Institutional Setting and Data

Our measure of substantive representation contrasts decisions made by politicians with the preferences of the electorate and female preferences for the same policy proposals, with identical wording. In referenda, Swiss citizens vote on the laws and constitutional amendments adopted by parliament. This allows us to analyze how male and female politicians respond to citizens’ preferences regarding the same legislative proposals. Because the gender of voting citizens remains unknown, we rely on post-referendum surveys to measure female voters’ preferences. The approach of measuring and comparing voter preferences and politicians’ decisions sets this article apart from other research on political representation.Footnote ³

We examine the final votes of politicians on legislative proposals during their time in office. Swiss Parliamentarians vote on laws, changes to laws, and constitutional amendments. Legislative proposals accepted by parliament do not necessarily turn into law. Citizens may demand a popular referendum on parliamentary decisions before laws are enacted by collecting 50,000 signatures within 100 days. A referendum is mandatory for any constitutional change. Citizens may propose constitutional amendments themselves through a federal initiative by collecting 100,000 signatures within 18 months (for details see Ahlfeldt et al., Reference Ahlfeldt, Maennig and Mueller2022; Bursztyn et al. Reference Bursztyn, Cantoni, Funk, Schönenberger and Yuchtman2023; Portmann, Stadelmann, and Eichenberger Reference Portmann, Stadelmann and Eichenberger2012; Stadelmann, Portmann, and Eichenberger Reference Stadelmann, Portmann and Eichenberger2019). Referenda reflect revealed preferences for real policies and their expected outcome as they permit constituents to rank them against the status quo (see Frey Reference Frey1994; Portmann, Stadelmann, and Eichenberger Reference Portmann, Stadelmann and Eichenberger2012; Schneider, Pommerehne, and Frey Reference Schneider, Pommerehne and Frey1981; Stadelmann, Portmann, and Eichenberger Reference Stadelmann, Portmann and Eichenberger2014; Stadelmann, Portmann, and Eichenberger Reference Stadelmann, Portmann and Eichenberger2019). We match data on referendum results for each constituency with its legislators’ final votes in parliament on the same legislative proposal.

We analyze the voting behaviour of 700 members of the Swiss National Council, which is the lower house of parliament with 200 seats, from 1996 to 2022 and 123 members of the Swiss Council of States, which is the upper house of parliament with 46 seats, from 2007 to 2022.Footnote ⁴ During the period of analysis, 28.5 per cent of legislators were women, increasing from 21.9 per cent in 1996 to 39.4 per cent in 2022. This is a similar trend as in other democracies (Inter-Parliamentary Union 2024). Legislators are elected in twenty-six constituencies, which correspond to the Swiss cantons. Members of the National Council are elected by proportional representation and members of the Council of States by majority rule.

To identify female preferences in the electorate, we rely on representative post-referendum surveys. These surveys, commonly known as VOX analyses, are based on a sample of 1,000 to 3,000 citizens and are conducted after each referendum.Footnote ⁵ The data obtained from the post-referendum surveys cover all 234 referenda held during our period of analysis. The surveys are highly respected, their results are frequently and prominently discussed in the newspapers, and have been successfully used in the literature (see Ahlfeldt et al., Reference Ahlfeldt, Maennig and Mueller2022; Bello and Galasso Reference Bello and Galasso2021; Stutzer, Baltensperger, and Meier Reference Stutzer, Baltensperger and Meier2019). Survey respondents report their socioeconomic characteristics, including gender, as well as their voting record. We use the share of female respondents who reported having voted ‘yes’ on a referendum to identify female preferences in the electorate. Using the surveys, we can identify the stated preferences of women after referenda via their survey responses on the same legislative proposals that men and women in parliament decided on. Post-referendum surveys for referenda offer a significant advantage over other surveys used in the literature. Employing them, we can exclusively rely on information related to voting behaviour, which entails real political consequences. Voters receive comprehensive information about the consequences of each referendum before their vote. Surveys are designed and weighted to be representative of the Swiss population, and referendum results correspond to the sample analyzed in the survey. Responses to post-referendum VOX surveys can thus be seen as reliable indicators of policy preferences. As we know voters’ preferences in all regional constituencies that elect legislators to the federal parliament as well as parties’ voting recommendations, we can explore whether male and female politicians represent female voters to a different extent, given constituency and party preferences.

Figure 1a shows that female and male preferences are positively correlated. Legislative proposals that receive higher levels of support from women in a referendum usually receive higher levels of support from the male electorate. However, in the context of referendum decisions, quantitatively relevant differences between female and male preferences emerge as shown in Fig. 1b. It displays the share of ‘yes’ votes at the national level for all referenda on the y-axis and the distribution of the absolute difference in the share of stated ‘yes’ votes between female and male respondents for each referendum on the x-axis. A box plot based on the differences between female and male preferences is superimposed in Fig. 1b, highlighting relevant gender gaps. On average, female and male preferences differ by about 5.7 percentage points. This is a non-negligible difference for referenda. By comparison, the median of the mean absolute difference in the share of ‘yes’ votes across constituencies is 6.3 percentage points. Moreover, in 20 out of 234 referenda (8.5 per cent), the majority of female survey respondents reported having rejected (accepted) the policy proposal, whereas a majority of male survey respondents reported having accepted (rejected) the proposal, depicted as triangles in Fig. 1. Given our institutional setting, different majorities could imply a different outcome of the referendum if only women or only men had voted and is, therefore, an indicator for particularly divergent preferences. We thus observe relevant differences in preferences between men and women in the electorate in all policy areas. Table A2 in the Appendix provides summary statistics on our measure of male and female voter preferences per referendum.

Figure 1.

Differences between female and male voters’ preferences.

Note: The left-hand graph shows the correlation between the share of ‘yes’ votes from female and male voters across all referenda. Triangles denote referenda with different majorities between men and women. Dots denote referenda in which the majority of women and men voted the same way. The grey reference band indicates a difference between female and male preferences of less than 5 percentage points. The right-hand graph shows the distribution of the absolute difference between female and male preferences for each referendum on the x-axis and the national share of ‘yes’ votes for each referendum on the y-axis.

Our data exhibits important advantages when compared to the existing literature. By considering voting behaviour in referenda, we observe constituents’ revealed policy preferences. In a parliamentary democracy, politicians usually have to infer voters’ policy preferences from experience or opinion polls when voting in parliament. Also in our setting, politicians in parliament have to make their decisions based on their inferences regarding voter preferences. The strength of our approach is that the true support for specific policies is revealed in the subsequent referendum, making preferences observable for the researcher ex-post. Thus, as politicians vote in parliament before their constituents (men or women) vote in referenda, they have to predict their constituents’ preferences just as in any other instances when they have to represent constituents’ preferences (see Brunner, Ross, and Washington Reference Brunner, Ross and Washington2013; Garrett Reference Garrett1999). This allows us to obtain some external validity of our setting because observed politicians cannot simply follow the revealed behaviour of their constituents. As in countries without referenda, politicians do not perfectly know what their constituency wants when they decide in parliament. Rather, they have to infer the preferences of their constituents based on, for example, experience, personal contacts, media responses, or surveys (see Brunner, Ross, and Washington Reference Brunner, Ross and Washington2013; Stadelmann, Portmann, and Eichenberger Reference Stadelmann, Portmann and Eichenberger2019).

Empirical Strategy for Analyzing Differences in Responsiveness

As discussed in the literature and theory section, there is evidence of gender differences in policymaking. However, it remains an empirical question whether female politicians pursue policies that are closer to the preferences of women in the electorate such that there are differences in substantive representation. To address this question, the preferences of male and female voters with respect to the policy decisions made by male and female legislators must be known. Studies analyzing gender differences in policymaking usually lack a direct measure of voter preferences. We contribute to closing this gap as we measure the preferences of female voters with respect to policy decisions made by male and female politicians in our sample. This allows us to directly investigate whether female politicians are more responsive to female preferences than male politicians.

In our institutional setting, constituents’ preferences for legislative proposals are revealed in referenda. The preferences of women in the electorate can be identified through post-referendum surveys. Legislators are either male or female.Footnote ⁶ We compare whether male or female politicians respond more closely to female preferences. Thus, we can investigate how female preferences in the electorate affect the legislative choices of male and female politicians.

We use the decision of a legislator i on the legislative proposal which led to referendum r as our dependent variable. Because this is a binary measure, we run the following logit model:

$$\eqalign{ MPYe{s_{i,r}} = & {\it \Lambda} (\alpha + {\beta _1}Femal{e_i} + {\beta _2}Femal{e_i}*FemalePreference{s_r} \cr & + {\beta _3}FemalePreference{s_r} + {\beta _4}ConstituencyPreference{s_{i,r}} \cr & + {\beta _5}PartyYe{s_{i,r}} + \gamma{X_{i,r}} + {\phi _i} + {\eta _r} + {\psi _r}) \cr}$$

where $MPYe{s_{i,r}}$ is an indicator equal to one if politician i votes ‘yes’ in parliament on the legislative proposal of referendum r. Female_i is an indicator equal to one if politician i is a woman. FemalePreferences _r is a continuous measure of female preferences in the electorate regarding referendum r. It represents the stated preferences of women in the post-referendum survey, measured as the share of female respondents who reported having voted ‘yes’ on referendum r.

β ₂ is our main coefficient of interest. It captures whether female legislators are, ceteris paribus, more responsive to female preferences than male legislators. In other words, it captures whether female politicians represent female preferences differently than their male counterparts. Specifically, if women in the electorate become more supportive of the legislative proposal (that is, the variable FemalePreferences _r increases), β ₂ indicates whether the probability of female politicians supporting the same legislative proposal increases differently compared to male politicians. Identification is based on different support levels across different referendum decisions. β ₂ could be interpreted as causal in the sense that if the sex of politicians is exogenous and female voters prefer a proposal in the referendum to the status quo, the coefficient β ₂ would causally identify whether male or female legislators vote ‘yes’ in parliament more often (see Nizalova and Murtazashvili Reference Nizalova and Murtazashvili2016). Our framework specifically assesses the extent to which female politicians respond to the revealed preferences of women in the electorate, and thereby broadly corresponds to typical conceptualizations of responsivenessFootnote ⁷ in the literature. β ₂ can also be interpreted as an indicator of differences in substantive representation that are associated with a politician’s gender. If β ₂ = 0, such differences in representation do not exist. Consequentially, gender is not a relevant indicator of differences in substantive representation. If β ₂ ≠ 0, differences exist between male and female politicians in the substantive representation of female preferences in the electorate.

To interpret β ₂ as a measure of how female politicians respond specifically to the preferences of women, we must also account for the preferences of the constituency the politicians are supposed to represent. As Figure B1 in the Appendix shows, this is necessary because the preferences of women in the electorate correlate with the overall preferences of the electorate, given that they reflect the preferences of both female and male voters. Thus, omitting the preferences of the constituency would yield an estimate of β ₂ that includes not only how female politicians respond to the preferences of women but also to those of the constituency, introducing a classical omitted variable bias. Therefore, we account for observed preferences of the constituency in Equation 1. ConstituencyPreferences _i , reflects the share of voters in politician i’s constituency who voted ‘yes’ on referendum r. β ₄ can therefore be interpreted as an indicator of how well politician i represented their constituents’ preferences in referendum r. We expect β ₄ > 0.

We measure female preferences using national post-referendum survey responses and control for the preferences of a politician’s constituency, which are the respective cantons. Thus, our measure of female preferences may pick up systematic differences between national and cantonal voting results. We assume that, if women are more supportive of a referendum at the national level, the same will be true at the constituency level. Thus, FemalePreferences _r represents a measure of female support for referenda not only at the national level but also within a constituency. Robustness tests using the overall national preferences of male and female voters, instead of the preferences of a constituency as a control variable, show that our results do not depend on the level at which the preferences of voters are measured (see Tables A3, A4 in the Appendix).

Another potential confounding factor that could affect our interpretations, if not adequately considered, is the influence of party voting recommendations (party lines). For Switzerland, in line with international evidence, it has been shown that female politicians are more likely to sort themselves into left-wing parties and female voters are more likely to support left-wing policies (Stadelmann, Portmann, and Eichenberger Reference Stadelmann, Portmann and Eichenberger2014). Thus, we introduce PartyYes_i,r, which is an indicator equal to one if politician i’s party recommended voting ‘yes’ on the legislative proposal of referendum r. By including PartyYes _i,r, β ₂ reflects whether female politicians are more responsive to female preferences while taking into account parties’ preferences. Thus, β ₅ can be seen as an indicator of how well politicians respond to their party. Legislators in Switzerland are not obliged to vote according to their party’s recommendation. The Federal Constitution states that they vote without any directives. Nevertheless, parties are a highly relevant principal for politicians. Therefore, we expect that politicians tend to vote with their party (β ₅ > 0).

Female preferences and the constituents’ preferences as well as party recommendations are likely to be partly linearly dependent. It is possible that party preferences are endogenous to voter preferences (that is, the two are correlated) and that both variables are endogenous to female preferences too. Thus, controlling for constituents’ preferences and party recommendations allows us to interpret our coefficient of interest as a pure indicator of how politicians respond to female preferences. Consequentially, β ₂ captures differences in responsiveness to female preferences independently of how politicians respond to constituency and party recommendations. Thus, our interpretations are ceteris paribus for constituents’ preferences and party preferences.

Finally, other covariates, such as personal characteristics, party affiliations, constituency, referendum type, and legislative periods, may be associated with legislative voting. X _i,r captures individual characteristics (age, education level, service length, party affiliation, and electoral rule) of legislator i at the time when the parliamentary vote is held. Φ _i denote constituency fixed effects, η _r denote referendum-type-fixed effects, and ψ _r denote legislative period fixed effects. Table A1 in the Appendix provides descriptive statistics for the relevant variables.

Results for all Policy Areas

We estimate Equation 1 for the entire dataset of 47,527 decisions from 1996 to 2022. Table 1 shows the results.

Table 1.

Effect of female voters’ preferences on decisions by female and male legislators

Note: Dependent variable is MPYes. The discrete effect of FemalePreferences is estimated for an increase in FemalePreferences from 45 per cent to 55 per cent while all other variables are held at their median values. Robust standard errors in parentheses are clustered at the MP level. Logistic models are estimated. ***p < 0.01; **p < 0.05; *p < 0.1.

Specification (1) presents logit estimates for the coefficient β ₂ of the interaction term of interest, Female_i ×FemalePreferences _r , and the two corresponding base effects without any controls. β ₂ is positive and statistically different from zero. However, specification (1) omits that political decision-making does not depend solely on the gender of the politician and the preferences of women in the electorate. Plausibly, the legislators’ decisions are also driven by the preferences of the constituency, made up of female and male citizens and the voting recommendation of their parties, as outlined above.

In specification (2), we additionally control for the preferences of a politician’s constituency. We do so by accounting for the share of ‘yes’ votes among male and female citizens in the respective constituency in a referendum decision. Doing so ensures that any potential differences in the substantive representation of female voters between female and male politicians captured by the interaction term Female_i×FemalePreferences_r are not driven by the preferences of a constituency. As expected, we find a positive and statistically significant effect of constituency preferences on politicians’ decisions (β ₄ > 0). Thus, the probability of a politician voting ‘yes’ increases with the share of voters in their constituency voting ‘yes’.Footnote ⁸ Our main coefficient of interest, β ₂, remains positive and statistically significant, meaning that, independently of constituency preferences, female legislators respond differently to higher support of women in the electorate for a legislative proposal than male legislators. However, specification (2) still omits the general correlation between women and parties’ preferences.

Separately controlling for party voting recommendations, reflected by PartyYes_{i ,r} , but excluding constituency preferences in specification (3), shows a positive and statistically significant effect for party voting recommendations on politicians’ decisions (β ₅ > 0). Thus, the probability of a politician voting ‘yes’ is higher if his or her party recommends a ‘yes’ vote. Once we control for party preferences, the coefficient β ₂ drops considerably and approaches zero. Thus, we no longer observe a statistically significant effect for differences in substantive representation between male and female politicians.

In specification (4) we control for both the preferences of a legislator’s constituency and their party. Both positively affect legislators’ decisions, as the statistically significant coefficients on ConstituencyPreferences _i,r and PartyYes _i,r show. Female legislators do not vote more in line with female preferences than male legislators, as evidenced by the insignificant coefficient β ₂ of the interaction term. Thus, once again, we find no support for female legislators responding more closely to female preferences in parliamentary decisions than male legislators.

In specification (5) we control for legislators’ personal characteristics (that is, age, age squared, service length, service length squared, having a master’s degree or a doctorate, being affiliated to either left or right parties, and the electoral rule). Again, the interaction term of interest remains insignificant. Similarly, there is no effect on the interaction term after controlling for district, referendum type, and legislative period fixed effects in specification (6). The interaction term always remains statistically insignificant and close to zero. The results are thus in line with H1.

To interpret the quantitative relevance of female politicians’ responsiveness to female preferences, we report the difference in the discrete effects of the variable FemalePreferences_r between female and male legislators in Table 1. Discrete effects represent the change in the probability of a politician voting ‘yes’ if the share of ‘yes’ votes among women in the electorate increases from 45 to 55 per cent while all other variables are held at their median values. Figure 2 presents how female and male legislators respond to an increase in the level of support of women in the electorate from 45 to 55 per cent for each of the models estimated in Table 1. When not controlling for party recommendations, female politicians are more responsive to female preferences, particularly when not accounting for constituency preferences. Once party recommendations are added and a full set of controls is estimated, there is no differential responsiveness to female preferences between female and male politicians.

Figure 2.

Discrete effect of an increase in female voters’ preferences on female and male legislators’ decisions.

Note: Discrete effect of an increase in FemalePreferences from 45 per cent to 55 per cent while holding all other variables at their median values and 95 per cent confidence intervals are estimated for female (Female = 1) and male (Female = 0) legislators for models (1) to (6) in Table 1. Confidence bands are based on robust standard errors clustered at the MP level.

Our results are consistent for various specification tests. As seen in Table A3 and the Appendix, we obtain the same results when we control for voters’ preferences at the national level measured as the share of ‘yes’ votes in the entire national electorate, instead of constituents’ preferences. Thereby, we rule out the possibility that systematic deviations in the differences between female and male preferences at the cantonal level from the national level drive our results. Table A4 in the Appendix shows that we also obtain the same results when controlling for voter preferences measured by post-referendum surveys.Footnote ⁹

There is evidence that women in Switzerland turn out less than men (see Goldberg and Sciarini Reference Goldberg and Sciarini2023; Gschwendt Reference Gschwendt2015). Note that this would make both male and female politicians less responsive to women’s preferences. As this article is about differences in female representation between male and female politicians, gender-specific turnout rates do not bias our results.Footnote ¹⁰

To summarize, the coefficient of the interaction term between being a female legislator and female preferences is statistically significant and positive when not controlling for party recommendations. Once party preferences are added, female politicians do not represent female preferences differently than their male counterparts. According to this finding, increasing descriptive representation of women does not automatically guarantee substantive representation of their preferences.

Results for Social and Gender Distinctive Policies

The literature provides evidence that female politicians tend to be more focused on redistribution than male politicians. This potentially leads to different provision of public goods and social policies when the number of female politicians increases (see Chattopadhyay and Duflo Reference Chattopadhyay and Duflo2004; Cunial Reference Cunial2021; Dutta and Maus Reference Dutta and Maus2021; Hessami and da Fonseca Reference Hessami and da Fonseca2020; O’Brien and Piscopo Reference O’Brien and Piscopo2019). If female politicians decide differently on social policy issues and if female voters’ preferences follow the same pattern, there should be differences in the responsiveness of female and male legislators to female voters’ preferences. We, therefore, examine differences in substantive representation of women’s preferences by male and female politicians, focusing on legislative proposals on social policy issues.

To classify referenda by topic, we rely on the categorization of Année Politique Suisse, a platform hosted by the Institute for Political Science at the University of Bern. They assign each referendum up to three out of twelve topic areas, which are defined by the Federal Statistical Office, with ‘Social Policy’ being one of them (Swissvotes 2023). We introduce an indicator variable equal to one if the referendum is assigned to ‘Social Policy’. Social policies represent one of the most important and thus most frequently voted-on topics. 102 of all 234 referenda observed are classified as being related to social policy.Footnote ¹¹

To test H2, we restrict our dataset to legislative proposals on social policies. The results of this exercise are presented in Table 2, specification (1). In specification (2), we exclude decisions on social policies and focus on all other legislative proposals. To make use of our entire dataset, we introduce a triple-interaction term between the three variables Female _i , FemalePreferences _r , and SocialPolicy _r in specification (3). This triple interaction captures the difference in the effect of whether female legislators respond more (less) closely to female preferences than male legislators for social policies, as compared to all other legislative proposals. We always estimate the most conservative specification, including constituency preferences, party recommendations, controls for politicians’ individual characteristics, and our array of fixed effects. Thus, we estimate the same model as in specification (6) in Table 1 but restrict the dataset in specifications (1) and (2), and introduce a further interaction term in specification (3), respectively. Estimating a less conservative setting yields similar results and interpretations.

Table 2.

Effect of female voters’ preferences on decisions by female and male legislators – social and gender-specific policies

Note: The dependent variable is MPYes. The dataset in specification (1) is limited to social policy referenda, that is, SocialPolicy _r = 1. The dataset is limited to non-social policies, that is, SocialPolicy _r = 0, in specification (2). In specification (4) the dataset is limited to referenda in which the majority of female and male survey respondents did not report the same voting record. In specification (5) the dataset is limited to referenda in which the majority of female and male survey respondents reported the same voting record. The discrete effect of FemalePreferences is estimated for an increase in FemalePreferences from 45 per cent to 55 per cent while all other variables are held at their median values. Robust standard errors in parentheses are clustered at the MP level. Logistic models are estimated. ***p < 0.01; **p < 0.05; *p < 0.1.

In contrast to Table 1, which concerns all policies, the results in Table 2 reveal that substantive representation of female preferences in the electorate differs between male and female legislators for social policies. Consistent with H2, the coefficient of interest for Female _i ×FemalePreferences _r is positive and statistically significant when estimating our model using decisions on only social policies in specification (1). We find differences in substantive representation of female preferences using our most conservative model, which controls for constituent and party preferences as well as constituency, legislative period, and referendum-type-fixed effects. This suggests that female politicians respond more closely to female preferences in the electorate regarding social issues. While the literature suggests that female politicians are often more redistribution-oriented, our results show that with such behaviour female legislators respond to the preferences of their female electorate. To interpret the quantitative relevance of female politicians’ responsiveness to female preferences, we report differences in the discrete effects of the variable FemalePreferences_r between female and male legislators as before.

Looking at the difference in the discrete effect between female and male legislators shows that female legislators are 2.29 percentage points more likely to vote ‘yes’ than their male counterparts if support for a referendum within the female electorate increases by 10 percentage points. To put this in perspective, a 10 percentage point increase in support for a referendum among the entire constituency increases the probability of a politician voting ‘yes’ by 6.2 percentage points. Thus, the difference in the responsiveness between female and male politicians to female preferences concerning social policies corresponds to about 37 per cent of the responsiveness of both male and female politicians to the preferences of their constituency.

Given that female politicians respond to female preferences mainly when social policy issues are at stake, we should observe that, for other policy areas, female politicians do not respond more closely to female preferences in the electorate than male politicians. This is precisely what we observe in specification (2) for decisions on non-social policies. We do not observe any statistically significant differences in substantive representation between male and female politicians when we consider issues unrelated to social policies. The coefficient β ₂ of the interaction term Female _i×FemalePreferences _r is even negative but not statistically significant. Thus, outside the social policy domain, there are no statistically relevant differences in substantive representation of female preferences between male and female politicians.

These findings are further supported by the empirical approach used in specification (3). If women are more responsive to female preferences in the electorate only regarding social policy issues, the coefficient of the triple-interaction term Female _i×FemalePreferences _r×SocialPolicy _r should be positive and significant, whereas the coefficient of the interaction term Female _i×FemalePreferences _r should be non-significant because in this setting the coefficient for Female _i×FemalePreferences _r represents substantive representation of female preferences by female politicians for non-social policies. As can be seen in specification (3), this is what we observe. Female politicians are more responsive to female preferences than male politicians with respect to social policies, but not in other policy areas.

The results show that although increasing the number of female politicians in parliament is unlikely to change the substantive representation of female preferences overall, a different picture emerges for legislative decisions on social issues. There is a significant differential effect between male and female politicians with respect to the representation of female preferences on social policies. We observe that female politicians respond more closely to female preferences than their male counterparts in parliament on these issues. Regarding non-social policies affecting female preferences, we do not observe that women in parliament represent female preferences better than male politicians.

Our data allow us to perform an even more detailed analysis based on differences in preferences of men and women in the electorate. The substantive representation of female preferences may differ for policy issues for which the preferences of men and women in the electorate are particularly different. To test H3, we limit our dataset to referenda in which the majority of female survey respondents reported a different voting record than the majority of male survey respondents; for example, 55 per cent of women reported having voted ‘yes’ and 45 per cent of men reported having voted ‘yes’. Different majorities were observed for twenty referenda. In line with H3, the results in specification (4) reveal that substantive representation of female preferences in the electorate differs between male and female legislators when the majority preferences of men and women in the electorate are different.Footnote ¹² Specification (5) shows that in the remaining referenda, responsiveness to female preferences does not differ between male and female legislators.

Heterogeneity and Robustness

So far, our results have shown that there are no differences in the substantive representation of female preferences between male and female politicians on average. However, differences emerge in the domain of social policies and policies with strongly diverging gender preferences. We now explore potential channels of heterogeneity with respect to the substantive representation of female preferences on social issues. In particular, we focus on the socialization of the politicians and potential links to social identity theory, as well as incentives and constraints according to theories of electoral competition. We capture heterogeneity in terms of politicians’ socialization by focusing on politicians’ age and the political incentives and constraints to which they are exposed by focusing on the historical role of women in their constituencies’ politics, the share of women in their party, and the electoral rule.

We restrict the analysis to the sample of legislative decisions on social policies as we only observe differences in the substantive representation of female preferences within this subfield.Footnote ¹³ We perform subsample estimates as outlined in Equation 1, defined by the variables mentioned above. We use the most conservative logit specification, including constituency preferences, party recommendations, personal characteristics control variables, referendum type, legislative period, and canton fixed effects. Figure 3 displays the logit estimate for the coefficient β ₂ of the interaction term Female _i ×FemalePreferences _r when considering social policies only for the above-mentioned split subsamples. The results for policies outside the domain of social policy are presented in Figure B2 in the Appendix for completeness. We obtain qualitatively identical results when performing a three-way interaction instead of a subsample analysis, as can be seen in Table A7 in the Appendix.

Figure 3.

Heterogeneity for social policy issues.

Note: Coefficient β₂ of the interaction term Female _i*FemalePreferences _r and 95 per cent confidence intervals for referred subsets are displayed. The dataset is restricted to social policies. The median age is 54 years. MPs elected by majority vote include all members of the Council of States and all members of the National Council from single-seat constituencies. Median share of women per party is 26.02 per cent. Coefficients for the entire model can be found in Table A5 in the Appendix. Confidence bands are based on robust standard errors clustered at the MP level.

Politicians’ socialization, which can be defined as how individuals learn and internalize the values, roles, beliefs, norms, and behaviours of their society, may affect how politicians respond to female preferences. The public role of women has changed considerably over time. This is reflected in the differential evolution of labour market participation (see Giudici and Schumacher Reference Giudici and Schumacher2017; Goldin Reference Goldin2006), education choices (see Becker and Zangger Reference Becker and Zangger2013; Federal Statistical Office 2023), and political participation (see Kostelka, Blais, and Gidengil Reference Kostelka, Blais and Gidengil2019) between men and women. How male and female politicians have internalized the role of women might, therefore, depend on their age. Younger politicians were socialized at a time when the issue of gender equality was already widely recognized and discussed. Young male politicians might show equal understanding and thus equal responsiveness as female politicians to women’s preferences. By contrast, older male politicians grew up in a time of greater gender inequality. The leftmost graph in Figure 3 displays the coefficient β ₂ of the interaction term Female _i ×FemalePreferences _r and the 95 per cent confidence intervals for legislators who are older, or younger than the median age of all members of parliament. As indicated by the overlapping confidence intervals, there are no differences in the substantive representation of female preferences between male and female politicians of different ages. Thus, there is no differential effect of representation of female preferences with respect to politicians’ age.

In Switzerland, female suffrage was introduced at the national level in 1971 by a referendum, allowing us to explore how responsiveness might depend on more conservative versus more progressive role models. While a national majority of male voters supported the referendum, it did not win a majority in all constituencies. In nine of the twenty-six constituencies, a majority of male voters rejected the right to grant women suffrage at the national level. The conservative attitude toward women expressed by the male electorate in these cantons in the past might still be reflected in the political landscape of these cantons today, shaping women’s life choices, role models in general, and thus their involvement in politics (Slotwinski and Stutzer Reference Slotwinski and Stutzer2022). Voter groups with greater involvement in politics are substantively better represented through various channels, such as better information exchange, larger campaign contributions, and higher voter turnout (see Grossman and Helpman Reference Grossman and Helpman2001). We distinguish between constituencies depending on the involvement of female voters and issues by dividing them based on whether they accepted or rejected the referendum on female suffrage in 1971. One hypothesis is that today’s female legislators from cantons, which rejected women’s suffrage, pay particular attention to the preferences of women in the electorate because they see themselves as advocates against conservative norms. However, it could also be argued that female politicians from cantons with a particularly strong influence of male voters may be under even more pressure to cater to the preferences of male voters and, therefore, not to those of female voters. We capture different incentives to respond to female preferences by considering subsets depending on the cantonal result in the vote on female suffrage in 1971 in the second graph in Figure 3. The results show that there are no differences in the substantive representation of female preferences between male and female legislators from cantons that accepted or rejected women’s suffrage in 1971. Only the coefficient for politicians from cantons accepting female suffrage is statistically significantly different from zero. Still, the confidence intervals overlap such that we cannot claim a statistically significant difference regarding the point estimates.

Next, we turn our attention to political incentives and constraints that may encourage or prevent politicians from closely adhering to the preferences of specific voter groups, such as women. The incentive to address female preferences in particular may arise when only a few party members focus on female voter’s preferences. This might be the case in parties with a low percentage of women. Women from these parties may have an incentive to specialize on women’s preferences and, thus, may represent female preferences more closely than male politicians. We split our sample into politicians from parties in which the share of women is above (below) the median share of women across all parties. The coefficients in the third graph of Fig. 3 support our expectations. There are differences in the substantive representation of female preferences but only among male and female politicians from parties with a small share of women. Women in parties with relatively larger shares of female politicians do not represent women in the electorate substantively differently than male politicians from such parties. But again, the confidence intervals overlap such that we cannot claim a statistically significant difference regarding the point estimates.

The electoral system may be important for the heterogeneity of politicians’ responsiveness to voters in general and female voters in particular (see Budge Reference Budge2012; Carey Reference Carey2007; Golder and Stramski Reference Golder and Stramski2010; Grofman Reference Grofman2004; Portmann, Stadelmann, and Eichenberger Reference Portmann, Stadelmann and Eichenberger2012; Stadelmann, Portmann, and Eichenberger Reference Stadelmann, Portmann and Eichenberger2019). In majority elections, centripetal forces drive politicians toward the position of the median voter. Conversely, in proportional elections, politicians can deviate more widely from the median voter and thus be more responsive to specific groups of voters, such as women. Therefore, whenever differences in the substantive representation of female preferences by male and female politicians occur, we expect them to be limited to politicians elected by proportional rule. The rightmost graph in Fig. 3 shows that only the point estimate for politicians appointed via proportional representation is positive and statistically significantly different from zero. Thus, as expected, differences in the substantive representation of female preferences are limited to politicians elected by proportional rule.

Finally, as an alternative to analysing responsiveness to female preferences, and for robustness purposes, we investigate the level of congruence between male and female politicians with female voters. There is an extended discussion in the literature on the differences between responsiveness and congruence (see Arnold and Franklin Reference Arnold and Franklin2012; Beyer and Hänni Reference Beyer and Hänni2018; Ferland Reference Ferland2020; Ferland Reference Ferland2021; Lax and Phillips Reference Lax and Phillips2012). While both approaches link voters’ and legislators’ positions, congruence in our setting simply reflects the level of agreement of a politician with the majority of the (female) electorate. Given our institutional setting, in which the majority decision is implemented, congruence is a relevant measure (see Kläy et al. Reference Kläy, Eichenberger, Portmann and Stadelmann2024a). It is theoretically possible that women in politics are more congruent with women in the electorate, even though they are not more responsive to female preferences. This is because congruence indicates whether decisions by female legislators mirror the majority of female preferences while responsiveness captures how female legislators, ceteris paribus, respond to differences in the level of support. Thus, we explore this possibility as a robustness check in Table A8 in the Appendix. We estimate the probability that a politician votes with the majority of women in the electorate (that is, whether MPVote _i,r = FemaleMajorityPreferences _r ). If there are no differences in congruence with the female electorate between male and female legislators, an indicator of the legislators’ gender, Female_i , which we use as our explanatory variable, should not be statistically significantly different from zero. Results in Table A8 reveal that the overall congruence of female politicians with female voters is not higher than the congruence of male politicians with female voters when all policy areas are considered. This is also the case when accounting for party preferences. Put differently, male and female politicians are equally congruent with female voters. When we only consider social and redistributive referenda or referenda with large differences in the preferences of male and female voters, we observe differences in the congruence of male and female politicians with female voters. We observe that female politicians have higher congruence levels than male politicians. However, the differences become statistically insignificant when accounting for party preferences.