Do Women Vote Less Correctly? The Effect of Gender on Ideological Proximity Voting and Correct Voting

Studies routinely find that women have lower levels of political knowledge than men. This gender gap is usually interpreted as troublesome for democracy, because a lack of political knowledge could imply that women’s participation in politics is less effective and that their interests will be represented less well than those of men. In this short article, we present a direct test of the assumption that women are less effective voters because of this lack of political knowledge. We make use of the Comparative Study of Electoral Systems data to study gender differences in proximity voting and correct voting. Our results do not suggest that women vote less correctly than men—a conclusion that prompts important questions about the role of different forms of political knowledge and the seemingly gendered nature of the vote choice.

Studies routinely find that women have lower levels of political knowledge than men. This gender gap is usually interpreted as troublesome for democracy, because a lack of political knowledge could imply that women's participation in politics is less effective and that their interests will be represented less well than those of men. In this short article, we present a direct test of the assumption that women are less effective voters because of this lack of political knowledge. We make use of the Comparative Study of Electoral Systems data to study gender differences in proximity voting and correct voting. Our results do not suggest that women vote less correctly than men-a conclusion that prompts important questions about the role of different forms of political knowledge and the seemingly gendered nature of the vote choice.
O ne of the most consistent findings in the field of gender and politics is the observation that women have lower levels of political knowledge than men (Dassonneville and McAllister 2018;Fraile 2014). This gender gap is found in a varied set of democracies (Fortin-Rittberger 2016; Fraile and Gomez 2017) and appears to be remarkably stable over time (Fraile 2014). While observed gender differences in political knowledge can be partly explained by measurement factors such as survey mode, question format (Fortin-Rittberger 2016; Mondak and Anderson 2004), or the content of the knowledge questions (Stolle and Gidengil 2010), differences in men's and women's levels of knowledge remain significant.
The gender gap in political knowledge is consistently interpreted as worrisome, given the importance of political knowledge; it enables effective participation and the representation of one's interests (Dolan 2011, 97). Political knowl-edge is indeed one of the strongest predictors of electoral participation (Smets and van Ham 2013) and, hence, influences whose voices will be heard. In addition, political knowledge influences the ways in which citizens decide what party or candidate to vote for (Singh and Roy 2014). Furthermore, higher levels of factual political knowledge increase the odds of choosing the most ideologically proximate party (Jessee 2010;Milic 2012). There are also indications that higher levels of knowledge strengthen the role of retrospective evaluations on the vote (de Vries and Giger 2014). Such findings suggest that higher levels of political knowledge increase the likelihood that voters will "make political decisions . . . based on rationally considered principles reflecting their own selfinterest and the common good" (Lau and Redlawsk 1997, 585). Several studies indeed confirm a strong and positive effect of political knowledge on correct voting (Dusso 2015;Lau, Andersen, and Redlawsk 2008).
Hence, if women generally possess a smaller amount of the "currency of citizenship," there is a risk that their interests will be represented less well. However, while the connection between political knowledge and voting in line with one's preferences is strong, political knowledge is only one of multiple factors that influence the likelihood of voting correctly (Lau et al. 2014). Because other determinants are important as well, the gender gap in political knowledge does not deterministically lead to lower levels of correct voting among women. The analogy with work on electoral participation is useful in this regard. Even though political interest and political knowledge are strong predictors of turnout (Smets and van Ham 2013), women's lower level of political interest and political knowledge does not imply a gender gap in electoral turnout (Inglehart and Norris 2003). To assess the implications of the gender gap in knowledge for women's political representation, a more direct test of gender differences in "correct" voting is needed.
We perform such a test and investigate gender differences in correct voting-operationalized as voting for the most proximate party and as a weighted measure that takes into account different determinants of the vote choice (ideology, retrospective evaluations, and partisanship). Our analyses draw on the data from the Comparative Study of Electoral Systems (CSES) project. While there is a substantial gender gap in political knowledge, we find that women are about as likely as men to vote for the most proximate party, or the "correct" party.

DATA AND MEASURES OF CORRECT VOTING
For analyzing gender differences in correct voting, we make use of the first four CSES modules, covering 134 elections . As the data set includes measures of political knowledge, it also allows validating whether there is a gender gap in political knowledge.
As a first indicator of correct voting, we focus on ideological proximity, in terms of Left-Right self-positioning. Respondents' Left-Right self-placement and their assessments of the ideological position of political parties in their country are included in most CSES surveys. We match respondents' ideological self-placement to the ideological position of parties. Parties' positions are estimated as the mean ideological position assigned to the parties by higher-educated respondents in the samples. Such a measure correlates well with other indicators of parties' positions, such as expert placements (Dalton and McAllister 2015). Across all elections, 31% of the respondents vote for the most proximate party (app. 1; apps. 1-9 are available online).
However, ideological proximity is only one indicator that can guide the vote choice of an informed voter. That is, ra-tional voters can increase their voting utility by means of other considerations as well, such as retrospective evaluations. We therefore complement the analysis of proximity voting with an investigation of gender differences in correct votingrelying on previous operationalizations of correct voting as the vote cast when "fully informed" (Lau et al. 2008). For these analyses, we rely on the coding from Lau et al. (2014). We present the results for their preferred measure of correct voting, but our results are robust to using alternative measures of correct voting (see app. 5). Across the election samples for which we have information, 71% of the respondents vote correctly (see app. 2).

RESULTS
We first validate whether male and female respondents have different levels of political knowledge. These analyses (app. 3) show a substantial gender gap in political knowledge. On average, when controlling for differences in survey mode and question format, women's level of political knowledge is about 10% lower than men's.
Subsequently, we proceed with an analysis of gender differences in proximity voting and correct voting. In a first step, we analyze voters' likelihood of choosing the ideologically most proximate party. Figure 1 presents the results of a series of separate logistic regression models, where we explain voting for the most proximate party by means of respondents' gender. We present the average marginal effect of being a woman (compared to a man) in each of the 134 election samples in the data set. Of these 134 estimates, 83 (62%) are in the expected negative direction. Only 11 out of these negative average marginal effects, however, are significant at conventional levels. Furthermore, the estimated effect of being female is positive in 51 samples, and five coefficients are significantly positive. In the language of meta-analysis, on a total of 134 tests, there were 11 successes, 118 failures, and 5 anomalies, implying a success rate of 8.2% only. In summary, our analyses offer very little evidence for the expectation that women are less likely to vote for the most proximate party.
The results in figure 1 show that women are about as likely as men to correctly identify the party that is ideologically closest. In a second step, we investigate gender differences in a more comprehensive measure of correct voting (Lau et al. 2008(Lau et al. , 2014. In figure 2 we present the average marginal effect of being female on voting correctly. These results offer even less evidence for the expectation that women cast lower-quality votes. The estimated effect of being female is negative in only 29 out of 58 samples (50%). Three of these effects are significant, while the estimated effect of being female is positive and significant in six surveys. This amounts to a success rate of 5.2% only.
The CSES data include information on a varied set of democracies for over two decades. The data are therefore well suited for getting a sense of general patterns in voting behavior but show no evidence that women cast votes that are of lower quality compared to those of men. The results presented here come from bivariate logistic regression models, in which gender is the only independent variable. However, adding more control variables does not alter our con-clusions and only further weakens the initial effect of gender (app. 4).

DISCUSSION
Much has been written about the apparently persistent gender gap in political knowledge. In the CSES data as well, we find that the knowledge score of women is significantly lower than the score of men. There is concern about this gender gap because it is assumed that women's lower levels of knowledge translate into suboptimal vote choices.  Performing a direct test of gender differences in proximity voting and correct voting, our analyses offer very little evidence of a disadvantage for women. On average, women are about as likely as men to vote for the most proximate party, or to vote correctly.
This apparent paradox has a number of possible explanations. First, perhaps women use their resources more efficiently than men. We pursue this possibility in appendix 6. We first focus on a limited number of important predictors of proximity voting and correct voting (summarized in app. 7). Subsequently we interact each of these predictors with gender, directly testing whether the effect of these predictors is different among women. The results in appendix 6 show that none of these interactive effects is significant, implying that we do not find evidence of women compensating for their lower level of political knowledge by means of a reliance on other factors.
Second, we considered the possibility that country and institutional contexts offer clues about what aids or hinders women's correct voting. From previous work in the field of gender and politics, we know that women's political attitudes are systematically correlated with contextual-level variables such as women's descriptive representation (Campbell and Wolbrecht 2006;Fraile and Gomez 2017). There is also evidence of long-term socialization effects on women's political attitudes and behaviors, with studies indicating that the timing of women's suffrage in a country correlates with the size of the gender gap in participation (Beauregard 2018). In addition, proportional electoral rules have been found to reduce the gender gap in turnout (Kittilson and Schwindt-Bayer 2012). In appendix 8, we report the results of additional analyses in which we seek to explain country-level difference in the gender gap in correct voting. Inspired by the literature, we evaluate the role of women's descriptive representation, the timing of women's suffrage, and electoral rules. The results (app. 8) show that none of these contextual-level variables is significantly associated with gender differences in correct voting.
Third, the knowledge items that are traditionally included in survey research might be less relevant for making informed vote choices. The knowledge questions in the CSES surveys focused mainly on knowledge of national institutions and international relations, but it is known that gender gaps are smaller or even reversed when the focus is on local politics (Coffé 2013) or practical political knowledge (Stolle and Gidengil 2010). Distinguishing between institutional knowledge, knowledge about politicians, knowledge of international relations, and policy knowledge in the questionnaires, we find that even when focusing on policy knowledge, a significant gender gap remains (app. 3). All these items, however, still cover "factual political knowledge." For voting for the ideologically most proximate party, or for voting correctly, it might be more important that citizens have a certain level of "ideological understanding" (Converse 1964;Jennings and Niemi 1981). We have pursued this possibility by creating a measure of respondents' ideological understanding-capturing the similarities between respondents' and experts' Left-Right placement of the three top parties in each election. From our supplemental analyses (app. 9), we draw two important conclusions. First, we find that in contrast to what holds for factual political knowledge, women have about the same level of ideological understanding as men. Second, when comparing the explanatory power of factual political knowledge and ideological understanding, the latter is much more important in explaining proximity voting and correct voting. As a result, the fact that women tend to get lower scores on the factual political knowledge questions, that are traditionally included in survey research, does not seem to be a disadvantage when voting. The key to choosing a proximate or correct party is ideological understanding, and here women do about as well as men.
Our results are important, for two main reasons. First, the fact that women are about as likely as men to vote correctly offers an important nuance to debates on the gender gap in factual political knowledge. Even though women have lower levels of factual political knowledge, this does not affect the quality of their vote choices-and therefore the quality of women's representation. We find evidence that women manage to do well in terms of correct voting because they have levels of ideological understanding that are similar to those of men. Second, while we do not deny that factual political knowledge is an important resource for citizens, our findings suggest that its importance should not be exaggerated. Lower levels of factual knowledge do not deterministically reduce the quality of voters' choices, as other resources and shortcuts can help voters to vote correctly. In this regard, voters' ideological understanding seems a particularly valuable resource.