A Man’s World? The Policy Representation of Women and Men in a Comparative Perspective

Abstract

Are the preferences of women and men unequally represented in public policies? This simple yet fundamental question has remained largely unexplored in the fast-growing fields of women’s representation and inequality in the opinion-policy link. Our study analyzes gender biases in policy representation using an original dataset covering 43 countries and four decades, with citizens’ preferences regarding more than 4,000 country-year policies linked to information about actual policy change. Our analysis reveals clear and robust evidence that women’s policy preferences are underrepresented compared to those of men. While this skew is fairly modest in terms of congruence, women’s representation is driven mostly by the high correlation of preferences with men. When there is disagreement, policy is more likely to align with men’s preferences. Our analyses further suggest that women’s substantive underrepresentation is mitigated in contexts with high levels of female descriptive representation and labor market participation. In sum, our study shows that gender inequality extends to the important realm of policy representation, but there is also meaningful variation in unequal representation across contexts.

One of the oldest and deepest fault lines in society is that between men and women. Despite variations across time and space and progress in some areas, women continue to face structural disadvantages in economic, political, and social life (Inter-Parliamentary Union 2021; Organization for Economic Cooperation and Development 2021; World Health Organization 2013). These many interwoven inequalities raise the question of whether they extend to and possibly have their roots in the policy process. More specifically, are the preferences of women and men unequally represented in public policies?

This question brings us to the intersection of two kinds of literature in the field of representation. On one side is the vast literature on women’s representation, where many studies analyze whether and particularly how women’s interests are represented in politics (substantive representation) (Childs and Krook 2006; Wängnerud 2009). For instance, a well-established finding is that women’s substantive representation is enhanced by the presence of female policymakers, i.e., what is referred to as descriptive representation (Chattopadhyay and Duflo 2004; Clayton et al. 2019; Lawless 2015; Thomas 1991; Weldon 2002). On the other side is a more recent but fast-growing literature on inequality in the opinion-policy link (Elsässer and Schäfer 2023; Erikson 2015). Most of these studies have been directed at differences between the poor and the affluent, showing that the preferences of the rich exert outsized influence on policy in many established democracies (Bartels 2016; Gilens 2012). In this study, we combine the themes of the first literature with the research design of the second to analyze potential gender inequality in policy representation.

To the limited extent that this approach has been used in previous studies, scholars have reached different conclusions. Some authors find that policy outcomes are more in line with the preferences of men than women (Homola 2019; Reher 2018; Weber 2020), while others conclude that women are not significantly underrepresented (Bernauer, Giger, and Rosset 2015; Dingler, Kroeber, and Fortin-Rittberger 2019; Griffin, Newman, and Wolbrecht 2012). We make several contributions to this nascent literature by covering many more countries, years, and policy issues than previous studies, by measuring actual policy changes rather than more indirect measures such as party positions, and by analyzing both congruence and responsiveness as indicators of unequal representation.

These contributions are possible thanks to an original dataset that combines existing survey data on preferences for policy changes with hand-coded information on the implementation of those same changes after five years. In total, our data cover 43 countries and four decades (1978 to 2017), with the preferences of nearly two million respondents linked to data on implementation (i.e., possible policy change) of 4,758 policy proposals.

Our analysis provides clear and robust evidence that women’s policy preferences are underrepresented relative to men’s. When it comes to congruence, this unequal representation is real but also very limited because of the close correlation between women’s and men’s preferences. Relative to men, however, women appear to have very little influence on policy, as the positive association between policy and the preferences of women disappears when controlling for the preferences of men. This means that much of women’s representation appears to be an “accidental” byproduct of their alignment with men.

In the second part of the analysis, we look more closely at contextual differences in unequal representation. Exploratory analyses examine whether these differences can be explained by women’s descriptive representation, socioeconomic status, political participation, and general cultural conventions. We find evidence that the first two factors mitigate gender inequalities, with representation differences being smaller in contexts where women are more present in politics and have higher labor market participation.

Overall, our study shows that gender inequality extends to the important area of political representation. From a normative perspective, this is certainly a troubling finding, especially if we believe that governments should give both women and men unbiased influence over policy. However, the fact that this inequality varies between contexts is good news from this perspective, as it suggests (however provisionally) that some avenues can be pursued to achieve more equal representation.

Theory and Hypotheses

Motivation

Gender inequality remains a universal phenomenon. Even in relatively egalitarian societies, women are often disadvantaged in economic, political, and social life. To cite just a few examples, women’s labor force participation is almost universally lower than men’s (United Nations Development Programme 2020), and women’s median earnings are also consistently lower (Organization for Economic Cooperation and Development 2021). Many women face significant household burdens, including the threat of domestic violence (World Health Organization 2013). And in 2021, there were only three countries in the world where a (slight) majority of politicians are women (Inter-Parliamentary Union 2021). So, regardless of the many details and differences, the general existence of gender inequality is undeniable.

For our purposes, however, the relevant question is whether these inequalities extend to and potentially have their roots in political representation. That is, are the policy preferences of women generally ignored by policymakers compared to the preferences of men?

The relevance of this question is threefold. First, and as mentioned earlier, potential gender-based inequality in policy representation is inherently important in a normative sense. It violates the democratic principle of political equality, under which all adult citizens should have equal opportunities and capabilities to influence their government and see their preferences reflected in policy (Ingham 2022; Kolodny 2014). Second, unequal policy representation would add to the relevance of other expressions of gender inequality, for instance in labor market participation, earnings, and political participation, as this suggests that these factors have political inequality as one of their consequences. And third, unequal representation could in turn reproduce or even reinforce other gender inequalities by suppressing policies that would benefit women.

Approach and Contribution

As mentioned earlier, our research question brings us to the intersection of two kinds of literature in the field of representation. On the one hand, there is research on the link between public opinion and policy (Canes-Wrone 2015; Erikson 2015; Shapiro 2011). For our purposes, the most relevant studies are related to inequality in this opinion-policy link on the basis of income (e.g., Gilens and Page 2014). On the other hand, we have research focusing on women’s substantive representation (Childs and Krook 2006; Lawless 2015; Wängnerud 2009). Here, relevant studies include those that consider the conditions affecting whether women’s interests are reflected in policy.

We will say more about the findings of these sets of studies later, but we mention them here to emphasize that we take inspiration from both research lines. Specifically, we combine the themes of the second literature with the research design of the first. We focus on the substantive representation of women and men, but measure representation as the correspondence between opinion and implemented policy. We do not make any assumptions about which policies are in the interest of women or men. Some studies have for instance argued that expanding reproductive rights are in women’s interest, and implementation of such policy, therefore, constitutes the representation of women (e.g., Franceschet and Piscopo 2008). Following the standard approach in the field, we instead consider citizens better represented when policies are implemented that are consistent with their stated preferences.

It should be noted that our approach is subject to potential criticism. First, there is skepticism about the ability of average citizens to develop informed preferences. One consequence of this is that the link between opinion and policy would at best pick up the adaptation of opinions to policy cues, rather than the other way around (Achen and Bartels 2016, 313). Second, and more specific to our topic, scholars studying women’s representation often point out that citizens’ interests can be “uncrystallized” at any point in time (Mansbridge 1999). It is in the process of representation that people develop an understanding of their interests, so the argument goes (Celis and Childs 2012; Saward 2010). Both criticisms question the value of citizens’ policy preferences and their impact on actual policy.

Space limitations prevent us from doing justice to these views and their own critics, but we mention a few considerations here that we believe justify our approach. First, we acknowledge the limitations of public opinion and its measurement through surveys (Berinsky 2017). At the same time, many studies show that aggregate opinion is stable, coherent, and capable of influencing policy (Erikson, Mackuen, and Stimson 2002; Hakhverdian 2010). Second, interests and preferences certainly change over time, but survey questions also measure the outcome of a process of “constructing” interests in the period before the survey (Converse 1987). Moreover, while preferences may be updated and refined over time, they are highly unlikely to reverse completely in the short to medium term, especially at the aggregate level (Page and Shapiro 1992). Thus, we believe that citizens’ preferences are a meaningful (though imperfect) indicator of their interests.

The main alternative—identifying women’s interests by assumption—also has major limitations. It is often unclear how “women’s issues” or “women’s interests” are established, and who has the authority to decide on them. In practice, different actors often have different understandings of women’s issues (Celis et al. 2014). More broadly, this approach simultaneously assumes too much by implying that all women have the same interests and assumes too little by implying that women do not have any interests beyond a relatively narrow set of women’s issues.

We may add that we do not consider public opinion to be a fully exogenous force in the policy process (Celis et al. 2008). In general, we know surprisingly little about opinion formation, and why men and women develop different preferences. However, it is likely that this is affected by their different experiences and social positions. In addition, to some extent preferences might be a product of men and women following different parties and taking up cues from their different platforms, given the general evidence for cue-taking among the public (Boudreau and Mackenzie 2014; Slothuus and Bisgaard 2020). Hence, while it is meaningful to ask whether policy reflects women’s and men’s preferences, we should simultaneously acknowledge the more complicated links between the two sides.

To our knowledge, few studies connect women’s and men’s preferences with policy (or other outcomes) in this way. And to the limited extent that previous studies have addressed this question, they have arrived at different answers. Some authors conclude that political outcomes are more consistent with men’s preferences than women’s (Homola 2019; Kopkin and Roberts 2022; Reher 2018; Weber 2020), while others find that women are not significantly underrepresented (Bernauer, Giger, and Rosset 2015; Dingler, Kroeber, and Fortin-Rittberger 2019; Ferland 2020; Griffin, Newman, and Wolbrecht 2012). Thus, further analysis that includes more data is needed to clarify this debate. We aim to contribute to this effort.

As briefly mentioned in the introduction, we make several (empirical) improvements over these studies of political representation by gender. To begin with, compared to previous analyses, our data cover many more countries, years, and policy issues, allowing us to draw more comprehensive conclusions about the extent of unequal representation. This larger dataset also provides a better opportunity to examine the contextual factors that moderate gender inequality in representation.

An additional contribution comes from the fact that representation is measured in terms of policy changes. Since government policies are what ultimately affect citizens the most, they are “arguably the ultimate metric of representation” (Caughey and Warshaw 2018, 250). In contrast, most previous studies use party positions as their dependent variable (exceptions include Kopkin and Roberts 2022; Reher 2018). However, this is an imperfect proxy for actual policy outcomes, especially since party stances come at a much earlier stage in the policy process. Some studies of the descriptive-substantive link have argued that women’s interests are often hampered at later stages of the policy process (Franceschet and Piscopo 2008; Volden, Wiseman, and Wittmer 2018). To the extent that this is true, analysis of party positions understates any inequality in substantive representation.

A final, important element of our approach relates to the meaning of policy representation. Here, we follow previous literature by theoretically distinguishing between congruence and responsiveness (Bartels 2021; Beyer and Hänni 2018). Congruence reflects who gets what in purely descriptive terms, while responsiveness is a causal process and captures whether policy adapts to changes in public opinion. Both concepts fall under the broader notion of substantive representation. ¹

It is important to add that while the distinction between congruence and responsiveness is an important one, we are limited in how far we can apply it empirically. Responsiveness implies a causal effect of preferences, but causality is very difficult to establish with the observational data usually employed in studies like this one. Still, one important distinction is whether to analyze the correspondence between opinion and policy for different groups independently from each other, or together. For instance, to establish congruence between women’s preferences and policy, we need only to compare the two. But to get at responsiveness, we should expect to see correspondence even when controlling for the preferences of men, that is, also when men and women disagree. While an analysis that controls each group’s preferences for the others still falls short of a causal interpretation, it provides a useful contrast to the bivariate analysis of congruence and goes some way towards measuring responsiveness. We return to this issue in the data section.

Theoretical Expectations and Hypotheses

Turning to our hypotheses, can we assume that women’s political representation is generally weaker than men’s? Several factors suggest that this is indeed the case, at least in terms of responsiveness, each of which is closely related to the broader manifestations of gender inequality noted earlier.

First, women are politically underrepresented almost everywhere in descriptive terms. Given that politicians’ backgrounds often affect whose viewpoints they know, understand, heed, and ultimately advocate (Butler 2014; Phillips 1995), it is likely that the lack of female politicians leads to unequal responsiveness. This is supported by many studies documenting a link between women’s descriptive and substantive representation (Chattopadhyay and Duflo 2004; Lawless 2015; Lowande, Ritchie, and Lauterbach 2019; Volden, Wiseman, and Wittmer 2018).

At the same time, this link has been shown to be more contingent and probabilistic than suggested by earlier research (cf. Weldon 2002). For example, Wängnerud argues that “substantial change—whatever that means—cannot be taken for granted just because a group, such as women, is taking part in decision making to a larger extent than before” (Wängnerud 2009, 59). According to the idea of “politics of presence”, representatives are likely to put forward the policy agenda and preferences of the group that they represent, but only if there are conditions where they can actively participate and have their perspectives and interests considered (Höhmann 2020; Phillips 1995). This is further illustrated by Childs and Krook (2009), who show that, beyond the sheer presence of female representatives, it also matters to what extent they can work in the interests of women, and to what extent they manage to form alliances with men with a similar policy agenda. They even suggest that “a lower proportion of women may be more effective than a higher number because female legislators may be able to specialize in women’s concerns without appearing to undermine male domination” (Childs and Krook 2009, 129).

Second, studies consistently find that female voters have lower levels of political knowledge than men (Delli Carpini and Keeter 1996, 135–77; Fraile 2014). This likely makes it more difficult for women to assert their political demands and influence policy (Adams and Ezrow 2009). Having said this, measurement issues make it hard to say exactly how large this gap is and research has found that to some extent the gap can be explained by men being more likely to guess (Mondak and Anderson 2004). Studies have also shown that the knowledge gap is not as large when applying a more expansive view on political knowledge that includes, for example, practical aspects of the welfare state system (Stolle and Gidengil 2010).

Third, women generally participate less in political activities (Burns, Schlozman, and Verba 2001; Ondercin and Jones-White 2011), which is in turn related to household power relations and bargaining power (Iversen and Rosenbluth 2006). Political participation, too, is an important prerequisite for steering politics in one’s preferred direction (Griffin and Newman 2005). A qualification here is that these inequalities are largest when it comes to voting, and smaller for other forms of participation (Kittilson 2016). Scholars working on gender and participation also disagree over whether the differences have to do with differences in resources, gender roles, or the political context. As Kittilson (2016) puts it, “to date, no single explanatory factor accounts for gender gaps across all types of participation, nor across different types of women. Certainly, these sets of explanation are interrelated and likely interact in complex (and to date largely untested) ways.”

Fourth and finally, women tend to have lower incomes than men. As mentioned earlier, many studies have tested whether higher levels of income correspond to better policy representation, and though there is no consensus on this point (Elkjær and Iversen 2020), most analyses find that affluence buys influence (Bartels 2016; Elsässer, Hense, and Schäfer 2021; Gilens and Page 2014; Persson and Sundell 2023).

To be clear, these four factors do not exhaust the mediating factors affecting gender-based inequality in policy representation. ² Naturally, the way that citizens’ input is processed depends on formal rules and elite actors, including the incumbent government (Annesley, Engeli, and Gains 2015; Htun and Weldon 2010). As we have also seen, all potential drivers of unequal representation come with qualifications. Overall, however, they suggest that both the input from citizens and the response to it from governments will tend to be biased against the views of women.

At the same time, this expectation of political inequality is somewhat weaker when we turn from responsiveness to congruence, since this presupposes differences in policy preferences. If, to take the extreme case, only men have a say in the political process but push for policies that women also support, the latter happen to get what they want (Enns 2015). Consistent with this hypothetical scenario, gender is generally not among the strongest predictors of policy preferences (Häusermann and Kriesi 2015). Nevertheless, gender does affect these preferences in significant ways. For example, women generally prefer more egalitarian and progressive policies than men. Women are also more in favor of reproductive rights and stringent environmental policies, while they are less in favor of tough-on-crime measures (Shapiro and Mahajan 1986; Yildirim 2022). In line with this, previous studies have suggested that women’s interests tend to be better represented by left ideologies (Erzeel and Celis 2016).

Summarizing the preceding considerations, our first hypothesis is the following:

H1: The preferences of women are underrepresented relative to the policy preferences of men in terms of both policy congruence and policy responsiveness.

But we also assume that unequal representation is more pronounced for the latter than for the former. In other words, women are likely to have limited political power compared to men, even if the consequences for the overall direction of policy will remain modest.

A natural follow-up question is: What macro-level factors moderate women’s political voice? In this second part of our analysis, we consider four sets of contextual factors that largely overlap with the mechanisms discussed earlier. First, and perhaps most obviously, we examine whether women’s substantive representation is improved when women have high levels of descriptive representation. As noted at the outset, a substantial body of research has argued and shown that a greater proportion of women in decision-making positions leads to better representation of women’s interests. At the same time, some previous studies incorporating public opinion have not found that women’s policy preferences gain more weight as the number of female parliamentarians increases (Dingler, Kroeber, and Fortin-Rittberger 2019; Reher 2018). While this makes our expectations somewhat uncertain, most of the evidence still suggests a link between descriptive and substantive representation. Hence our second hypothesis is the following:

H2: The underrepresentation of women’s preferences is smaller in contexts with higher levels of women’s descriptive representation.

The second set of factors is socioeconomic and based on known class differences in representation (Erikson 2015; Peters 2018). Here, the reasonable expectation is that women are better represented in contexts where they are higher on the socioeconomic ladder in terms of income, education, and status. This leads to our third hypothesis:

H3: The underrepresentation of women’s preferences is smaller in contexts where women have higher levels of socioeconomic status.

Third, we consider political behavior. Assuming that “politicians and officials are not compelled to pay much attention to classes and groups of citizens who do not vote” (Key 1949, 527), political inequality may be mitigated by widespread and equal political participation (Leighley and Oser 2018). Hence our fourth hypothesis is:

H4: The underrepresentation of women’s preferences is smaller in contexts where women have higher levels of political participation.

Fourth and finally, it is plausible that cultural conventions moderate women’s substantive representation (Matland 1998). That is, if the importance of women’s views in the political process is less constrained by public opinion or law, we would expect those views to carry more weight. Thus, our fifth and final hypothesis is the following:

H5: The underrepresentation of women’s preferences is smaller in contexts where women have more liberties to express their political views.

Data and Empirical Strategy

Data

To measure the policy representation of men and women, we combine existing survey data from 43 countries with hand-coded information on the implementation of policy proposals asked about in the surveys. The survey questions were taken from cross-country survey instruments from the Eurobarometer, the International Social Survey Program, the European Social Survey, the European Values Study, the Comparative Study of Electoral Systems, and the World Values Survey. To be considered, a question had to concern a specific policy proposal and be answerable in an appropriate manner regarding its implementation or lack thereof. The questions needed to be clear enough to allow coders to know what information to look for to check implementation. Questions relating to priorities, importance, or conditions were not included. Nevertheless, inquiries could encompass absolute changes (such as joining NATO or shutting down all nuclear power plants) or relative changes (increasing or decreasing spending in various domains). These questions could pertain to new propositions or already implemented policies (the status quo).

In total, we use 152 unique questions from these surveys, covering a wide range of topics including redistribution, public spending, abortion, and immigration. Each of these questions is asked in multiple countries, and many are asked at multiple points in time. Therefore, at the aggregate level, our dataset consists of 4,758 observations that are survey questions nested in countries and years. The dataset contains information on 43 countries, 40 years, and nearly two million respondents who answered an average of 5.6 questions. Fifty-three percent of respondents are women. The distribution of countries, years, and sources in our data is shown in figures A1-A2 and table A1 of the online appendix, while table A2 provides more information on the policy areas covered by our survey questions.

Variables

The dependent variable in our analysis is dichotomous and indicates whether a particular policy was changed or not. To determine this, a team of research assistants checked whether the policy was implemented after five years. The variables were coded 1 if a government decision was made or a proposed outcome occurred. If there was no change, if a decision was made in the opposite direction, or if no information was found after thorough research, the proposal was classified as not implemented (0). In cases where the official implementation of the policy differed from actual implementation, the team relied on official factual data (de jure). For relative change statements, the team looked at differences between the level in the survey year and the level five years later. If there was a relative change in the direction of the policy proposal, the implementation variable was coded 1; otherwise, it was coded 0.

The coders received clear instructions to code whether the policy proposal was implemented at the specified time points. To determine implementation, coders had to consult various sources based on the type of policy. For instance, if a survey question specifically asked about policy decisions, the team referred to the national parliament minutes. On the other hand, for questions that focused on proposal implementation, the team referred to relevant sources for the topic in question, such as budgets, administrative files, or documentation on closure of nuclear power plants or infrastructure projects.

Our main independent variables are the proportion of men or women who favor a policy change on a survey question. We include all respondents in this calculation, instead of, say, only voters. Many questions ask respondents about the status quo; in these cases, we treat opposition to existing policy as support for policy change. For example, in 2003, 66% of men and 61% of women supported EU membership in France. In this case, our independent variables are equal to 0.34 and 0.39, respectively. ³

Overall, male and female support for proposals is highly correlated in our data with r = 0.97 (within each country, the correlation is the same). The distribution of support for policy change is shown in figure 1. This shows that women tend to prefer more policy change than men, though the average difference between the distributions is small (0.7 percentage points).

Figure 1 Density plot of support for policy change among men and women

Although these correlations are quite striking, they should not be too surprising for several reasons. First, this pattern of high agreement between men and women has already been found in other studies (Schakel and Van der Pas 2021). Second, gender does not stratify preferences as broadly and strongly as, say, social class. Accordingly, previous studies have found somewhat more modest correlations between preferences of different income groups within and across countries (Elsässer, Hense, and Schäfer 2021; Schakel 2021). Third, it is important to note that much of this correlation, which undoubtedly reflects a “true” overlap in preferences, also could be inflated by correlated measurement error arising from the fact that men’s and women’s preferences are based on the same survey questions. For example, some questions may be worded or arranged to elicit strong agreement among all respondents in the survey, while other questions elicit predominantly disagreement, regardless of the topic. Thus, like most other studies of unequal representation, our data somewhat underestimate differences in preferences.

Figure 2 shows the distribution of the simple average difference in support for the proposals in our dataset, across all countries and years, with points to the right being more supported by men, on average, and points to the left more supported by women. The vertical position in the graph does not signify anything. The proposal with the largest gender difference (averaged across all years and countries) is support for nuclear power. Men are also more in favor of joining the European currency and the European Union itself, whereas women are more supportive of gay couples’ right to adopt and a proposal to limit imports. On the classical left-right issue of supporting progressive taxation, women and men are equally supportive. To reiterate, an analysis such as ours that builds on actual preference differences among genders will thus potentially give different results than one that looks at the implementation of proposals assumed to be in the interest of women.

Figure 2 Beeswarm plot of differences in support for policy proposals

In the second part of the analysis, we introduce several potential moderators of gender inequality in representation, all measured at the time of the survey. First, descriptive representation is measured as the proportion of female parliamentarians (Inter-Parliamentary Union 2021) or female ministers (Teorell et al. 2021) in each country-year. Second, the socioeconomic indicators included in the analysis are the proportion of women who are in the labor market and have a college degree (Teorell et al. 2021). Third, political participation is measured as voter turnout (Teorell et al. 2021), since turnout becomes more equal—less skewed—among groups the more widespread it is, and as women’s civic participation, using expert surveys from the Varieties of Democracy (V-Dem) project (Coppedge et al. 2021). Fourth and finally, we measure cultural conventions using survey data from the World Values Survey (WVS) and the European Values Survey (EVS) or, more specifically, as an index of citizens’ views on the justifiability of abortion, divorce, and homosexuality. ⁴ We also consider here women’s civil liberties (including their right to freedom of movement and property), again based on V-Dem data.

Measures

Using the (dichotomous) policy change indicator and (continuous) indicators of support for policy change, we construct several measures of congruence and responsiveness.

Starting with congruence, the simplest strategy to measure this is to assess whether the majority preference in a subgroup is congruent with implemented policy. Returning to the example of French EU membership, this dichotomous congruence measure would equal one for both women and men, since France did not leave the EU in subsequent years and hence a majority of both groups (61% and 66%, respectively) got what they wanted. However, a limitation of this measure is that it ignores anything above or below the 50% threshold (for example, it does not distinguish between 51% and 100% agreement), and it may exaggerate small differences around this threshold that are likely to be affected by sampling and measurement error. Therefore, we also use a second approach, suggested by Bartels (2021), which uses a continuous measure that assesses the proportion of women and men who got what they wanted for each policy. This measure equals 0.61 for women and 0.66 for men in this example.

To measure responsiveness, we regress policy changes on the preferences of men and women in the same model. We also include fixed effects for countries and years to alleviate some omitted variable bias (however, results are not substantively affected by their exclusion). Following the recommendations of the econometric literature, we estimate our dependent variable using linear probability models with heteroskedastic consistent standard errors (Angrist and Pischke 2008). While we are limited in our ability to make causal inferences, as noted before, this is a complementary approach to the entirely descriptive congruence approach and gives some indication of the relative influence of the two groups. A second specification is similar to the first, but here we regress policy change on the non-absolute difference in preferences between men and women while controlling for the average level of support. Positive differences indicate that men are more in favor of policy change than women, while negative differences indicate the opposite. This variable is particularly useful to test whether responsiveness is significantly stronger for one group relative to the other, since a positive (negative) coefficient points to biased representation in favor of men (women).

Results

We present our results as follows. First, we discuss the main results regarding the representation of men and women in terms of congruence and responsiveness. Next, we turn to the potential moderators of gender inequality in representation.

The Link between Preferences and Policy

Table 1 shows the degree of policy congruence for women and men. Both the dichotomous and continuous measures can be interpreted as percentages, though they indicate a different quantity. Dichotomous congruence indicates the percentage of times a majority of a group received the policy it preferred, while continuous congruence indicates the percentage of each group that received its preferred outcome on average.

Table 1 Policy congruence of women and men

Note: ^** p < 0.01, ^* p < 0.05; confidence intervals in brackets.

With this in mind, table 1 has a twofold message. First, there are inequalities in congruence between women and men in both dichotomous and continuous terms. For example, policy matches the preferences of the majority of men 62.7% of the time, while this is the case for women only 61.3% of the time. As the right-most column shows, this difference easily clears the threshold of statistical significance; the same is true for continuous agreement. ⁵ The second message, however, is that these congruence gaps are essentially small, with differences around or below one percentage point. One way to illustrate this is to say that the dichotomous congruence would be equal if 33 of our 4,758 country issues had gone the other way. The continuous congruence could be balanced if 136 country-year-issues switched from policy change to no change or vice versa, which is slightly more, but still a small fraction of the data. Consequently, women and men are unequally represented in terms of congruence, but only slightly.

This result was anticipated by the high correlation between women’s and men’s preferences: if both groups want more or less the same thing, there is no way for one group to get much more than the other. Again, a caveat is that the similarities in preferences—and hence the similarities in congruence—are somewhat exaggerated by common measurement error. But even if we could correct for this, the level of congruence would probably still be very similar for women and men.

Turning to responsiveness, table 2 shows the results of our regression analyses. In model 1 of table 2, we regress policy change on the preferences of women and men. Since all variables range from zero to one, the coefficients can be interpreted as the change in the probability of policy change when one group moves from outright opposition to total support for change, holding the preferences of the other group constant. We see a large positive effect for men, suggesting that they exert substantial influence on policy. On the other hand, the effect of women’s preferences is negative, suggesting that policy change becomes less likely the more they favor it.

Table 2 Linear probability models of five-year policy change by citizen support

Note: * p < 0.05, ^** p < 0.01; confidence intervals in brackets. All models include country- and year-fixed-effects, with robust standard errors.

In models 2 and 3 of table 2 we instead include the difference between men and women in support for the proposal. The positive coefficient in model 2 shows that proposals that are more supported by men are more likely to be implemented, all else equal. This result also holds while controlling for average support of the proposal (model 3). To interpret these effects, figure 3 illustrates the predicted probability of policy change based on the preference gap in model 3, with the other variables held at their mean, together with a histogram of the distribution of preference differences. A proposal for policy change that for instance is supported by five percentage points more men than women has a 7.9% higher likelihood of being implemented than a proposal that is supported by five percentage points more women than men. While the effect might seem modest, it is important to keep in mind that this result is based on a model that controls for average support. The two proposals in this example thus have the same electoral potential, but one is still more likely to become policy. ⁶

Figure 3 Predicted probability of policy change by gender difference in preferences

Distribution of preference differences shown with bars at bottom

To delve more into patterns of unequal responsiveness, figure 4 divides the data into deciles of support for policy change among men and women and plots the percentage of proposals within each cell that saw policy change happen within five years. The graph summarizes three main findings discussed so far: first, preference gaps between men and women are generally small (the observations cluster across the diagonal); second, policy change is more common at higher levels of support (top right portion of the graph); and third, policy change varies differently with the support of men and women (policy change is more common above the diagonal than under it).

Figure 4 Mean policy change (%) of proposals with different levels of support among men and women

Number of observations in parentheses. Percentages only shown for cells with at least 10 observations.

If we compare cells down-to-up, that is, increasing the support of men while holding women’s support constant, policy change becomes more common. On average, going from the cell below the diagonal to the cell above it increases the likelihood of policy change by 13 percentage points. ⁷ In contrast, if we compare left-to-right, that is, increasing the support of women while holding the support of men constant, the likelihood of policy change decreases by three percentage points on average. The latter result may again point to negative responsiveness. The clearer and more important result, however, is that policy responsiveness is much weaker for women than it is for men. In general, women’s policy preferences correspond quite well to implemented policy, but when men and women disagree, women more often draw the shorter straw.

Due to the high correlation between preferences, only two cells in the graph represent instances where majorities among men and women disagree. For 127 observations in the data, between 50% and 60% of women favored change, while only 40% to 50% of men did the same. A mere 13% of these observations saw policy change. Across the diagonal, we find 67 observations where between 50% and 60% of men favored change, compared to between 40% and 50% support for women; 22% of these observations saw policy change happen. There is thus, as noted in previous research (Gilens 2012), a clear status quo bias. In fact, in only two cells does the share of observations with policy change exceed 50%. But the relative probability of change taking place is clearly much higher for proposals preferred only by a majority of men (22%) compared to those only preferred by a majority of women (13%).

In terms of how gender differences in responsiveness varies across policy areas, the largest differences in responsiveness are naturally found for issues and areas where gender differences in preferences also are sizeable. For instance, figure 2 shows that men in general were more positive to EU membership and the EU common currency. Given that many countries in our sample indeed are members of these organizations, there is thus greater responsiveness to men on these issues. In contrast, women tend to hold more positive views on issues related to welfare and public health and see more responsiveness than men on these issues. However, the survey questions were not selected in order to facilitate comparison between policy areas, and we therefore leave the issue of comparing responsiveness across domains to future research.

Robustness Check: National Datasets

Given that the results are intimately bound to the types of proposals used in the analysis, there is a risk that differences in the policy representation of men and women depend on the nature of our data. On the one hand, we include a wide range of countries, but the broad scope also means that the issues asked about in the comparative surveys might not be the most salient in each country. As a robustness check, we therefore replicate our main analysis on four national-level datasets compiled in previous research to document economic inequality in representation. These datasets, from the United States, Germany, the Netherlands, and Sweden are directly comparable to the cross-national data in terms of independent and dependent variables, but they are based on country-specific survey questions, which were often asked in response to public or political debates. ⁸ Observing the same inequalities in these different datasets therefore greatly increases the generalizability of our findings. Table 3 replicates model 1 from table 2, in which men’s and women’s support for policy change are entered into the same model.

Table 3 Linear probability models of four-year policy change by citizen support, in four national datasets

Note: * p < 0.05, ^** p < 0.01; confidence intervals in brackets. All models include robust standard errors.

Even when using four different datasets, collected by different researchers, and including different questions, the results are remarkably similar: the coefficients for men’s support are positive, and those for women’s support are negative. We relegate the other specifications—dichotomous and continuous congruence, and responsiveness to preference gaps—to the online appendix (table B1-table B3), but they also reveal the same picture as the one presented earlier. As in the cross-national data, congruence gaps tend to be small due to overlapping preferences, though policy is more congruent overall with men than with women. When there is disagreement, however, the regression analyses show that policy tends to follow what men want.

Contextual Variation

Having established that women and men are unequally represented, we turn to the question of how and why this gender gap varies across contexts. To further explore macro-level variation in unequal representation, we interact the preference gap parameter with a set of contextual variables. The interaction term is shown in figure 5 for each of our eight moderators. For clarity, each interaction is included in a separate model; refer to tables C1 and C2 in the online appendix for the full results. To make the results easier to interpret, all moderating variables have been standardized so that the numbers presented in figure 5 show how the effect of the preference gap on policy changes with a one standard deviation increase in the variables listed on the y-axis. All models include country- and year-fixed effects. ⁹

Figure 5 Contextual moderators of the impact of gender differences in preferences on policy change with 95 percent confidence intervals

Notes: Negative values indicate that higher values of the moderating variable reduce the relative impact of men’s preferences. Standard errors clustered on country-year combinations.

The first thing that emerges from figure 5 is that most of the interactions are negative, which is consistent with our expectations. The exceptions are the coefficients for civil liberties and civil society participation, but they are close to zero and statistically insignificant. As was shown in figure 3, the probability of policy change increases when men are more positive to it relative to women. A negative interaction thus means that this bias is mitigated in contexts scoring higher on the moderating variable.

A second message, however, is that some interactions are more pronounced than others. More specifically, we find an indication of the importance of descriptive representation, as increases in the number of female ministers and parliamentarians are associated with more equal substantive representation. While the confidence intervals are quite wide, the effects are substantial. The estimates in the graph represent the moderating effect of a one-standard deviation increase in the contextual variable. For instance, when the percentage for women minister is at its mean (22%), the coefficient for the impact of the difference between men’s and women’s preferences is 0.62. A one-standard deviation increase in the percentage of women ministers (an increase of 15 percentage points) implies that the coefficient decreases with 0.27 (the point estimate shown in the graph), almost halving the bias towards men.

Moreover, the share of women with college education and female labor force participation is associated with lower levels of political inequality, indicating the importance of socioeconomic factors, though the coefficients fall just shy of statistical significance (with α = 0.05). In contrast, cultural conventions (measured as women’s culturally progressive values or civil liberties) and political participation (women’s electoral participation and civic participation) do not significantly moderate unequal influence. In sum, unequal representation of women and men varies in meaningful ways across time and space, in part along the lines of descriptive representation and socioeconomic inequalities. However, it is important to keep in mind that the analysis here builds on observational data and should be interpreted as such.

Conclusion

This research has revealed three main findings. First, there is widespread and systematic gender inequality in policy representation, such that women are substantively underrepresented compared to men. Second, this inequality is quite modest in terms of congruence, but when controlling each group’s preferences for the other, men are consistently better represented in terms of responsiveness. Third, unequal representation varies substantially on the macro level. Exploratory analyses suggest that women’s underrepresentation is mitigated in contexts with higher levels of descriptive representation and, potentially, socioeconomic parity.

What are we to make of these findings? First and foremost, the inequality we document is normatively troubling and therefore demonstrates a flaw in established democracies and beyond (Ingham 2022). From this perspective, however, the contextual variation in unequal representation is good news because it shows that it is not an inevitable phenomenon. Societies may indeed be able to achieve a more balanced political process with factors under their control, including descriptive representation.

At the same time, the contrast between the finding that both men and women are well represented in terms of congruence and the fact that women’s congruence often seems to be entirely driven by their agreement with men, raises issues that have previously been discussed regarding economic inequality in representation. Women, just like low-income voters, mainly seem to receive “coincidental representation” (Gilens 2015). On the one hand, one could argue that our results are of only minor concern because men and women receive the policies they want at roughly the same rate. On the other hand, as we have emphasized throughout the paper, unbiased influence is an important normative good, regardless of policy effects (Kolodny 2014), and the results suggest that women’s influence is substantially smaller than men’s (with the now-familiar caveats about causal inferences).

Another, more empirical implication of our study concerns the dependent variable in studies of (women’s) substantive representation. We have already noted that previous studies have found mixed evidence suggesting that women’s preferences are underrepresented, but it is notable that most of these studies use party positions (Bernauer, Giger, and Rosset 2015; Dingler, Kroeber, and Fortin-Rittberger 2019; Weber 2020), or roll-call votes (Griffin, Newman, and Wolbrecht 2012) as dependent variables. The two studies that explicitly analyze policy conclude, as we do, that women’s views are disadvantaged relative to men’s (Kopkin and Roberts 2022; Reher 2018). Combined with our results, then, this strongly suggests that gender inequality is exacerbated in the latter stages of the policy process. This is also a methodological lesson for students of representation, as it points to the dangers of using proxies for policy.

The results also point us in the direction of future studies. We note here four suggestions related to between-issue variation, salience, between-country variation, and mechanisms. First, much of the variation in preferences and policy is between policy issues in the same country-years. Thus, there is great potential for unequal representation to vary across issues, which future research can explore using datasets such as those analyzed in this study (see also Kopkin and Roberts 2022). Second, and related to the first point, our study is similar to many analyses of the relationship between opinion and policy in that it treats all policies equally and does not include the possibility that some are more important to (groups of) citizens than others. At the same time, it is both empirically and normatively significant if citizens are underrepresented on issues that affect them most, even if all other issues have different patterns of representation.

Third, due to the limitations of survey availability, most of our data comes from industrialized democracies, especially in Europe. While we uncover significant and meaningful variation in unequal representation between contexts, this variation is still limited as a result. To fully test whether factors such as economic development or cultural conventions affect substantive representation, we need to look for other and perhaps new data sources (see, for example, approaches used by Clayton et al. 2019 and Lupu and Warner 2022).

Fourth, and again related to the last point, this literature should pay more attention to the causal mechanisms that lead to gender bias in representation. Our analysis of potential moderators is a first step in this direction, but we acknowledge that these causal inferences are somewhat weak, in part because many contextual factors are highly correlated and therefore difficult to separate. It is also clear that the gender bias we describe here overlaps with the bias against less affluent citizens mapped in previous research, as women on average tend to earn less and have less wealth than men. To some extent, gender differences in responsiveness could possibly be attributed to gender inequality in income. However, research on income-based bias in responsiveness has so far also been unable to determine the exact causal mechanisms behind that bias. Furthermore, there are many more potential sources of unequal political influence that could result in biased responsiveness that correlate with both gender and income, such as level of education, occupational sector, area of residence, and so on. It is beyond the scope of the group-level analyses we perform here to determine the relative impact of each of these factors, but future research would do well to include individual-level data to disentangle the mechanisms behind political inequality.