equal rights of women and men (results not shown), where feminists were mobilized to a greater extent. 29 There, female preferences might be well represented even despite a slightly lower turnout. insert Table 2 here 3.4 The nature of the gender gaps So far, we focused solely on di erences between female and male policy preferences. Whether these gender gaps occur out of di erences between women s and men s characteristics (e.g. di erences in income, the level of education, location etc.), or rather because women are di erent (biology, values or attitudes) 30 remained unchecked. Furthermore, even though gender gaps have been discovered in policy issues, quite well. For instance, in a comparison of Swedish Attitudes, gender di erences in economic, environmental and social issues have been found, with women opposing nuclear energy, and favoring medical care, environmental protection and sex equality (Eduards (1982)). Also, Norris (1988) found for European Countries that women were more left on policies such as nuclear energy, unemployment and defense. Finally, our result that women seem to be more social than men, con rms what other studies have found: Schlesinger & Heldman (2001) attribute part of women s higher support for compassionate domestic policies to di erent values in altruism and egalitarism; similarly, Canadian women seemed to oppose the free-trade agreement with the States because they had more egalitarian values and were less persuaded about the virtues of competition (see Gidengil (1995)). Finally, sex di erences in racial attitudes have also been discovered in Seltzer & Newman (1997), where women have expressed more positive feelings towards blacks. 29 Schumaker & Burns (1998) found gender gaps between policy makers to be bigger than within the citizen population. In our case, a similar e ect seems to be going on between citizens and voters in the area of Equal Rights. 30 It has been argued in the literature that women s nurturing and kin-keeping role gives rise to di erent policy preferences (see Schlesinger & Heldmann (2001)). 12

14 multiple dimensions, they might be explained by an underlying one-dimensional party-ideology variable. If so, this would justify the focus on di erences in electoral gender gaps as a collapsed form of gender gaps in policy preferences. Table 3 rst row replicates the pure gender gap in the di erent policy areas. Row (2) controls for the demographic variables age, education, religion and language. Since women are generally less well educated than men (see Appendix A3 for Summary Statistics), gender di erences should disappear if di erent education is the driver behind the gender gaps. However, as can be seen from the results in row 2, only certain redistributive issues (reduction unemployment bene ts, subvention health premia) are a ected by the inclusion of demographics. A similar picture emerges, if controls for employment, civil status and house-ownership (as a proxy for income) are added (Row 3). Most gender gaps cannot be explained by di erences in characteristics. Row (4) adds a measure for self-stated party-ideology. 31 The gender gaps disappear only for the three social votes longer maternity leave, reduction unemployment bene ts, subvention health premia and the votes on military expenses. Therefore, a one-dimensional left-right -indicator at best captures some gender di erences regarding re-distribution and size of the army, but leaves out many other dimensions of the gender gaps in voter preferences. Finally, row (5) adds geographic area as well as Time- and Canton- Fixed E ects to the previously described controls. 32 As can be seen from the estimated coe cients, signi cant gender gaps prevail in the area of environmental and nuclear policy, agricultural policy, certain social issues, military and cultural policy. insert Table 3 here 31 This left-right -measure takes values from 0 to 10, with 10 being maximally right. Self-stated party ideology is immune to changes in party-de nition and -structure and therefore most commonly used in the literature (see e.g. Edlund & Pande (2002)). 32 The indicator Party-Ideology is dropped, because there are quite a few missing observations. 13

15 Summing up, we note that gender di erences in policy preferences evolve around issues involving solidarity and risk: while women solidarize more with certain local groups (the disabled, the aged), solidarity is also present in a more global context (peace, protection of the environment). Gender di erences in risky issues were found in votes on the security on the streets, or health risks from the consumption of alcohol or tobacco. These di erent gender gaps cannot be collapsed into one dimension. An indicator for partyideology captures at best preferences for re-distribution, but leaves out most other dimensions. Furthermore, only the gender gap for redistribution can (partly) be explained by di erences in women s and men s characteristics. Gender gaps in other areas seem to arise out of di erent values, since adding characteristics does not make them disappear. 4 Women Su rage and Government Expenditures at the Federal level While the last section described and explained the nature of the gender gaps, this section studies the policy impacts of having female voters. Due to data availability, we focus on voting outcomes. However, when women received the right to vote, they also obtained other possibilities of political in uence such as the right to sign initiatives and referenda. Therefore, women most likely had an impact on the types of votes that were held as well. However, since approval of the vote is needed to have an impact on government policy, a focus on voting outcomes seems a natural starting point. 4.1 Approval of Costly Projects So far, votes in di erent policy areas were studied. While certain votes had scal implications involved, others did not. The goal of this section is to asses gender di erences in the approval of costly projects. The data set consists of a sub-sample of propositions, which had unambiguous implications for 14

16 scal policy (i.e. increase or decrease expenditures, increase or decrease taxes etc.). The government (federal council) prepares detailed comments for each initiative before the vote, which include implied expenditures and taxes in the case of electorate approval. Likewise, the government publishes details and scal consequences for each new law and executive order. 33 With this information, we were able to identify 71 propositions, where the scal consequences were straightforward (see Appendix A4 for an overview). For instance, we excluded propositions where a proposed increase in expenditures was accompanied by a proposed tax to nance it. Only propositions, which clearly led to an increase or decrease in expenditures, subsidies, grants or taxes were taken. For the sake of overview, we de ne a variable More Government, which comprises propositions implying an increase in federal expenditures, taxes, debt, grants or subsidies. Similarly, the variable Less Government contains the propositions, which acceptance leads to a decrease in expenditures, taxes, grants/subsidies or debt. The econometric model is similar to the one before, i.e. probit estimates of the e ect of female sex on the approval of projects in di erent scal categories. Table 4 columns 1 and 2 show the e ect of female voting on the size of government. As can be seen therefrom, women were slightly more supportive of projects leading to an expansion of the government. Overall, they were two percent more likely to say yes to votes with a positive e ect on the size of government. More important than the nancial consequences might be the subject of the vote and will therefore be examined in more depth. In order to understand better, which types of expenditures women were supporting, we further categorized the 71 propositions with predictable nancial consequences. In particular, we distinguished votes a ecting social expenditures, environmental expenditures, agricultural expenditures, cultural expenditures, public transport expenditures and 33 The parliamentary debates and arguments of the initiative s supporters are also often included in the comments on initiatives. All comments by the federal council and parliamentary resolutions can be viewed online at 15

17 defense expenditures. As can be seen from the rest of table 4, the gender gap is the biggest for environmental expenditures, followed by public transport, social, agricultural, 34 cultural and defense expenditures. Women supported higher expenditures, if they were invested in environmental protection, in public transport or social security, but opposed expenditures on military a airs or subsidies for agriculture. insert Table 4 here Therefore, women seem to have di erent preferences in many dimensions, in particular regarding the environment, social security, agriculture, defense and public transport. As for the scal implications, the e ects seem to depend on the type of expenditure. While environmental and social expenditures would have risen, if only women had voted, expenditures for defense and agriculture had declined. Therefore, female voters had a bigger impact on the structure rather than the size of government. 4.2 Women as pivotal voters Even though women supported increases in certain types of expenditures, they only a ected expenditures, if they changed the voting outcome. Furthermore, due to lower participation, women s impact might be smaller than suspected from the approval rates. Since the data present a representative sample, we can roughly evaluate whether women actually changed the voting outcome. 35 Table 5 depicts all the votes, where women and men had accepted di erent voting outcomes. From these 15 votes, women changed the outcome in 4 cases (i.e. the 34 Women s greater opposition towards subsidizing agriculture partly stems from the fact that women more often live in cities. Therefore, the gender e ect turns smaller as soon as a control for urban location is added. 35 Obviously, untruthful reporting might cause some distortions. 16

18 nal result corresponded to the one preferred result by women, but men would have preferred the opposite). As such, in roughly 2 percentage of all federal votes, women actually changed the result. As for the nancial consequences, in one case, women caused an increase in 70 Mio. franks yearly, since they opposed a reduction of unemployment bene ts, and in the other case, they voted for cancelling subsidies for parking spaced, which led to yearly savings of 20 Mio. Swiss franks. In the other two of the four proposals, where women changed the outcome, the nancial consequences are unclear. 36 Compared to total government expenditures of 164 billions in 2003, the scal impact of women voting in federal elections was small. insert Table 5 here 5 Women Su rage and Policy E ects on the Cantonal level The federal data bear the great advantage that voting choices of women and men can be directly compared for a wide range of policy issues. However, the data are only available since the beginning of the 80 s and restricted to voting at the polls. Since women are likely to have in uenced politics since the beginning of su rage, and possibly also through other channels (like policy making and voting at elections), we re-estimate the reduced-form su rage e ects with Cantonal data. The panel data range from and allow to relate women su rage to government expenditures over a longer time-horizon. 5.1 Background information Su rage adoption occurred late, on a national as well as on a Cantonal level. However, there is 36 At a rst glance, women s lower acceptance rates in the votes on Ecological and modern agriculture and Easier access to real estate by foreigners might seem in contrast to what we previously found. However, a detailed analysis of the votes shows that the opponents of the rst vote feared that the reform did not go far enough, and in the second vote, ecological concerns (fear of too many new buildings) played a major role. See: 17

19 substantial disparity among the Canton s timing of introduction. Table 6, rst column shows the year of women su rage adoption on a Cantonal level. 37 As can be seen therefrom, the french-speaking Cantons (Waad, Neuenburg and Geneva) gave women the right to vote around 1960, while the Cantons Appenzell-Ausserrhoden and Appenzell-Innerrhoden introduced su rage roughly 30 years later. As for the Canton Appenzell-Ausserrhoden, it is the only Canton were su rage adoption was involuntarily, and mandated by the federal court in Table 6, columns 2 and 3 show the fraction voting yes in the two referendums by Canton. Obviously, there is a strong positive correlation between the share yes on the federal referenda and the Cantonal timing of introduction. Cantons are clearly ranked according to their support for women su rage and the ranking is roughly stable over time. Between the two referenda in 1959 and 1971, the percentage change of voting yes increased between 20 and 30 percent in all cantons. insert Table 6 here 5.2 Estimation Strategy and Data The purpose of this section is to investigate the e ect of su rage on policy outcomes. Using a panel of 25 Cantons 38 from , we are interested in which form female su rage a ected revenues and expenditures. Since policy outcomes might be a ected by other factors than female voters, we include a broad range of socio-demographic and economic control variables. (DD): The average impact of women su rage on outcome Y is estimated by di erence-in-di erence 37 Similar to the federal level, voting rights for women at the cantonal level had also to be approved by the male electorate in a Cantonal referendum. 38 The Canton Jura was founded in the year 1977 and is dropped from the sample. 18

20 Y st = s + t + b 1 (Dummyfs st F emale st ) + b Z st + u st We capture the su rage e ect by a variable Dummyfs F emale, which is constructed as the product of an indicator variable for su rage adoption (1 if su rage is granted, 0 otherwise) multiplied with the fraction of women older than 20. Intuitively, for su rage having an e ect, women must be eligible to vote (and be older than 20). 39 Yet, the main e ects do not hinge on this particular speci cation. As control variables Z, we include socio-demographic variables (e.g. age structure, share foreigners, religion, education, divorce rates), economic variables (unemployment rate, female labor force participation) and subsidies / revenue shares from the federal level. Canton- xed-e ects pick up time-invariant heterogeneity between the Cantons, and time- xed e ects control for common trends in the development of revenues and expenditures. Table 7 presents summary statistics for early and late introducers. Early introducers are the Cantons which adopted Cantonal su rage before it was adopted on a federal level (i.e. Cantonal acceptance before or at the same day of the national referendum). As can be seen therefrom, early and late introducers di er considerably, and most obviously in terms of language and use of direct democracy. As will be discussed in the robustness Section, these institutional and cultural di erences prove to be the strongest predictors for the Cantons timing of su rage adoption and mitigate concerns of endogeneity. insert Table 7 here 39 The voting age was reduced to 18 in 1991, but we only have 5-year age-classes. 19

21 5.3 Results Table 8 depicts the estimation results. As can be seen therefrom, the average e ect of giving women the right to vote was signi cantly negative for most types of expenditures and revenues. Only for culture, transport and environment expenditures, insigni cant positive e ects are observed. Concerning the economic signi cance of the e ects, let us interpret the e ect of su rage on overall expenditures. With an average share of 51.4 percent women in the total adult population, the impact of su rage on real per capita expenditures was a reduction of 961 Sfr. (=e (0:514 0:076) 1000). Evaluated at the mean level of expenditures, the reduction of 961 Sfr. corresponds to a decrease of nearly 20 percent. As such, su rage had a statistically and economically signi cant negative impact on overall expenditures and most individual expenditure categories. Not only expenditures, but also revenues were negatively a ected by su rage adoption. As for the controls, most of the coe cients seem to be plausible. More federal revenues induced the Cantons to spend more (signi cant positive e ects of share revenues and subsidies), a higher share of elderly people had a large e ect on health expenditures, and a higher share of unemployed increased Cantonal expenditures. A bit puzzling seems to be the negative coe cient before higher education in the education-expenditure regression. While there is a strong positive correlation between a Canton s share of higher educated youngsters and education expenditure, the positive e ect disappears as soon as time- xed e ects and the controls are added. Yet the sign of the coe cient is not that big to pose a serious concern. insert Table 8 here Overall, the data suggest that granting su rage to women exerted a sizeable negative e ect on nearly all types of revenues and expenditures. How can we reconcile these ndings with the results from the federal data? 20

22 Note rst that from the policy preferences discovered from the federal data, we found the greatest gender gaps in the expenditure categories Environment, Defense, Public Transport and Social A airs. While Defense and Public Transport are mostly federal duties (see Table A1 in the Appendix), the general directions in the eld of environment are also taken on the federal level (see footnote 14). Furthermore, the sub-category Environment was built later than the other categories (i.e. in the year 1970), so that less variation might also be a cause for an insigni cant, albeit positive e ect. The biggest puzzle poses the signi cant negative e ects of women su rage on social expenditures (welfare and education). Since identi cation in the estimations stems from the period between 1959 and 1990, the earlier time period might be a possible explanation (remember Edlund & Pande s (1999) model, which predicts a change in the female demand for re-distribution over time). Whether the gender gap on social issues widened over time will be more carefully examined in the robustness section. 6 Robustness So far, the aggregate estimates more than con rm the view that allowing women to politically participate did not lead to an in ation of the government. However, while the micro-data are unambiguous to interpret, di erence-in-di erence estimates (as all estimates with aggregate data) are not immune to potential endogeneity bias. Furthermore, we discovered a di erent female impact on social expenses on a Cantonal level between 1959 and 1990 and on a federal level after Whether the di erent time-period can explain this gap remains to be seen. 21

23 6.1 Endogeneity of Adoption Endogeneity and Di erence-in-di erence Estimation A strict textbook endogeneity problem (causality from expenditures to su rage adoption) is obviously not possible, since male voters decided on su rage adoption. If we talk about an endogeneity problem in di erence-in-di erence estimations, we usually refer to some omitted variable that a ects both, su rage adoption and expenditures. Yet, Di erence-in-di erence-estimation is so popular precisely because endogeneity is less of a concern. Since Canton-Fixed-E ects absorb any time-invariant heterogeneity between the Cantons, omitted variables that simultaneously a ect the expenditure level and su rage adoption do not pose a problem. Only if there is an omitted variable that causes di erent trends in expenditures between early and late adopters, the estimated coe cients might be biased Estimating Cantonal Voter Preferences Since su rage adoption occurred after the majority of the male voters approved it, male preferences determined su rage adoption, but might have had an independent e ect on expenditures. Therefore, scal voter preferences are the key candidate for omitted variable bias. 40 Again, if male voters preferences di er between early and late adopters, but the gap is constant over time, it does not bias the estimations. Only if there is a di erent trend between early and late adopters male preferences, if might bias the estimated su rage e ect. In addition, the amount of the gender gap itself might di er between early and late introducers, e.g. because a large di erence between female and male preferences might have delayed adoption. Then, the long-term e ects of women su rage might di er from the estimated short-term e ect Voter preferences unrelated to spending do not pose a problem. Since we control for a whole bunch of Canton characteristics, scal voter preferences are likely to be partly captured by those. 41 Di erence-in-di erence estimation basically compares early introducers expenditure-change after su rage adoption with the respective expenditures of a control group of non-adopters (i.e. the 22

24 With our unique data set of federal issue votes, we can recover Cantonal male preferences from (remember that su rage adoption occurred in 1971 at the federal level), and total Cantonal voter preferences thereafter. Since we have information about the Cantons share yes on all federal votes (only the surveys are restricted to the votes after 1981), we can build an intuitive measure for scal voter preferences. For the ve decades between 1950 and 2000, we measure the Cantons scal voter preferences as the average share yes on proposals with an implied increase in expenditures. 42 Since this measure of voter preferences is extensively discussed in another paper, 43 we are very brief here. We would only like to mention that in many areas, the Cantonal and federal level have shared responsibilities, and that preferences measured from federal voting data are a good proxy for voter preferences at the Cantonal level. Furthermore, since all Cantons voted on the same federal issues, these voter preferences are directly comparable between the Cantons. We will use our measure for voter preferences to check, whether early introducers had a di erent trend in male voter preferences between , and whether the gender gap di ers between early and late introducers. 44 If both is not the case, concerns for biases are minimal. Table 9 rst column shows that early introducers were scally less conservative than late introducers. Remembering that early introducers had signi cantly higher expenditures than late introducers, this comes at no surprise. Whether early introducers had a di erent trend in male voter preferences is investigated by including a linear time trend and an interaction term time early adopter. As can be seen from column 2, there was no trend in scal voter preferences between 1950 and 1970, and neither a di erence between early and late introducers. As such, it seems safe to assume that there is no di erent trend in male voters preferences, which could dislate introducers). Therefore, the period of identi cation ends when the latest adopter introduces su rage. 42 The classi cation into votes with implied scal consequences is analog to the previous analysis in section 4, where the federal votes with survey data were analyzed. 43 See Funk & Gathmann (2005). 44 Since Cantonal and federal su rage adoption di ered in time, we cannot just plug in the omitted variable into the regressions. 23