Does women’s descriptive representation matter for policy preferences? The role of political parties

Although the presence of women has been increasing in several parliaments around the world, we still do not know much about the consequences that their presence has for policy representation. Relying on a rich comparative dataset on prospective MPs’ policy preferences in 12 countries and 87 political parties collected between 2006 and 2012 within the Comparative Candidates Survey, this article aims to understand how political parties interplay with prospective MPs’ sex to affect the latter’s policy preferences. Our results show that the descriptive representation of women makes a difference for policy representation, (i) mainly (though not only) when issues that particularly affect women are at stake and (ii) only concerning issues around which political parties do not yet have settled positions (i.e. uncrystallized issues). There are therefore empirical grounds to support an imposed representation of minority groups to deal with issues that are new on the political agenda.


Introduction
In the last decades, we have been assisting an increase in the number of women in political power, to a great extent as a consequence of the international trend towards the adoption of gender quotas (Franceschet et al., 2012). Whereas in 2000, the global average percentage of women in single or lower houses was 13.5%, 17 years later it is 23.6%. Demands for a more equilibrated representation of women and men are based on several arguments, 1 one of which is that, in certain situations, being physically present might make a difference in terms of policy (Phillips, 1995;Mansbridge, 1999). In particular, there remain overlooked interests that might get more attention when there are more female representatives in public office and hence democratic deliberation can improve (Dovi, 2007: 309). This does not imply that all women share the same preferences and goals -a topic we further elaborate below. In fact, if there were instructions on how to represent women (or any other group), there might be fewer grounds for insisting on having descriptive representatives (Phillips, 1998: 72). However, this argument does claim that the representation of people strictly based on their expressed ideas -and not at all based on who they are -is also unsatisfactory (Phillips, 1995: 157). This is the case because 'a descriptive representative can draw on elements of experiences shared with constituents to explore the uncharted ramifications of newly presented issues' (Mansbridge, 1999: 644). Or, as argued by Phillips, when unanticipated problems and issues emerge during the mandate or when some ideas or concerns reach the political agenda for the first time, then a significant under-representation of disadvantaged groups at the point of final decision can and does have serious consequences (1995: 44), that is, in some situations, the gender composition of parliament might have an impact on the policies approved.
Nevertheless, policy preferences are first and foremost connected to political parties and to their programmatic visions, and in established parliamentary democracies, 'the vast majority of the MPs vote with their party the vast majority of the time' (Willumsen, 2017: 137). In these systems, party discipline tends to be very strong and therefore most MPs behave mainly as party delegates (Andeweg and Thomassen, 2005). Consequently, representatives' party affiliations usually outstrip the strength of MPs' feelings regarding their gender or any other personal characteristics. For this reason, this article places parties at the centre of the analysis and aims to understand how political parties interplay with prospective MPs' sex to affect the latter's policy preferences, depending on the kinds of issues involved.
Scholarship on policy preferences among female and male MPs tends to find differences between both sexes, mainly concerning feminist issues (e.g. Campbell et al., 2010;Childs, 2004;Childs and Webb, 2012;Conway et al., 1997;Diaz, 2002;Kittilson, 2008, among many others). Regarding other issues, documented differences are less common. Although most of these studies do control for ideology before reporting gender differences, they use ideology simply as a control variable. Instead, we argue that parties, and in particular whether or not they have clear positions on certain issues (irrespective of direction), play determinant roles. Therefore, we contend that the descriptive representation of women makes a difference for policy preferences if the issues at stake are not crystallized.
Whereas most previous studies are either case studies (for instance, Campbell et al., 2010), comparative studies with a small N (Schwindt- Bayer, 2006) or studies with a large N but only focused on one issue (Kittilson, 2008), this research relies on a rich comparative data set on prospective MPs' policy preferences in 12 countries and 87 political parties collected between 2006 and 2012 within the Comparative Candidates Survey (CCS).

The relative importance of descriptive representation for policy preferences
There are a great variety of studies about the potential implications that descriptive representation of parliaments -understood as the compositional similarity between representatives and the represented (Pitkin, 1967) -might have on MPs' policy preferences in diverse regions of the world, namely in Western Europe, the United States and Latin America. Nevertheless, there has been a paradigm change. Up until recently, there was an equation of the substantive representation of women with a feminist substantive representation of women Childs and Krook, 2008). In other words, adopting the gender equality position was usually perceived as substantially representing women and defending feminist ideas was conceived as fighting for women's interests.
However, women are not a homogeneous group: 'women differ when they have children or do not, are divorced or not, have been raped or not, are straight or gay, obese or thin, Muslim or Christian, menopausal or prepubescent' (Dovi, 2007: 311), aside from their class and ethnic identities (Childs, 2004: 23). They are not uniform in their needs and desires (Celis et al., 2014: 171); instead, they are a diverse group of people who may have less in common with each other than with similar groups of men (Campbell et al., 2010: 174) and sometimes improving the descriptive representation of some women can even come at a cost to other vulnerable subgroups of women (Dovi 2007: 311). Therefore, representing women does not mean being feminist (Celis and Childs, 2012). For instance, there are conservative women representatives in the United Kingdom (and elsewhere) who claim to act for women but promote socially conservative policies (Campbell and Childs, 2015: 157).
A recent wave of literature has been seeking to rethink the study of women's substantive representation (e.g. Celis, 2012;Celis et al., 2008;Jónasdó ttir and Jones, 2009; among many others). There is now a consensus around the idea that there is no straightforward relationship between women's (or any other group's) descriptive and substantive representation; instead, this relationship has been portrayed as complicated, mediated and probabilistic (Celis and Childs, 2014: 3). As a consequence, the notion of 'women's interests' and the existence of a universal set of women's issues have now also been put into question (Celis, 2013;Celis and Childs, 2012;Schwindt-Bayer and Taylor-Robinson, 2011). Alternately, an inductive approach is recommended in order to map the diverse visions of 'what women need' .
Although, as stated before, feminism cannot be confused with representing women, most empirical studies do find that female politicians (vis-à-vis their male peers) tend to act in a more feminist direction, that is, adopting the gender equality position (Campbell et al., 2010;Childs, 2004;Childs and Webb, 2012;Conway et al., 1997;Diaz, 2002;Kittilson, 2008;Lovenduski and Norris, 2003;Macdonald and O'Brien, 2011;Wängnerud and Sundell, 2012 and so on). However, there are a few studies that conclude otherwise (for instance, Tremblay and Pelletier, 2000;Studlar and McAllister, 2002).
Several other authors have aimed to investigate differences between female and male MPs' attitudes that are not directly related to gender issues. For example, women are often pictured as more liberal (Evans, 2005;Swers, 2002) and more left-leaning (Greene and O'Brien, 2016) than men. In particular, they are more supportive of the view that the government should provide services and assistance to those who are less fortunate (Conway et al., 1997: 37), whether they are children (Jones, 1997) or seniors (Giles-Sims et al., 2012). Women are also more likely than men to back gun control, oppose the use of force to resolve conflicts and support decreased government funding for military programs (Conway et al., 1997: 37). However, in Latin America, no gender differences in attitudes towards education, health or the economy were found among legislators (Schwindt-Bayer, 2006).

Theory and hypotheses: Bringing in the parties
Political parties are experiencing hard times. There is a clear decline in party membership (Van Biezen et al., 2012), volatility is higher than ever (Dassonneville and Hooghe, 2015) and voter turnout has been decreasing in general -although there are some signs of a reversing tendency in countries such as Germany, Austria and the Netherlands, among others. 2 Furthermore, there is a consensus around the idea that the representative functions of parties do not work as they used to (see, e.g. Mair, 2013). Regardless, most MPs are party members, parliamentary work is organized within partisan groups and, depending on the electoral system, parties may play a determinant role in placing candidates on lists and therefore enable or prevent their electoral success. So, even if they are no longer the only vehicles of citizens' representation, parties remain key actors in shaping MPs' policy preferences.
However, there is some evidence that when the issues at stake are connected with the MPs' personal characteristics (for instance with their religious faith or personal health experiences), those characteristics have an impact on MPs' legislative behaviour, namely speechmaking and bill co-sponsorship (Burden, 2007). Looking at gender studies, the simple evidence that women MPs adopt the gender equality position more often than male MPs (as previously mentioned) is also an example of how the linkage between MPs' characteristics and the issues being discussed plays a role.
Therefore, we argue that the degree to which parties shape MPs' policy preferences depends on the kinds of issues we are considering. Whilst in this piece, we have intentionally avoided the controversial concept of women's interests or issues, as we have also circumvented discussing the substantive representation of women, we do claim that there are some issues -such as abortion, maternity, gender quotas, gender-based violence and so on -that are clearly not gender-neutral, since their legal frameworks are of greater concern to women than to men (Phillips, 1995: 68). Because they have more direct consequences on women's than men's lives, we have named them issues that particularly affect women and in the analyses performed, we compare these issues to the remaining ones.
Although it has been shown that party explains more than sex at the elite level, even on attitudes towards gender equality (Campbell et al., 2010;Lovenduski and Norris, 2003;Kittilson, 2006), we also know that the more formal and exposed the act at stake, the heavier the weight of the party. By contrast, in more informal acts, the representatives' feelings regarding their gender show up more often. For instance, while party overwhelmingly accounts for MPs' voting direction in the United Kingdom, sex plays a greater role when it comes to parliamentary debates (Childs and Webb, 2012: 147). Whether the MP is male or female seems even more determinant when the acts are performed behind the scenes, namely within committees and constituencies -where party discipline, identity and inter-party conflicts are less important (Childs, 2004) -or away from legislative bodies entirely (Weldon, 2002). Since we are dealing with policy preferences revealed anonymously in the present research, that is, completely out of sight of the public, we do not expect party to totally trump sex, particularly concerning some issues.
Our argument is then that when it comes to issues that particularly affect women, the role played by political parties is more often overcome by the strength of descriptive representation, that is, by the sex of the candidate, than when other kinds of issues are at stake. To be sure, parties are stronger than sex; however, female prospective MPs are expected to act less often as party delegates and more often as trustees in regards to issues that particularly affect women than in regards to other kinds of issues. So, our first hypothesis poses that issues that particularly affect women are less permeable to party influence than other issues (hypothesis 1). As a consequence, we anticipate that there are more gender differences in those issues overall (than in others) and that when parties are introduced in the analysis, not only is there a decrease in the number of issues where gender differences are visible (which is quite an obvious expectation), but also differences remain mainly in issues that particularly affect women.
Parties differ among themselves in several ways, namely in their formal organization, behavioural norms and programmatic commitments (Gunther and Diamond, 2003). We contend that even if many parties have been going through a de-ideologization process for some decades (Kirchheimer, 1990), programmatic commitment (or more generally, party ideology) stands out as the most important party characteristic in terms of political representation. Also, irrespective of the fact that not all current parties can be integrated into an ideological family (Mair, 2007;Ware, 1996), those families remain an understandable way of classifying parties (von Beyme, 1985). Most importantly for the subject of this research, parties vary on the issues they 'own' (Budge and Farlie, 1983;Petrocik, 1996). Issue ownership refers to 'the policy areas where [a party] has a long-standing reputation for handling the issue well and prioritising the resolution of key challenges' (Wagner andMeyer, 2014: 1020). Usually, the issues that parties own are crystallized for those parties, that is, the parties have clear positions on them (irrespective of direction). By contrast, uncrystallized issues consist of topics that have not been on the political agenda long, on which candidates have not taken public stances and around which political parties are not organized (Mansbridge, 1999: 643).
Following Phillips, we argue that when issues have crystallized in the form of policy proposals, it may be relatively unimportant who the politicians are, as long as they follow their programs (Phillips, 1995: 159). However, when some ideas or concerns reach the political agenda for the first time and parties choose to 'ride the wave' (Wagner and Meyer, 2014), '( . . . ) descriptive representatives are, other things being equal, more likely than non-descriptive representatives to act as their descriptive constituents would like them to act' (Mansbridge, 1999: 646). Furthermore, looking at party members in Nordic countries, it has been suggested that the greatest differences between women and men appear in issues that are not part of a party's ideological core (Heidar and Pedersen, 2006).
Following this line of thought, we expect that in parties where the issues that particularly affect women are crystallized, there are fewer differences between female and male legislative candidates on those issues (hypothesis 2). Concerning the remaining issues, we do not expect to find any particular gender effect. In these cases, there might be higher heterogeneity among prospective MPs when the issues are not crystallized, and more homogeneity when they are crystallized, but there should be no particularly significant gender differences in either case.
According to the new paradigm of gender studies in this area, as previously depicted, the association between descriptive and substantive representation is not expected to follow a simple or binary pattern; representing women does not simply entail being feminist and the notion of 'women's interests' is very contested. Therefore, we have deliberately avoided making any inference concerning the direction of sex differences. We could follow an inductive approach  and elaborate on the differences between women and men within each party but that would lead to an entirely different paper.

The data
We drew on data on issue positions or policy preferences of 7264 female and male prospective MPs collected between 2006 and 2012 within the CCS (2015) and distributed by FORS, Lausanne. 3 Table 1 presents the 12 countries included in the analysis. 4 The number of legislative candidates inquired in each country varies from a minimum of 170 in the Netherlands and a maximum of 1741 in Sweden. 5 Female candidates comprise 37% of the overall sample (vs. 63% male candidates) and in all national samples except for Ireland, women constitute at least 30%. 6 Table 2 presents the 10 indicators of policy preferences available in our data set. Seven of those items have been grouped into two indices, whereas the remaining three items were kept separate on both theoretical and methodological (factor analysis 7 and reliability scaling) grounds. The first index concerns 'immigration and harshness' and  Although the index is less reliable in these latter cases, we have good overall reliability for this instrument because it is usually equal to or greater than 0.600 (see DeVellis, 1991). The second index, 'social equality and environment', includes three items, namely one about fighting inequalities, one on the role of government in social protection and a final one concerning environmental protection. Its reliability is 0.622, which again is a good result. In this case, only two countries have reliability figures below 0.535: Ireland (0.373) and Greece (0.415). Thus, we have reasonable indices in both situations. The remaining indicators that were kept separate are: (1) on same-sex marriage and (2) operationalized issues that particularly affect women, namely positive discrimination towards women (in the labour market) and women's freedom to choose in the case of abortion. 8 All items were coded in a way that implies that lower values always mean 'liberal/progressive, left-wing and/or libertarian' positions, and higher values always mean 'conservative, right-wing and/or authoritarian' positions (on a scale from 1 to 5).

Crystallization of issues
In order to be able to identify whether or not a certain kind of issue is crystallized in a given party, the existent political parties in the data set were classified according to two wellknown party typologies. First, relying on the party family typology (see Mair and Mudde, 1998), the 87 political parties were clustered into six types: new left and greens (i.e. left-wing libertarian parties, reformed communists and greens; 18 parties), communists (i.e. orthodox communists and non-reformed communists; 3 parties), socialists and social democrats (16 parties), liberals (12 parties), conservatives 9 (23 parties) and new radical right (8 parties). Thus, conservatives correspond to the largest number of parties in the sample, followed by socialists and the new left/greens. The smallest numbers of parties in the sample belong to the radical right and to the (non-reformed) communists, while the liberals fall in between. A second classification that works equally well as a proxy to issue crystallization is party left-right ideology (see Benoit and Laver, 2006). This has three categories: left (35 parties), centre (22 parties) and right (30 parties). In order to classify the parties according to both party typologies, we relied mainly on country experts (see acknowledgement) and on the database led by Pilet and Cross (2014). 10 There is a great consensus in the literature around the fact that parties to the left of the political spectrum have tended to take the initiative in introducing measures for women's social and political equality (for instance, Phillips, 1995: 42). More precisely, left-wing parties (especially new left and green parties, but also socialists and, though perhaps to a smaller extent, communists) are usually on the front lines of defending measures of reconciling work and family life, as well as pro-choice stances on abortion issues; therefore, we consider the two issues previously defined as particularly affecting women to be crystallized for left-wing parties. Although it is true that pro-life stances (on abortion issues) are also characteristic of at least some conservative parties, namely the Christian democratic ones, two points should be considered. First, in some secularized societies, conservative parties are no longer as firm on abortion and lifestyle issues as they once were (see, e.g. the Tories under the leadership of David Cameron). Second, in some societies, those issues have always been polarized within Christian democratic parties, which has left room for secularized conservatives to represent the more liberal constituencies on these topics (see, e.g. the case of Sweden). Therefore, we tend to consider that issues that particularly affect women are less crystallized for the conservative party family and/or for right-wing parties than for left-wing parties.
The same argument can be made concerning 'social equality and environment' issues that can be attributed to left-wing parties in general. While socialists and communists clearly own social equality topics and the new left and the greens typically prioritize environmental issues. On the contrary, issues related to immigration and harshness, that is, on how stiff sentences should be or on whether torturing prisoners is ever acceptable are particularly present in radical right-wing parties and, to a certain extent, in conservative parties' discourse. Therefore, we argue that they are particularly crystallized in those parties.

The models
The analysis proceeded in two steps. In the first step, the differences between women and men's average positions in each of the five indicators/indexes previously described (Table 2) were tested. In order to do this, either overall or within parties, we used the t-test (for independent samples) or its corresponding non-parametric alternative (Mann-Whitney test) when the assumptions for the t-test were not met.
In the second step of our analysis, the party became the unit of analysis. Linear regressions were performed with the male-female absolute differences (i.e. without the signs) in policy preferences as dependent variables and with party family and party ideology as main independent variables. Both have been operationalized as sets of dummies. Furthermore, two control variables were added to the model, namely party quotas for women and government status of the party. The inclusion of the first control builds upon the critical mass theory, according to which a certain threshold of representation of any group is needed before major changes in several aspects of legislatures can be noticed (Dahlerup, 2006). Although this theory has been widely criticized (Celis et al., 2008;Childs and Krook, 2008), following Heidar and Pedersen (2006), we believe that in parliamentary parties with relatively high number of women, women as a group probably feel stronger and more comfortable pronouncing their opinions loudly -even if they differ from the mainstream's (or most men's) party positions -which might lead to more pronounced differences between men and women in policy preferences. As an operational measure for this, we used the party quota for women whenever there is one (i.e. the minimum percentage of candidates of each sex required on party lists) or the national quota (when it is higher than the party quota). Parties that have no quota, neither partisan nor national, were coded with zero. 11 The rationale to include government status of the party as a control variable was as follows. Having executive power has some costs, namely it implies great exposure, not only to citizens but also to all opposition parties. Any internal contradiction that becomes public tends to have strong negative effects on the party's image, specifically on the level of public government support, as it suggests leadership problems and disorganization. Overall, more discipline is necessary in order to govern efficiently and effectively, ensure cabinet stability and contribute to good governance (Alderman, 1967;Patzelt, 2003). For all these reasons, we expected party discipline to be higher within parties in government and therefore differences in policy preferences between women and men to be smaller. This variable was measured as a dummy variable: 1, when the party was in government after the election (due to the fact that the CCS are post-electoral and done within 6 months or more after an election) and 0, otherwise.

Testing for differences between women and men in policy preferences
In this first section of the empirical analysis, we concentrated on mapping the (absolute) differences between men and women in terms of policy preferences. According to our first hypothesis, we anticipated that issues that particularly affect women are less permeable to party influence than other issues (hypothesis 1). Hence, when we moved from an analysis where parties were considered all together to an analysis within parties, we expected a significant decrease in the number of issues where gender differences are visible, except for the issues that particularly affect women, namely the issues of abortion liberalization and affirmative action for women in the labour market.
Looking at Table 3, we can say that hypothesis 1 is confirmed. First, in terms of immigration and harshness and social equality and environmental issues, not only do we find very few countries where there are significant  Yes 2 differences between women and men (three and four countries, respectively), but also we find they are very scarce when we control for political party (four parties for both issues, out of 87 parties). On the contrary, when we consider issues that particularly affect women (affirmative action and abortion), we find significant differences in 9 and 8 countries (respectively) and within 16 and 14 parties, respectively. The same-sex marriage case falls in between: there are relevant overall differences in 6 countries and the differences remain within 11 parties. Therefore, although in any issue domain, the norm is the lack of significant differences in policy preferences between women and men once parties are brought into the analysis -and thus confirming the primacy of the party over MPs' sex (Campbell et al., 2010;Lovenduski and Norris, 2003;Kittilson, 2006) -this occurs much less often in issues that particularly affect women and, to a certain extent, when the same-sex marriage issue is considered.
There are some cross-national differences worth noting. Most Nordic countries -notably Sweden, Iceland, Finland and Germany -present relatively high numbers of sex differences both before and after parties' introduction in the analysis, mainly for the issues that particularly affect women; however, many other countries (Australia, Ireland, the Netherlands, Portugal and so on) have remarkably few differences. Similar results have often been found for the Nordic region (e.g. Heidar and Pedersen, 2006;Svaleryd, 2009;Wängnerud, 2009), possibly because it may be easier for women to keep their deviating opinions when they are present in large numbers (Heidar and Pedersen, 2006). More cross-national research is nevertheless necessary to explain these country discrepancies.
Our next task is to explain variation across parties according to whether or not the issues are crystallized within parties.

Explaining party variation in differences between men and women in policy preferences
Using parties as the unit of analysis and two party-level factors (party family and party ideology) as proxy for the crystallization of issues, in this section, we tested explanations for variations in absolute differences between women and men in policy preferences, for which we have built a specific data set (N ¼ 87 parties). We tested the different party typologies (party family and party ideology) separately in Tables 4 and 5 to avoid multicollinearity problems.
We started by analysing the results we got for issues that particularly affect women. Following our second hypothesis, we expected that left-leaning parties vis-à-vis the right (the latter is the reference group in party left-right ideology -see Table 4) would tend to show significantly smaller differences between women and men due to higher crystallization of issues that particularly affect women. The same was predicted for the new left/greens, the communists and socialists 12 vis-à-vis the conservatives (which is the reference group in party family -see Table 5). In both Tables 4 and 5, the cells where we would expect to find a significant effect have been highlighted in grey.
The results offer a mixed picture of the issues that particularly affect women, justifying the relevance of keeping both items separated. In terms of (female) positive discrimination at work, being a left-wing party has no significant effect (Table 4). Similarly, when party families are considered (Table 5), neither belonging to the socialist or communist families nor to the new left or green families plays a significant role. Hence, looking at this policy preference, we do not confirm hypothesis 2. This means that although there are significant sex differences on this issue (see again  Table 3), 13 they are spread throughout all kinds of parties. A more careful look at the 16 parties where a significant (1) Reference category ¼ dummy right.
(2) In the cells, non-standardized regression coefficients are shown; standardized beta coefficients are inside the brackets.
sex difference was found for this issue in Table 3 reveals that they belong to several party families (data not shown). 14 Taken all together, these findings suggest that this issue is not crystallized in any party family or ideological party position. Indeed, while there is a long tradition of political gender quotas, gender quotas in the economic sphere are much more recent and less common (Holli, 2011;Meier, 2013). Our expectations measure up much better when the abortion issue is analysed. Concerning the ideological spectrum (Table 4), left-wing parties show smaller gender differences than right-wing parties. In other words, the leftist orientation -where we argue abortion issue is more crystallized -depresses the differences between women and men (À0.200). Turning now to the ideological families (Table 5), the findings also sustain hypothesis 2. The impact of the party dummies is very relevant in the case of abortion (R 2 of 31.7%). Namely, the leftist parties (including socialists, communists and the new left and greens) show fewer gender differences than the conservative parties.
The second part of hypothesis 2 states that we do not expect other issues besides (female) positive discrimination at work and position on abortion to be related to significant gender effects. Although that is true for both 'immigration and harshness' and 'social equality and environment', it does not apply at all to attitudes towards same-sex marriage.
Looking at the results for same-sex marriage, we observe a pattern very similar to the one we expected for issues that particularly affect women and much like the one we described for the position on abortion: the 'leftist' parties (À0.226) vis-à-vis the 'right-wing' parties depress sex differences in terms of attitudes towards same-sex marriage (R 2 ¼ 25.8%, Table 4). In addition, the socialist and communist (À0.299) and new left and green parties (À0.263) present significantly smaller sex differences in policy preferences than conservative parties (R 2 ¼ 29%, Table 5). A deeper review of the data demonstrates that although there are slightly fewer parties with significant sex differences for abortion than for same-sex marriage (in Table 3), in the former case those differences are slightly more prominent (data not shown). This result challenges our expectation of finding particular outcomes for issues that particularly affect women. Two explanations are likely and further research including more issues is needed to confirm which one of them applies. One possibility is that women are particularly supportive of same-sex marriage, an outcome that has been observed before (Herek, 2002;Sherkat et al., 2010). Another possibility is that there is a modern gap, which applies more to the authoritarian-libertarian scale than to the traditional left-right scale, in the vein of the results found by Campbell (2004). In fact, and although that was not the goal of this article, in all three issues analysed that may be included in the latter scale (namely affirmative action towards women in the labour market, abortion and same-sex marriage), women's preferences are clearly more leftist than men's positions. Although the same can be said for the remaining two issues, the differences between women and men are smaller (data not shown).

Conclusions
Although women have been gaining increasing representation in several parliaments around the world in recent (1) Reference category ¼ the conservatives.
(2) In the cells, non-standardized regression coefficients are shown; standardized beta coefficients are inside the brackets.
decades, we still do not know much about the consequences of that increase in terms of policy representation. This article aims to understand how political parties interplay with the legislative candidates' sex to affect the latter's policy preferences. More specifically, we seek to empirically test two main arguments. The first one poses that since some issues have more direct consequences on women than men's lives, that is, they particularly affect women, we expect them to have more prominent sex differences in policy preferences than other issues. Following Mansbridge (1999) and Phillips (1995), our second argument establishes that the descriptive representation of women makes a difference in policy preferences for issues that are not crystallized within the political party. In a first step, using individual level data on policy preferences across five sets of issues, we arrived at the following conclusions. Significant sex differences in policy preferences are rather scarce and usually do not remain when we control for the party. However, in the issues that particularly affect women (affirmative actions in the labour market and abortion) and -though to a smaller extent -in the same-sex marriage issue, the differences between male and female prospective MPs are much more common and still remain (within a reasonable number of cases/parties) when we control for the party list. These findings confirm our first argument, although they start pointing to same-sex marriage as a peculiar issue -a question we address further in the second part of the article.
In the latter step, having the party as unit of analysis and using party family and party ideology as proxies for the crystallization of issues, we address our second argument, which to our knowledge had never been empirically tested. The analysis performed here totally backs the crystallization theory, since we observe that descriptive representation of women does make a difference for policy representation, but only for uncrystallized issues. In contrast, when a specific issue is at the core of a party policy proposal, this seems to determine a high level of homogeneity within the party and therefore it depresses the differences between women and men. Thus, there is evidence here that supports the theoretical assumption that an imposed representation of minority groups could be justified in order to deal with issues that are new on the political agenda and around which political parties are not yet organized. In those cases, who the representatives are can make a difference.
However, the results we get in the second part of this article further challenge our first argument that descriptive representation of women only matters for issues that particularly affect the female citizens. Indeed, we obtain very similar results for the abortion and same-sex marriage issues, despite the fact that it is groundless to argue that the latter particularly affects women. This suggests that the strength of descriptive representation for policy preferences is not only confined to issues that directly relate to the group, although it is particularly strong in those issues. The contradictory results that previous studies have achieved on trying to identify the policy areas where women and men differ (for instance, Conway et al., 1997;Evans, 2005;Schwindt-Bayer, 2006;Swers, 2002) confirm that this is shaky ground. In any case, additional innovative research is necessary to further elucidate this finding.
All in all, these results suggest that bringing parties into the centre of the analysis on the impact of descriptive representation on policy preferences is absolutely necessary, particularly when considering the level of crystallization of issues within each party.
5. This variation of N is not a matter of concern in this article since the analysis is either done within each country, within each party or has the political party as the unit of analysis. 6. This article is supported by supplementary online material, organized in three appendices (A, B and C). Online Appendix A has more information on the candidates for each country. 7. Online Appendix B provides detailed information on the factor analysis. 8. There is a positive but weak correlation between the two items that particularly affect women (r s ¼ 0.180; p ¼ 0.000). 9. The Christian democrats and the conservatives were grouped together due to the lack of the former parties in many of the countries in our sample. 10. For the country-by-country classification of the different parties, see Online Appendix C. 11. These data were collected from the quota project website: http://www.quotaproject.org/. A few details were completed afterwards using complementary sources. 12. Due to the lack of orthodox communist parties in many of the 12 countries, in the empirical analysis we grouped the socialists and the communists together (see Table 5). 13. In fact, this is the issue where the differences between women and men are bigger. 14. Contrary to the issues of abortion or same-sex marriage, where there is a concentration around the conservative family.