The Gendered Cost of Politics

Abstract

We investigate how sexism and harassment affect political candidates’ preferences for political positions by deploying a conjoint experiment among political candidates in the 2021 Danish local elections. We find that, compared to men candidates, women candidates experience far more sexism and harassment, and assess their risk of victimization as being far higher. Correspondingly, the conjoint experiment reveals that women candidates state stronger preferences for equal working environments in politics than men, while holding similar preferences for formal working conditions like political positions, remuneration, and workload. Substantively, women’s willingness to lower their remuneration and increase their workload to avoid sexism in politics is more than double the size of men’s willingness. Our approach provides us with highly accurate descriptions of candidates’ preferences for political jobs, which are often assumed rather than measured directly. This lets us quantify the magnitude of an important working condition in politics with significant repercussions for women.

Introduction

Sexism and harassment in politics are increasingly recognized to cause gender inequality in political representation. The #MeToo movement has highlighted the widespread nature of this issue, affecting politicians at both national and local levels (Dhrodia 2018; Jonsdottir et al. 2022). Women, especially those in powerful positions, are more often exposed to sexism and harassment (Håkansson 2021; Thomas et al. 2019; Collignon and Rüdig 2021; Kosiara-Pedersen 2023). Some argue that sexism and harassment are instruments that deliberately keep women out of politics (Biroli 2018). Research indicates that these behaviours may also have negative effects on political representation, as victims tend to change their behaviour in public spaces, and some even consider voluntary retirement (Herrick and Thomas, 2002 b; Erikson, Håkansson and Josefsson 2021; Bardall, Bjarnegård and Piscopo 2020; Krook 2020; Herrick and Franklin 2019). However, research also suggests that the marginalization of women can make women’s candidacy emerge because staying out of politics becomes perceived as too politically costly (Clayton, O’Brien and Piscopo 2023; Dittmar 2020). Hence, it is important to consider the negative emotions that result from sexism and harassment. Victims of these actions may still choose to run for office, but their preferences for political positions may differ from those who have not experienced such mistreatment. Therefore, we ask how preferences for political positions vary in the candidate pool between women and men, and victims and non-victims of sexism and harassment.

We argue that the working environment in politics is shaped by gendered norms and practices, resulting in a fundamentally higher incidence and risk of sexism and harassment for women, who consequently have stronger preferences against working environments with previous incidents of sexism. We test this argument on the entire pool of political candidates for the 2021 Danish local election using a survey with an embedded conjoint experiment fielded just after the election. ¹

Our conjoint experiment presents hypothetical political positions to candidates running for the local council, while randomly changing the content of four attributes of the positions: 1) level of influence, 2) remuneration, 3) workload, and 4) working environment. By randomly varying the attributes of the positions, we validly estimate the candidates’ preferences for different positions based on the choices they make. This approach provides an innovative way of studying the preferences of political candidates for political positions. By studying people actually competing for political positions, we study the relevant population to these hypothetical positions. Our respondents are therefore highly familiar with the job traits portrayed in the hypothetical positions, which limits the potential risk of hypothetical bias that conjoint experiments can suffer from (Folke and Rickne 2022; Mas and Pallais 2020).

Our experimental results show no gender gaps in candidates’ preferences for higher political positions and higher levels of remuneration or workload, but when it comes to the working environment, women – compared to men – have a stronger preference against holding positions in a council where several have experienced sexism and harassment, and a larger preference for positions in a council where the members enjoy mutual respect. This is complemented descriptively, where we find that women candidates are more frequently targets of harassment and that women perceive their risk of victimization as higher than men.

To substantiate the gender gap in political candidates’ dispreference for sexism in politics, we estimate the candidates’ willingness to pay and willingness to work. While both men and women candidates are willing to be remunerated less and willing to work more to avoid sexism in politics, we find that women’s willingness is double the size of men’s.

Some candidates likely quit politics due to harassment and are, consequently, less likely to be part of our sample. We believe this makes candidates with stronger preferences against sexual harassment less present in our sample, and therefore expect our estimates to be lower bounds of the candidates’ true preferences.

We make several contributions to the existing literature. First, we contribute to the literature on sexism and harassment in politics (Clayton, O’Brien and Piscopo 2023; Dittmar 2020; Håkansson 2021) and the literature about the negative, personal consequences that disproportionately hit women in office (Fiva and King 2022; Campbell and Childs 2014; Folke and Rickne 2020b ). Second, we also offer a new way to test existing explanations rooted in women’s lower career ambitions or weaker preferences for monetary remuneration. Third, we provide the first empirical evidence for political candidates’ preferences for the political environment they work in. We thereby contribute to the literature on political selection which researches how features of political offices connect to the officeholders’ utility of doing politics (Besley 2005; Dal Bó et al. 2017; Dal Bó and Finan 2018; Gulzar 2021).

Concerning external validity, we consider Denmark a least likely case. Denmark ranks second in the EU’s gender equality index (Barbieri et al. 2020), but sexism and harassment prevail in politics (Kosiara-Pedersen 2023; Pedersen et al. 2021; The Danish Institute For Human Rights 2022). Gender gaps persist in Danish local councils (Kjøller et al. 2024), and following the 2021 election, women make up 36 per cent of councillors and only 20 per cent of mayors. As gender equality is comparatively high, the effects found in our study compose a lower bound and likely exist in places with less gender equality.

Research on gender and political positions

A persistent explanation for gender inequality in political representation has been gender gaps in political ambition, defined as a nascent interest in running for or holding elected office. Women are argued to lack political ambition, hold low political efficacy, not be interested in politics, or not be willing to run due to conflicts with family life (Fox and Lawless 2010; Bonneau and Kanthak 2020; Norris and Lovenduski 1995; Kanthak and Woon 2015; Dahl and Nyrup 2021). By contrast, the feminist institutionalist approach argues that the gender gap cannot be explained by individual-level factors alone (Allen and Cutts 2020; Pruysers, Thomas and Blais 2020). Factors at the individual level interact with institutional and contextual factors, which shape candidate emergence and selection in gendered ways (Bjarnegård and Kenny 2016; Kenny and Verge 2016; Josefsson 2020; Erikson and Josefsson 2019). Among these are: persistent gender roles (Eagly and Karau 2002), conscious and unconscious bias against women in leadership positions (Erikson and Josefsson 2023), men-dominated cultures and stereotyping (Schneider et al. 2016), gendered networks in politics (Lovenduski and Norris 2003; Lovenduski 2016), women’s own perceptions of their individual abilities (Dahl and Nyrup 2021; Fox and Lawless 2011; Fox and Lawless 2010), unequal access to key resources and unequal family obligations (Fiva and King 2022; Folke and Rickne 2020 b) and the lack of women role models (Wolbrecht and Campbell 2017; Campbell and Wolbrecht 2006; Gilardi 2015; Beaman et al. 2012). Thus, the complex calculus of preferences for political positions among men and women candidates is shaped by many (Muller and Tömmel 2022). Recent research adds to this by investigating how sexism and harassment are linked to political positions.

Sexism, harassment, and political positions

Sexual harassment can have severe negative consequences on mental and physical health, even if it occurs infrequently or involves less severe behaviours (Leskinen, Cortina and Kabat 2011; Sojo, Wood and Genat 2016). According to Schneider, Swan and Fitzgerald (1997), sexual harassment is more severe than other aggressive peer behaviours at work, even at lower frequencies of sexual harassment. Due to these detrimental effects, sexism and harassment can be considered workplace hazards (Fitzgerald and Cortina 2018; Folke and Rickne 2022).

The literature on violence against women in politics (VAWIP) indicates that violence is often perpetrated against women exactly because they are women and in politics (Krook 2020). Indeed, some argue that gendered violence is an instrument to intentionally keep women out of politics (Krook 2017; Krook and Sanín 2020). However, such intentionality can be hard to uncover and much sexism and harassment may fall outside this category but still be detrimental to gender equality by having a gendered form such as sexualized language, imagery, or content (Bardall, Bjarnegård and Piscopo 2020), and will ultimately matter for women’s political representation.

Correnspondingly, Folke and Rickne (2022) find the cost of harassment in the labour market to be much higher for women than for men. This might certainly also be the case for women in politics who operate in a traditionally men-dominated context (Bardall 2013; Berdahl 2007; Fitzgerald et al. 1997; Håkansson 2021). For instance, women running in districts with more traditional gender norms report more violence, which men do not (Herrick and Thomas 2022 a). The literature does indeed suggest that people are rewarded for complying with gender norms, while they inflict costs on both themselves and their working environment when failing to comply (Folke and Rickne 2022; Pina, Gannon and Saunders 2009). This reward system is not limited to the occupational choice but also extends to the performance of the occupation itself. Women politicians, for example, face additional challenges due to the historical association of politics with men and the gender norms that accompany this association (Eagly and Karau 2002).

Danish local councils are well-suited for investigating how sexism and harassment influence political ambitions. Although gender equality is comparatively high, 46 per cent of women and 24 per cent of men in Danish youth parties have been exposed to sexism or sexual harassment within the last two years (Sex & Samfund og Børns Vilkår 2021), and Kosiara-Pedersen (2023) find that a third of the local councillors report to having experienced at least one type of sexual harassment. Women councillors are more often targeted for sexism and harassment than men (Pedersen et al. 2021). Adding to this, 21 per cent of local councillors who are victims of sexism or harassment, also report that this has made them consider voluntary retirement (Pedersen et al. 2021).

To sum up, since women are targeted considerably more with sexism and harassment than men, we expect that women state stronger preferences for a working environment void of sexism and harassment and that victims of sexual harassment state stronger preferences about the quality of the working environment than non-victims, and also that women more than men tend to decrease their preferences for political positions when the risk of sexism and harassment is high.

Institutional positions and political influence

A politician’s political influence depends on the institutional positions they occupy, like chairmanships. This job attribute is distinct from holding a political office, as the political influence linked to holding an institutional position is a job attribute that cannot easily (if at all) be reaped outside of the political realm. The preference for political influence can also be regarded as expressing political ambition. In our case here, this will be equivalent to intra-institutional ambition, the desire for a political position in one’s current chamber (Schlesinger 1966).

Historically, the distribution of institutional positions has been highly gendered across the globe (IPU Inter-Parliamentary Union 2021), and women are underrepresented even more in hierarchical positions. Some argue that this is due to women having lower ambitions for taking on political leadership positions (Elprana et al. 2015; Croson and Gneezy 2009), as they perceive political power as less desirable (Schneider et al. 2016). When women make it to the political top, they are evaluated in more gender-stereotyped ways (Reyes-Housholder 2020). This follows role congruity theory, which highlights how masculine leadership ideals clash with female gender roles to create role incongruity (Eagly and Karau 2002; Schneider, Bos and DiFilippo 2022). This can explain the discrimination women face when taking on publicly visible and powerful positions (Okimoto and Brescoll 2010; Håkansson 2021). It is thus well-established in existing research that women political leaders are disadvantaged in various gendered ways (Dassonneville, Quinlan and McAllister 2021). On the one hand, they are perceived as not fitting into traditional men’s occupations, and on the other hand, they violate shared beliefs concerning how women should behave when they try to adapt to masculine leadership ideals (Eagly and Karau 2002). Research further finds that men’s norms in politics are highly institutionalized and often remain intact in spite of gender-equality initiatives (Krook and Mackay 2011; Erikson and Josefsson 2023). Hence, unequal opportunities exist even at higher levels of women representation because of informal institutions’ stickiness (Franceschet and Waylen 2017). However, recent research indicates that feminine political leadership ideals can be recoded in more gender-equal settings, while masculine practises remain persistent (Erikson and Josefsson 2023). Overall, both individual-level and institutional facts indicate that women are likely to state weaker preferences for leadership positions.

Remuneration

Working in a political position is not an ordinary job, especially when it comes to the pay received. Some politicians work for higher rewards (Dahlgaard, Kjøller and Kristensen 2025), while others are likely to speak out against increasing politicians’ pay due to ideological positions and distance from institutional positions (Pedersen, Pedersen and Bhatti 2018). According to research, women councillors in Denmark have more negative attitudes toward remuneration – at least for Mayors (Pedersen, Pedersen and Bhatti 2018, 685). Additionally, voters strongly disapprove of pay raises for politicians and often sanction them accordingly if they increase their pay (Hibbing and Theiss-Morse 2002; Pedersen, Dahlgaard and Pedersen 2019; Pedersen, Pedersen and Bhatti 2018). Hence, pay for local councillors remains at a modest level. Local councillors are compensated for their loss of income while serving as councillors (Kjøller 2024). Being a local councillor is only a full-time job for mayors, while the remuneration of local councillors in Denmark varies according to the different institutional positions (Pedersen, Pedersen and Bhatti 2018).

Denmark has one of the world’s highest rates of women’s labour market participation, but labour market participation is highly gendered and a heavily institutionalized wage gap persists. The different gender roles in relation to maternity leave and parenthood imply the largest gender wage gap for mothers compared to the rest of the population (Gallen, Lesner and Vejlin 2019). Hence, women’s greater obligations in childbearing and housework may lead to women selecting jobs with certain job attributes (such as flexibility) by accepting lower wages, and men are given higher earnings to compensate for negative job attributes (such as inflexibility). (Polachek 1985; Zellner 1975). However, people’s preferences are different from the outcomes produced by gendered labor markets. A gender wage gap does not imply that women have low preferences for higher pay. But aspiration for pay is likely contingent on the pay actually received. Hence, men candidates are expected to state stronger preferences for higher remuneration than women candidates.

Working hours

Workload is an important job characteristic that can be traded off for pay and positions. The job level is positively related to work-life/family conflict, such that individuals with higher leadership positions also experience more conflict with their family life (Pedersen, Hjelmar and Bhatti 2018). This tension between work-life and family obligations has traditionally been more pertinent for women due to a higher involvement in childbearing and childcare (Zellner 1975; Polachek 1985). Correspondingly, recent research finds a child penalty for women in politics (Fiva and King 2022; Campbell and Childs 2014). Furthermore, women, on average, carry out more domestic work (Sheppard 2018). This can make women do ‘a double day’ when they also hold a full-time job. The unequal distribution of workloads between genders in the domestic realm makes it reasonable to expect women to have lower preferences for taking on higher workloads in the professional and political spheres. Even though women politicians are highly selective, they report more than men that politics take time and energy away from their family life (Pedersen et al. 2021; Bhatti et al. 2017). Therefore, we expect women to state preferences for smaller workloads than men.

Expectations

In sum, the expectations raised are as follows:

• H1: Men candidates state stronger preferences for positions with political influence than women candidates.

• H2: Men candidates state stronger preferences for higher remuneration than women candidates.

• H3: Women candidates state stronger preferences for smaller workloads in political jobs than men candidates.

• H4: Women candidates state stronger preferences about the quality of the working environment than men candidates.

• H5: Victims of sexual harassment state stronger preferences about the quality of the working environment than non-victims.

• H6: Women candidates are more hesitant than men candidates to strive for political power when harassment is prevalent in the council.

Setting, Method, and Data

Danish local councils and local elections

Danish local elections are held every fourth year and elect the local councils that govern the ninety-eight Danish municipalities. The local councils have between nine and thirty-one members (Copenhagen, 55) and each municipality constitutes an electoral district (proportional representation) with several parties running, including both national parties and local lists. The councils are headed by a mayor appointed by the local councillors by a simple majority.

The Danish municipalities hold high levels of autonomy and responsibility, as they govern the central welfare provision. Their spending covers 50 per cent of the public spending in Denmark and made up to 20 per cent of the Danish GDP in 2021. Each municipality has a committee on finance responsible for the budget and usually four to six other committees. Each committee is chaired by a councillor and the levels of influence vary markedly between members and committee chairs, and also between chairpersons, according to which committee they are chairing (Pedersen 2014). As such, there is a strong link between the institutional position and the level of political influence. The mayor is the only member of the council who is a full-time officeholder. The remaining councillors hold office as a part-time job, but about a third of the councillors report that they spend more than twenty hours a week holding office (Pedersen et al. 2021; Dahlgaard et al. 2009; Bhatti et al. 2017).

Sampling

The Danish local election of 2021 was held on November 16 and had 9,177 candidates running for office in total. Out of these, we obtained emails from 7,688, equivalent to 83.7 per cent of all candidates. Prior to the election, the Danish Institute for Human Rights manually collected a dataset from sources publicly available on the candidates, which contains the email address, party affiliation, municipality, and gender. We are thankful that they agreed to share this dataset with us. Candidates with missing email addresses tend to be from local party lists. Therefore, the municipalities where our share of emails acquired are the smallest tend to be municipalities where local lists are dominant. We believe this implies that potential respondents are among the most professional candidates who have a higher probability of winning. We have a slightly smaller share of email addresses for women than men candidates (82.2 per cent vs. 84.5 per cent, p $ \lt $ 0.01). Although statistically significant, we consider this difference of 2.3 percentage points to be substantially small and therefore unproblematic. Find further description of the candidate pool and the sample in Appendix B and Appendix C.

We invited all 7,688 candidates to participate in our study. The survey company Epinion set up the survey and the embedded experiment and sent out the invitations. The first email invitation was sent on 2 December 2021, and follow-up emails were sent on 7 December 2021 and 14 December 14, 2021. The data collection ended on December 21, 2021. Thus, our study was conducted shortly after the election and before the new councils had organized themselves into councils, but while they were discussing how the positions should be distributed. Hence, the timing of our study entails a degree of realism that is rarely found in survey experiments. ²

Our survey first measures the municipality of candidacy, marital status, year of birth, highest educational degree completed, winning seat in 2021, and scale for left-right placement from 0–10. Then follows a sequence of seven conjoint experiments. Finally, we measure self-reported workload in politics, experiences of sexism and harassment, and their self-assessed risk of harassment.

In total, 1,938 candidates completed the survey in time, which corresponds to a response rate of 26.2 per cent (women: 24.9 per cent, men: 26.8 per cent) of the invited candidates (7,688) and 22.1 per cent (women: 20.5 per cent, men: 22.9 per cent) of the total candidate pool (9,177). In Appendix B we show different sample descriptives, and in Appendix D we compare the representativeness of distributions of demographics for our sample (all, men, and women) and registry data for the entire candidate pool. Candidates with longer education and winners are more present in our sample than the candidate pool, but our sample is representative in terms of age and the share who are married, which is important as prior research suggests that young, single women are the most vulnerable in politics (Kosiara-Pedersen 2023).

Conjoint experiment

Our conjoint experiment has a paired choice design. The political candidates choose between two hypothetical political positions where four attributes are randomly varied, which have three levels each: 1) institutional position (political influence), 2) remuneration, 3) workload, and 4) a working environment. The experiment thus consists of ${3^4} = 81$ unique political positions. The attribute levels are assigned by simple randomization, and since all four attributes have three levels, each attribute level occurs by a probability of ${1 \over 3}$ . Table 1 summarizes the attributes and levels of the experiment and shows examples of how an experimental situation could look to the respondent. Since our experiment is conducted on actual political candidates, the risk of hypothetical bias is limited (Folke and Rickne 2022; Mas and Pallais 2020), as our respondents are highly familiar with the job traits portrayed. We follow Andersson and Pearson (1999) and define a good working environment as one where colleagues mutually respect one another. In Appendix A, we elaborate on the ethics of our study.

Table 1. The experiment’s attributes and levels, and examples of how two job choices are presented

Note: All levels are assigned by simple randomization, which means that every level occurs by a probability of ${1 \over 3}$ since all attributes have three levels.

To increase power, each political candidate responds to a sequence of seven situations of choice, one after the other. As the conjoint experiment presents two options for political positions and is repeated seven times for each candidate; we have an effective sample size of $2 \times 7 \times 1,938 = 27,132$ choices, which is our unit of analysis. For all respondents, the top feature is fixed to be ‘position in the local council’, while the order of the other four attributes is randomized between subjects but held constant within subjects. Thus, the order of features is random but constant for a particular respondent throughout the seven repeated experiments. This is to mitigate order effects on the one hand while easing the cognitive demand of answering the sequence of experiments on the other, and to avoid satisficing, that is, respondents finding shortcuts to reduce the cognitive costs of completing several experiments (Krosnick 1991; Bansak et al. 2018). We account for repeated experiment choices and fixed attribute order by clustering the standard errors at the subject level.

In Appendix E, we investigate the potential effects of survey fatigue and show that the candidates’ sequence of choices is not serially correlated, which suggests no carry-over effects. Further, we test for balance in A/B preferences. We cannot reject the null, and hence find no evidence that the candidates systematically chose the left/right choice more or less in the experiments(p $ \gt $ 0.1). Both tests suggest that satisficing is not a problem. In Appendix F, we examine the survey attrition and show that it is as good as even across every possible treatment condition. ³

Estimation

We test our hypotheses by estimating marginal means for men and women candidates as well as their differences in marginal means, which is the optimal strategy for estimating subgroup preferences in conjoint experiments (Leeper, Hobolt and Tilley 2020). Marginal means express candidates’ average favourability (from 0 to 1) toward political jobs with a particular attribute level, ignoring all other attributes (Leeper, Hobolt and Tilley 2020). Since half of all jobs shown to the candidates are chosen, marginal means significantly larger than 0.5 indicate favourability toward the attribute level, while marginal means smaller than 0.5 indicate disfavourability for the attribute level ignoring all other attributes of the jobs. By estimating the differences in marginal means, we test whether men and women candidates hold different favourability, that is, preferences for a certain feature of the political job.

Measuring victims of sexism and harassment

After the experiments, we ask them to assess their risk of being victims of sexual harassment in relation to their political activities now and in the future. Further, we ask the candidates if they have experienced five specific types of harassment in politics: 1) receiving degrading comments about looks or gender, 2) being called degrading names, 3) being told to have a high-pitched voice or that one talked too much, 4) being sent or shown unwanted sexual content, and 5) unwanted physical contact. The instrument has previously been applied in a Danish context and the results are robust to our findings (Kosiara-Pedersen 2023; Pedersen et al. 2021). ⁴ The validity of self-reported data on sexism and harassment should be considered. The questions give a limited number of behavioural examples, which is likely to bias incidence rates downward as respondents may have experienced other types of harassment than those listed and as they need to recall and classify their experiences (Kosiara-Pedersen 2023). Another important caveat is that men are known to underreport sexual harassment (Folke and Rickne 2022).

As per Berdahl’s (2007) definition of sex-based harassment, it can happen between individuals of any gender, be it men/women, men/men, women/women, or women/men. The crucial aspect is that the behaviour is driven by a desire to protect one’s social status, which is a motivation shared by both men and women. However, the behaviours that we evaluate in our study focus on specific actions rather than the motives behind them. Although many people engage in harassing behaviours to safeguard their social standing, this doesn’t necessarily mean that gender is the only driving factor. Other factors, such as education or financial status, can also be used by individuals to maintain their social status. Therefore, it can be challenging to determine whether certain behaviours are motivated by sexism or not. As a result, we use the term ‘sexism and harassment’ instead of Berdahl’s term ‘sex-based harassment’. While we cannot be sure of the true motives behind the reported behaviours, we can confirm that the respondents experienced them.

We deliberately placed the questions related to sexual harassment after the experiment to avoid respondents being primed before the experiment. However, this is not flawless, as they, in principle, are measured post-treatment and therefore could be products of treatments. This is a trade-off between two competing concerns; demand effects or priming of effects versus post-treatment bias, and we believe the former was the greatest concern (Klar, Leeper and Robison 2020). Respondents choose between hypothetical political positions with varying attributes and since we want to infer their preferences from these choices, we must expose them as little as possible to any attribute before the actual experiment. Otherwise, we risk affecting their choices and thereby inject bias into their estimated preferences. In addition, we ask about their experiences of sexism and harassment, which are logically pre-treatment experiences. ⁵ Altogether, we assess our decision to be the optimal solution.

Figure 1 shows the distribution of candidates, men, and women for each of the five types of harassment in total. For each category, women report yes more often than men, and ‘unwanted or degrading comments about own body or sex’ and ‘unwanted physical contact’ stand out as harassment types that women are manifold more likely to have experienced than men. Both types of harassment are examples of very direct, explicit sexual harassment.

Figure 1. Share of victims in total and across harassment types: Women candidates are targets of sexism and harassment more than men.

Note: Distributions of victims across the five questions about experiences of harassment and in total. Victims are defined as candidates who answer yes to one or more of the five questions about experiences of sexual harassment.

We define victims of harassment as candidates who answer yes to one or more of the five questions. Based on this definition, 29.4 per cent of our sample are victims and, consequently, 70.6 per cent are non-victims; 21.9 per cent of men candidates and 44.5 per cent of women candidates are victims (see the bottom-right corner panel of Figure 1). In Appendix G, we show the distribution when victims are defined by 2+ and 3+ types experienced and further. Our results for estimated differences in preferences between victims and non-victims hold under these stricter definitions of victimhood.

Candidates’ self-assessed risk of sexism and harassment

In Figure 2, we show how men, women, and all candidates assess their risk of being victims of sexism and harassment. Women candidates correctly evaluate their risk markedly higher than men candidates; for example, 70 per cent of men candidates assess a very low risk while only 30 per cent of the women candidates assess this. Meanwhile, 19 per cent of women candidates assess a high or very high risk while only 3 per cent of the men candidates assess this.

Figure 2. Women correctly assess their risk of being harassed or sexually assaulted in political work as higher than men.

Note: The self-assessed risk of being victims of sexual harassment in political work among all candidates, men, and women respectively.

Results

Above, we have shown that women are markedly more likely to be victims of harassment and sexism than men. This is consistent across all of our five items of experiences, and, altogether, women candidates are twice as likely to be victims than men. This illustrates again that sexism and harassment in politics have disproportionate gendered consequences (Figure 1). These disproportionate consequences turn into different risk assessments among candidates, where women clearly and correctly assess their risk of sexual harassment as higher than men (Figure 2).

Experimental results

In Figure 3, the left-hand side panels show men and women candidates’ marginal means for the attributes, and the right-hand side panels show the estimated differences between men and women. The figure therefore holds the answers to H1–H4. ⁶

Figure 3. Men and women hold similar preferences for political positions, remuneration, and workload, but women hold markedly stronger preferences than men for the quality of working environments in politics.

Note: The left-hand panels show men and women candidates marginal means for the attributes with 83 per cent confidence intervals for subgroup means - non-overlapping intervals indicate significant group differences (Goldstein and Healy 1995). Estimates below the dashed line (grand mean) indicate disfavourability and estimates above indicate favourability. The right-hand panels show the estimated differences in marginal means with 95 per cent confidence intervals. N = 27,132, N _men = 18,200, N _women = 8,932. Standard errors are clustered at the candidate level. The estimates and standard errors are shown in Table L.1 in Appendix L.

Figure 3 reveals that men versus women candidates state similar preferences regarding the position, remuneration, and workload, and therefore, we find no support for H1–H3. As the bottom facet of Figure 3 shows, we do find support for H4. Thus, we find that women candidates hold stronger preferences for a good working environment compared to men candidates. Both men and women favour working environments characterized by mutual respect, and disfavour working environments with sexism and harassment, as their marginal means are higher (lower) than 0.5. The differences in marginal means show that this favourability (disfavorability) is stronger for women than for men candidates. This difference in stated preferences is well-aligned with women correctly assessing their risk of being harassed as being higher (Figure 2). This is also apparent from men candidates’ marginal means being approximately similar for working environments where several experienced harassment and several experienced sexism and harassment, while women candidates clearly distinguish between these two options in the conjoint. In Appendix I, we present the results when interacting with gender and experience, which provide similar results.

Next, we turn to H5, where we test victims and non-victims’ preferences for the quality of working environments in political positions. Figure 4 shows the marginal means for the working environment attribute for victims and non-victims. Here, we see the same pattern for men and women, as victims state stronger preferences for a working environment void of sexism and harassment than non-victims - the magnitudes are smaller, however.

Figure 4. Victims hold stronger preferences for the quality of working environments than non-victims.

Note: The left-hand panels show victim and non-victim candidates’ marginal means for the three levels working environment with 83 per cent confidence intervals for subgroup means - non-overlapping intervals indicate significant group differences (Goldstein and Healy 1995). Estimates below the dashed line (grand mean) indicate disfavourability and estimates above indicate favourability. The right-hand side shows the estimated differences in marginal means with 95 per cent confidence intervals. N = 27,132, N _victims = 7,980, N _non-victims = 19,152. Standard errors are clustered at the candidate level. The estimates and standard errors are shown in Table L.2 in Appendix L.

Next, we turn to H6, which states that women candidates are more hesitant than men candidates to strive for political power when harassment is prevalent in the council. We test this by estimating the marginal means for the interaction between the political position and working environment and then estimate the difference for men and women. The marginal means for the interaction essentially express the favourability a candidate has for the combined levels of the two attributes, political position and working environment. By estimating the difference in marginal means for men and women, we therefore estimate gender differences in the favourability of, say, a position with large political power in a working environment void of sexism and harassment.

In Figure 5, the left-hand side shows men and women candidates’ estimated marginal means for combinations of political positions and working environments, with the levels of political positions at the x-axis and each level of the working environment in a facet. The right-hand panel shows the estimated difference in men’s and women’s marginal means. Note that the men and women candidates’ differences in marginal means remain almost constant within the quality of working environments – as apparent from the approximately constant difference in marginal means on the left-hand side and the overlapping confidence intervals on the right-hand side. If women candidates shy away from power to a larger degree than men candidates when harassment is prevalent, men and women’s marginal means would have an increasing distance across political positions. This is not what we see, however, and we therefore do not find any evidence in favour of H6. Rather, what we find is that women prefer holding political power in good working environments, but when harassment is prevalent, women do not shy away from power more than men.

Figure 5. Higher levels of sexism and harassment in the working environment do not increase the differences in women and men’s preferences for positions with political influence.

Note: The left-hand panels show the marginal means for men and women for the interaction between the levels of the working environment and political position with 83 per cent confidence intervals for subgroup means – non-overlapping intervals indicate significant group differences (Goldstein and Healy 1995). Estimates below the dashed line (grand mean) indicate disfavourability and estimates above indicate favourability. The right-hand panels show the differences in marginal means estimated as a three-way interaction between 1) position, 2) working environment, and 3) candidate gender: 95 per cent confidence intervals N = 27,132, N _men = 18,200, N _women = 8,932. Standard errors are clustered at the candidate level. The estimates and standard errors are shown in Table L.3 in Appendix L.

Willingness to pay and willingness to work to avoid sexism in politics

To substantiate the gender difference in candidates’ preferences for a working environment void of sexism, we can calculate men and women candidates’ willingness to decrease their remuneration and willingness to increase their workload to avoid sexism in politics. The former is normally labelled willingness to pay (Eriksson and Kristensen 2014; Folke and Rickne 2022), and we will label the latter willingness to work – as workloads make up another type of cost of doing politics because candidates’ time is a finite good.

We follow the procedure of Folke and Rickne (2022) and use OLS to regress the dummy variable ${Y_{ipr}}$ equal to 1 if candidate ${\it{i}}$ selected a specific political position, ${\it{p}}$ , in experiment round ${\it{r}}$ . Under the assumptions presented by Hainmueller, Hopkins and Yamamoto (2014) OLS provides the so-called Average Marginal Component Effects (AMCE), which capture one attribute level’s effect on a position being selected compared to the attributes’ reference category while keeping all else constant. We include dummy variables for each attribute level, except for one for each attribute to make up the attributes’ reference category:

$$\eqalign{{Y_{ipr}} = & \alpha + {\beta _1}P_{ipr}^{li} \!+ \!{\beta _2}P_{ipr}^{la} + {\beta _3}R_{ipr}^{ - 10} + {\beta _4}R_{ipr}^{ + 10} + {\beta _5}W_{ipr}^{ - 10} \!+ \!{\beta _6}W_{ipr}^{ + 10} \!+ \!{\beta _{sh}}S{H_{ipr}} \cr & + \beta _{mr}M{R_{ipr}} + {\varepsilon _{ipr}}}$$

For working environment, we use ‘several experienced harassment’ as our reference category because men and women candidates hold similar (dis) preferences to this category which makes it a good baseline, and because the prevalence of sexism is the only element that makes it different to the sexism-level. Followingly, ${\beta _{sh}}$ will express candidates’ propensity to select political positions where several experienced sexism and harassment compared to the reference category, that is, a political position where several experienced harassment. For the attributes remuneration and workload, we set ‘normal remuneration’ and ‘normal workload’ as the reference category respectively. The dummies for 10 per cent higher and 10 per cent lower remuneration or workload then show the percentage point difference in position choices between these levels and the reference category.

We run this OLS model for women and men separately, and then calculate their willingness to pay and willingness to work by dividing the coefficient for sexism, ${\beta _{sh}}$ by a weighted average of the two remuneration coefficients and two workload coefficients respectively. For remuneration, the equations will look like this:

$${\it{Willingness}\,{{\it{To}}\,{{\it{Pay}}_{sh}} = {{{\beta _{sh}}} \over {{1 \over 2}( - {{{\beta _3}} \over {0.1}} + {{{\beta _4}} \over {0.1}})}} \,{and}\, {\it{Willingness}\,{{\it{To}}\,{{\it{Work}}_{sh}} = {{{\beta _{sh}}} \over {{1 \over 2}( - {{{\beta _5}} \over {0.1}} + {{{\beta _6}} \over {0.1}})}}.}}}}$$

These ratios express the remuneration or workload change in per cent, that makes the average woman or man candidate indifferent between political jobs where ‘several experienced sexual harassment’ and jobs where ‘several experienced harassment’, that is, how much they are willing to be paid less or work more in per cent to avoid sexism.

Figure 6 presents women and men candidates’ willingness to pay and willingness to work to avoid sexism in politics and the estimated gender difference. We obtain standard errors by bootstrapping with 1,000 replications. The left-hand estimates indicate that women candidates are willing to be remunerated 15.7 per cent less and work 15.8 per cent more to avoid sexism in politics, while men candidates are willing to be remunerated 7.9 per cent less and work 6.8 per cent more to avoid sexism. On both willingness measures, the gender difference is significant as shown in the right-hand column of Figure 6. Hence, while both men and women candidates are willing to pay and willing to work to avoid sexism in politics, women’s willingness is double the size of men’s.

Figure 6. Candidates’ willingness to be remunerated less and willingness to work more to avoid sexism in politics.

Note: Candidates’ willingness to lower their remuneration or increase their workload in per cent in order to avoid sexism in politics: 83 per cent confidence intervals for subgroup estimates and 95 per cent confidence intervals for gender differences. Standard errors obtained using bootstrapping with 1,000 replications – the bootstrap distribution is slightly skewed, and therefore the confidence intervals are not symmetrical around the estimate.

Robustness: Removal of dominant choice options and inattentive respondents

Since the four job attributes are randomly varied, some choice situations will by chance imply a dominant option, that is, a political job that has better or equal values than the other job on all attributes. If we employ this definition, 38.4 per cent of the choices had dominant options. Since dominant options do not imply any trade-off between attributes, a methodological concern is whether dominant options are the driving force of our results, which would consequently hamper our ability to infer the candidates’ preferences.

We test for this by rerunning our analyses without the choices of dominant options. This decreases the sample size from 27,132 to 16,716 and, hence, obviously decreases the estimates’ precision. In Figure 7, the panel displays the marginal means estimates from Figure 3 and Figure 4, which rely on all choices (solid line and shape) and the same marginal mean estimates when choices with dominant options are excluded (dashed confidence interval and hollow shape). On the right-hand side, we again show the corresponding differences in marginal means between men and women with and without choices with dominant options.

Figure 7. The differences in preferences between 1) men and women and 2) victims and non-victim candidates remain when disregarding dominant choices.

Note: Dashed confidence intervals and hollow shapes for estimates where dominant options are excluded. The left-hand side has 83 per cent confidence intervals for subgroup means – non-overlapping intervals indicate significant group differences (Goldstein and Healy 1995). The right-hand side has 95 per cent confidence intervals for the estimated difference in marginal means. Standard errors are clustered at the candidate level.

Overall, our results are unchanged, when only including choice situations without dominant options. The estimates for differences in marginal means (right side) are strikingly stable with and without dominant options. This is the case for the comparison of men and women as well as victims and non-victims. The subgroup marginal means (left side) do change when dominant options are disregarded; however, the change is similar for men and women, hence the stability of their differences in marginal means.

Interestingly, when disregarding dominant options, the candidates’ preferences for position and working environment remain similar, while their preferences for remuneration and workload oddly suggest that they prefer less remuneration and more work. This shift indeed sounds counterintuitive at first glance, but remember that we are only looking at choice situations entailing trade-offs between attributes here. Since candidates’ preferences for positions and working environments at large do not change, this suggests that candidates – both men and women – care about their institutional position and working environment more than the remuneration and workload. In line with our section above about willingness to pay and willingness to work, the remaining difference in mens and women’s candidates’ marginal means for working environment suggests that women hold stronger preferences for a working environment void of sexual harassment also when this entails compromising remuneration and working hours.

In Appendix J, Figure J.1 shows that our results are as good as unchanged when removing both the inattentive respondents and all choice situations with dominant options.

Discussion

So far, we have shown that women hold higher personal costs in politics and that this is expressed in their preferences for jobs, their willingness to pay, and their willingness to work. But while women suffer more from sexism in politics, victimization can also mobilize and make victims seek political influence in response to sexism and harassment. In the following, we discuss these dual mechanisms.

Victimization and stronger preferences for political influence

The additional costs for women to run made us expect women to shy away from power. Therefore, it is noteworthy that H6 is not only disconfirmed but also the preferences for power are actually stronger among victims than non-victims, as shown in Figure 8. Candidates with political influence may be targeted more frequently (Håkansson 2021), and hence the association between victimhood and political position is a matter of selection. Note that, since our conjoint-experiment has no temporal variation, our data cannot capture such a causal relationship of victimization. However, it is noteworthy that the finding aligns with recent research, which points to another mechanism. Evidence shows that violence, in general, causes political mobilization (Sønderskov et al. 2022), and sexually-based violence and gendered policy threats are specifically found to increase the political mobilization of women (Agerberg and Kreft 2020; Kreft 2019; Clayton, O’Brien and Piscopo 2023).

Figure 8. Men and women victims hold stronger preferences for political power.

Note: Marginal means and differences in marginal means are conditional on gender and victimhood. The left-hand panels show victim and non-victim candidates’ marginal means with 83 per cent confidence intervals for subgroup means – non-overlapping intervals indicate significant group differences (Goldstein and Healy 1995). The right-hand side shows the estimated differences in marginal means with 95 per cent confidence intervals. N = 27,132, N _{men, non-victims} = 14,196, N _{women, non-victims} = 4,956, N _{men, victims} = 4,004, N _{women, non-victims} = 3,976. Standard errors are clustered at the candidate level. Marginal means for remuneration and workload are shown in Figure K.1 in Appendix K. The estimates and standard errors are shown in Table L.5 in Appendix L.

Crisis moments – and the emotions they stir – can act as tipping points in some women’s calculus, whereby the costs of not running become too high (Clayton, O’Brien and Piscopo 2023). Indeed, after the 2016 election of Donald Trump as US president, findings reveal that Democratic non-incumbent women candidates were most likely to describe negative incumbents, including feelings of urgency, anger, and/or threat as motivating candidacy (Dittmar 2020). In the same way that political events can motivate candidacy, incidents of sexism and harassment are events that instil emotions of anger, unfairness, and indignation. Hence, they can create mobilization and not just despair. It is important to note again that since we only study one point in time, such mobilization in response to sexism and harassment may be overestimated in the sample if less resilient candidates have already left politics due to harassment. We discuss these dual dynamics at more length below. In sum, our results suggest that some women do not shy away from political candidacy in response to sexism and harassment, but instead stay in political candidacy and hold strong preferences for political influence even if they are victimized.

Seniority and stronger preferences for political influence

Candidates’ preferences might also differ by their political experience. Preferences for remuneration, workload, and a working environment void of sexism and harassment are similar between the candidates with and without political experience. However, there is a difference in preferences stated for political influence; experienced candidates have stronger preferences for political influence than inexperienced ones. This could, first, be due to selection, as re-running incumbents hold stable political ambitions (Cirone, Cox and Fiva 2021) and hence have ambitions to regain their seat and hope to climb the political ladder by achieving a position as committee chair. Thus, the politicians with experience who appear in our sample might be crowded in because of their preferences for power. A precondition for this is that re-elected candidates, through their experience, have updated their priorities and learned that chairpersons in fact exercise considerable political influence. Figure 9 shows the marginal means for politically experienced and inexperienced candidates.

Figure 9. Experienced political candidates hold stronger preferences for political power than inexperienced candidates.

Note: Marginal means and differences in marginal means for politically experienced and inexperienced candidates. Political experience is defined as having held a political office previously. The left-hand panels show men and women candidates’ marginal means for the attributes with 83 per cent confidence intervals for subgroup means – non-overlapping intervals indicate significant group differences (Goldstein and Healy 1995). The right-hand panels show the estimated differences in marginal means with 95 per cent confidence intervals. N = 27,132, N _experienced = 18,438, N _{inexperienced} = 8,694. Standard errors are clustered at the candidate level.

Women’s descriptive underrepresentation is traditionally presented as a ladder where women are discarded as they climb toward the top (Lovenduski and Norris 2003; Lovenduski 2016). However, recent research provides an alternative hourglass pattern, where women’s presence decreases in the early phases but increases in the later phases (Kjaer and Kosiara-Pedersen 2019). Our findings for seniority and political influence suggest that the women in the candidate pool have ambitions for higher hierarchical positions just as much as men. While this is, of course, a matter of initial selection into candidacy, it also emphasizes that this stage in the political pipeline has a supply of women who have preferences to reach the higher rungs of the political ladder. Consequently, with similar motivations and preferences between men and women candidates, women’s absence from higher hierarchical leadership positions in politics may be attributed to the demand side of representation.

Gender gaps in seniority

While victimization can lead to political mobilizations, sexism and harassment can also make women consider voluntarily retirement from holding a political position, or make women disappear from politics in other ways (Kosiara-Pedersen 2023; Krook and Sann 2020). Even if victimized, women candidates do not shy away from power, overall, and in the longer run, women obtain lower seniority in politics as shown in Figure 10. There can be multiple reasons for this, but we believe the gendered cost of politics is a significant one.

Figure 10. Women candidates and incumbents rerun less than men candidates from 1997–2017.

Note: The share of women and men candidates and women and men incumbents that rerun. The gap between the black line and the grey line expresses the gender difference. The two lines have similar trends over time, but the grey line (for women) is always placed under the black line (for men). Based on data from Statistics Denmark.

While role congruity theory suggests that discrimination stems from women not adhering to the normative ideals about being a woman and being a leader, recent research indicates that even when ideals change, practices that disadvantage women can remain very persistent. Even in very gender-equal settings – like Sweden and, in this study, Denmark – women leaders are to a higher extent than men questioned, critiqued, and harassed – which can undermine their legitimacy and competence (Erikson and Josefsson 2023). Hence, even in gender-equal contexts where normative ideals on gender and leadership are changing, women remain disadvantaged in practice. The results presented here add to the discussion on how, even in more gender-equal societies, sexism and harassment impact women’s possibilities and preferences for holding political positions. We believe that dual dynamics are involved and that future research should try to uncover when one mechanism is triggered rather than the other.

Conclusion

We contribute to the literature with the first attempt to measure political candidates’ preferences for political job attributes and provide new evidence about how candidates’ preferences for political positions vary between women and men, and victims and non-victims of sexism and harassment. We argue that because women in politics disproportionately face sexism and harassment, women candidates will have different preferences regarding attributes for political positions, and the circumstances under which they govern. We test this argument on the pool of candidates at the Danish local election in 2021, through a survey that identified candidates as victims of sexism and harassment, and estimated the candidates’ preferences through a conjoint experiment in which the candidates were selected between two political positions with randomly varied attributes.

We find that women candidates experience far more sexism and harassment than men candidates, and also correctly assess their risk of victimization as being far higher than men. In line with this, the conjoint experiment revealed that women candidates state stronger preferences against sexism and for an equal and respectful working environment than men, while women and men hold similar stated preferences on political positions, remuneration, and workload.

To substantiate the gender gap in political candidates’ dispreference for sexism in politics, we estimate the candidates’ willingness to pay and willingness to work. While both men and women candidates are willing to be remunerated less and willing to work more to avoid sexism in politics, we find that women’s willingness is double the size of men’s.

With no gender gap in preferences for political positions, our results show that women who lean into political candidacy are equally ambitious as men. Existing research often explains gender inequality using factors on the supply side of political representation. However, when we investigated those who are interested in politics and who do run, the preferences stated are surprisingly similar between men and women. Hence, the results counteract explanations rooted in women candidates’ lack of ambition on the supply side, as the descriptive under-representation of women in higher political positions. Rather, our results call for a re-orientation toward the institutional mechanisms of candidate selection and deselection. This, suggests that the working conditions of politicians once elected may be a place to look for the mechanisms generating inequality in representation.

The results have a certain duality. On one hand, they suggest that men and women in the candidate pool are highly selective, not just in terms of demographics as previously found (Dal Bó et al. 2017; Carnes and Lupu 2023), but also in regards to their preferences for several working conditions of a political position. On the other hand, despite candidates being a highly selected pool, men and women candidates still state stronger preferences against a working environment void of sexism and harassment which is prevalent in politics and disproportionately hit women. When we compare victims of sexism and harassment and non-victims, we also found that the difference in preferences for working environment remained, although smaller in magnitude. We found no evidence, however, that women candidates shy away from political power when it entails an increased risk of sexism and harassment. By contrast, we found prior-victims to state stronger preferences for positions with political influence than non-victims. Hence, victimization may have dual effects. It is likely that some victims have left politics and are, therefore, not present in the candidate pool, but among the remaining, there are signs of mobilization as their preferences for political influence are higher compared to non-victims.

However, men victims’ preferences for working environments void of sexism and harassment are not as strong as women victims, which indicates that variation in preferences does not arise from personal experiences alone. By contrast, we find that victims state preferences according to their gender, rather than according to victimization. Gender is indeed commonly recognized as a main fault line for social identities and an important basis for within-group loyalty and out-group hostility (Tajfel 1974). As such, gender works as an institutional carrier, which mediates between individual experiences and group norms. However, it is remarkable that the same mediation is not found between men victims and their gender group. One reason may be that social acceptance of being a victim accords less with masculine norms and, hence, victimization is subject to greater taboo. Therefore, out-group hostility and lack of out-group altruism may further reduce the internalization of the disutility of the victims among men, even if men also experience harassment (Folke and Rickne 2022). Furthermore, women assess their risk of being subject to sexism and harassment as being much higher. These factors together may lead to weak sanctions for inappropriate behaviour and an under-demand for working environments void of sexism and harassment. The gender gap in candidates’ preferences for working environments, underlines how working conditions in politics are crucial to the attraction and attrition of women candidates and hence for gender equality in political representation.

Our approach is methodologically innovative, as the majority of studies on the political class either rely on assumptions about its preferences or capture these preferences, like political ambitions, through measures prone to social desirability bias. Future attempts to apply this to other contexts would be great advancements. We only study Danish candidates and, due to Denmark’s comparatively high gender equality, we expect the gendered costs of politics to be even higher in contexts with less gender equality. Future applications on national parliamentary candidates would also be valuable, although conjoint experiments require many observations, which can rarely be obtained outside local government settings.