The Way She Moves: Political Repositioning and Gender Stereotypes

Abstract

Research on policy shifts has found that repositioning can be costly as it affects candidates’ perceived honesty, reliability, and competence. It remains unclear, however, whether a politician’s gender affects perceptions of repositioning. Research on gender stereotypes has found that while male politicians are viewed as more competent, decisive, and displaying strong leadership, female politicians are believed to be more honest. Applying expectancy-violation theory, I test the hypothesis that the reputational cost of repositioning is more pronounced for female politicians in a preregistered survey experiment fielded on a large convenience sample in Flanders, Belgium (n = 6,957). I find that while frequent repositioning is evaluated negatively, the negative effect of repositioning is not more pronounced for female candidates than for male candidates on most outcome measures.

Introduction

Research on party policy shifts and political repositioning has found that it can be costly for politicians to change their policy stances (Andreottola 2021; Doherty et al. 2016; Sigelman and Sigelman 1986; Tomz and Van Houweling 2016). Candidates who change their positions are seen as less honest, less reliable, and less competent (McCaul et al. 1995). Studies in social psychology have called this the “waffle-phenomenon”: when a person changes their mind – i.e., when they “waffle” – they are perceived by others to be weaker, less honest, and less sincere (Allgeier et al. 1979).

It remains unclear, however, how the politician’s gender affects citizens’ appraisal of repositioning. In this paper, I test the hypothesis that women are punished more strongly for frequent repositioning. Given limited time and resources, citizens rely on heuristic cues for opinion formation about political candidates, such as the candidates’ partisanship and gender. Research on gender stereotypes in politics has found that voters ascribe different personality and competence characteristics to female and male politicians (Dolan and Lynch 2014). While male politicians are viewed as more decisive, and displaying strong leadership, female politicians are believed to be more honest, expressive, and compassionate (Alexander and Andersen 1993; Huddy and Terkildsen 1993). These gender-trait stereotypes have been shown to affect candidate perceptions (Huddy and Terkildsen 1993). Personality traits perceived as “male” are moreover considered by citizens to be particularly important for political office (Lawless 2004).

Given that political repositioning has been found to impact precisely those personality characteristics affected by gender stereotypes, it is plausible that repositioning activates gender stereotypes about politics (Bauer 2015a; Kunda and Spencer 2003). It is worth examining whether policy change by female politicians is evaluated differently than policy change by male politicians. If repositioning is associated with dishonesty, is a female’s politician violation of the gender stereotype that women are more honest particularly harmful? (Cassese and Holman 2018). This study sets out the answer to this question.

What is more, a secondary aim of this study is to examine whether repositioning by politicians is also perceived as a negatively valenced phenomenon in a European context. The majority of studies examining the reputational implications of repositioning have focused on the United States (but see Christensen and Fernandez-Vazquez 2022; Lupu 2014). Multiparty systems are marked by more inter-party compromise and coalition governments requiring a higher degree of policy flexibility of politicians. A number of recent studies also show that voters in European multiparty democracies tend to punish parties that have been willing to compromise in government coalitions (Fortunato and Adams 2015; Fortunato 2019; Klüver and Spoon 2020). This suggests that citizens in multiparty democracies also perceive party policy change negatively. Conclusive evidence is lacking, however. After all, although politicians who change their positions can be dismissed as insincere, a willingness to change positions can also be welcomed as a sign of flexibility and adaptiveness.

To this end, I conduct a preregistered survey experiment in Flanders, Belgium in which I manipulate both the gender of the political candidate and whether they regularly change their position, or not – without priming respondents with the candidates' party affiliation or policy positions. ¹ To explore whether gender stereotypes are mechanisms through which respondents evaluate the repositioning of male and female politicians, I explore whether perceived personality characteristics of honesty, decisiveness, and competence drive these findings.

As one of the first experimental studies on candidate repositioning in the European context, I find that candidates who frequently change their policy positions are evaluated less positively than those who reposition only infrequently. Hence, a distaste for repositioning is clearly not something limited to citizens in majoritarian systems, such as the US. However, this negative effect of repositioning is overall not more pronounced for female candidates than for male candidates. What is more, no differences are found in the perceptions of honesty, decisiveness, and competence for male and female repositioning candidates. This suggests that, at least in the case of Flanders, neither candidates’ gender nor cues about frequent (or infrequent) repositioning, as such, are sufficient to activate gender stereotypes in candidate evaluation.

In what follows, I will present the hypotheses of the study, followed by a discussion of the experimental design. Subsequently, the findings are presented after which I close with a conclusion.

Expectations

I expect that female politicians are punished more severely for frequent repositioning than male politicians. Previous research has shown that repositioning affects perceptions’ of candidate traits, such as honesty. Repositioning is likely to be perceived to go against the positively-valanced feminine stereotype of honesty and reliability (McCaul et al. 1995). Kunda and Spencer (2003) posit that stereotypes need activation before they can be applied. Bauer (2015a) finds that gender stereotypes only play a role in the evaluation of political candidates once they have been activated. While stereotype activation can result from cues of gendered traits, it is also possible that situational cues and gendered behavioral expectations can activate cues Bauer (2015a, p. 705). As such, it is plausible that repositioning activates gendered stereotypes about honesty, precisely because repositioning is associated with honesty, and honesty with feminine traits (see Kunda and Thagard 1996). Expectancy-violation theory, in turn, suggests that candidates are evaluated more negatively when they violate stereotypical assumptions about their “group” (Brown et al. 2018; Cassese and Holman 2018). Therefore, I expect that the reputational cost of repositioning is more negative for female politicians than for male politicians.

These reflections lead me to posit the following two preregistered hypotheses:

Hypothesis 1: A candidate who changes positions more often is evaluated more negatively than a candidate who changes positions less often.

Hypothesis 2: The negative effect of repositioning (vs. remaining steadfast) is stronger for female politicians than for male politicians.

An alternative expectation puts the hypothesized relationship on its head. While the preregistered hypotheses posit that expectancy-violation theory points to harsher evaluation of repositioning female candidates, repositioning by male politicians could also be regarded as a violation of male stereotypes concerning decisiveness. This could mean that male politicians can also suffer from stereotype-violation, albeit for a different associated trait. Given the prevalence of stereotype-based attacks against female politicians (Cassese and Holman 2018) and pervasive sexism (Cassese and Holman 2019; Mansell and Gatto 2022), however, I expect that stereotype violation has a more pronounced negative effects on evaluations of female than male representatives.

To explore whether, and how, gender stereotypes affect candidate evaluations, I also examine whether the interaction effect between the gender of the politician and repositioning is determined by the personality traits perceptions of honesty, decisiveness, and competence.

Research design

To test my expectations, I conducted a preregistered survey experiment in Flanders, Belgium in December 2022. ²

Participants were recruited from a Flemish panel of users of the voting advice application “Election Compass,” and were aged 18 or older. The panel is constituted on opt-in basis with Voting Advice Application (VAA) users willing to be recontacted. While non-probability samples can produce biased samples, it has been found that there are no marked differences between effects found in survey experiments administered in convenience samples and in population samples (Krupnikov et al. 2021).

The preregistration stated that 4000 respondents would be recruited, based on a power analysis conducted prior to the experiment (see Appendix A.2 in Supplementary material). To attenuate selection bias, the preregistration noted that quotas for age, gender, and education would be implemented. Data collection deviated from the preregistration, however, as quotas for age, gender, and education could not all be filled. Instead, the collected sample includes 6957 respondents and contains an over-representation of males and higher-educated individuals. In the main analyses, the sample average treatment effects are estimated with this sample. To address the issue of representativeness, post-stratification weights are used for robustness purposes (Franco et al. 2017; Miratrix et al. 2018, see). More information on the distribution of respondents in the weighted and unweighted samples as well as the post-stratification procedure can be found in Appendix A.3 in Supplementary Material.

Balance checks ascertain whether certain demographic groups are over-represented in one of the experimental treatment groups. While unbalanced covariates were found, analyses including unbalanced covariates did not yield different results (see Appendix A.7 in Supplementary Material).

The case: Flanders, Belgium

The experimental study is conducted in Flanders, the Dutch-speaking region of Belgium. The choice for Flanders was based on a “least-likely” case selection strategy. A “least-likely” case is a type of so-called “crucial cases,” used for hypothesis testing (Gerring 2006). “Least-likely” cases are useful if one finds an effect in a context that is geared toward not finding effect, it provides strong support for the theory. In other words, it presents a strong (i.e., “most-difficult”) test of the theory.

Flanders is a suitable “least likely case” for two reasons. First, stringent gender quotas, introduced in Belgium in 1994 and 2002, radically affected the composition of the various parliaments in Belgium and its three regions, Brussels, Flanders, and Wallonia (Vandeleene 2014). In 2019, 46 percent of the Members of Parliament in Flanders were female, compared to an EU average of 33 percent). Second, recent work has found that gender stereotypes are not strongly prevalent in Flanders (Devroe and Wauters 2018; Devroe 2021). In contrast to cases like the United States, Devroe (2021) found that Flemish voters hold stereotyped views only to a limited extent.

As such, while the “least likely” character of the Flemish case provides for a “hard” test of theory, the implication is that if an effect is found, this is likely broadly generalizable to context in which gender stereotypes are more prevalent, such as the US (Dolan 2010).

Experimental design

The survey is conducted online in Dutch, the official language of Flanders. After reading an informed consent message, participants were forwarded to the main questionnaire upon acceptance. ³ Pretreatment, respondents are asked about how important they find the three issues that appear in the treatments.

Subsequently, respondents were presented with two attention checks after which the experimental treatment was applied. The attention checks are modeled after Berinsky et al. (2014) and adapted to the Flemish context. If a respondent fails the first attention check, a warning appears and the respondent can only continue with the survey once the respondent has correctly answered the question correctly. In the second attention check, respondents could continue the survey when they do not comply. In total, 85.93 percent of the respondents successfully completed the second attention check, while 14.07 percent failed it. Analyses in the appendix show that controlling for attentiveness as well as split-sample analysis of attentive and non-attentive respondents does not affect the substantive findings (see Table A.5 and Figure A.6 in Supplementary Material). Respondents who failed the attention checks were not dropped from the analysis as this could induce undue bias because inattentiveness is structured around respondents’ education levels, age, and gender (Berinsky et al. 2014).

The experiment is a 2 × 2 factorial experiment with two conditions: position change (frequent repositioning vs. infrequent repositioning) and candidate gender (male vs. female). This results in four different conditions. Of theoretical interest is thus the comparison between candidates who change positions frequently and those who change infrequently, as well as between male and female candidates. This means that the experimental test is comparative between these conditions. As such, the design does not include a “pure” control condition.

Respondents were presented with a fictitious candidate profile in an unspecified news source. The treatments are simulated “report cards” of politicians performance. In Flanders, the newspaper Het Nieuwsblad published similar articles in the run-up to the 2019 elections. ⁴ The treatments are headed by some general candidate information (city, place on the list, region) to increase experimental realism.

The profile signals the gender of the candidate by means of a fictitious photo and by means of the first name (“Koen” [male] or “Eva” [female]). The photo is generated with artificial intelligence from the platform Generated Photos. Using machine learning trained on 30,000 real photos taken in a controlled environment, Generated Photos produces faces of non-existent people. The photos used in the experiment are constant over number of photo-based and facial characteristics such as head pose, age, emotion, skin tone, hair color, and ethnic traits. Only the gender of photo is varied. ⁵

Position change is operationalized as relative repositioning. The repositioning scenario notes that the candidate has changed their position in the last years more frequently on three issues than other representatives without specifying the direction of the policy change. This approach allows us to specifically examine the valence implications of repositioning. Previous research has shown that policy direction may trump valence assessments of repositioning (Doherty et al. 2016; Hoffman and Carver 1984).

The text of the repositioning scenarios notes that the candidate particularly changed positions on three issues: childcare, climate policy, and immigration policy. The treatments in which the candidate does not frequently change position note that the candidates have changed their positions less than other representatives, being particularly steadfast on the same three policy issues. I specify specific issues for experimental realism. Given the existence of gender stereotypes on policy competence, the policy domains address a variety of issues with different associated stereotypes. Climate change policy is a domain in which no gendered competence stereotypes have been previously found (Devroe 2021; Swim and Geiger 2018). Immigration policy is seen as a stereotypically masculine policy domain (Dolan 2014; Dolan and Lynch 2016). Childcare policy, by contrast, is often stereotyped as a feminine policy domain (Dolan 2014; Devroe and Wauters 2018). What is more, the selected issues cover the relevant issue dimensions in Flemish politics related to socioeconomic policy, immigration, and the environment (Walgrave et al. 2012). The full stimulus text in English is presented in Figure 1. Figure 2 shows a translated example of a treatment.

Figure 1. Stimulus text translated to English.

Note: Treatment conditions shown in brackets with the different conditions separated with a forward slash.

Figure 2. Example of a treatment (translated).

After receiving the treatment, a number of dependent variables are measured. The main variables on which the preregistered hypotheses are tested are the overall evaluation of the candidate, the perceived trustworthiness of the candidate, and respondents’ vote intention for the candidate. Candidate evaluation and trustworthiness are measured on an 11-point scale from 0 to 10. Voten intention is measured as a dichotomous variable: Rather yes or Rather no. In order to explore the mechanisms of gendered reputational effects of repositioning, I additionally measure respondents’ assessment of the candidate’s honesty, decisiveness, and competence on an 11-point scale from 0 to 10. The vote intention variable is arguably a more stringent test of the hypotheses.

After the dependent variables, I ascertain whether the experimental treatment has worked using two manipulation checks. The first manipulation check asks the respondent whether the candidate in the scenario changed their position, or not. The second manipulation check asks which of which gender the political candidate was. Analyses in the appendix show that the findings are driven by respondents who succeeded in the manipulation check. Given that approximately 90 percent of the respondents succeeded in the manipulation check, this did not affect the findings in the main model.

The hypotheses will be tested using Ordinary Least Squares (OLS) regression analysis, I will use an $\alpha $ -value of 0.05 as the threshold for statistical significance.

Results

Is there a repositioning penalty on political candidates, and is it more pronounced for female politicians than male politicians? This section discusses the evidence from the survey experiment as well as several robustness tests.

Table 1 shows the treatment effects of the candidate’s gender and their repositioning using OLS regression analysis for the dependent variables evaluation trust, and vote intention. ⁶ The first two columns show that respondents evaluate parliamentary candidates who reposition more frequently than others much less positively. Candidates who reposition frequently are evaluated and trusted less by a whopping 10.22 percent and 11.12 percent, respectively (p value = 0.000). The effect on vote intention is smaller, with candidates who frequently change positions receiving ca. 2 percent fewer votes. This allows us to accept Hypothesis 1, which stipulates that candidates who reposition more than others bear higher reputational costs. Figure A.2 in Supplementary Material shows that this conclusion also applies to the other measured dependent variables (honesty, decisiveness, and competence), despite substantial variation between the different dependent variables.

Table 1. The effect of candidate gender and repositioning on candidate reputation

Standard errors in parentheses.

* $p \lt 0.05$ , ** $p \lt 0.01$ , *** $p \lt 0.001$ .

What is more, the first two columns of Table 1 suggest that – contrary to assumption of widespread sexism in politics – respondents generally evaluate the female politician more positively (p value = 0.029 [evaluation] and 0.000 [trust]). This, however, seems to be driven by the sample composition. Table A.3 in Supplementary Material shows that this effect is not present in a replication of the analysis with post-stratification weights.

The third and fourth columns of Table 1 test Hypothesis 2, which expected that respondents evaluate repositioning more negatively for female than male politicians. As the statically insignificant interaction effect between candidate gender and repositioning shows, this hypothesis has to be rejected (p value = 0.318 [evaluation], 0.365 [trust], and 0.185 [vote intention]). Although the coefficient of the interaction is negative as hypothesized, the evidence suggests that respondents do not evaluate repositioning differently for female and male candidates. Figure 3 shows that there are no significant effects for the interaction effect between candidate gender and repositioning for honesty (p value = 0.220), decisiveness (p value = 0.05), and competence (p value = 0.272). ⁷ I moreover do not find statistically significant differences between the various dependent variables. This is suggestive evidence that gender stereotypes were not a driving factor in candidate evaluation.

Figure 3. Coefficient plot for all six dependent variables.

These results are robust to a number of alternative model specifications. Balance tests show that treatment assignment was not completely random for a set of variables ⁸ . Figure A.7 in Supplementary Material shows that results are substantively the same when controlling for these covariates. Table A.5 in Supplementary Material shows moreover that controlling for attentiveness does not affect the results and Figure A.5 in Supplementary Material shows that the null finding is most pronounced in the attentive sample. Subsequently, I examine whether meeting the manipulation check, or not, affects the results. Table A.6 in Supplementary Material demonstrates that controlling for succeeding or failing the manipulation check does not alter the substantive findings. Figure A.6 in Supplementary Material shows that the main results are driven by respondents who succeeded in the manipulation check.

Since the raw sample strays far from representativeness, additional analyses using post-stratification weights on a reduced sample are conducted. Tables A.3 and A.4 in Supplementary Material show that the results as well as the size of the coefficients remain substantively similar when post-stratification weights are applied. Note also that the effect size of the negative effect of repositioning is comparable in the analyses with post-stratification weights.

Finally, I probe whether respondents’ gender and perceived issue importance of the three issues in the treatment affect the results. Figure A.3 in Supplementary Material shows a split-sample analysis for female and male respondents. The results show broadly comparable results across male and female respondents. Figure A.4 in Supplementary Material shows that the results are substantively similar for respondents who believe the three issues (childcare policy, climate policy, and immigration policy) to be less and more important. ⁹

All in all, the results point to a robust negative effect of repositioning, which is not moderated by the candidate’s gender.

Conclusion

This study examined the reputational effects of repositioning in the case of Flanders, Belgium, and whether these effects differ between female and male politicians. Repositioning has been shown to negatively affect citizens' evaluations of political candidates in the United States. To my knowledge, it presents the first study into the relationship between gender stereotypes and appraisals for repositioning candidates. What is more, this is one of the first studies experimentally examining the effect of policy change in the European context (see also Christensen and Fernandez-Vazquez 2022; Nasr and Hoes 2023).

The hypothesis was posited that repositioning negatively affects the valence image of a political candidate. In addition, the hypothesis formulated the expectation that a female politician would suffer from greater reputational costs of repositioning than her male counterpart. This hypothesis was rooted in the idea that repositioning would violate the widely held gender stereotype that women are more “honest” than men (Alexander and Andersen 1993; Dolan 2014; Huddy and Terkildsen 1993).

The results show that repositioning is, as hypothesized, very costly for politicians. A political candidate who changes their position on policy regarding childcare, climate change, and immigration more frequently than their colleagues is evaluated less positively, is considered to be less trustworthy, receives less votes, and is perceived as less honest, less decisive, and less competent. The sizeable effect size of repositioning is in line with previous studies (Christensen and Fernandez-Vazquez 2022). As hypothesized, respondents in Flanders perceive policy change to be a negatively valenced phenomenon. To be sure, since the candidate’s party affiliation nor the policy direction of the policy change was cued, respondents were arguably particularly likely to cast a valence-based judgment on repositioning. This finding suggests that also in multiparty democracies, which are characterized by consensus-building and compromise (Lijphart 1999), politicians changing their positions are perceived less positively. This implies that voters are not necessarily more forgiving for candidates who operate in political contexts in which parties frequently bargain and compromise in parliament and in governing coalitions. This bolsters recent findings that voters in multiparty democracies tend to punish parties that compromised on their electoral pledges (Fortunato 2019; Klüver and Spoon 2020).

That said, the Belgian political system is strongly party-centered, being marked by flexible-list proportional representation and strong party discipline. It is possible that the party-centered nature of Flemish politics means that real-world Flemish voters would be less aware of the repositioning of individual candidates, making valence-based evaluations less likely. In contrast, it is plausible that the repositioning penalty found is even stronger on more candidate-centered political systems.

The hypothesized negative effect of repositioning is found to be not more pronounced for female candidates than male candidates. Overall, this is good news for representative democracy as it suggests that female politicians’ policy track records are evaluated along the same yardstick as those of male politicians. Voters respond to the substantive informational cues about the candidate (i.e., their tendency to change positions) rather than to candidates’ gender. Since feminine stereotyping can affect citizens’ voting decisions (Bauer 2015b), this is an important finding. It echoes previous findings that gender stereotypes are not always decisive for candidate evaluations or voting decisions (Dolan 2014; Devroe 2021).

Crucially, these findings do not mean that gender stereotypes are not at all relevant (anymore) in a context like Flanders. As Bauer (2015a) argued, voters’ reliance on gender stereotypes depends on whether they are activated or not (Kunda and Thagard 1996; Kunda and Spencer 2003, see also). It is plausible that gender stereotypes continue to play a role when they have been explicitly activated during political campaigns. The experimental stimulus cued respondents with information on whether the candidate changed its positions frequently or not: a characteristic that is associated with personality traits such as a lack of honesty and reliability. These traits are in turn often associated with gendered personality expectations. While the research on stereotype activation notes that such behavioral cues could be sufficient to trigger stereotypical thinking (Kunda and Thagard 1996), the results suggest that these cues were not sufficient to activate gender stereotypes. Future studies should examine whether evaluations of repositioning candidates are gendered when gender stereotypes related to “honesty”, “reliability”, but also “decisiveness” are explicitly prompted.

Flanders was chosen as a plausible “least likely” case, i.e., a context in which the political context made the plausibility of gender stereotypes in evaluations of repositioning unlikely. In part due to the presence of gender quota, Flanders has particularly low levels of female under-representation (Vandeleene 2014). In line with the findings of this study, research by Devroe has previously documented that gender stereotypes in politics are less pronounced in Flanders than elsewhere (Devroe and Wauters 2018; Devroe 2021). The aim of choosing a “least likely” case was that if I were to find an effect of gender stereotypes on reactions to repositioning in a context that is unfavorable to finding such an effect, this effect likely held across contexts marked by higher degrees of gender stereotypes. Hence, it is possible that gender stereotypes do affect responses to candidate repositioning in context in which such stereotypes are more pronounced, such as in the US (Dolan 2010). Future research should therefore address whether gender stereotypes affect appraisals of repositioning in contexts with more pronounced gender gaps in politics. On the flip side, it is likely that these findings do hold in contexts with similar levels of gender stereotyping as Flanders. As far as I am aware, comparative studies on the prevalence of gender stereotypes across countries and political contexts are scarce, which makes it hard to gauge in which countries citizens are prone to resort to gender stereotyping in their candidate evaluations. We therefore need more research into the prevalence of gender stereotypes in politics across countries and how they affect political beliefs and behavior.

This study examined frequent vs. infrequent repositioning across a bundle of three issues. Yet, previous research has found that citizens ascribe different issue competencies to male and female politicians as issues themselves can be perceived as “feminine” and “masculine” domains (Dolan 2014; Huddy and Terkildsen 1993). It is worth examining therefore whether evaluations of repositioning differ for more “feminine” and “masculine” policy areas, and whether female politicians are punished more harshly for changing positions on stereotypically “feminine” issues.

All in all, this study finds that, in the case of Flanders, political candidates who reposition frequently face significant reputational penalties, and that this effect is not gendered. This suggests that citizens perceive repositioning – at face value – as a negative phenomenon, allegations of which politicians are best advised to avoid.